research paper dementia

Diagnosis and Management of Dementia: Review

Affiliations.

• 1 Rush Alzheimer's Disease Center, Rush University Medical Center, Chicago, Illinois.
• 2 Department of Neurological Sciences, Rush University Medical Center, Chicago, Illinois.
• 3 Department of Family Medicine, Rush University Medical Center, Chicago, Illinois.
• PMID: 31638686
• PMCID: PMC7462122
• DOI: 10.1001/jama.2019.4782

Importance: Worldwide, 47 million people live with dementia and, by 2050, the number is expected to increase to 131 million.

Observations: Dementia is an acquired loss of cognition in multiple cognitive domains sufficiently severe to affect social or occupational function. In the United States, Alzheimer disease, one cause of dementia, affects 5.8 million people. Dementia is commonly associated with more than 1 neuropathology, usually Alzheimer disease with cerebrovascular pathology. Diagnosing dementia requires a history evaluating for cognitive decline and impairment in daily activities, with corroboration from a close friend or family member, in addition to a thorough mental status examination by a clinician to delineate impairments in memory, language, attention, visuospatial cognition such as spatial orientation, executive function, and mood. Brief cognitive impairment screening questionnaires can assist in initiating and organizing the cognitive assessment. However, if the assessment is inconclusive (eg, symptoms present, but normal examination findings), neuropsychological testing can help determine whether dementia is present. Physical examination may help identify the etiology of dementia. For example, focal neurologic abnormalities suggest stroke. Brain neuroimaging may demonstrate structural changes including, but not limited to, focal atrophy, infarcts, and tumor, that may not be identified on physical examination. Additional evaluation with cerebrospinal fluid assays or genetic testing may be considered in atypical dementia cases, such as age of onset younger than 65 years, rapid symptom onset, and/or impairment in multiple cognitive domains but not episodic memory. For treatment, patients may benefit from nonpharmacologic approaches, including cognitively engaging activities such as reading, physical exercise such as walking, and socialization such as family gatherings. Pharmacologic approaches can provide modest symptomatic relief. For Alzheimer disease, this includes an acetylcholinesterase inhibitor such as donepezil for mild to severe dementia, and memantine (used alone or as an add-on therapy) for moderate to severe dementia. Rivastigmine can be used to treat symptomatic Parkinson disease dementia.

Conclusions and relevance: Alzheimer disease currently affects 5.8 million persons in the United States and is a common cause of dementia, which is usually accompanied by other neuropathology, often cerebrovascular disease such as brain infarcts. Causes of dementia can be diagnosed by medical history, cognitive and physical examination, laboratory testing, and brain imaging. Management should include both nonpharmacologic and pharmacologic approaches, although efficacy of available treatments remains limited.

Publication types

• Research Support, N.I.H., Extramural
• Research Support, Non-U.S. Gov't
• Alzheimer Disease / diagnosis
• Alzheimer Disease / therapy
• Cholinesterase Inhibitors / adverse effects
• Cholinesterase Inhibitors / therapeutic use
• Dementia / diagnosis*
• Dementia / therapy*
• Excitatory Amino Acid Antagonists / adverse effects
• Excitatory Amino Acid Antagonists / therapeutic use
• Memantine / adverse effects
• Memantine / therapeutic use
• Neuroimaging
• Neuropsychological Tests
• Cholinesterase Inhibitors
• Excitatory Amino Acid Antagonists

Grants and funding

• P30 AG010161/AG/NIA NIH HHS/United States
• R01 AG040039/AG/NIA NIH HHS/United States
• R01 NS084965/NS/NINDS NIH HHS/United States
• RF1 AG059621/AG/NIA NIH HHS/United States

Advertisement

Machine Learning for Dementia Prediction: A Systematic Review and Future Research Directions

• Original Paper
• Open access
• Published: 01 February 2023
• Volume 47 , article number  17 , ( 2023 )

Cite this article

You have full access to this open access article

• Ashir Javeed 1 , 2   na1 ,
• Ana Luiza Dallora 2   na1 ,
• Johan Sanmartin Berglund 2 ,
• Arif Ali 3 ,
• Liaqata Ali 4 &
• Peter Anderberg 2 , 5  

14k Accesses

25 Citations

47 Altmetric

Explore all metrics

Nowadays, Artificial Intelligence (AI) and machine learning (ML) have successfully provided automated solutions to numerous real-world problems. Healthcare is one of the most important research areas for ML researchers, with the aim of developing automated disease prediction systems. One of the disease detection problems that AI and ML researchers have focused on is dementia detection using ML methods. Numerous automated diagnostic systems based on ML techniques for early prediction of dementia have been proposed in the literature. Few systematic literature reviews (SLR) have been conducted for dementia prediction based on ML techniques in the past. However, these SLR focused on a single type of data modality for the detection of dementia. Hence, the purpose of this study is to conduct a comprehensive evaluation of ML-based automated diagnostic systems considering different types of data modalities such as images, clinical-features, and voice data. We collected the research articles from 2011 to 2022 using the keywords dementia, machine learning, feature selection, data modalities, and automated diagnostic systems. The selected articles were critically analyzed and discussed. It was observed that image data driven ML models yields promising results in terms of dementia prediction compared to other data modalities, i.e., clinical feature-based data and voice data. Furthermore, this SLR highlighted the limitations of the previously proposed automated methods for dementia and presented future directions to overcome these limitations.

Similar content being viewed by others

Early detection of dementia using artificial intelligence and multimodal features with a focus on neuroimaging: A systematic literature review

Ovidijus Grigas, Rytis Maskeliunas & Robertas Damaševičius

Data Mining and Machine Learning Methods for Dementia Research

A Comparative Analysis of Classification Algorithms for Dementia Prediction

Avoid common mistakes on your manuscript.

Introduction

Over a period of time, the advancements made in the field of medical science helped to increase the lifespan in the modern world [ 1 ]. This increased life expectancy raised the prevalence of neurocognitive disorders, affecting a significant part of the older population as well as global economies. In 2010, it was estimated that $604 billion have been spent on dementia patients in the USA alone[ 2 ]. The number of dementia patients is rapidly increasing worldwide, and statistical projections suggest that 135 million people might be affected by dementia by 2050 [ 3 ]. There are several risk factors that contribute to the development of dementia, including aging, head injury, and lifestyle. While age is the most prominent risk factor for dementia; figures suggest that a person at the age of 65 years old has 1–2% risk of developing dementia disease. By the age of 85 years old, this risk can reach to 30% [ 4 ].

Dementia is a mental disorder that is characterized by a progressive deterioration of cognitive functions that can affect daily life activities such as memory, problem solving, visual perception, and the ability to focus on a particular task. Usually, older adults are most vulnerable to dementia, and people take it as an inevitable consequence of aging, which is perhaps the wrong perception. Dementia is not a part of the normal ageing process; however, it should be considered a serious form of cognitive decline that affects your daily life. Actually, the primary cause for the development of dementia is the several diseases and injuries that affect the human brain [ 5 ]. Dementia is ranked on the seventh place in the leading causes of deaths in the world [ 6 ]. Furthermore, it is the major cause of disability and dependency among older people globally [ 6 ]. A change in the person’s ordinary mental functioning and obvious signs of high cognitive deterioration are required for a diagnosis of dementia [ 7 ]. Figure 1 presents the progression of dementia with age.

Progression of dementia disease with ageing

Types of dementia

Dementia is not a single disease, but, it is used as a generic term for several different cognitive disorders. Figure 2 provides the overview of different types of dementia along with the percentage of particular dementia type occurrence in the patients [ 8 ]. To have a better idea about dementia, we have studied common types of dementia for better problem awareness.

Types of dementia disease

Alzheimer’s disease

Alzheimer’s disease (AD) is thought to develop when abnormal amounts of amyloid beta (A \(\beta\) ) build up in the brain, either extracellularly as amyloid plaques, tau proteins or intracellularly as neurofibrillary tangles, affecting neuronal function, connectivity and leading to progressive brain function loss [ 9 ]. This diminished ability to eliminate proteins with ageing is regulated by brain cholesterol [ 10 ] and is linked to other neurodegenerative illnesses [ 11 ]. Except for 1–2% of cases where deterministic genetic anomalies have been discovered, the aetiology of the majority of Alzheimer’s patients remains unexplained [ 12 ]. The amyloid beta (A \(\beta\) ) hypothesis and the cholinergic hypothesis are two competing theories presented to explain the underlying cause of AD [ 13 ].

Vascular dementia

Vascular dementia (VaD) is a subtype of dementia caused by problems with the brain’s blood flow, generally in the form of a series of minor strokes, which results in a slow decline of cognitive capacity [ 14 ]. The VaD refers to a disorder characterized by a complicated mix of cerebrovascular illnesses that result in structural changes in the brain, as a result of strokes and lesions, which lead to cognitive impairment. A chronological relationship between stroke and cognitive impairments is necessary to make the diagnosis [ 15 ]. Ischemic or hemorrhagic infarctions in several brain areas, such as the anterior cerebral artery region, the parietal lobes, or the cingulate gyrus, are associated with VaD. In rare cases, infarcts in the hippocampus or thalamus might cause dementia [ 16 ]. A stroke increases the risk of dementia by 70%, whereas a recent stroke increases the risk by almost 120% [ 17 ]. Brain vascular lesions can also be caused by diffuse cerebrovascular disease, such as small vessel disease [ 18 ]. Risk factors for VaD include age, hypertension, smoking, hypercholesterolemia, diabetes mellitus, cardiovascular disease, and cerebrovascular sickness; geographic origin, genetic proclivity, and past strokes are also risk factors [ 19 ]. Cerebral amyloid angiopathy, which develops when beta amyloid accumulates in the brain, can occasionally lead to vascular dementia.

Lewy body dementia

Lewy body dementia (LBD) is a subtype of dementia characterized by abnormal deposits of the protein alpha-synuclein in the brain. These deposits, known as Lewy bodies, affect brain chemistry, causing problems with thinking, movement, behavior, and mood. Lewy body dementia is one of the most common causes of dementia [ 20 ]. Progressive loss of mental functions, visual hallucinations, as well as changes in alertness and concentration are prevalent in persons with LBD. Other adverse effects include tight muscles, delayed movement, difficulty walking, and tremors, all of which are also signs and symptoms of Parkinson’s disease [ 21 ]. LBD might be difficult to identify. Early LBD symptoms are commonly confused with those of other brain diseases or mental problems. Lewy body dementia can occur alone or in conjunction with other brain disorders [ 22 ]. It is a progressive disorder, which means that symptoms emerge gradually and worsen with time. A timespan of five to eight years is averaged, although it can last anywhere from two to twenty years for certain people [ 23 ]. The rate at which symptoms arise varies greatly from person to person, depending on overall health, age, and the severity of symptoms.

Frontotemporal dementia

Frontotemporal Dementia (FTD) is a subtype of dementia characterized by nerve cell loss in the frontal and temporal lobes of the brain [ 24 ]. As a result, the lobes contract. FTD can have an impact on behavior, attitude, language, and movement. This is one of the most common dementias in people under the age of 65. FTD most commonly affects persons between the ages of 40 and 65; however, it may also afflict young adults and older individuals [ 25 ]. The lobes decrease, and behavior, attitude, language, and mobility can all be affected by FTD. FTD affects both men and women equally. Dissociation from family, extreme oniomania, obscene speech, screaming, and the inability to regulate emotions, behavior, personality, and temperament are examples of social display patterns caused by FTD [ 26 ]. The symptoms of FTD appeared several years prior to visiting a neurologist [ 27 ].

Mixed Dementia (MD)

Mixed dementia occurs, when more than one kind of dementia coexists in a patient, and it is estimated to happen in around 10% of all dementia cases [ 6 ]. AD and VaD dementia are the two subtypes that are most common in MD [ 28 ]. This case is usually associated with factors such as old age, high blood pressure, and brain blood vessel damage [ 29 ]. Because one dementia subtype often predominates, MD is difficult to identify. As a result, the individuals affected by MD are rarely treated and miss out on potentially life-changing medicines. MD can cause symptoms to begin earlier than the actual diagnosis of the disease and spread swiftly to affect the most areas of the brain [ 30 ].

Recently, numerous automated methods have been developed based on machine learning for early the prediction of different diseases [ 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 ]. This systematic literature review (SLR) presented hereby, investigates machine learning-based automated diagnostic systems that are designed and developed by scientists to predict dementia and its subtypes, such as AD, VaD, LBD, FTD and MD. We used the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) criteria to conduct this SLR [ 49 , 50 ]. A comprehensive search was conducted to retrieve the research articles that contain ML approaches to predict the development of dementia and its subtypes using three different types of data modalities (images, clinical-variables, voice).

Aim of the study

SLRs are done to synthesize current evidence, to identify gaps in the literature, and to provide the groundwork for future studies [ 51 ]. Previous, SLRs studies have been done on automated diagnostic systems for dementia prediction based on ML approaches, which focused on a single sort of data modality. These SLR investigations did not emphasize the limits of previously published automated approaches for dementia prediction. The SLR presented herein assesses the previously proposed automated diagnostic systems based on deep learning (DL) and ML algorithms for the prediction of dementia and its common subtypes (e.g. AD, VaD, FTD, MD). The aim of this SLR is to analyse and evaluate the performance of automated diagnostic systems for dementia prediction using different data modalities. The main question is decomposed in the following sub-research questions:

What types of ML and DL techniques have been used by researchers to diagnose dementia?

Examine the methods of feature extraction or selection used by the researchers.

Analyze the different performance evaluation measures that are adopted by the researcher to validate the effectiveness of the proposed diagnostic system for demetnia.

Analyze the performance of ML models on various data types.

Identification of weaknesses in previously proposed ML models for dementia prediction.

Flow diagram of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses)

Article selection

For this SLR study, the research articles were selected based on keywords such as ML, DL, dementia and its subtypes (AD, VaD, FTD, and MD). For the collection of research articles, we conducted an electronic search from different online databases such as ScienceDirect, PubMed, IEEE Xplore Digital Library, Springer, Hindawi, and PLOs, which helped to gather 450 research studies on the specific topic. After reviewing the title and abstract in each study, 120 publications were found to be ineligible for processing, while 330 articles were selected for further processing. Following the deduplication of data, 125 full-text publications were retrieved for further processing after the screening phase of the article selection, with 205 of them being eliminated due to not satisfying the article selection criteria of the screening phase. Finally, 50 research articles were eliminated due to not fulfilling the eligibility criteria for article selection. The final set of selected papers consisted of 75 research papers, among these final selected articles, each of the data modalities (image, clinical-variables, voice) contained 25 papers. After rerunning the database searches in May 2022, no further suitable research article was found for the selection. Figure 3 presents the workflow for article selection, which includes the four PRISMA guidelines-recommended steps such as identification, screening, eligibility, and inclusion [ 49 , 50 ]. In recent years, ML scientists have shown a strong interest in designing and developing ML-based automated diagnostic systems for dementia prediction. Therefore, the number of research articles in this research area has been increased and it can be depicted from Fig. 4 where research articles are published years wise with regarding data modality. The publications utilized in this study were selected based on the following criteria:

Studies that present automated diagnostic systems for dementia and its common subtypes (AD,VaD, FTD, MD).

Studies published between 2011 and 2022.

Studies employing ML approaches for dementia diagnosis.

Studies which have utilized several data modalities.

Studies published in the English language.

Selected research articles which are published from 2011 to 2022 regarding data modality

Machine learning for dementia

Over the years, the increasing use and availability of medical equipment has resulted in a massive collection of electronic health records (EHR) that might be utilized to identify dementia using developing technologies such as ML and DL [ 52 ]. These EHRs are one of the most widely available and used clinical datasets. They are a crucial component of contemporary healthcare delivery, providing rapid access to accurate, up-to-date, comprehensive patient information while also assisting with precise diagnosis and coordinated, efficient care [ 53 ]. Laboratory tests, vital signs, drugs, and other therapies, as well as comorbidities, can be used to identify the people at risk of dementia using the EHRs’ data [ 54 ]. In some situations, patients may also be subjected to costly and invasive treatments such as neuroimaging scans i.e., magnetic resonance imaging (MRI) and position emission tomography (PET)) and cerebrospinal fluid (CSF) collection for biomarker testing [ 55 , 56 , 57 ]. These tests’ findings may also be found in the EHR. According to researchers, such longitudinal clinical EHR data can be used to track the advancement of AD dementia over time [ 58 ]. Recently, several automated diagnostic systems for different diseases, such as Parkinson’s disease [ 59 ], hepatitis [ 47 ], carcinoma [ 41 ], and heart failure [ 60 , 61 , 62 ] prediction have been designed by employing ML and DL techniques. Inspired by this fact, the unmet demand for dementia knowledge, along with the availability of relevant huge datasets, has motivated scientists to investigate the utility of artificial intelligence (AI), which is gaining a prominent role in the area of healthcare innovation [ 63 ]. ML, a subset of AI, can model the relationship between input quantities and clinical outcomes, identify hidden patterns in enormous volumes of data, and draw conclusions or make decisions that help with more accurate clinical decision-making [ 51 ]. However, computational hypotheses generated by ML models must still be confirmed by subject matter experts in order to achieve enough precision for clinical decision-making [ 64 ].

In this SLR, we have included studies that have used ML predictive models (supervised and unsupervised) for dementia prediction and excluded studies that have used statistical methods for cohort summarization and hypothesis testing (e.g., odds ratio, chi-square distribution, Kruskal-Wallis test, and Kappa-Cohen test). Furthermore, we have referenced the data modality-based study [ 65 ] for this literature review, where we have categorized the three data modality types such as image, clinical-variable and voice. Thus, we have studied each modality-based automated diagnostic system for dementia prediction that has been proposed in the past using ML and DL.

This section explains the datasets that were used in the selected research papers for experiments and performance evaluation of the proposed automated diagnostic systems designed by the researchers using ML algorithms for dementia and its subtypes. A total of 61 datasets were studied from the selected research articles. These datasets are compiled from a wide range of organizations and hospitals throughout the world. Only a few datasets are openly available to the public, while others are compiled by researchers from various hospitals and healthcare institutes. We have only included datasets that have been used to diagnose AD, VaD, FTD, MD, and LBD using ML and DL techniques. On the basis of data modality, we have categorised the dataset into three types: images, clinical_variables and voice datasets. The datasets differ in terms of the number of variables (features) and samples. As a result, we examined each modality of the dataset one by one.

Image modality based datasets

There are several image datasets based on brain imaging, such as magnetic resonance imaging (MRI), collected by the researchers for the diagnosis of dementia. From the Table 1 , it can be depicted that Open Access Series of Imaging Studies (OASIS) and Alzheimer’s Disease Neuroimaging Initiative (ADNI) datasets are mostly used by the researchers for the experimental purpose. OASIS aims to make neuroimaging datasets available to the scientific community for free. By gathering and openly disseminating this multimodal dataset produced by the Knight ADRC and its related researchers, they had used different samples and variables of the datasets in their research work. ADNI researchers acquire, validate, and use data such as MRI, PET imaging, genetics, cognitive assessments, CSF, and blood biomarkers as disease predictors. The ADNI website contains research information and data from the North American ADNI project, which includes Alzheimer’s disease patients, people with mild cognitive impairment, and older controls. Table 1 provides us with the following information: dataset_id, dataset name, number of samples in the particular dataset, variables in the dataset, and finally, the type of dementia.

Clinical-variables modality based datasets

Throughout the course of time, the growing usage and availability of medical devices have resulted in an overwhelming collection of clinical EHR data. Furthermore, the patient’s medical history consists of medical tests and clinical records that can be used for the prediction of diseases. Thus, the importance of clinical data emerges as a vital tool for proactive management of disease. The dataset based on clinical variables for dementia consists of medical tests that are used by doctors to check the dementia status in patients, such as the Mini Mental Status Exam (MMSE), the Montreal Cognitive Assessment (MoCA), the Telephone Interview for Cognitive Status (TICS), and the Brief Interview for Mental Status (BIMS). Clinical-variables based datasets consist of information about these medical tests along with patient personal information, i.e., age, sex, and marital status. Hereby, Table 2 provides the information regarding clinical-variables modality-based datasets that are used by the researchers for the design and development of automated diagnostic systems for dementia patients based on ML. Table 2 presents the dataset_id, dataset name, number of samples in the particular dataset, variables in the dataset, and finally the type of dementia.

Voice modality based datasets

Speech analysis is a useful technique for clinical linguists in detecting various types of neurodegenerative disorders affecting the language processing areas. Individuals suffering from Parkinson’s disease (PD, deterioration of voice quality, unstable pitch), Alzheimer’s disease (AD, monotonous pitch), and the non-fluent form of Primary Progressive Aphasia (PPA-NF, hesitant, non-fluent speech) may experience difficulties with prosody, fluency, and voice quality. Besides imaging and clinical-variables data, the researchers employed voice recording data to identify dementia using ML and DL algorithms. The data collection process for voice data varies from dataset to dataset, for example, in a few datasets, patients were requested to answer a prepared set of questions (interview) in a specific time interval. In a few datasets, selected neuropsychological tests were carried out, the description of each neuropsychological test was played and was followed by an answering window. Table 3 presents the dataset_id, dataset name, number of samples in the particular dataset, variables in the dataset, and finally the subtype of dementia.

Data sharing challenges

In this digital era, public health decision-making has grown progressively complicated, and the utilization of data has become critical [ 66 ]. Data are employed at the local level to observe public health and target interventions; at the national scale for resource allocation, prioritization, and planning; and at the global scale for disease burden estimates, progress in health and development measurement, and the containment of evolving global health threats [ 67 , 68 ]. Van Panhuis et al. have adequately described the challenges to exchanging health data [ 69 ]. Based on our initial analysis, we built on this taxonomy to identify the hurdles related to data sharing in global public health, and we have highlighted how they may apply to each typology as given below.

Lack of complete data, lost data, restrictive as well as conflicting data formats, a lack of metadata and standards, a lack of interoperability of datasets (e.g., structure or “language”), and a lack of appropriate analytic solutions are examples of technical barriers encountered by health information management systems.

Individuals and organizations face motivational challenges when it comes to sharing data. These impediments include a lack of incentives, opportunity costs, apprehension about criticism, and disagreements over data usage and access.

The potential and present costs of sharing data are both economic hurdles.

Political obstacles are those that are built into the norms of local health governance and often emerge as regulations and guidelines. They can also entail trust and ownership difficulties.

Legal issues that arise as a result of data collection, analysis, and usage include questions regarding who owns or controls the data, transparency, informed permission, security, privacy, copyright, human rights, damage, and stigma.

Ethical constraints include a lack of perceived reciprocity (i.e., the other side will not disclose data) and proportionality (i.e., deciding not to share data based on an assessment of the risks and benefits). An overall concern is that frameworks, rules, and regulations have not kept up with technological changes that are transforming how data is collected, analyzed, shared, and used.

ML based diagnostic models for dementia: Image modality

In recent years, researchers have designed many ML and DL algorithms for the detection of dementia and its subtypes using MRI images of the brain. For example, Dashtipour et al. [ 70 ] proposed a ML based method for the prediction of Alzheimer’s disease. In their proposed model, they used DL techniques to extract the features from brain images, and for classification purposes, they deployed SVM and bidirectional long short-term memory (BiLSTM). Through their proposed model, they had reported the classification accuracy of 91.28%. Moreover, for early detection of the AD, a DL based approach was proposed by Helaly et al. In their proposed work, they employed convolutional neural networks (CNN). The Alzheimer’s disease spectrum is divided into four phases. Furthermore, different binary medical image classifications were used for each two-pair class of Alzheimer’s disease stages. Two approaches were used to categorize medical images and diagnose Alzheimer’s disease. The first technique employs basic CNN architectures based on 2D and 3D convolution to cope with 2D and 3D structural brain images from the ADNI dataset. They had achieved highly promising accuracies for 2D and 3D multi-class AD stage classification of 93.61% and 95.17%, respectively. The VGG19 pre-trained model had been fine-tuned and obtained an accuracy of 97% for multi-class AD stage classification [ 71 ]. Vandenberghe et al. had proposed a method for binary classification of 18F-flutemetramol PET using ML techniques for AD and mild cognitive impairment (MCI). They had tested whether support vector machines (SVM), a supervised ML technique, can duplicate the assignments made by blindfolded visual readers, as well as which image components had the highest diagnostic value according to SVM and how 18F-fluoromethylamol-based SVM classification compares to structural MRI-based SVM classification in the same cases. Their F-flutemetamol based classifier was able to replicate the assignments obtained by visual read with 100% accuracy [ 72 ]. Odusami et al. proposed a novel method for the detection of early-stage dementia from functional brain changes in MRI using a fine-tuned ResNet-18 network. Their research work presents a DL based technique for predicting MCI, early MCI, late MCI, and Alzheimer’s disease (AD). The ADNI fMRI dataset was used for analysis and consisted of 138 participants. On EMCI vs. AD, LMCI vs. AD, and MCI vs. AD, the fine-tuned ResNet18 network obtained classification accuracy of 99.99%, 99.95%, and 99.95%, respectively [ 73 ]. Zheng et al. had presented a ML based framework for differential diagnosis between VaD and AD using structural MRI features. The least absolute shrinkage and selection operator (LASSO) was then used to build a feature set that was fed into SVM for classification. To ensure unbiased evaluation of model performance, a comparative analysis of classification models was conducted using different ML algorithms to discover which one had better performance in the differential diagnosis between VaD and AD. The diagnostic performance of the classification models was evaluated using quantitative parameters derived from the receiver operating characteristic curve (ROC). The experimental finding had shown that the SVM with RBF performed well for the differential diagnosis of VaD and AD, with sensitivity (SEN), specificity (SPE), and accuracy (ACC) values of 82.65%, 87.17%, and 84.35%, respectively (AUC = 86.10–95%, CI = 0.820–0.902) [ 74 ]. Basheer et al. [ 75 ] had presented an innovative technique by making improvements in capsule network design for the best prediction outcomes. The study used the OASIS dataset with dimensions (373 X 15) to categorize the labels as demented or non-demented. To make the model swifter and more accurate, several optimization functions were performed on the variables, as well as the feature selection procedure. The claims were confirmed by demonstrating the correlation accuracy at various iterations and layers with an allowable accuracy of 92.39%. L. K. Leong and A. A. Abdullah had proposed a method for the prediction of AD based on ML techniques with the Boruta algorithm as a feature selection method. According to the Boruta algorithm, Random Forest Grid Search Cross Validation (RF GSCV) outperformed other 12 ML models, including conventional and fine-tuned models, with 94.39% accuracy, 88.24% sensitivity, 100.00% specificity, and 94.44% AUC even for the small OASIS-2 longitudinal MRI dataset [ 76 ]. Battineni et al. had presented a SVM based ML model for the prediction of dementia. Their proposed model had achieved an accuracy and precision of 68.75% and 64.18% using the OASIS-2 dataset [ 77 ]. Mathotaarachchi et al. had analyzed the amyloid imaging using ML approaches for the detection of dementia. To overcome the inherent unfavorable and imbalance proportions between persons with stable and progressing moderate cognitive impairment in a short observation period. The innovative method had achieved 84.00% accuracy and an AUC of 91.00% for the ROC [ 78 ]. Aruna and Chitra had presented a ML approach for the identification of dementia from MRI images, where they had deployed Independent Component Analysis (ICA) to extract the features from the images, and for classification purposes, SVM with different kernels is used. Through their proposed method, they had obtained an accuracy of 90.24% [ 79 ] (Fig. 5 ).

Accuracy comparison of different ML models based on image modality

Supervised ML techniques and CNNs were examined by Herzog and Magoulas. They had achieved the accuracy of 92.5% and 75.0% for NC vs EMCI, 93.0% and 90.5% for NC vs. AD, respectively [ 80 ]. Battineni et al. had comprehensive applied ML model on MRI to predict Alzheimer’s disease (AD) in older subjects, and they had proposed two ML models for AD detection. In the first trial, manual feature selection was utilized for model training, and ANN produced the highest AUC of 81.20% by ROC. The NB had earned the greatest AUC of 94.20% by ROC in the second trial, which included wrapping approaches for the automated feature selection procedure [ 81 ]. Ma et al. had conducted a study where they compared feature-engineered and non-feature-engineered ML methods for blinded clinical evaluation for dementia of Alzheimer’s type classification using FDG-PET. The highest accuracy of 84.20% was obtained through CNN’s [ 82 ]. Bidani et al. had presented a novel approach in the field of DL that combines both the deep convolutional neural network (DCNN) model and the transfer learning model to detect and classify dementia. When the features were retrieved, the dementia detection and classification strategy from brain MRI images using the DCNN model provided an improved classification accuracy of 81.94%. The transfer learning model, on the other hand, had achieved an accuracy of 68.13% [ 83 ].

Moscoso et al. had designed a predictive model for the prediction of Alzheimer’s disease using MRI images. Their proposed model had obtained the highest accuracy of 84.00% [ 84 ]. Khan and Zubair had presented an improved multi-modal based ML approach for the prognosis of AD. Their proposed model had a five-stage ML pipeline, where each stage was further categorized into different sub-levels. Their proposed model had reported the highest accuracy of 86.84% using RF [ 85 ]. Mohammed et al. had evaluated the two CNN models (AlexNet and ResNet-50) and hybrid DL/ML approaches (AlexNet+SVM and ResNet-50+SVM) for AD diagnosis using the OASIS dataset. They had found that RF algorithm had attained an overall accuracy of 94%, as well as precision, recall, and F1 scores of 93%, 98%, and 96%, respectively [ 86 ]. Salvatore et al. had developed a ML method for early AD diagnosis using magnetic resonance imaging indicators. In their proposed ML model, they used PCA for extracting features from the images and SVM for the classification of dementia. They had achieved a classification accuracy of 76% using a 20-fold cross validation scheme [ 87 ]. Katako et al. had identified the AD related FDGPET pattern that is also found in LBD and Parkinson’s disease dementia using ML approaches. They studied different ML algorithms, but SVM with an iterative single data algorithm produced the best performance, i.e., sensitivity 84.00%, specificity 95.00% through 10-fold cross-validation [ 88 ]. Gray et al. had presented a system in which RF proximities were utilized to learn a low-dimensional manifold from labelled training data and then infer the clinical labels of test data that translated to this space. Their proposed model, voxel-based (FDG-PET), obtained an accuracy of 87.9% using ten-fold cross-validation [ 89 ]. Table 4 provides the overall performance evaluation of the ML models that were presented by the researchers for the prediction of dementia and its subtypes by using image data as a modality.

ML based diagnostic models for dementia: Clinical-variable modality

Aside from image-based ML techniques for dementia prediction, several research studies have utilized clinical-variable data with ML algorithms to predict dementia and its subtypes. For instance, Chiu et al. had designed a screening instrument to detect MCI and dementia using ML techniques. They had developed a questionnaire to assist neurologists and neuropsychologists in the screening of MCI and dementia. The contribution of 45 items that matched the patient’s replies to questions was ranked using feature selection through information gain (IG). Among the 45 items, 12 were ranked the highest in feature selection. The ROC analysis showed that AUC in test group was 94.00% [ 96 ]. Stamate et al. had developed a framework for the prediction of MCI and dementia. Their proposed framework was based on the ReliefF approach paired with statistical permutation tests for feature selection, model training, tweaking, and testing using ML algorithms such as RF, SVM, Gaussian Processes, Stochastic Gradient Boosting, and eXtreme Gradient Boosting. The stability of model performances was studied using computationally expensive Monte Carlo simulations, and the results of their proposed framework were given as for dementia detection, the accuracy was 88.00%, sensitivity was 93.00%, and the specificity was 94.00%, whereas moderate cognitive impairment had a sensitivity of 86.00% and a specificity of 90% [ 97 ]. Stamate et al. developed a system for detecting dementia subtypes (AD) in blood utilizing DL and other supervised ML approaches such as RF and extreme gradient boosting. The AUC for the proposed DL method was 85% (0.80–0.89), for XGBoost it was 88% (0.86–0.89), and for RF it was 85% (0.83–0.87). In comparison, CSF measurements of amyloid, p-tau, and t-tau (together with age and gender) gave AUC values of 78%, 83%, and 87%, respectively, by using the XGBoost [ 98 ]. Bansal1 et al. had performed the comparative analysis of the different ML methods for the detection of dementia using clinical-variables. In their experiments, they exploited the performance of four ML models, such as J48, NB, RF, and multilayer perceptrons. From the results of experiments, they had concluded that j48 outperformed the rest of the ML models for the detection of dementia [ 99 ]. Nori et al. had experimented the lasso algorithm on a big dataset of patient and identify the 50 variables by ML model with an AUC of 69.30% [ 100 ]. Alam et al. [ 101 ]used signal processing on wearable sensor data streams (e.g., electrodermal activity (EDA), photoplethysmogram (PPG), and accelerometer (ACC)) and machine learning techniques to measure cognitive deficits and their relationship with functional health deterioration.

Gurevich et al. had used SVM and neuropsychological test for the classification of AD from other causes of cognitive impairment. The highest classification accuracy they had achieved through their proposed method was 89.00% [ 102 ]. Karaglani et al. had proposed a ML based automated diagnosis system for AD by using blood-based biosignatures. In their proposed method, they used mRNA-based statistically equivalent signatures for feature ranking and a RF model for classification. Their proposed automated diagnosis system had reported the accuracy of 84.60% using RF [ 103 ]. Ryzhikova et al. had analyzed cerebrospinal fluid for the diagnosis of AD by using ML algorithms. For classification purposes, artificial neural networks (ANN) and SVM discriminant analysis (SVM-DA) statistical methods were applied, with the best findings allowing for the distinguishing of AD and HC participants with 84.00% sensitivity and specificity. The proposed classification models have a high discriminative power, implying that the technique has a lot of potential for AD diagnosis [ 104 ]. Cho and Chen had designed a double layer dementia diagnosis system based on ML where fuzzy cognitive maps (FCMs) and probability neural networks (PNNs) were used to provide initial diagnoses at the base layer, and Bayesian networks (BNs) were used to provide final diagnoses at the top layer. Diagnosis results, “proposed treatment,” and “no treatment required” might be used to provide medical institutions with self-testing or secondary dementia diagnosis. The highest accuracy reported by their proposed system was 83.00% [ 105 ]. Facal et al. had studied the role of cognitive reserve in the conversion from MCI to dementia using ML. Nine ML classification algorithms were tried in their study, and seven relevant performance parameters were generated to assess the prediction accuracy for converted and non-converted individuals. The use of ML algorithms on socio-demographic, basic health, and CR proxy data allowed for the prediction of dementia conversion. The Gradient Boosting Classifier (ACC = 0.93; F1 = 0.86 and Cohen’s kappa = 0.82) and RF Classifier (ACC = 92%; F1 = 0.79 and Cohen’s kappa = 0.71) performed the best [ 106 ]. Jin et al. had proposed automatic classification of dementia from learning of clinical consensus diagnosis in India using ML techniques. All viable ML models exhibited remarkable discriminative skills (AUC >90%) as well as comparable accuracy and specificity (both around 95%). The SVM model beat other ML models by obtaining the highest sensitivity (0.81), F1 score (0.72), kappa (.70, showing strong agreement), and accuracy (second highest) (0.65). As a consequence, the SVM was chosen as the best model in their research work [ 107 ]. James et al. had evaluated the performance of ML algorithms for predicting the progression of dementia in memory clinic patients. According to their findings, ML algorithms outperformed humans in predicting incident all-cause dementia within two years. Using all 258 variables, the gradient-boosted trees approach had an overall accuracy of 92% , sensitivity of 0.45, specificity of 0.97, and an AUC of 0.92. Analysis of variable significance had indicated that just 6 variables were necessary for ML algorithms to attain an accuracy of 91% and an AUC of at least 89.00% [ 108 ]. Bougea et al. had investigated the effectiveness of logistic regression (LR), K-nearest neighbours (K-NNs), SVM, the Naive Bayes classifier, and the Ensemble Model to correctly predict PDD or DLB. The K-NN classification model exhibited an overall accuracy of 91.2% based on 15 top clinical and cognitive scores, with 96.42% sensitivity and 81% specificity in distinguishing between DLB and PDD. Based on the 15 best characteristics, the binomial logistic regression classification model had attained an accuracy of 87.5%, with 93.93% sensitivity and 87% specificity. Based on the 15 best characteristics, the SVM classification model had achieved an accuracy of 84.6% of overall instances, 90.62% sensitivity, and 78.58% specificity. A model based on NB classification obtained an accuracy of 82.05%, sensitivity of 93.10%, and a specificity of 74.41%. Finally, an ensemble model, which was constructed by combining the separate ones, attained 89.74% accuracy, 93.75% sensitivity, and 85.73% specificity [ 109 ] (Fig. 6 ).

Accuracy comparison of different ML models based on clinical-variable modality

Salem et al. had presented a regression-based ML model for the prediction of dementia. In their proposed method, they had investigated ML approaches for unbalanced learning. In their suggested supervised ML approach, they started by intentionally oversampling the minority class and undersampling the majority class, in order to reduce the bias of the ML model to be trained on the dataset. Furthermore, they had deployed cost-sensitive strategies to penalize the ML models when an instance was misclassified in the minority class. According to their findings, the balanced RF was the most resilient probabilistic model (with just 20 features/variables) with an F1 score of 0.82, a G-Mean of 0.88, and an AUC of 0.88 using ROC. With a F1-score of 0.74 and an AUC of 0.80 by ROC, the calibrated-weighted SVM was their top classification model for the same number of features [ 110 ]. Gutierrez et al. had designed an automated diagnosis system for the detection of AD and FTD by using feature engineering and genetic algorithms. Their proposed system had obtained the accuracy of 84% [ 111 ]. Mirzaei and Adeli had analyzed the state-of-the-art ML techniques used for the detection and classification of AD [ 112 ]. Hsiu et al. had studied ML algorithms for early identification of cognitive impairment. Their proposed model had obtained the accuracy of 70.32% by threefold cross-validation scheme [ 113 ]. Several classification models were constructed using various ML and feature selection methodologies to automate MCI detection using gait biomarkers. They had demonstrated, however, that dual-task walking differentiated between MCI and CN individuals. The ML model used for MCI pre-screening based on inertial sensor-derived gait biomarkers achieved 71.67% accuracy and 83.33% sensitivity, respectively, as reported by Shahzad et al. [ 114 ]. Hane et al. investigated the use of deidentified clinical notes acquired from multiple hospital systems over a 10-year period to enhance retrospective ML models predicting the risk of developing AD. The AUC improved from 85.00% to 94.00% by utilizing clinical notes, and the positive predictive value (PPV) rose from 45.07% (25,245/56,018) to 68.32% (14,153/20,717) in the model at the beginning of disease [ 115 ]. Table 5 provides the overall performance evaluation of the ML models that were presented by the researchers for the prediction of dementia and its subtypes by using clinical-variable data as a modality.

ML based diagnostic models for dementia: Voice modality

Similar to the image and clinical-variable modalities, researchers had also developed automated diagnostic systems based on voice data for the prediction of dementia. Hereby, we have reviewed the research work done by the scientists in detail. For example, Chlasta and Wolk had worked on the computer-based automated screening of dementia patients by spontaneous speech analysis using DL and ML techniques. In their work, they used neural networks to extract the features from the voice data; the extracted features were then fed into a linear SVM for classification purposes. Their SVM model had obtained the accuracy of 59.1% while CNN based ML model had reported the accuracy of 63.6% [ 121 ]. Chien et al. had presented an ML model for the assessment of AD using speech data. Their suggested model included a feature sequence that was used to extract the features from the raw audio data, as well as a recurrent neural network (RNN) for classification. Their proposed ML model had reported an accuracy of 83.80% based on the ROC curve [ 122 ]. Shimoda et al. had designed an ML model that identified the risk of dementia based on the voice feature in telephone conversations. Extreme gradient boosting (XGBoost), RF, and LR based ML models were used, with each audio file serving as one observation. The predictive performance of the constructed ML models was tested by characterizing the ROC curve and determining the AUC, sensitivity, and specificity [ 123 ]. Nishikawa et al. had developed an ensemble discriminating system based on a classifier with statistical acoustic characteristics and a neural network of transformer models, with an F1-score of 90.70% [ 124 ]. Liu et al. had introduced a new technique for recognizing Alzheimer’s disease that used spectrogram features derived from speech data, which aided families in comprehending the illness development of patients at an earlier stage, allowing them to take preventive measures. They used ML techniques to diagnose AD using speech data collected from older adults who displayed the attributes described in the speech. Their proposed method had obtained the maximum accuracy of 84.40% based on LogisticRegressionCV [ 125 ]. Searle et al. had created a ML model to assess spontaneous speech, which might potentially give an efficient diagnostic tool for earlier AD detection. Their suggested model was a fundamental Term Frequency-Inverse Document Frequency (TF-IDF) vectorizer as input into an SVM model, and the top performing models were a pre-trained transformer-based model ’DistilBERT’ when used as an embedding layer into simple linear models. The proposed model had obtained the highest accuracy of 82.00% [ 126 ]. Zhu et al. had suggested an ML model that employed the speech pause as an effective biomarker in dementia detection, with the purpose of reducing the detection, model’s confidence levels by adding perturbation to the speech pauses of the testing samples. They next investigated the impact of the perturbation in training data on the detection model using an adversarial training technique. The proposed model had achieved an accuracy of 84.00% [ 127 ]. Ossewaarde et al. had proposed ML model based on SVM for the classification of spontaneous speech of individuals with dementia based on automatic prosody analysis. Their findings suggest that the classifier can distinguish some dementia types (PPA-NF, AD), but not others (PD) [ 128 ]. Xue et al. had developed an ML model based on DL for the detection of dementia by using voice recordings. In their ML model, long short-term memory (LSTM) network and the convolutional neural network (CNN) utilized audio recordings to categorize whether the recording contained a participant with either NC or only DE and to discriminate between recordings belonging to those with DE and those without DE (i.e., NDE (NC+MCI)) [ 129 ]. Weiner et al. had presented two pipelines of feature extraction for dementia detection: the manual pipeline used manual transcriptions, while the fully automatic pipeline used transcriptions created by automatic speech recognition (ASR). The acoustic and linguistic features that they had extracted need no language specific tools other than the ASR system. Using these two different feature extraction pipelines, they had automatically detect dementia [ 130 ] (Fig. 7 ).

Accuracy comparison of different ML models based on voice modality

Furthermore, Sadeghian et al. had presented the empirical evidence that a combination of acoustic features from speech, linguistic features were extracted from an automatically determined transcription of the speech including punctuation, and results of a mini mental state exam (MMSE) had achieved strong discrimination between subjects with a probable AD versus matched normal controls [ 131 ]. Khodabakhsh et al. had evaluated the linguistic and prosodic characteristics in Turkish conversational language for the identification of AD. Their research suggested that prosodic characteristics outperformed linguistic features by a wide margin. Three of the prosodic features had helped to achieve a classification accuracy of more than 80%, However, their feature fusion experiments did not improve classification performance any more [ 132 ]. Edwards et al. had analyzed the text data at both the word level and phoneme level, which leads to the best-performing system in combination with audio features. Thus, the proposed system was both multi-modal (audio and text) and multi-scale (word and phoneme levels). Experiments with larger neural language models had not resulted in improvement, given the small amount of text data available [ 133 ]. Kumar et al. had identified speech features relevant in predicting AD based on ML. They had deployed neural network for the classification and obtained the accuracy of 92.05% [ 134 ]. Ossewaarde et al. had built ML model based on SVM for the classification from spontaneous speech of individuals with dementia by using automatic prosody [ 128 ]. Luz et al. had developed an ML approach for analyzing patient speech in dialogue for dementia identification. They had designed a prediction model, and the suggested strategy leveraged additive logistic regression (ML boosting method) on content-free data gathered through dialogical interaction. Their proposed model obtained the accuracy of 86.50% [ 135 ]. Sysed et al. had designed a multimodal system that identified linguistic and paralinguistic traits of dementia using an automated screening tool. Their proposed system had used bag-of-deep-feature for feature selection and ensemble model for classification [ 136 ]. Moreover, Sarawgi et al. had used multimodal inductive transfer learning for AD detection and severity. Their proposed system further achieved state-of-the-art AD classification accuracy of 88.0% when evaluated on the full benchmark DementiaBank Pitt database. Table 6 provides the overall performance evaluation of the ML models that were presented by the researchers for the prediction of dementia and its subtypes by using voice-modality data.

In this SLR, we examined the research work that employed ML and DL algorithms to analyze clinical data in order to identify variables that might help predict dementia. We studied 75 research articles that were published in the last 10 years that used image, clinical-variable, and voice data to predict dementia and its subtypes. Nowadays, the healthcare industry creates a vast quantity of data on patients’ health; this data is used by researchers to enhance individual health by utilizing developing technologies such as ML and DL. As a result, researchers can not only distinguish dementia patients from healthy people with high accuracy, but also forecast the disease progression of MCI patients.

Accuracy comparison of ML models based on data modality

Therefore, researchers have expressed a strong interest in designing and developing automated diagnostic systems based on ML and DL techniques. As seen in Fig. 4 ., there has been an exponential increase in the number of such research publications that use ML algorithms for dementia prediction and detection in the previous four years. We investigated the selected papers using significant performance assessment criteria for ML and DL approaches such as data attributes, computational methodologies, and study emphasis. In this SLR, we have uncovered research gaps in the present literature as well as anticipated future research opportunities. Additionally, in Fig. 8 model comparison, we examined the performance of multiple ML algorithms for dementia prediction based on three types of data modalities: image, clinical-variable, and voice. The accuracy gained by image-based ML algorithms is higher when compared to clinical-variable and voice modalities, as shown in Fig. 8 model comparison. Moreover, the researchers’ suggested SVM, RF, and ANN-based ML techniques outperformed the rest of the ML algorithms in terms of performance. According to Fig. 8 model comparison, voice modality-based ML models show worse accuracy when compared to image and clinical-variable modality data. As a result, there is still a performance gap for researchers to close in order to improve the performance of ML algorithms for the prediction of dementia using voice data. Hence, researchers have shown a strong interest in the creation of automated diagnosis systems for dementia prediction utilizing speech data and ML algorithms, as illustrated in Fig. 4 .

Sensitivity and specificity comparison of ML based on modality

The ML and DL models are likely prone to problems such as poor quality of data, poor selection of ML model, Bias Variance tradeoff and training too complex models. Thus, scientists have developed various evaluation metrics (i.e., ROC, AUC, MCC, F1-score, K-fold) and methods to avoid these problems. The data is a crucial element in ML because ML models work only with numeric data; therefore, poor data quality results in lower performance of ML models. Moreover, imbalance classes in the dataset also cause the bias results from the ML models. Thus, this problem can be overcome by oversampling or undersampling the training data. There are different techniques that are used by the AI engineers for oversampling, such as random oversampling and the synthetic minority oversampling technique (SMOTE). To evaluate the bias researchers’ work, use sensitivity and specificity as an evaluation metric to measure the bias of the ML model. Higher values of sensitivity and specificity means model is free from the biasness while having either one parameter value higher and other one is lower means there is biasness exist. Thus, we have also studied the sensitivity and specificity, along with the accuracy, of the previously proposed ML models for dementia prediction. Figure 9 Comparison provides a brief description of the sensitivity and specificity of the ML models for the detection of dementia based on different data modalities. From Fig. 9 , we can observe that ML models have higher values for sensitivity and specificity when using image data as compared to clinical-variable and voice modality data. In comparison to accuracy from Fig. 8 to sensitivity and specificity from Fig. 9 , we have noted that the results obtained from image based modality are more reliable and precise using ML and DL algorithms in spite of clinical and voice modality.

Furthermore, the correlation between sensitivity and specificity would help us understand the efficacy of the ML models, which are designed for automated disease prediction. The mathematical terms “sensitivity” and “specificity” indicate the accuracy of a test that reports the presence or absence of a disease. Individuals who meet the requirement are labelled “positive,” while those who do not are considered “negative”. The chance of a positive test, conditioned on being actually positive, is referred to as sensitivity (the true positive rate), while specificity (true negative rate) is the likelihood of a negative test if it is actually negative. Sensitivity and specificity are inversely proportional, which means that as sensitivity rises, specificity falls, and vice versa. Mathematically, sensitivity and specificity are given as:

On the other hand, accuracy is a ratio of number of correct assessments / number of all assessments. The proportion of genuine positive outcomes (both true positive and true negative) in the selected population is represented by the numerical value of accuracy. The test result is accurate 99% of the time, whether positive or negative. For the most part, this is right. However, it is worth noting that the equation of accuracy means that even if both sensitivity and specificity are high, say 99%, this does not imply that the test’s accuracy is also high. In addition to sensitivity and specificity, accuracy is determined by the prevalence of the illness in the target population. A diagnosis for a rare ailment in the target group may have high sensitivity and specificity but low accuracy. However, for a balanced dataset, ML models with higher sensitivity and specificity result in higher accuracy. Hence, accuracy must be interpreted carefully. The mathematical formula for accuracy is given as:

where, TP stands for the number of true positives, FP stands for the number of false positives, TN stands for the true negative, and FN stands for the false negative.

Accuracy comparison of ML models along with number of sample in the dataset based on data modality

We classified all datasets that were used by researchers to test the performance of their proposed ML models for the prediction of dementia (AD, VaD, MCI, and FTD) into three types: image, clinical-variable, and voice. A total of 61 datasets were examined in terms of the number of samples and variables in the datasets. In image modality datasets from the Table 1 , it can be observed that the ADNI dataset has a significant number of samples, which is 750, while the NINDS-AIREN dataset has more variables as compared to the rest of the datasets in the image modality data. Moreover, from the Table 2 of clinical-variable modality datasets, it can be noticed that the ADRD dataset has the highest number of samples (44945) as compared to the rest of the dataset, while the Raman spectral dataset has the highest number of variables (366). In the last, Table 3 of voice data modality elaborated the dataset of voice modality where FHS dataset has highest number of samples of 5449 while VBSD dataset had highest variables of 254 as compared to rest of the datasets in voice modality. The type of data and the size of the dataset are two important factors that have a significant influence on the performance of ML models. Thus, we have also studied this factor by comparing the accuracy along with the number of samples in the dataset with respect to data modalities. From Fig. 10 , it can be observed that the majority of the ML models that used image data have higher accuracy along with a higher number of samples in the dataset. There are few ML models that show poor performance when the number of samples in the dataset is large. While, clinical-variable and voice modalities show prominent performance when the number of samples in the dataset is small.

Overall percentage of ML models used in the selected research articles regardless of data modality

Moreover, we examined the effectiveness of ML classifiers utilized by the researchers in their proposed automated diagnostic systems for dementia prediction and classification. According to the selected studies of this SLR, SVM is the most commonly used ML classifier by researchers for the classification of patients and normal subjects using three data modalities (i.e., image, clinical-variable, voice), RF is the second most commonly used ML classifier by researchers, and CNN is the third most commonly used ML classifier by researchers. It can be observed from the Fig. 11 . SVMs are the most powerful tools for the binary classification task, along with RF. From Fig. 8 , we can see that SVM also obtained the highest average accuracy based on three types of data modalities. Hence, this factor also encourages the scientists to employ SVM as a binary classifier for dementia prediction or other disease prediction systems. From Fig. 11 , we can observe the percentage of other ML classifiers that were used by the researchers in selected research articles for the automated diagnosis of dementia.

There are several evaluation metrics that are used for the performance assessment of ML models, such as F1score, AUC, ROC, Matthew’s correlation coefficient (MCC), cross-validation, K-fold, specificity, sensitivity, and accuracy. Each evaluation metric has its own pros and cons. Thus, the selection of appropriate evaluation metrics for the assessment of the ML model is essential to understanding its efficiency and performance. For instance, when data plays a vital role in ML models for decision-making and a dataset has unbalanced classes, it may be possible that results from the ML predictive model might be biassed due to the unbalanced nature of the data in the dataset. Thus, here evaluation metrics help to eliminate the factor of biasness in the results, i.e., the k-fold. The F1-score evolution metric is suitable for the classification of multiple classes in the dataset. while ROC tells us how well the ML model can differentiate binary classes. As a result, AUC and ROC reveal how effectively the probabilities from the positive classes are separated from the probabilities from the negative classes. From Fig. 12 , it can be depicted that cross validation is mostly used in the studies that were selected for this SLR to evaluate the performance of proposed ML models. MCC is the second most used evaluation metric, while ROC is in third place. The proposition of other evaluation metrics used by the researchers to validate the efficiency of their proposed ML models can be observed from Fig. 12 .

Overall percentage of evaluation metrics of ML models used by the researchers in the selected research articles

Limitations in the previously proposed ML models

ML algorithms have been effectively applied to a broad range of real-world challenges, including banking, cybersecurity, transportation, and robots. They do, however, have fundamental limitations that make them inappropriate for every problem. In the clinical domain, researchers have concentrated on the supervised learning approach, developing various automated diagnostics for AD, MCI, and dementia prediction using supervised machine algorithms. From the Figs.  8 and 11 , It can be noticed that supervised ML classifiers are mostly used by the researchers in the selected past research articles. Because supervised machine learning approaches have various limitations, automated diagnostic methods for dementia prediction based on supervised techniques suffer from some, if not all, of these constraints. In this part, we have examined the drawbacks of supervised ML-based techniques for dementia prediction, which are as follows:

The model overfitting problem affects the performance of ML models. As previously indicated, several researchers have used the k-fold cross-validation approach to evaluate the efficacy of their constructed diagnostic system. However, because of data leaks, it may result in highly biassed findings.

To deal with problem of imbalance classes in the dataset, Researchers and scientists had devised several techniques to eliminate the problem of imbalance classes such as random oversampling example (ROSE), synthetic minority over-sampling technique (SMOTE) and random over sampling (ROS) etc. Unfortunately, in the selected study, the researchers had not considered this factor to deal with the problem of imbalanced classes in the dataset that cause problems of bias.

Supervised ML models require training on a dataset; nevertheless, training on a large quantity of significant data is a hard and time-consuming job, especially for slow learning algorithms like kNN.

For training and testing of the ML models, researchers had used different data partitioning methods, which resulted in inconsistent comparisons of accuracy and other evaluation metrics among the proposed ML models for dementia prediction. Thus, standard data partition schemes should be adopted (holdout) for the comparison of ML models developed by the researcher for dementia prediction.

Another challenge with ML-based automated diagnostic systems for dementia is the time complexity of the proposed ML algorithms. The time complexity means the overall time require to complete all the computational tasks by the ML model for making a prediction. The ML model can forecast results only after it has been trained on the training data, which takes time to analyze. Furthermore, ML models include a large number of parameters that must be manually modified in the case of supervised learning. As a result, it takes a significant amount of effort and time to fine-tune the hyperparameters of the ML model in order to get higher performance.

DL technology has demonstrated cutting-edge performances for the prediction of various diseases in the recent years. However, DL technology needs a massive quantity of data for model training, which is a time-consuming and tough task. Due to the complexity of data models, training is quite costly. Furthermore, DL necessitates the use of pricey GPUs and hundreds of workstations, which are not effective in terms of economics.

Future research directions

In recent years, several ML models have been presented for the prediction of AD and MCI; nevertheless, there are still certain areas that need to be explored by academics and experts. In this section, we have discussed different research areas and the future prospects of ML algorithms for dementia detection. We infer from this study that the following major parameters have a role in the efficient identification of dementia and its forms.

Data is extremely important in the case of ML-based automated detection of dementia, especially when DL models are considered. Many of the publicly available datasets, however, are modest in size. But future research should concentrate on gathering a huge number of samples for the datasets. In this SLR, we studied ML-based automated diagnostic systems for dementia prediction using three different kinds of data modalities (image, clinical_variable, voice). From Fig. 10 , it can be observed that only the image modality based ML model obtained the higher accuracy along with the large size of the dataset, while the voice modality based ML model obtained the higher accuracy on a small dataset. Thus, for the researchers, there is still room available for designing and developing the automated prediction of dementia and its sub-types by using voice data. Therefore, the interest of researchers have been tremendously raised for the development of automated diagnostic systems for dementia prediction using voice data modality and this trend can be confirmed from the Fig. 4 . There is still a lot room available for the improvement in design and construction of automated diagnostic systems for the dementia using clinical-variable data modality for the researchers. Because, the ML model was developed in the past using clinical-variable data, it displays mix performance by using clinical_variable modality, i.e., when the number of samples is lower in the dataset, the ML shows lower accuracy. Thus. In the future, we need to increase the number of samples in the dataset so that we have larger datasets for experimental purposes and the designed ML model can be effectively evaluated.

In selected studies of this SLR, the majority of ML algorithms belong to the supervised category of learning. While few researchers used an unsupervised ML approach for the prediction of dementia and its subtypes, Altough, unspervsied learning approaches suffer from the limitation such as less accuracy, more expensive in term of computational etc. Therefore, it will encourage scientists and researchers to design and construct new techniques and methods using supervised ML algorithms that are more precise and accurate for the prediction of dementia and its subtypes. Moreover, in this SLR, we have analysed the various ML models based on three data modalities (image, clinical-variable, and voice), and we have comprehensively compared previously proposed ML-based systems in terms of various evaluation metrics, but with different data modalities, it would be suggested that multimodal processing techniques based on ML would provide more reliable and efficient results. Hence, in the future, researchers should exploit multimodal approaches based on ML for a better prediction of dementia and its subtypes.

In contrast to earlier SLR studies that examined numerous ML techniques proposed for the automated diagnosis of dementia and its subtypes (AD, VaD, FTD, and MCI) using one type of data modality, this study reviewed ML methods for dementia considering different types of data modalities such as image data, clinical variables, and voice data. The research articles published from 2011 to 2022 were gathered using different databases. It was pointed out that ML approaches based on image data modality has shown better performance compared with ML methods trained on clinical variables based data and voice data modality. Furthermore, this study critically evaluated the previously proposed methods and highlighted limitations in these methods. To overcome these limitations, this study presented future research directions in the domain of automated dementia prediction using ML approaches. We hope that this SLR will be helpful for AI and ML researchers and medical practitioners who are working in the domain of automated diagnostic systems for dementia prediction.

Data Availability

Not applicable.

Code Availability

Menéndez, G.: La revolución de la longevidad: cambio tecnológico, envejecimiento poblacional y transformación cultural. Revista de Ciencias Sociales 30 (41), 159–178 (2017)

Prince, M.J., Wimo, A., Guerchet, M.M., Ali, G.C., Wu, Y.-T., Prina, M.: World alzheimer report 2015-the global impact of dementia: An analysis of prevalence, incidence, cost and trends (2015)

Vrijsen, J., Matulessij, T., Joxhorst, T., de Rooij, S.E., Smidt, N.: Knowledge, health beliefs and attitudes towards dementia and dementia risk reduction among the dutch general population: a cross-sectional study. BMC public health 21 (1), 1–11 (2021)

Widiger, T.A., Costa, P.T., Association, A.P., et al : Personality Disorders and the Five-factor Model of Personality. JSTOR, (2013)

Lo, R.Y.: The borderland between normal aging and dementia. Tzu-Chi Medical Journal 29 (2), 65 (2017)

WHO: Dementia. World Health Organization. https://www.who.int/news-room/fact-sheets/detail/dementia

Budson, A.E., Solomon, P.R.: Memory Loss E-book: A Practical Guide for Clinicians. Elsevier Health Sciences, (2011)

friendly wyoming, D.: Types of dementia. Dementia friendly wyoming. https://www.dfwsheridan.org/types-dementia

Tackenberg, C., Kulic, L., Nitsch, R.M.: Familial alzheimer’s disease mutations at position 22 of the amyloid \(\beta\) -peptide sequence differentially affect synaptic loss, tau phosphorylation and neuronal cell death in an ex vivo system. PloS one 15 (9), 0239584 (2020)

Wang, H., Kulas, J.A., Wang, C., Holtzman, D.M., Ferris, H.A., Hansen, S.B.: Regulation of beta-amyloid production in neurons by astrocyte-derived cholesterol. Proceedings of the National Academy of Sciences 118 (33) (2021)

Vilchez, D., Saez, I., Dillin, A.: The role of protein clearance mechanisms in organismal ageing and age-related diseases. Nature communications 5 (1), 1–13 (2014)

Breijyeh, Z., Karaman, R.: Comprehensive review on alzheimer’s disease: Causes and treatment. Molecules 25 (24), 5789 (2020)

Long, J.M., Holtzman, D.M.: Alzheimer disease: an update on pathobiology and treatment strategies. Cell 179 (2), 312–339 (2019)

Román, G.C.: Vascular dementia may be the most common form of dementia in the elderly. Journal of the neurological sciences 203 , 7–10 (2002)

Gold, G., Bouras, C., Canuto, A., Bergallo, M.F., Herrmann, F.R., Hof, P.R., Mayor, P.-A., Michel, J.-P., Giannakopoulos, P.: Clinicopathological validation study of four sets of clinical criteria for vascular dementia. American Journal of Psychiatry 159 (1), 82–87 (2002)

Lina, R.: Atrial fibrillation in aging; methodological aspects and the relation to dementia and cerebral vascular disease (2022)

Kuźma, E., Lourida, I., Moore, S.F., Levine, D.A., Ukoumunne, O.C., Llewellyn, D.J.: Stroke and dementia risk: a systematic review and meta-analysis. Alzheimer’s & Dementia 14 (11), 1416–1426 (2018)

Tay, J., Morris, R.G., Tuladhar, A.M., Husain, M., de Leeuw, F.-E., Markus, H.S.: Apathy, but not depression, predicts all-cause dementia in cerebral small vessel disease. Journal of Neurology, Neurosurgery & Psychiatry 91 (9), 953–959 (2020)

Morton, R.E., St. John, P.D., Tyas, S.L.: Migraine and the risk of all-cause dementia, alzheimer’s disease, and vascular dementia: A prospective cohort study in community-dwelling older adults. International journal of geriatric psychiatry 34 (11), 1667–1676 (2019)

Sanford, A.M.: Lewy body dementia. Clinics in geriatric medicine 34 (4), 603–615 (2018)

Taylor, J.-P., McKeith, I.G., Burn, D.J., Boeve, B.F., Weintraub, D., Bamford, C., Allan, L.M., Thomas, A.J., T O’Brien, J.: New evidence on the management of lewy body dementia. The Lancet Neurology 19 (2), 157–169 (2020)

McKEITH, I.G., Galasko, D., Wilcock, G.K., Byrne, E.J.: Lewy body dementia–diagnosis and treatment. The British Journal of Psychiatry 167 (6), 709–717 (1995)

Surendranathan, A., Kane, J.P., Bentley, A., Barker, S.A., Taylor, J.-P., Thomas, A.J., Allan, L.M., McNally, R.J., James, P.W., McKeith, I.G., et al.: Clinical diagnosis of lewy body dementia. BJPsych open 6 (4) (2020)

Finger, E.C.: Frontotemporal dementias. Continuum: Lifelong Learning in Neurology 22 (2 Dementia), 464 (2016)

Rabinovici, G.D., Miller, B.L.: Frontotemporal lobar degeneration. CNS drugs 24 (5), 375–398 (2010)

Organization, W.H., et al.: The ICD-10 classification of mental and behavioural disorders. Clinical descriptions and diagnostic guidelines; 2004. Geneva: WHO (2021)

Ghetti, B., Buratti, E., Boeve, B., Rademakers, R.: Frontotemporal Dementias vol. 320. Springer, (2021)

Custodio, N., Montesinos, R., Lira, D., Herrera-Pérez, E., Bardales, Y., Valeriano-Lorenzo, L.: Mixed dementia: A review of the evidence. Dementia & neuropsychologia 11 , 364–370 (2017)

Garcia-Ptacek, S., Kåreholt, I., Cermakova, P., Rizzuto, D., Religa, D., Eriksdotter, M.: Causes of death according to death certificates in individuals with dementia: a cohort from the swedish dementia registry. Journal of the American Geriatrics Society 64 (11), 137–142 (2016)

Arvanitakis, Z., Shah, R.C., Bennett, D.A.: Diagnosis and management of dementia. Jama 322 (16), 1589–1599 (2019)

Ullah, H., Bin Heyat, M.B., AlSalman, H., Khan, H.M., Akhtar, F., Gumaei, A., Mehdi, A., Muaad, A.Y., Islam, M.S., Ali, A., et al.: An effective and lightweight deep electrocardiography arrhythmia recognition model using novel special and native structural regularization techniques on cardiac signal. Journal of Healthcare Engineering 2022 (2022)

Javeed, A., Ali, L., Mohammed Seid, A., Ali, A., Khan, D., Imrana, Y.: A clinical decision support system (cdss) for unbiased prediction of caesarean section based on features extraction and optimized classification. Computational Intelligence and Neuroscience 2022 (2022)

Imrana, Y., Xiang, Y., Ali, L., Abdul-Rauf, Z., Hu, Y.-C., Kadry, S., Lim, S.: \(\chi\) 2-bidlstm: A feature driven intrusion detection system based on \(\chi\) 2 statistical model and bidirectional lstm. Sensors 22 (5), 2018 (2022)

Ali, L., Zhu, C., Zhao, H., Zhang, Z., Liu, Y.: An integrated system for unbiased parkinson’s disease detection from handwritten drawings. In: Advances in Intelligent Systems and Computing, pp. 3–13. Springer, (2022)

Ali, L., Niamat, A., Khan, J.A., Golilarz, N.A., Xingzhong, X., Noor, A., Nour, R., Bukhari, S.A.C.: An optimized stacked support vector machines based expert system for the effective prediction of heart failure. IEEE Access 7 , 54007–54014 (2019)

Ali, L., Zhu, C., Zhou, M., Liu, Y.: Early diagnosis of parkinson’s disease from multiple voice recordings by simultaneous sample and feature selection. Expert Systems with Applications 137 , 22–28 (2019)

Ali, L., Khan, S.U., Golilarz, N.A., Yakubu, I., Qasim, I., Noor, A., Nour, R.: A feature-driven decision support system for heart failure prediction based on statistical model and gaussian naive bayes. Computational and Mathematical Methods in Medicine 2019 (2019)

Mehbodniya, A., Khan, I.R., Chakraborty, S., Karthik, M., Mehta, K., Ali, L., Nuagah, S.J.: Data mining in employee healthcare detection using intelligence techniques for industry development. Journal of Healthcare Engineering 2022 (2022)

Ali, L., Khan, S.U., Arshad, M., Ali, S., Anwar, M.: A multi-model framework for evaluating type of speech samples having complementary information about parkinson’s disease. In: 2019 International Conference on Electrical, Communication, and Computer Engineering (ICECCE), pp. 1–5 (2019). IEEE

Ahmad, F.S., Ali, L., Khattak, H.A., Hameed, T., Wajahat, I., Kadry, S., Bukhari, S.A.C., et al : A hybrid machine learning framework to predict mortality in paralytic ileus patients using electronic health records (ehrs). Journal of Ambient Intelligence and Humanized Computing 12 (3), 3283–3293 (2021)

Ali, L., Wajahat, I., Golilarz, N.A., Keshtkar, F., Bukhari, S.A.C.: Lda–ga–svm: improved hepatocellular carcinoma prediction through dimensionality reduction and genetically optimized support vector machine. Neural Computing and Applications 33 (7), 2783–2792 (2021)

Ali, L., Zhu, C., Zhang, Z., Liu, Y.: Automated detection of parkinson’s disease based on multiple types of sustained phonations using linear discriminant analysis and genetically optimized neural network. IEEE journal of translational engineering in health and medicine 7 , 1–10 (2019)

Ali, L., He, Z., Cao, W., Rauf, H.T., Imrana, Y., Heyat, M.B.B.: Mmdd-ensemble: A multimodal data–driven ensemble approach for parkinson’s disease detection. Frontiers in Neuroscience 15 (2021)

Rehman, A., Khan, A., Ali, M.A., Khan, M.U., Khan, S.U., Ali, L.: Performance analysis of pca, sparse pca, kernel pca and incremental pca algorithms for heart failure prediction. In: 2020 International Conference on Electrical, Communication, and Computer Engineering (ICECCE), pp. 1–5 (2020). IEEE

Ahmed, F.S., Ali, L., Joseph, B.A., Ikram, A., Mustafa, R.U., Bukhari, S.A.C.: A statistically rigorous deep neural network approach to predict mortality in trauma patients admitted to the intensive care unit. Journal of Trauma and Acute Care Surgery 89 (4), 736–742 (2020)

Imrana, Y., Xiang, Y., Ali, L., Abdul-Rauf, Z.: A bidirectional lstm deep learning approach for intrusion detection. Expert Systems with Applications 185 , 115524 (2021)

Akbar, W., Wu, W.-p., Saleem, S., Farhan, M., Saleem, M.A., Javeed, A., Ali, L.: Development of hepatitis disease detection system by exploiting sparsity in linear support vector machine to improve strength of adaboost ensemble model. Mobile Information Systems 2020 (2020)

Ali, L., Bukhari, S.: An approach based on mutually informed neural networks to optimize the generalization capabilities of decision support systems developed for heart failure prediction. Irbm 42 (5), 345–352 (2021)

McInnes, M.D., Moher, D., Thombs, B.D., McGrath, T.A., Bossuyt, P.M., Clifford, T., Cohen, J.F., Deeks, J.J., Gatsonis, C., Hooft, L., et al : Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the prisma-dta statement. Jama 319 (4), 388–396 (2018)

Liberati, A., Altman, D.G., Tetzlaff, J., Mulrow, C., Gøtzsche, P.C., Ioannidis, J.P., Clarke, M., Devereaux, P.J., Kleijnen, J., Moher, D.: The prisma statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaboration. Journal of clinical epidemiology 62 (10), 1–34 (2009)

Martí-Juan, G., Sanroma-Guell, G., Piella, G.: A survey on machine and statistical learning for longitudinal analysis of neuroimaging data in alzheimer’s disease. Computer methods and programs in biomedicine 189 , 105348 (2020)

McGinnis, J.M., Olsen, L., Goolsby, W.A., Grossmann, C., et al : Clinical Data as the Basic Staple of Health Learning: Creating and Protecting a Public Good: Workshop Summary. National Academies Press, (2011)

Vaughn, V.M., Linder, J.A.: Thoughtless design of the electronic health record drives overuse, but purposeful design can nudge improved patient care. BMJ Publishing Group Ltd (2018)

Doody, R., Stevens, J., Beck, C., Dubinsky, R., Kaye, J., Gwyther, L., Mohs, R., Thal, L., Whitehouse, P., DeKosky, S., et al : Practice parameter: Management of dementia (an evidence-based review): Report of the quality standards subcommittee of the american academy of neurology. Neurology 56 (9), 1154–1166 (2001)

Chi, C.-L., Zeng, W., Oh, W., Borson, S., Lenskaia, T., Shen, X., Tonellato, P.J.: Personalized long-term prediction of cognitive function: Using sequential assessments to improve model performance. Journal of biomedical informatics 76 , 78–86 (2017)

Johnson, K.A., Minoshima, S., Bohnen, N.I., Donohoe, K.J., Foster, N.L., Herscovitch, P., Karlawish, J.H., Rowe, C.C., Carrillo, M.C., Hartley, D.M., et al : Appropriate use criteria for amyloid pet: a report of the amyloid imaging task force, the society of nuclear medicine and molecular imaging, and the alzheimer’s association. Alzheimer’s & Dementia 9 (1), 1–16 (2013)

Shaw, L.M., Arias, J., Blennow, K., Galasko, D., Molinuevo, J.L., Salloway, S., Schindler, S., Carrillo, M.C., Hendrix, J.A., Ross, A., et al : Appropriate use criteria for lumbar puncture and cerebrospinal fluid testing in the diagnosis of alzheimer’s disease. Alzheimer’s & Dementia 14 (11), 1505–1521 (2018)

Mills, K.L., Tamnes, C.K.: Methods and considerations for longitudinal structural brain imaging analysis across development. Developmental cognitive neuroscience 9 , 172–190 (2014)

Ali, L., Zhu, C., Golilarz, N.A., Javeed, A., Zhou, M., Liu, Y.: Reliable parkinson’s disease detection by analyzing handwritten drawings: construction of an unbiased cascaded learning system based on feature selection and adaptive boosting model. Ieee Access 7 , 116480–116489 (2019)

Javeed, A., Rizvi, S.S., Zhou, S., Riaz, R., Khan, S.U., Kwon, S.J.: Heart risk failure prediction using a novel feature selection method for feature refinement and neural network for classification. Mobile Information Systems 2020 (2020)

Javeed, A., Zhou, S., Yongjian, L., Qasim, I., Noor, A., Nour, R.: An intelligent learning system based on random search algorithm and optimized random forest model for improved heart disease detection. IEEE Access 7 , 180235–180243 (2019)

Ali, L., Rahman, A., Khan, A., Zhou, M., Javeed, A., Khan, J.A.: An automated diagnostic system for heart disease prediction based on chi2 statistical model and optimally configured deep neural network. IEEE Access 7 , 34938–34945 (2019)

Maddox, T.M., Rumsfeld, J.S., Payne, P.R.: Questions for artificial intelligence in health care. Jama 321 (1), 31–32 (2019)

Chen, P.-H.C., Liu, Y., Peng, L.: How to develop machine learning models for healthcare. Nature materials 18 (5), 410–414 (2019)

Javeed, A., Khan, S.U., Ali, L., Ali, S., Imrana, Y., Rahman, A.: Machine learning-based automated diagnostic systems developed for heart failure prediction using different types of data modalities: A systematic review and future directions. Computational and Mathematical Methods in Medicine 2022 (2022)

Thacker, S.B., Qualters, J.R., Lee, L.M., for Disease Control, C., Prevention, et al : Public health surveillance in the united states: evolution and challenges. MMWR Suppl 61 (3), 3–9 (2012)

Boerma, J.T., Stansfield, S.K.: Health statistics now: are we making the right investments? The Lancet 369 (9563), 779–786 (2007)

Walport, M., Brest, P.: Sharing research data to improve public health. The Lancet 377 (9765), 537–539 (2011)

Van Panhuis, W.G., Paul, P., Emerson, C., Grefenstette, J., Wilder, R., Herbst, A.J., Heymann, D., Burke, D.S.: A systematic review of barriers to data sharing in public health. BMC public health 14 (1), 1–9 (2014)

Dashtipour, K., Taylor, W., Ansari, S., Zahid, A., Gogate, M., Ahmad, J., Assaleh, K., Arshad, K., Imran, M.A., Abbai, Q.: Detecting alzheimer’s disease using machine learning methods. In: EAI (2021)

Helaly, H.A., Badawy, M., Haikal, A.Y.: Deep learning approach for early detection of alzheimer’s disease. Cognitive Computation, 1–17 (2021)

Vandenberghe, R., Nelissen, N., Salmon, E., Ivanoiu, A., Hasselbalch, S., Andersen, A., Korner, A., Minthon, L., Brooks, D.J., Van Laere, K., et al : Binary classification of 18f-flutemetamol pet using machine learning: comparison with visual reads and structural mri. Neuroimage 64 , 517–525 (2013)

Odusami, M., Maskeliūnas, R., Damaševičius, R., Krilavičius, T.: Analysis of features of alzheimer’s disease: Detection of early stage from functional brain changes in magnetic resonance images using a finetuned resnet18 network. Diagnostics 11 (6), 1071 (2021)

Zheng, Y., Guo, H., Zhang, L., Wu, J., Li, Q., Lv, F.: Machine learning-based framework for differential diagnosis between vascular dementia and alzheimer’s disease using structural mri features. Frontiers in Neurology, 1097 (2019)

Basheer, S., Bhatia, S., Sakri, S.B.: Computational modeling of dementia prediction using deep neural network: Analysis on oasis dataset. IEEE Access 9 , 42449–42462 (2021)

Leong, L.K., Abdullah, A.A.: Prediction of alzheimer’s disease (ad) using machine learning techniques with boruta algorithm as feature selection method. In: Journal of Physics: Conference Series, vol. 1372, p. 012065 (2019). IOP Publishing

Battineni, G., Chintalapudi, N., Amenta, F.: Machine learning in medicine: Performance calculation of dementia prediction by support vector machines (svm). Informatics in Medicine Unlocked 16 , 100200 (2019)

Mathotaarachchi, S., Pascoal, T.A., Shin, M., Benedet, A.L., Kang, M.S., Beaudry, T., Fonov, V.S., Gauthier, S., Rosa-Neto, P., Initiative, A.D.N., et al : Identifying incipient dementia individuals using machine learning and amyloid imaging. Neurobiology of aging 59 , 80–90 (2017)

Aruna, S., Chitra, S.: Machine learning approach for identifying dementia from mri images. International Journal of Computer and Information Engineering 9 (3), 881–888 (2016)

Herzog, N.J., Magoulas, G.D.: Brain asymmetry detection and machine learning classification for diagnosis of early dementia. Sensors 21 (3), 778 (2021)

Battineni, G., Chintalapudi, N., Amenta, F., Traini, E.: A comprehensive machine-learning model applied to magnetic resonance imaging (mri) to predict alzheimer’s disease (ad) in older subjects. Journal of Clinical Medicine 9 (7), 2146 (2020)

Ma, D., Yee, E., Stocks, J.K., Jenkins, L.M., Popuri, K., Chausse, G., Wang, L., Probst, S., Beg, M.F.: Blinded clinical evaluation for dementia of alzheimer’s type classification using fdg-pet: A comparison between feature-engineered and non-feature-engineered machine learning methods. Journal of Alzheimer’s Disease 80 (2), 715–726 (2021)

Bidani, A., Gouider, M.S., Travieso-González, C.M.: Dementia detection and classification from mri images using deep neural networks and transfer learning. In: International Work-Conference on Artificial Neural Networks, pp. 925–933 (2019). Springer

Moscoso, A., Silva-Rodríguez, J., Aldrey, J.M., Cortés, J., Fernández-Ferreiro, A., Gómez-Lado, N., Ruibal, Á., Aguiar, P., Initiative, A.D.N., et al : Prediction of alzheimer’s disease dementia with mri beyond the short-term: Implications for the design of predictive models. NeuroImage: Clinical 23 , 101837 (2019)

Khan, A., Zubair, S.: An improved multi-modal based machine learning approach for the prognosis of alzheimer’s disease. Journal of King Saud University-Computer and Information Sciences (2020)

Mohammed, B.A., Senan, E.M., Rassem, T.H., Makbol, N.M., Alanazi, A.A., Al-Mekhlafi, Z.G., Almurayziq, T.S., Ghaleb, F.A.: Multi-method analysis of medical records and mri images for early diagnosis of dementia and alzheimer’s disease based on deep learning and hybrid methods. Electronics 10 (22), 2860 (2021)

Salvatore, C., Cerasa, A., Battista, P., Gilardi, M.C., Quattrone, A., Castiglioni, I.: Magnetic resonance imaging biomarkers for the early diagnosis of alzheimer’s disease: a machine learning approach. Frontiers in neuroscience 9 , 307 (2015)

Katako, A., Shelton, P., Goertzen, A.L., Levin, D., Bybel, B., Aljuaid, M., Yoon, H.J., Kang, D.Y., Kim, S.M., Lee, C.S., et al : Machine learning identified an alzheimer’s disease-related fdg-pet pattern which is also expressed in lewy body dementia and parkinson’s disease dementia. Scientific reports 8 (1), 1–13 (2018)

Gray, K.R., Aljabar, P., Heckemann, R.A., Hammers, A., Rueckert, D.: Random forest-based manifold learning for classification of imaging data in dementia. In: International Workshop on Machine Learning in Medical Imaging, pp. 159–166 (2011). Springer

De Bruijne, M.: Machine learning approaches in medical image analysis: From detection to diagnosis. Elsevier (2016)

Tong, T., Wolz, R., Gao, Q., Hajnal, J.V., Rueckert, D.: Multiple instance learning for classification of dementia in brain mri. In: International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 599–606 (2013). Springer

Akhila, J., Markose, C., Aneesh, R.: Feature extraction and classification of dementia with neural network. In: 2017 International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT), pp. 1446–1450 (2017). IEEE

Chen, Y., Pham, T.D.: Development of a brain mri-based hidden markov model for dementia recognition. Biomedical engineering online 12 (1), 1–16 (2013)

Patil, M., Yardi, A.: Ann based dementia diagnosis using dct for brain mr image compression. In: 2013 International Conference on Communication and Signal Processing, pp. 451–454 (2013). IEEE

Gulhare, K.K., Shukla, S., Sharma, L.: Deep neural network classification method to alzheimer’s disease detection. International Journals of Advanced Research in Computer Science and Software Engineering 7 (6), 1–4 (2017)

Chiu, P.-Y., Tang, H., Wei, C.-Y., Zhang, C., Hung, G.-U., Zhou, W.: Nmd-12: A new machine-learning derived screening instrument to detect mild cognitive impairment and dementia. PloS one 14 (3), 0213430 (2019)

Stamate, D., Alghamdi, W., Ogg, J., Hoile, R., Murtagh, F.: A machine learning framework for predicting dementia and mild cognitive impairment. In: 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), pp. 671–678 (2018). IEEE

Visser, P.J., Lovestone, S., Legido-Quigley, C.: A metabolite-based machine learning approach to diagnose alzheimer-type dementia in blood: Results from the european medical information framework for alzheimer disease biomarker discovery cohort (2019)

Bansal, D., Chhikara, R., Khanna, K., Gupta, P.: Comparative analysis of various machine learning algorithms for detecting dementia. Procedia computer science 132 , 1497–1502 (2018)

Nori, V.S., Hane, C.A., Martin, D.C., Kravetz, A.D., Sanghavi, D.M.: Identifying incident dementia by applying machine learning to a very large administrative claims dataset. PLoS One 14 (7), 0203246 (2019)

Alam, M.A.U., Roy, N., Holmes, S., Gangopadhyay, A., Galik, E.: Automated functional and behavioral health assessment of older adults with dementia. In: 2016 IEEE First International Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE), pp. 140–149 (2016). IEEE

Gurevich, P., Stuke, H., Kastrup, A., Stuke, H., Hildebrandt, H.: Neuropsychological testing and machine learning distinguish alzheimer’s disease from other causes for cognitive impairment. Frontiers in aging neuroscience 9 , 114 (2017)

Karaglani, M., Gourlia, K., Tsamardinos, I., Chatzaki, E.: Accurate blood-based diagnostic biosignatures for alzheimer’s disease via automated machine learning. Journal of clinical medicine 9 (9), 3016 (2020)

Ryzhikova, E., Ralbovsky, N.M., Sikirzhytski, V., Kazakov, O., Halamkova, L., Quinn, J., Zimmerman, E.A., Lednev, I.K.: Raman spectroscopy and machine learning for biomedical applications: Alzheimer’s disease diagnosis based on the analysis of cerebrospinal fluid. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 248 , 119188 (2021)

Cho, P.-C., Chen, W.-H.: A double layer dementia diagnosis system using machine learning techniques. In: International Conference on Engineering Applications of Neural Networks, pp. 402–412 (2012). Springer

Facal, D., Valladares-Rodriguez, S., Lojo-Seoane, C., Pereiro, A.X., Anido-Rifon, L., Juncos-Rabadán, O.: Machine learning approaches to studying the role of cognitive reserve in conversion from mild cognitive impairment to dementia. International journal of geriatric psychiatry 34 (7), 941–949 (2019)

Jin, H., Chien, S., Meijer, E., Khobragade, P., Lee, J., et al : Learning from clinical consensus diagnosis in india to facilitate automatic classification of dementia: Machine learning study. JMIR Mental Health 8 (5), 27113 (2021)

James, C., Ranson, J.M., Everson, R., Llewellyn, D.J.: Performance of machine learning algorithms for predicting progression to dementia in memory clinic patients. JAMA network open 4 (12), 2136553–2136553 (2021)

Bougea, A., Efthymiopoulou, E., Spanou, I., Zikos, P.: A novel machine learning algorithm predicts dementia with lewy bodies versus parkinson’s disease dementia based on clinical and neuropsychological scores. Journal of geriatric psychiatry and neurology, 0891988721993556 (2021)

Salem, F.A., Chaaya, M., Ghannam, H., Al Feel, R.E., El Asmar, K.: Regression based machine learning model for dementia diagnosis in a community setting. Alzheimer’s & Dementia 17 , 053839 (2021)

Garcia-Gutierrez, F., Delgado-Alvarez, A., Delgado-Alonso, C., Díaz-Álvarez, J., Pytel, V., Valles-Salgado, M., Gil, M.J., Hernández-Lorenzo, L., Matías-Guiu, J., Ayala, J.L., et al.: Diagnosis of alzheimer’s disease and behavioural variant frontotemporal dementia with machine learning-aided neuropsychological assessment using feature engineering and genetic algorithms. International journal of geriatric psychiatry 37 (2) (2022)

Mirzaei, G., Adeli, H.: Machine learning techniques for diagnosis of alzheimer disease, mild cognitive disorder, and other types of dementia. Biomedical Signal Processing and Control 72 , 103293 (2022)

Hsiu, H., Lin, S.-K., Weng, W.-L., Hung, C.-M., Chang, C.-K., Lee, C.-C., Chen, C.-T.: Discrimination of the cognitive function of community subjects using the arterial pulse spectrum and machine-learning analysis. Sensors 22 (3), 806 (2022)

Shahzad, A., Dadlani, A., Lee, H., Kim, K.: Automated prescreening of mild cognitive impairment using shank-mounted inertial sensors based gait biomarkers. IEEE Access (2022)

Hane, C.A., Nori, V.S., Crown, W.H., Sanghavi, D.M., Bleicher, P.: Predicting onset of dementia using clinical notes and machine learning: case-control study. JMIR medical informatics 8 (6), 17819 (2020)

Aschwanden, D., Aichele, S., Ghisletta, P., Terracciano, A., Kliegel, M., Sutin, A.R., Brown, J., Allemand, M.: Predicting cognitive impairment and dementia: A machine learning approach. Journal of Alzheimer’s Disease 75 (3), 717–728 (2020)

Ryu, S.-E., Shin, D.-H., Chung, K.: Prediction model of dementia risk based on xgboost using derived variable extraction and hyper parameter optimization. IEEE Access 8 , 177708–177720 (2020)

de Langavant, L.C., Bayen, E., Yaffe, K., et al : Unsupervised machine learning to identify high likelihood of dementia in population-based surveys: development and validation study. Journal of medical Internet research 20 (7), 10493 (2018)

Fouladvand, S., Mielke, M.M., Vassilaki, M., Sauver, J.S., Petersen, R.C., Sohn, S.: Deep learning prediction of mild cognitive impairment using electronic health records. In: 2019 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 799–806 (2019). IEEE

Balea-Fernandez, F.J., Martinez-Vega, B., Ortega, S., Fabelo, H., Leon, R., Callico, G.M., Bibao-Sieyro, C.: Analysis of risk factors in dementia through machine learning. Journal of Alzheimer’s Disease 79 (2), 845–861 (2021)

Chlasta, K., Wołk, K.: Towards computer-based automated screening of dementia through spontaneous speech. Frontiers in Psychology, 4091 (2021)

Chien, Y.-W., Hong, S.-Y., Cheah, W.-T., Yao, L.-H., Chang, Y.-L., Fu, L.-C.: An automatic assessment system for alzheimer’s disease based on speech using feature sequence generator and recurrent neural network. Scientific Reports 9 (1), 1–10 (2019)

Shimoda, A., Li, Y., Hayashi, H., Kondo, N.: Dementia risks identified by vocal features via telephone conversations: A novel machine learning prediction model. PloS one 16 (7), 0253988 (2021)

Nishikawa, K., Akihiro, K., Hirakawa, R., Kawano, H., Nakatoh, Y.: Machine learning model for discrimination of mild dementia patients using acoustic features. Cognitive Robotics (2021)

Liu, L., Zhao, S., Chen, H., Wang, A.: A new machine learning method for identifying alzheimer’s disease. Simulation Modelling Practice and Theory 99 , 102023 (2020)

Searle, T., Ibrahim, Z., Dobson, R.: Comparing natural language processing techniques for alzheimer’s dementia prediction in spontaneous speech. arXiv preprint arXiv:2006.07358 (2020)

Zhu, Y., Tran, B., Liang, X., Batsis, J.A., Roth, R.M.: Towards interpretability of speech pause in dementia detection using adversarial learning. arXiv preprint arXiv:2111.07454 (2021)

Ossewaarde, R., Jonkers, R., Jalvingh, F., Bastiaanse, R.: Classification of spontaneous speech of individuals with dementia based on automatic prosody analysis using support vector machines (svm). In: The Thirty-Second International Flairs Conference (2019)

Xue, C., Karjadi, C., Paschalidis, I.C., Au, R., Kolachalama, V.B.: Detection of dementia on voice recordings using deep learning: a framingham heart study. Alzheimer’s research & therapy 13 (1), 1–15 (2021)

Weiner, J., Engelbart, M., Schultz, T.: Manual and automatic transcriptions in dementia detection from speech. In: INTERSPEECH, pp. 3117–3121 (2017)

Sadeghian, R., Schaffer, J.D., Zahorian, S.A.: Speech processing approach for diagnosing dementia in an early stage (2017)

Khodabakhsh, A., Yesil, F., Guner, E., Demiroglu, C.: Evaluation of linguistic and prosodic features for detection of alzheimer’s disease in turkish conversational speech. EURASIP Journal on Audio, Speech, and Music Processing 2015 (1), 1–15 (2015)

Edwards, E., Dognin, C., Bollepalli, B., Singh, M.K., Analytics, V.: Multiscale system for alzheimer’s dementia recognition through spontaneous speech. In: INTERSPEECH, pp. 2197–2201 (2020)

Kumar, Y., Maheshwari, P., Joshi, S., Baths, V.: Ml-based analysis to identify speech features relevant in predicting alzheimer’s disease. arXiv preprint arXiv:2110.13023 (2021)

Luz, S., de la Fuente, S., Albert, P.: A method for analysis of patient speech in dialogue for dementia detection. arXiv preprint arXiv:1811.09919 (2018)

Syed, Z.S., Syed, M.S.S., Lech, M., Pirogova, E.: Automated recognition of alzheimer’s dementia using bag-of-deep-features and model ensembling. IEEE Access 9 , 88377–88390 (2021)

Orsulic-Jeras, S., Sanders, B., Powers, S., Ejaz, F., Cordell, A., Wilk, C.: Developing a reminiscence therapy platform-lifebio memorytm-with a novel machine-learning-based application that transfers speech to text and generates life stories for nursing home residents with dementia. Alzheimer’s & Dementia 17 , 052281 (2021)

Sarawgi, U., Zulfikar, W., Soliman, N., Maes, P.: Multimodal inductive transfer learning for detection of alzheimer’s dementia and its severity. arXiv preprint arXiv:2009.00700 (2020)

Calzà, L., Gagliardi, G., Favretti, R.R., Tamburini, F.: Linguistic features and automatic classifiers for identifying mild cognitive impairment and dementia. Computer Speech & Language 65 , 101113 (2021)

Haider, F., De La Fuente, S., Luz, S.: An assessment of paralinguistic acoustic features for detection of alzheimer’s dementia in spontaneous speech. IEEE Journal of Selected Topics in Signal Processing 14 (2), 272–281 (2019)

Lopez-de-Ipiña, K., Alonso, J.B., Solé-Casals, J., Barroso, N., Henriquez, P., Faundez-Zanuy, M., Travieso, C.M., Ecay-Torres, M., Martinez-Lage, P., Eguiraun, H.: On automatic diagnosis of alzheimer’s disease based on spontaneous speech analysis and emotional temperature. Cognitive Computation 7 (1), 44–55 (2015)

Orimaye, S.O., Wong, J.S.-M., Golden, K.J.: Learning predictive linguistic features for alzheimer’s disease and related dementias using verbal utterances. In: Proceedings of the Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality, pp. 78–87 (2014)

Santander-Cruz, Y., Salazar-Colores, S., Paredes-García, W.J., Guendulain-Arenas, H., Tovar-Arriaga, S.: Semantic feature extraction using sbert for dementia detection. Brain Sciences 12 (2), 270 (2022)

Download references

Acknowledgements

The first author’s learning process was supported by the National E-Infrastructure for Aging Research (NEAR), Sweden. NEAR is working on improving the health condition of older adults in Sweden.

Open access funding provided by Blekinge Institute of Technology. This research received no external funding.

Author information

Ashir Javeed and Ana Luiza Dallora contributed equally to this work.

Authors and Affiliations

Aging Research Center, Karolinska Institutet, Tomtebodavagen, Stockholm, 17165, Solna, Sweden

Ashir Javeed

Department of Health, Blekinge Institute of Technology, Valhallavägen 1, Karlskrona, 37141, Blekinge, Sweden

Ashir Javeed, Ana Luiza Dallora, Johan Sanmartin Berglund & Peter Anderberg

Department of Computer Science, University of Science and Technology Bannu, Township, Bannu, 28100, Khyber-Pakhtunkhwa, Pakistan

Department of Electrical Engineering, University of Science and Technology Bannu, Township, Bannu, 28100, Khyber-Pakhtunkhwa, Pakistan

Liaqata Ali

School of Health Sciences, University of Skovde, Högskolevägen 1, Skövde, SE-541 28, Skövde, Sweden

Peter Anderberg

You can also search for this author in PubMed   Google Scholar

Contributions

Conceptualization by Peter Anderber, Data curation by Liaqat ALi, Formal analysis by Ana Luiza Dallora, Write up and Methodology by Ashir Javeed, Proofread by Arif Ali, Supervised by Johan Sanmartin Berglund. If any of the sections are not relevant to your manuscript.

Corresponding author

Correspondence to Johan Sanmartin Berglund .

Ethics declarations

Ethics approval.

This study was carried out in accordance with the Declaration of Helsinki and was approved by the Research Ethics Committee at Blekinge Institute of Technology (BTH).

Consent to Participate

Consent for publication, conflict of interest.

The authors declare no conflict of interest.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Javeed, A., Dallora, A.L., Berglund, J.S. et al. Machine Learning for Dementia Prediction: A Systematic Review and Future Research Directions. J Med Syst 47 , 17 (2023). https://doi.org/10.1007/s10916-023-01906-7

Download citation

Received : 16 July 2022

Accepted : 03 January 2023

Published : 01 February 2023

DOI : https://doi.org/10.1007/s10916-023-01906-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Dementia prediction
• Feature selection
• Machine learning
• Deep learning
• Find a journal
• Publish with us
• Track your research

• Introduction
• Conclusions
• Article Information

Screenshots of the smartphone cognitive tasks developed by Datacubed Health and included in the ALLFTD Mobile App. Details about the task design and instructions are included in the eMethods in Supplement 1. A, Flanker (Ducks in a Pond) is a task of cognitive control requiring participants to select the direction of the center duck. B, Go/no-go (Go Sushi Go!) requires participants to quickly tap on pieces of sushi (go) but not to tap when they see a fish skeleton (no-go). C, Card sort (Card Shuffle) is a task of cognitive flexibility requiring participants to learn rules that change during the task. D, The adaptative, associative memory task (Humi’s Bistro) requires participants to learn the food orders of several restaurant tables. E, Stroop (Color Clash) is a cognitive inhibition paradigm requiring participants to inhibit their tendency to read words and instead respond based on the color of the word. F, The 2-back task (Animal Parade) requires participants to determine whether animals on a parade float match the animals they saw 2 stimuli previously. G, Participants are asked to complete 3 testing sessions over 2 weeks. Shown in dark blue, they have 3 days to complete each testing session with a washout day between sessions on which no tests are available. Session 2 always begins on day 5 and session 3 on day 9. Screenshots are provided with permission from Datacubed Health.

Forest plots present internal consistency and test-retest reliability results in the discovery and validation cohorts, as well as an estimate in a combined sample of discovery and validation participants. ICC indicates interclass correlation coefficient.

A and B, Correlation matrices display associations of in-clinic criterion standard measures and ALLFTD mobile App (mApp) test scores in discovery and validation cohorts. Below the horizontal dashed lines, the associations among app tests and between app tests and demographic characteristics convergent clinical measures, divergent cognitive tests, and neuroimaging regions of interest can be viewed. Most app tests show strong correlations with each other and with age, convergent clinical measures, and brain volume. The measures show weaker correlations with divergent measures of visuospatial (Benson Figure Copy) and language (Multilingual Naming Test [MINT]) abilities. The strength of convergent correlations between app measures and outcomes is similar to the correlations between criterion standard neuropsychological scores and these outcomes, which can be viewed by looking across the rows above the horizontal black line. C and D, In the discovery and validation cohorts, receiver operating characteristics curves were calculated to determine how well a composite of app tests, the Uniform Data Set, version 3.0, Executive Functioning Composite (UDS3-EF), and the Montreal Cognitive Assessment (MoCA) discriminate individuals without symptoms (Clinical Dementia Rating Scale plus National Alzheimer’s Coordinating Center FTLD module sum of boxes [CDR plus NACC-FTLD-SB] score = 0) from individuals with the mildest symptoms of FTLD (CDR plus NACC-FTLD-SB score = 0.5). AUC indicates area under the curve; CVLT, California Verbal Learning Test.

eMethods. Instruments and Statistical Analysis

eResults. Participants

eTable 1. Participant Characteristics and Test Scores in Original and Validation Cohorts

eTable 2. Comparison of Diagnostic Accuracy for ALLFTD Mobile App Composite Score Across Cohorts

eTable 3. Number of Distractions Reported During the Remote Smartphone Testing Sessions

eTable 4. Qualitative Description of the Distractions Reported During Remote Testing Sessions

eFigure 1. Scatterplots of Test-Retest Reliability in a Mixed Sample of Adults Without Functional Impairment and Participants With FTLD

eFigure 2. Comparison of Test-Retest Reliability Estimates by Endorsement of Distractions

eFigure 3. Comparison of Test-Retest Reliability Estimates by Operating System

eFigure 4. Correlation Matrix in the Combined Cohort

eFigure 5. Neural Correlates of Smartphone Cognitive Test Performance

eReferences

Nonauthor Collaborators

Data Sharing Statement

See More About

Sign up for emails based on your interests, select your interests.

Customize your JAMA Network experience by selecting one or more topics from the list below.

• Academic Medicine
• Acid Base, Electrolytes, Fluids
• Allergy and Clinical Immunology
• American Indian or Alaska Natives
• Anesthesiology
• Anticoagulation
• Art and Images in Psychiatry
• Artificial Intelligence
• Assisted Reproduction
• Bleeding and Transfusion
• Caring for the Critically Ill Patient
• Challenges in Clinical Electrocardiography
• Climate and Health
• Climate Change
• Clinical Challenge
• Clinical Decision Support
• Clinical Implications of Basic Neuroscience
• Clinical Pharmacy and Pharmacology
• Complementary and Alternative Medicine
• Consensus Statements
• Coronavirus (COVID-19)
• Critical Care Medicine
• Cultural Competency
• Dental Medicine
• Dermatology
• Diabetes and Endocrinology
• Diagnostic Test Interpretation
• Drug Development
• Electronic Health Records
• Emergency Medicine
• End of Life, Hospice, Palliative Care
• Environmental Health
• Equity, Diversity, and Inclusion
• Facial Plastic Surgery
• Gastroenterology and Hepatology
• Genetics and Genomics
• Genomics and Precision Health
• Global Health
• Guide to Statistics and Methods
• Hair Disorders
• Health Care Delivery Models
• Health Care Economics, Insurance, Payment
• Health Care Quality
• Health Care Reform
• Health Care Safety
• Health Care Workforce
• Health Disparities
• Health Inequities
• Health Policy
• Health Systems Science
• History of Medicine
• Hypertension
• Images in Neurology
• Implementation Science
• Infectious Diseases
• Innovations in Health Care Delivery
• JAMA Infographic
• Law and Medicine
• Leading Change
• Less is More
• LGBTQIA Medicine
• Lifestyle Behaviors
• Medical Coding
• Medical Devices and Equipment
• Medical Education
• Medical Education and Training
• Medical Journals and Publishing
• Mobile Health and Telemedicine
• Narrative Medicine
• Neuroscience and Psychiatry
• Notable Notes
• Nutrition, Obesity, Exercise
• Obstetrics and Gynecology
• Occupational Health
• Ophthalmology
• Orthopedics
• Otolaryngology
• Pain Medicine
• Palliative Care
• Pathology and Laboratory Medicine
• Patient Care
• Patient Information
• Performance Improvement
• Performance Measures
• Perioperative Care and Consultation
• Pharmacoeconomics
• Pharmacoepidemiology
• Pharmacogenetics
• Pharmacy and Clinical Pharmacology
• Physical Medicine and Rehabilitation
• Physical Therapy
• Physician Leadership
• Population Health
• Primary Care
• Professional Well-being
• Professionalism
• Psychiatry and Behavioral Health
• Public Health
• Pulmonary Medicine
• Regulatory Agencies
• Reproductive Health
• Research, Methods, Statistics
• Resuscitation
• Rheumatology
• Risk Management
• Scientific Discovery and the Future of Medicine
• Shared Decision Making and Communication
• Sleep Medicine
• Sports Medicine
• Stem Cell Transplantation
• Substance Use and Addiction Medicine
• Surgical Innovation
• Surgical Pearls
• Teachable Moment
• Technology and Finance
• The Art of JAMA
• The Arts and Medicine
• The Rational Clinical Examination
• Tobacco and e-Cigarettes
• Translational Medicine
• Trauma and Injury
• Treatment Adherence
• Ultrasonography
• Users' Guide to the Medical Literature
• Vaccination
• Venous Thromboembolism
• Veterans Health
• Women's Health
• Workflow and Process
• Wound Care, Infection, Healing

Get the latest research based on your areas of interest.

Others also liked.

• Download PDF
• X Facebook More LinkedIn

Staffaroni AM , Clark AL , Taylor JC, et al. Reliability and Validity of Smartphone Cognitive Testing for Frontotemporal Lobar Degeneration. JAMA Netw Open. 2024;7(4):e244266. doi:10.1001/jamanetworkopen.2024.4266

Manage citations:

© 2024

• Permissions

Reliability and Validity of Smartphone Cognitive Testing for Frontotemporal Lobar Degeneration

• 1 Department of Neurology, Memory and Aging Center, Weill Institute for Neurosciences, University of California, San Francisco
• 2 Department of Neurology, Columbia University, New York, New York
• 3 Department of Neurology, Mayo Clinic, Rochester, Minnesota
• 4 Department of Quantitative Health Sciences, Division of Biomedical Statistics and Informatics, Mayo Clinic, Rochester, Minnesota
• 5 Department of Neurology, Case Western Reserve University, Cleveland, Ohio
• 6 Department of Neurosciences, University of California, San Diego, La Jolla
• 7 Department of Radiology, University of North Carolina, Chapel Hill
• 8 Department of Neurology, Indiana University, Indianapolis
• 9 Department of Neurology, Vanderbilt University, Nashville, Tennessee
• 10 Department of Neurology, University of Washington, Seattle
• 11 Department of Psychiatry and Psychology, Mayo Clinic, Rochester, Minnesota
• 12 Department of Neurology, Institute for Precision Health, University of California, Los Angeles
• 13 Department of Neurology, Knight Alzheimer Disease Research Center, Washington University, Saint Louis, Missouri
• 14 Department of Psychiatry, Knight Alzheimer Disease Research Center, Washington University, Saint Louis, Missouri
• 15 Department of Neuroscience, Mayo Clinic, Jacksonville, Florida
• 16 Department of Neurology, University of Pennsylvania Perelman School of Medicine, Philadelphia
• 17 Division of Neurology, University of British Columbia, Musqueam, Squamish & Tsleil-Waututh Traditional Territory, Vancouver, Canada
• 18 Department of Neurosciences, University of California, San Diego, La Jolla
• 19 Department of Neurology, Nantz National Alzheimer Center, Houston Methodist and Weill Cornell Medicine, Houston Methodist, Houston, Texas
• 20 Department of Neurology, UCLA (University of California, Los Angeles)
• 21 Department of Neurology, University of Colorado, Aurora
• 22 Department of Neurology, David Geffen School of Medicine, UCLA
• 23 Department of Neurology, University of Alabama, Birmingham
• 24 Tanz Centre for Research in Neurodegenerative Diseases, Division of Neurology, University of Toronto, Toronto, Ontario, Canada
• 25 Department of Neurology, Massachusetts General Hospital and Harvard Medical School, Boston
• 26 Department of Epidemiology and Biostatistics, University of California, San Francisco
• 27 Department of Psychological & Brain Sciences, Washington University, Saint Louis, Missouri

Question   Can remote cognitive testing via smartphones yield reliable and valid data for frontotemporal lobar degeneration (FTLD)?

Findings   In this cohort study of 360 patients, remotely deployed smartphone cognitive tests showed moderate to excellent reliability comparedwith criterion standard measures (in-person disease severity assessments and neuropsychological tests) and brain volumes. Smartphone tests accurately detected dementia and were more sensitive to the earliest stages of familial FTLD than standard neuropsychological tests.

Meaning   These findings suggest that remotely deployed smartphone-based assessments may be reliable and valid tools for evaluating FTLD and may enhance early detection, supporting the inclusion of digital assessments in clinical trials for neurodegeneration.

Importance   Frontotemporal lobar degeneration (FTLD) is relatively rare, behavioral and motor symptoms increase travel burden, and standard neuropsychological tests are not sensitive to early-stage disease. Remote smartphone-based cognitive assessments could mitigate these barriers to trial recruitment and success, but no such tools are validated for FTLD.

Objective   To evaluate the reliability and validity of smartphone-based cognitive measures for remote FTLD evaluations.

Design, Setting, and Participants   In this cohort study conducted from January 10, 2019, to July 31, 2023, controls and participants with FTLD performed smartphone application (app)–based executive functioning tasks and an associative memory task 3 times over 2 weeks. Observational research participants were enrolled through 18 centers of a North American FTLD research consortium (ALLFTD) and were asked to complete the tests remotely using their own smartphones. Of 1163 eligible individuals (enrolled in parent studies), 360 were enrolled in the present study; 364 refused and 439 were excluded. Participants were divided into discovery (n = 258) and validation (n = 102) cohorts. Among 329 participants with data available on disease stage, 195 were asymptomatic or had preclinical FTLD (59.3%), 66 had prodromal FTLD (20.1%), and 68 had symptomatic FTLD (20.7%) with a range of clinical syndromes.

Exposure   Participants completed standard in-clinic measures and remotely administered ALLFTD mobile app (app) smartphone tests.

Main Outcomes and Measures   Internal consistency, test-retest reliability, association of smartphone tests with criterion standard clinical measures, and diagnostic accuracy.

Results   In the 360 participants (mean [SD] age, 54.0 [15.4] years; 209 [58.1%] women), smartphone tests showed moderate-to-excellent reliability (intraclass correlation coefficients, 0.77-0.95). Validity was supported by association of smartphones tests with disease severity ( r range, 0.38-0.59), criterion-standard neuropsychological tests ( r range, 0.40-0.66), and brain volume (standardized β range, 0.34-0.50). Smartphone tests accurately differentiated individuals with dementia from controls (area under the curve [AUC], 0.93 [95% CI, 0.90-0.96]) and were more sensitive to early symptoms (AUC, 0.82 [95% CI, 0.76-0.88]) than the Montreal Cognitive Assessment (AUC, 0.68 [95% CI, 0.59-0.78]) ( z of comparison, −2.49 [95% CI, −0.19 to −0.02]; P  = .01). Reliability and validity findings were highly similar in the discovery and validation cohorts. Preclinical participants who carried pathogenic variants performed significantly worse than noncarrier family controls on 3 app tasks (eg, 2-back β = −0.49 [95% CI, −0.72 to −0.25]; P  < .001) but not a composite of traditional neuropsychological measures (β = −0.14 [95% CI, −0.42 to 0.14]; P  = .32).

Conclusions and Relevance   The findings of this cohort study suggest that smartphones could offer a feasible, reliable, valid, and scalable solution for remote evaluations of FTLD and may improve early detection. Smartphone assessments should be considered as a complementary approach to traditional in-person trial designs. Future research should validate these results in diverse populations and evaluate the utility of these tests for longitudinal monitoring.

Frontotemporal lobar degeneration (FTLD) is a neurodegenerative pathology causing early-onset dementia syndromes with impaired behavior, cognition, language, and/or motor functioning. 1 Although over 30 FTLD trials are planned or in progress, there are several barriers to conducting FTLD trials. Clinical trials for neurodegenerative disease are expensive, 2 and frequent in-person trial visits are burdensome for patients, caregivers, and clinicians, 3 a concern magnified in FTLD by behavioral and motor impairments. Given the rarity and geographical dispersion of eligible participants, FTLD trials require global recruitment, 4 particularly for those that are far from expert FTLD clinical trial centers. Furthermore, criterion standard neuropsychological tests are not adequately sensitive until symptoms are already noticeable to families, limiting their usefulness as outcomes in early-stage FTLD treatment trials. 4

Reliable, valid, and scalable remote data collection methods may help surmount these barriers to FTLD clinical trials. Smartphones are garnering interest across neurological conditions as a method for administering remote cognitive and motor evaluations. Preliminary evidence supports the feasibility, reliability, and/or validity of unsupervised smartphone cognitive and motor testing in older adults at risk for Alzheimer disease, 5 - 8 Parkinson disease, 9 and Huntington disease. 10 The clinical heterogeneity of FTLD necessitates a uniquely comprehensive smartphone battery. In the ALLFTD Consortium (Advancing Research and Treatment in Frontotemporal Lobar Degeneration [ARTFLD] and Longitudinal Evaluation of Familial Frontotemporal Dementia Subjects [LEFFTDS]), the ALLFTD mobile Application (ALLFTD-mApp) was designed to remotely monitor cognitive, behavioral, language, and motor functioning in FTLD research. Taylor et al 11 recently reported that unsupervised ALLFTD-mApp data collection through a multicenter North American FTLD research network was feasible and acceptable to participants. Herein, we extend that work by investigating the reliability and validity of unsupervised remote smartphone tests of executive functioning and memory in a cohort with FTLD that has undergone extensive phenotyping.

Participants were enrolled from ongoing FTLD studies requiring in-person assessment, including participants from 18 centers from the ALLFTD study study 12 and University of California, San Francisco (UCSF) FTLD studies. To study the app in older individuals, a small group of older adults without functional impairment was recruited from the UCSF Brain Aging Network for Cognitive Health. All study procedures were approved by the UCSF or Johns Hopkins Central Institutional Review Board. All participants or legally authorized representatives provided written informed consent. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology ( STROBE ) reporting guideline.

Inclusion criteria were age 18 years or older, having access to a smartphone, and reporting English as the primary language. Race and ethnicity were self reported by participants using options consistent with the National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS) and were collected to contextualize the generalizability of these results. Participants were asked to complete tests on their own smartphones. Informants were encouraged for all participants and required for those with symptomatic FTLD (Clinical Dementia Rating Scale plus NACC FTLD module [CDR plus NACC-FTLD] global score ≥1). Recruitment targeted individuals with CDR plus NACC-FTLD global scores less than 2, but sites had discretion to enroll more severely impaired participants. Exclusion criteria were consistent with the parent ALLFTD study. 12

Participants were enrolled in the ALLFTD-mApp study within 90 days of annual ALLFTD study visits (including neuropsychological and neuroimaging data collection). Site research coordinators (including J.C.T., A.B.W., S.D., and M.M.) assisted participants with app download, setup, and orientation and observed participants completing the first questionnaire. All cognitive tasks were self-administered without supervision (except pilot participants, discussed below) in a predefined order with minor adjustments throughout the study. Study partners of participants with symptomatic FTLD were asked to remain nearby during participation to help navigate the ALLFTD-mApp but were asked not to assist with testing.

The baseline participation window was divided into three 25- to 35-minute assessment sessions occurring over 11 days. All cognitive tests were repeated in every session to enhance task reliability 6 , 13 and enable assessment of test-retest reliability, except for card sort, which was administered once every 6 months due to expected practice effects. Adherence was defined as the percentage of all available tasks that were completed. Participants were asked to complete the triplicate of sessions every 6 months for the duration of the app study. Only the baseline triplicate was analyzed in this study.

Replicability was tested by dividing the sample into a discovery cohort (n = 258) comprising all participants enrolled until the initial data freeze (October 1, 2022) and a validation cohort (n = 102) comprising participants enrolled after October 1, 2022, and 18 pilot participants 11 who completed the first session in person with an examiner present during cognitive pretesting. Sensitivity analyses excluded this small pilot cohort.

ALLFTD investigators partnered with Datacubed Health 14 to develop the ALLFTD-mApp on Datacubed Health’s Linkt platform. The app includes cognitive, motor, and speech tasks. This study focuses on 6 cognitive tests developed by Datacubed Health 11 comprising an adaptive associative memory task (Humi’s Bistro) and gamified versions of classic executive functioning paradigms: flanker (Ducks in a Pond), Stroop (Color Clash), 2-back (Animal Parade), go/no-go (Go Sushi Go!), and card sort (Card Shuffle) ( Figure 1 and eMethods in Supplement 1 ). Most participants with symptomatic FTLD (49 [72.1%]) were not administered Stroop or 2-back, as pilot studies identified these as too difficult. 11 The app test results were summarized as a composite score (eMethods in Supplement 1 ). Participants completed surveys to assess technological familiarity (daily or less than daily use of a smartphone) and distractions (present or absent).

Criterion standard clinical data were collected during parent project visits. Syndromic diagnoses were made according to published criteria 15 - 19 based on multidisciplinary conferences that considered neurological history, neurological examination results, and collateral interview. 20

The CDR plus NACC-FTLD module is an 8-domain rating scale based on informant and participant report. 21 A global score was calculated to categorize disease severity as asymptomatic or preclinical if a pathogenic variant carrier (0), prodromal (0.5), or symptomatic (1.0-3.0). 22 A sum of the 8 domain box scores (CDR plus NACC-FTLD sum of boxes) was also calculated. 22

Participants completed the UDS Neuropsychological Battery, version 3.0 23 (eMethods in Supplement 1 ), which includes traditional neuropsychological measures and the Montreal Cognitive Assessment (MoCA), a global cognitive screen. Executive functioning and processing speed measures were summarized into a composite score (UDS3-EF). 24 Participants also completed a 9-item list-learning memory test (California Verbal Learning Test, 2nd edition, Short Form). 25 Most (339 [94.2%]) neuropsychological evaluations were conducted in person. In a subsample (n = 270), motor speed and dexterity were assessed using the Movement Disorder Society Uniform Parkinson Disease Rating Scale 26 Finger Tapping subscale (0 indicates no deficits [n = 240]).

We acquired T1-weighted brain magnetic resonance imaging for 199 participants. Details of image acquisition, harmonization, preprocessing, and processing are provided in eMethods in Supplement 1 and prior publications. 27 Briefly, SPM12 (Statistical Parametric Mapping) was used for segmentation 28 and Large Deformation Diffeomorphic Metric Mapping for generating group templates. 29 Gray matter volumes were calculated in template space by integrating voxels and dividing by total intracranial volume in 2 regions of interest (ROIs) 30 : a frontoparietal and subcortical ROI and a hippocampal ROI. Voxel-based morphometry was used to test unbiased voxel-wise associations of volume with smartphone tests (eMethods in Supplement 1 ). 31 , 32

Participants in the ALLFTD study underwent genetic testing 33 at the University of California, Los Angeles. DNA samples were screened using targeted sequencing of a custom panel of genes previously implicated in neurodegenerative diseases, including GRN ( 138945 ) and MAPT ( 157140 ). Hexanucleotide repeat expansions in C9orf72 ( 614260 ) were detected using both fluorescent and repeat-primed polymerase chain reaction analysis. 34

Statistical analyses were conducted using Stata, version 17.0 (StataCorp LLC), and R, version 4.4.2 (R Project for Statistical Computing). All tests were 2 sided, with a statistical significance threshold of P < .05.

Psychometric properties of the smartphone tests were explored using descriptive statistics. Comparisons between CDR plus NACC-FTLD groups (ie, asymptomatic or preclinical, prodromal, and symptomatic) for continuous variables, including demographic characteristics and cognitive task scores (first exposure to each measure), were analyzed by fitting linear regressions. We used χ 2 difference tests for frequency data (eg, sex and race and ethnicity).

Internal consistency, which measures reliability within a task, was estimated for participants’ first exposure to each test using Cronbach α (details in eMethods in Supplement 1 ). Test-retest reliability was estimated using intraclass correlation coefficients for participants who completed a task at least twice; all exposures were included. Reliability estimates are described as poor (<0.500), moderate (0.500-0.749), good (0.750-0.890), and excellent (≥0.900) 35 ; these are reporting rules of thumb, and clinical interpretation should consider raw estimates. We calculated 95% CIs via bootstrapping with 1000 samples.

Validity analyses used participants’ first exposure to each test. Linear regressions were fitted in participants without symptoms with age, sex, and educational level as independent variables to understand the unique contribution of each demographic factor to cognitive test scores. Correlations and linear regression between the app-based tasks and disease severity (CDR plus NACC-FTLD sum of boxes score), neuropsychological test scores, and gray matter ROIs were used to investigate construct validity in the full sample. Demographic characteristics were not entered as covariates because the primary goal was to assess associations between app-based measures and criterion standards, rather than understand the incremental predictive value of app measures. To address potential motor confounds, associations with disease severity were evaluated in a subsample without finger dexterity deficits on motor examination (using the Movement Disorder Society Uniform Parkinson Disease Rating Scale Finger Tapping subscale). To complement ROI-based neuroimaging analysis based on a priori hypotheses, we conducted voxel-based morphometry (eMethods in Supplement 1 ) to uncover other potential neural correlates of test performance. 31 , 32 Finally, we evaluated the association of the number of distractions and operating system with reliability and validity, controlling for age and disease severity, which are predictive factors associated with test performance in correlation analyses.

To evaluate the app’s ability to select participants with prodromal or symptomatic FTLD for trial enrollment, we tested discrimination of participants without symptoms from those with prodromal and symptomatic FTLD. To understand the app’s utility for screening early cognitive impairment, we fit receiver operating characteristics curves testing the predictive value of the app composite, UDS3-EF, and MoCA for differentiating participants without symptoms and those with preclinical FTLD from those with prodromal FTLD; areas under the curves (AUC) for the app and MoCA were compared using the DeLong test in participants with results for both predictive factors.

We compared app performance in preclinical participants who carried pathogenic variants with that in noncarrier controls using linear regression adjusted for age (a predictive factor in earlier models). For this analysis, we excluded those younger than 45 years to remove participants likely to be years from symptom onset based on natural history studies. 4 We analyzed memory performance in participants who carried MAPT pathogenic variants, as early executive deficits may be less prominent. 34 , 36

Of 1163 eligible participants, 360 were enrolled, 439 were excluded, and 364 refused to participate (additional details are provided in the eResults in Supplement 1 ). Participant characteristics are reported in Table 1 for the full sample. The discovery and validation cohorts did not significantly differ in terms of demographic characteristics, disease severity, or cognition (eTable 1 in Supplement 1 ). In the full sample, there were 209 women (58.1%) and 151 men (41.9%), and the mean (SD) age was 54.0 (15.4) years (range, 18-89 years). The mean (SD) educational level was 16.5 (2.3) years (range, 12-20 years). Among the 358 participants with racial and ethnic data available, 340 (95.0%) identified as White. For the 18 participants self-identifying as being of other race or ethnicity, the specific group was not provided to protect participant anonymity. Among the 329 participants with available CDR plus NACC-FTLD scores ( Table 1 ), 195 (59.3%) were asymptomatic or preclinical (Global Score, 0), 66 (20.1%) were prodromal (Global score, 0.5), and 68 (20.7%) were symptomatic (global score, 1.0 or 2.0). Of those with available genetic testing results (n = 222), 100 (45.0%) carried a pathogenic familial FTLD pathogenic variant, including 63 of 120 participants without symptoms and with available results. On average, participants completed 78% of available smartphone measures over a mean (SD) of 2.6 (0.6) sessions.

Descriptive statistics for each task are presented in Table 2 . Ceiling effects were not observed for any tests. A small percentage of participants were at the floor for flanker (19 [5.3%]), go/no-go (13 [4.0%]), and card sort (9 [3.3%]) scores. Floor effects were only observed in participants with prodromal or symptomatic FTLD.

Except for go/no-go, internal consistency estimates ranged from good to excellent (Cronbach α range, 0.84 [95% CI, 0.81-0.87] to 0.99 [95% CI, 0.99-0.99]), and test-retest reliabilities were moderate to excellent (interclass correlation coefficient [ICC] range, 0.77 [95% CI, 0.69-0.83] to 0.95 [95% CI, 0.93-0.96]), with slightly higher estimates in participants with prodromal or symptomatic FTLD ( Table 2 , Figure 2 , and eFigure 1 in Supplement 1 ). Go/no-go reliability was particularly poor in participants without symptoms (ICC, 0.10 [95% CI, −0.37 to 0.48]) and was removed from subsequent validation analyses except the correlation matrix ( Figure 3 A and B). The 95% CIs for reliability estimates overlapped in the discovery and validation cohorts ( Figure 2 ). Reliability estimates showed overlapping 95% CIs regardless of distractions (eFigure 2 in Supplement 1 ) or operating systems (eFigure 3 in Supplement 1 ), with a pattern of slightly lower reliability estimates when distractions were endorsed for all comparisons except Stroop (Cronbach α).

In 57 participants without symptoms who did not carry pathogenic variants, older age was associated with worse performance on all measures (β range,  − 0.40 [95 CI, −0.68 to −0.13] to −0.78 [95 CI, −0.89 to −0.52]; P ≤ .03), except card sort (β = −0.22 [95% CI, −0.54 to 0.09]; P  = .16) and go-no/go (β = −0.15 [95% CI, −0.44 to 0.14]; P  = .31), though associations were in the expected direction. Associations with sex and educational level were not statistically significant.

Cognitive tests administered using the app showed evidence of convergent and divergent validity (eFigure 4 in Supplement 1 ), with very similar findings in discovery ( Figure 3 A) and validation cohorts ( Figure 3 B). App–based measures of executive functioning were generally correlated with criterion standard in-person measures of these domains and less with measures of other cognitive domains ( r range, 0.40-0.66). For example, the flanker task was associated with the UDS3-EF composite (β = 0.58 [95% CI, 0.48-0.68]; P  < .001) and measures of visuoconstruction (β for Benson Figure Copy, 0.43 [95% CI, 0.32-0.54]; P  = .01) and naming (β for Multilingual Naming Test, 0.25 [95% CI, 0.14-0.37]; P  < .001). The app memory test was associated with criterion standard memory and executive functioning tests.

Worse performance on all app measures was associated with greater disease severity on CDR plus NACC-FTLD ( r range, 0.38-0.59) ( Table 1 , Figure 3 , and eFigure 4 in Supplement 1 ). The same pattern of results was observed after excluding those with finger dexterity issues. Except for go/no-go, performance of participants with prodromal FTLD was statistically significantly worse than that of participants without symptoms on all measures ( P  < .001).

The AUC for the app composite to distinguish participants without symptoms from those with dementia was 0.93 (95% CI, 0.90-0.96). The app also accurately differentiated participants without symptoms from those with prodromal or symptomatic FTLD (AUC, 0.87 [95% CI, 0.84-0.92]). Compared with the MoCA (AUC, 0.68 [95% CI, 0.59-0.78), app composite performance (AUC, 0.82 [95% CI, 0.76-0.88]) more accurately differentiated participants without symptoms and with prodromal FTLD ( z of comparison, −2.49 [95% CI, −0.19 to −0.02]; P  = .01), with similar accuracy to the UDS3-EF (AUC, 0.81 [95% CI, 0.73-0.88]); highly similar results (eTable 2 in Supplement 1 ) were observed in the discovery ( Figure 3 C) and validation ( Figure 3 D) cohorts.

In 56 participants without symptoms who were older than 45 years, those carrying GRN , C9orf72 , or another rare pathogenic variants performed significantly worse on 3 of 4 executive tests compared with noncarrier controls, including flanker (β = −0.26 [95% CI, −0.46 to −0.05]; P  = .02), card sort (β = −0.28 [95% CI, −0.54 to −0.30]; P  = .03), and 2-back (β = −0.49 [95% CI, −0.72 to −0.25]; P  < .001). The estimated scores of participants who carried pathogenic variants were on average lower than those of carriers on a composite of criterion standard in-person tests, but the difference was not statistically significant (UDS3-EF β = −0.14 [95% CI, −0.42 to 0.14]; P  = .32). Participants who carried preclinical MAPT pathogenic variants scored higher than noncarriers on the app Memory test, though the difference was not statistically significant (β = 0.21 [95% CI, −0.50 to 0.58]; P  = .19).

In prespecified ROI analyses, worse app executive functioning scores were associated with lower frontoparietal and/or subcortical volume ( Figures 3 A and B) (β range, 0.34 [95% CI, 0.22-0.46] to 0.50 [95 CI, 0.40-0.60]; P < .001 for all) and worse memory scores with smaller hippocampal volume (β = 0.45 [95% CI, 0.34-0.56]; P  < .001). Voxel-based morphometry (eFigure 5 in Supplement 1 ) suggested worse app performance was associated with widespread atrophy, particularly in frontotemporal cortices.

Only for card sort were distractions (eTables 3 and 4 in Supplement 1 ) associated with task performance; those experiencing distractions unexpectedly performed better (β = 0.16 [95% CI, 0.05-0.28]; P  = .005). The iPhone operating system was associated with better performance on 2 speeded tasks: flanker (β = 0.16 [95% CI, 0.07-0.24]; P  < .001) and go/no-go (β = 0.16 [95% CI, 0.06-0.26]; P  = .002). In a sensitivity analysis, associations of all app tests with disease severity, UDS3-EF, and regional brain volumes remained after covarying for distractions and operating system, as did the models differentiating participants who carried preclinical pathogenic variants and noncarrier controls.

There is an urgent need to identify reliable and valid digital tools for remote neurobehavioral measurement in neurodegenerative diseases, including FTLD. Prior studies provided preliminary evidence that smartphones collect reliable and valid cognitive data in a variety of age-related and neurodegenerative illnesses. This is the first study, to our knowledge, to provide analogous support for the reliability and validity of remote cognitive testing via smartphones in FTLD and preliminary evidence that this approach improves early detection relative to traditional in-person measures.

Reliability, a prerequisite for a valid clinical trial end point, indicates measurements are consistent. In 2 cohorts, we found smartphone cognitive tests were reliable within a single administration (ie, internally consistent) and across repeated assessments (ie, test-retest reliability) with no apparent differences by operating system. For all measures except go/no-go, reliability estimates were moderate to excellent and on par with other remote digital assessments 5 , 6 , 10 , 37 , 38 and in-clinic criterion standards. 39 - 41 Go/no-go showed similar within- and between-person variability in participants without symptoms (ie, poor reliability), and participant feedback suggested instructions were confusing and the stimuli disappeared too quickly. Those endorsing distractions tended to have lower reliability, though 95% CIs largely overlapped; future research detailing the effect of the home environment on test performance is warranted.

Construct validity was supported by strong associations of smartphone tests with demographics, disease severity, neuroimaging, and criterion standard neuropsychological measures that replicated in a validation sample. These associations were similar to those observed among the criterion standard measures and similar to associations reported in other validation studies of smartphone cognitive tests. 5 , 6 , 10 Associations with disease severity were not explained by motor impairments. The iPhone operating system was associated with better performance on 2 time-based measures, consistent with prior findings. 6

A composite of brief smartphone tests was accurate in distinguishing dementia from cognitively unimpaired participants, screening out participants without symptoms, and detecting prodromal FTLD with greater sensitivity than the MoCA. Moreover, carriers of preclinical C9orf72 and GRN pathogenic variants performed significantly worse than noncarrier controls on 3 tests, whereas they did not significantly differ on criterion standard measures. These findings are consistent with previous studies showing digital executive functioning paradigms may be more sensitive to early FTLD than traditional measures. 42 , 43

This study has some limitations. Validation analyses focused on participants’ initial task exposure. Future studies will explore whether repeated measurements and more sophisticated approaches to composite building (current composite assumes equal weighting of tests) improve reliability and sensitivity, and a normative sample is being collected to better adjust for demographic effects on testing. 24 Longitudinal analyses will explore whether the floor effects in participants with symptomatic FTLD will affect the utility for monitoring. The generalizability of the findings is limited by the study cohort, which comprised participants who were college educated on average, mostly White, and primarily English speakers who owned smartphones and participated in the referring in-person research study. Equity in access to research is a priority in FTLD research 44 , 45 ; translations of the ALLFTD-mApp are in progress, cultural adaptations are being considered, and devices have been purchased for provisioning to improve the diversity of our sample.

The findings of this cohort study, coupled with prior reports indicating that smartphone testing is feasible and acceptable to patients with FTLD, 11 suggest that smartphones may complement traditional in-person research paradigms. More broadly, the scalability, ease of use, reliability, and validity of the ALLFTD-mApp suggest the feasibility and utility of remote digital assessments in dementia clinical trials. Future research should validate these results in diverse populations and evaluate the utility of these tests for longitudinal monitoring.

Accepted for Publication: February 2, 2024.

Published: April 1, 2024. doi:10.1001/jamanetworkopen.2024.4266

Open Access: This is an open access article distributed under the terms of the CC-BY License . © 2024 Staffaroni AM et al. JAMA Network Open .

Corresponding Author: Adam M. Staffaroni, PhD, Weill Institute for Neurosciences, Department of Neurology, Memory and Aging Center, University of California, San Francisco, 675 Nelson Rising Ln, Ste 190, San Francisco, CA 94158 ( [email protected] ).

Author Contributions: Dr Staffaroni had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Staffaroni, A. Clark, Taylor, Heuer, Wise, Forsberg, Miller, Hassenstab, Rosen, Boxer.

Acquisition, analysis, or interpretation of data: Staffaroni, A. Clark, Taylor, Heuer, Sanderson-Cimino, Wise, Dhanam, Cobigo, Wolf, Manoochehri, Mester, Rankin, Appleby, Bayram, Bozoki, D. Clark, Darby, Domoto-Reilly, Fields, Galasko, Geschwind, Ghoshal, Graff-Radford, Hsiung, Huey, Jones, Lapid, Litvan, Masdeu, Massimo, Mendez, Miyagawa, Pascual, Pressman, Ramanan, Ramos, Rascovsky, Roberson, Tartaglia, Wong, Kornak, Kremers, Kramer, Boeve, Boxer.

Drafting of the manuscript: Staffaroni, A. Clark, Taylor, Heuer, Wolf, Lapid.

Critical review of the manuscript for important intellectual content: Staffaroni, Taylor, Heuer, Sanderson-Cimino, Wise, Dhanam, Cobigo, Manoochehri, Forsberg, Mester, Rankin, Appleby, Bayram, Bozoki, D. Clark, Darby, Domoto-Reilly, Fields, Galasko, Geschwind, Ghoshal, Graff-Radford, Hsiung, Huey, Jones, Lapid, Litvan, Masdeu, Massimo, Mendez, Miyagawa, Pascual, Pressman, Ramanan, Ramos, Rascovsky, Roberson, Tartaglia, Wong, Miller, Kornak, Kremers, Hassenstab, Kramer, Boeve, Rosen, Boxer.

Statistical analysis: Staffaroni, A. Clark, Taylor, Heuer, Sanderson-Cimino, Cobigo, Kornak, Kremers.

Obtained funding: Staffaroni, Rosen, Boxer.

Administrative, technical, or material support: A. Clark, Taylor, Heuer, Wise, Dhanam, Wolf, Manoochehri, Forsberg, Darby, Domoto-Reilly, Ghoshal, Hsiung, Huey, Jones, Litvan, Massimo, Mendez, Miyagawa, Pascual, Pressman, Ramanan, Kramer, Boeve, Boxer.

Supervision: Geschwind, Miyagawa, Roberson, Kramer, Boxer.

Conflict of Interest Disclosures: Dr Staffaroni reported being a coinventor of 4 ALLFTD mobile application tasks (not analyzed in the present study) and receiving licensing fees from Datacubed Health; receiving research support from the National Institute on Aging (NIA) of the National Institutes of Health (NIH), Bluefield Project to Cure FTD, the Alzheimer’s Association, the Larry L. Hillblom Foundation, and the Rainwater Charitable Foundation; and consulting for Alector Inc, Eli Lilly and Company/Prevail Therapeutics, Passage Bio Inc, and Takeda Pharmaceutical Company. Dr Forsberg reported receiving research support from the NIH. Dr Rankin reported receiving research support from the NIH and the National Science Foundation and serving on the medical advisory board for Eli Lilly and Company. Dr Appleby reported receiving research support from the Centers for Disease Control and Prevention (CDC), the NIH, Ionis Pharmaceuticals Inc, Alector Inc, and the CJD Foundation and consulting for Acadia Pharmaceuticals Inc, Ionis Pharmaceuticals Inc, and Sangamo Therapeutics Inc. Dr Bayram reported receiving research support from the NIH. Dr Domoto-Reilly reported receiving research support from NIH and serving as an investigator for a clinical trial sponsored by Lawson Health Research Institute. Dr Bozoki reported receiving research funding from the NIH, Alector Inc, Cognition Therapeutics Inc, EIP Pharma, and Transposon Therapeutics Inc; consulting for Eisai and Creative Bio-Peptides Inc; and serving on the data safety monitoring board for AviadoBio. Dr Fields reported receiving research support from the NIH. Dr Galasko reported receiving research funding from the NIH; clinical trial funding from Alector Inc and Esai; consulting for Esai, General Electric Health Care, and Fujirebio; and serving on the data safety monitoring board of Cyclo Therapeutics Inc. Dr Geschwind reported consulting for Biogen Inc and receiving research support from Roche and Takeda Pharmaceutical Company for work in dementia. Dr Ghoshal reported participating in clinical trials of antidementia drugs sponsored by Bristol Myers Squibb, Eli Lilly and Company/Avid Radiopharmaceuticals, Janssen Immunotherapy, Novartis AG, Pfizer Inc, Wyeth Pharmaceuticals, SNIFF (The Study of Nasal Insulin to Fight Forgetfulness) study, and A4 (The Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease) trial; receiving research support from Tau Consortium and the Association for Frontotemporal Dementia; and receiving funding from the NIH. Dr Graff-Radford reported receiving royalties from UpToDate; reported participating in multicenter therapy studies by sponsored by Biogen Inc, TauRx Therapeutics Ltd, AbbVie Inc, Novartis AG, and Eli Lilly and Company; and receiving research support from the NIH. Dr Grossman reported receiving grant support from the NIH, Avid Radiopharmaceuticals, and Piramal Pharma Ltd; participating in clinical trials sponsored by Biogen Inc, TauRx Therapeutics Ltd, and Alector Inc; consulting for Bracco and UCB; and serving on the editorial board of Neurology . Dr Hsiung reported receiving grant support from the Canadian Institutes of Health Research, the NIH, and the Alzheimer Society of British Columbia; participating in clinical trials sponsored by Anavax Life Sciences Corp, Biogen Inc, Cassava Sciences, Eli Lilly and Company, and Roche; and consulting for Biogen Inc, Novo Nordisk A/S, and Roche. Dr Huey reported receiving research support from the NIH. Dr Jones reported receiving research support from the NIH. Dr Litvan reported receiving research support from the NIH, the Michael J Fox Foundation, the Parkinson Foundation, the Lewy Body Association, CurePSP, Roche, AbbVie Inc, H Lundbeck A/S, Novartis AG, Transposon Therapeutics Inc, and UCB; serving as a member of the scientific advisory board for the Rossy PSP Program at the University of Toronto and for Amydis; and serving as chief editor of Frontiers in Neurology . Dr Masdeu reported consulting for and receiving research funding from Eli Lilly and Company; receiving personal fees from GE Healthcare; receiving grant funding and personal fees from Eli Lilly and Company; and receiving grant funding from Acadia Pharmaceutical Inc, Avanir Pharmaceuticals Inc, Biogen Inc, Eisai, Janssen Global Services LLC, the NIH, and Novartis AG outside the submitted work. Dr Mendez reported receiving research support from the NIH. Dr Miyagawa reported receiving research support from the Zander Family Foundation. Dr Pascual reported receiving research support from the NIH. Dr Pressman reported receiving research support from the NIH. Dr Ramos reported receiving research support from the NIH. Dr Roberson reported receiving research support from the NIA of the NIH, the Bluefield Project, and the Alzheimer’s Drug Discovery Foundation; serving on a data monitoring committee for Eli Lilly and Company; receiving licensing fees from Genentech Inc; and consulting for Applied Genetic Technologies Corp. Dr Tartaglia reported serving as an investigator for clinical trials sponsored by Biogen Inc, Avanex Corp, Green Valley, Roche/Genentech Inc, Bristol Myers Squibb, Eli Lilly and Company/Avid Radiopharmaceuticals, and Janssen Global Services LLC and receiving research support from the Canadian Institutes of Health Research (CIHR). Dr Wong reported receiving research support from the NIH. Dr Kornak reported providing expert witness testimony for Teva Pharmaceuticals Industries Ltd, Apotex Inc, and Puma Biotechnology and receiving research support from the NIH. Dr Kremers reported receiving research funding from NIH. Dr Kramer reported receiving research support from the NIH and royalties from Pearson Inc. Dr Boeve reported serving as an investigator for clinical trials sponsored by Alector Inc, Biogen Inc, and Transposon Therapeutics Inc; receiving royalties from Cambridge Medicine; serving on the Scientific Advisory Board of the Tau Consortium; and receiving research support from NIH, the Mayo Clinic Dorothy and Harry T. Mangurian Jr. Lewy Body Dementia Program, and the Little Family Foundation. Dr Rosen reported receiving research support from Biogen Inc, consulting for Wave Neuroscience and Ionis Pharmaceuticals, and receiving research support from the NIH. Dr Boxer reported being a coinventor of 4 of the ALLFTD mobile application tasks (not the focus of the present study) and previously receiving licensing fees; receiving research support from the NIH, the Tau Research Consortium, the Association for Frontotemporal Degeneration, Bluefield Project to Cure Frontotemporal Dementia, Corticobasal Degeneration Solutions, the Alzheimer’s Drug Discovery Foundation, and the Alzheimer’s Association; consulting for Aeovian Pharmaceuticals Inc, Applied Genetic Technologies Corp, Alector Inc, Arkuda Therapeutics, Arvinas Inc, AviadoBio, Boehringer Ingelheim, Denali Therapeutics Inc, GSK, Life Edit Therapeutics Inc, Humana Inc, Oligomerix, Oscotec Inc, Roche, Transposon Therapeutics Inc, TrueBinding Inc, and Wave Life Sciences; and receiving research support from Biogen Inc, Eisai, and Regeneron Pharmaceuticals Inc. No other disclosures were reported.

Funding/Support: This work was supported by grants AG063911, AG077557, AG62677, AG045390, NS092089, AG032306, AG016976, AG058233, AG038791, AG02350, AG019724, AG062422, NS050915, AG032289-11, AG077557, K23AG061253, and K24AG045333 from the NIH; the Association for Frontotemporal Degeneration; the Bluefield Project to Cure FTD; the Rainwater Charitable Foundation; and grant 2014-A-004-NET from the Larry L. Hillblom Foundation. Samples from the National Centralized Repository for Alzheimer’s Disease and Related Dementias, which receives government support under cooperative agreement grant U24 AG21886 from the NIA, were used in this study.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Group Information: A complete list of the members of the ALLFTD Consortium appears in Supplement 2 .

Data Sharing Statement: See Supplement 3 .

Additional Contributions: We thank the participants and study partners for dedicating their time and effort, and for providing invaluable feedback as we learn how to incorporate digital technologies into FTLD research.

Additional Information: Dr Grossman passed away on April 4, 2023. We want to acknowledge his many contributions to this study, including data acquisition, and design and conduct of the study. He was an ALLFTD site principal investigator and contributed during the development of the ALLFTD mobile app.

• Register for email alerts with links to free full-text articles
• Access PDFs of free articles
• Manage your interests
• Save searches and receive search alerts

Study finds 3 big risk factors for dementia

Diabetes, air pollution and alcohol consumption could be the biggest risk factors for dementia, a study has found.

Researchers compared modifiable risk factors for dementia — which is characterized by the impairment of memory, thinking and reasoning — and studied how these factors appear to affect certain brain regions that are already particularly vulnerable to Alzheimer’s disease and schizophrenia.

The research, based on brain scans of nearly 40,000 adults, between ages 44 and 82, in Britain was published Wednesday in Nature Communications.

These vulnerable regions of the brain develop during adolescence and help the brain process and integrate “bits of information across different modalities, across different senses,” said Gwenaëlle Douaud , an associate professor at the University of Oxford and co-author of the study. But “they’re the first ones to go when we start aging.”

“What we’re trying to do is say: What are the common risk factors for dementia that are affecting these regions?” Douaud said. “These are the three most harmful but then, obviously, the others, they have an effect.”

• Researchers investigated the genetic and modifiable risk factors that contribute to the vulnerability of the “most fragile parts of the brain” by studying the brain scans of nearly 40,000 relatively healthy participants from the U.K. Biobank.
• The study examined 161 modifiable risk factors, including blood pressure, cholesterol, diabetes, weight, alcohol consumption, smoking, mood, inflammation, pollution, hearing, sleep, socialization, diet, physical activity and education.
• A diagnosis of diabetes, the amount of nitrogen dioxide in the air and how often someone drinks alcohol — from never to daily, or nearly every day — were found to be the three most detrimental risk factors to these regions of the brain, Douaud said.
• Diabetes, air pollution and alcohol consumption each has an effect that is about twice as much as the other leading risk factors, Douaud said. The next major risk factors are sleep, weight, smoking and blood pressure.
• Researchers identified seven genetic clusters that affect these vulnerable parts of the brain, some of which are also associated with Parkinson’s and Alzheimer’s diseases. Douaud said the genetic and modifiable risk factors are not comparable.

More than 55 million people live with dementia around the world, and that figure is expected to increase to 153 million by 2050, according to the World Health Organization.

Get Well+Being tips straight to your inbox

Dementia is the loss of cognitive function, and symptoms result from brain neurons losing their connection to other brain cells and eventually dying, according to the National Institute on Aging . Everyone loses neurons over time, but the loss is more significant in dementia patients.

Diabetes and alcohol consumption “have been consistently shown to be associated with both cerebral and cognitive decline,” the researchers wrote in the Nature Communications study. And there is growing evidence that exposure to air pollution is a risk factor for cognitive decline and dementia.

A 2020 Lancet report on dementia found that a dozen modifiable risk factors such as hypertension, hearing impairment, smoking and obesity together account for up to 40 percent of dementia cases worldwide.

A person’s age, genes, family history, a traumatic brain injury or a stroke also are potential risk factors .

What other experts say

Gill Livingston , a professor of psychiatry of older people at University College London and the lead author of the 2020 Lancet report, said that the new study was “very interesting” but that the participants in the U.K. Biobank are a “much healthier” and highly motivated group. The findings may not be applicable to a broader population.

Still, they show that people can make decisions to reduce their risk of cognitive decline as they age, Livingston said.

“There’s quite a lot people can do in their everyday life to maintain cognitive health ,” she said. “This just reinforces it.”

What you can do about it

Try to eat a healthy, varied diet to help lower your blood sugar, take measures to protect against “traffic-related pollution” and drink alcohol in moderation, Douaud said in an email.

“Of course, some of these should not only be down to individuals, and the burden should also be shared with (local) governments devising helpful policies,” she said.

Livingston said social and physical activity — talking with friends and exercising — “make a huge difference.” And stimulating experiences, walking outside , “seeing different things” can be beneficial, she said.

Hearing loss , which is more likely as a person ages, can take away opportunities for conversations that lead to “rapid brain stimulation,” Livingston said.

“And if you smoke, stop,” she said.

Marlene Cimons contributed to this report.

Read more from Well+Being

Well+Being shares news and advice for living well every day. Sign up for our newsletter to get tips directly in your inbox.

Are you taking your meds wrong ? Many patients make these common mistakes.

Centenarians give their advice about everything.

The wall sit is a simple exercise that can lower your blood pressure.

Tart cherries — more specifically, tart cherry juice — may help with inflammation and pain.

Do you self-sabotage ? Here’s how to stop.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Indian J Psychol Med
• v.43(5 Suppl); 2021 Sep

Ethical Issues in Dementia Research

Mina chandra.

1 Centre of Excellence in Mental Health, Atal Bihari Vajpayee Institute of Medical Sciences and Dr Ram Manohar Lohia Hospital, New Delhi, Delhi, India.

Vijay Harbishettar

2 Health Heal Geriatric Healthcare Services, Bengaluru, Karnataka, India.

Harbandna Sawhney

Shabbir amanullah.

3 Woodstock General Hospital, Woodstock, Ontario, Canada.

Dementia is a global public health issue with an urgent need for developing newer and more effective treatment strategies. Research in the area of dementia, however, poses unique ethical and legal challenges. Epidemiological studies, studies on pharmacological and non-pharmacological interventions have to deal with obtaining consent from persons with cognitive impairments, those from diverse cultural groups and need to contend with privacy and confidentiality issues. The caregiver support intervention research has not yet translated into policy change and effective clinical care. Biomedical research that involves invasive procedures may not translate into short- or long-term therapeutic benefits but is necessary research. Palliative care research in dementia has to deal with ethical issues involving people at end-of-life research. Proposed research may not receive approval, citing necessary safeguards to the vulnerable older people against invasive studies even when it is least invasive. This article aims to review the ethical aspects for safeguarding vulnerable older people with dementia and the potential challenges in conducting dementia research from a researcher’s perspective.

Some of the safeguards for ethical research include determining capacity to consent, obtaining advanced directives in early stages and proxy consent from caregivers, obtaining informed consent in cognitively impaired individuals. Future research policies need to consider the logistics of involving older people in research, enhancing caregiver support, and encouraging supportive decision-making. It will also need to address developing capacity assessment tools while addressing advanced care planning that will ensure the well-being of subjects in research.

Background:

Dementia has become a global public health issue, with hospitalization rates being 65% higher in seniors with dementia than others. 1 , 2 The pressures on healthcare systems mean an urgent need to develop robust preventive and treatment strategies for dementia, which requires multidisciplinary research. However, the patient’s stage of illness and ability to engage in discussions around the merits of participating in research and caregiver concerns is an important aspect of dementia research.

Hence, dementia research poses unique ethical challenges compared to populations with other diseases, which has led to the evolution of an ethical framework for dementia research. This article aims to review and give a viewpoint on the ethical aspects for safeguarding vulnerable older people with dementia and the potential challenges in conducting dementia research from a researcher’s perspective.

Materials and Methods

Systematic Review of PubMed was performed using the following search string to obtain studies on ethical issues in dementia research:

(dementia[Title] OR ncd[Title] OR neurocognitive[Title] OR alzheimer[Title] OR cognitive[Title] or OR cognition[Title] OR neuropsychology[Title] OR neuropsychological[Title] OR pick[Title] OR lewy[Title] OR frontotemporal[Title] OR huntington[Title] OR prion[Title]) AND (ethics[Title] OR ethical[Title]) AND (research[Title])

Seventy six results were obtained, which included publications from 1984 to 2021. Results were filtered for the availability of free full text and, subsequently, 15 studies were left. References within the articles were reviewed to obtain a comprehensive review.

These articles were reviewed for relevance, and all were found suitable for inclusion in this review.

These articles were evaluated in detail (authors MC and HS ) to create the first draft. The second draft incorporated the clinical and research experience of the authors who had varying levels of experience. An in-depth discussion was held with all authors regarding various aspects of the paper on a virtual platform. The inputs from all authors were combined to formulate a viewpoint on ethical issues in dementia research.

The methodology is illustrated in detail in Figure 1 .

Ethical Issues Across Specific Types of Dementia Research

Epidemiological studies in dementia.

Epidemiological studies have dealt with consent issues in cognitively impaired populations with low baseline educational attainment or culturally diverse populations. Furthermore, there are often privacy and confidentiality issues in field settings, issues around medical care, and rehabilitative measures for identified persons with dementia (PwD). This raises questions about basic ethical issues of beneficence and nonmaleficence. Another ethical aspect of dementia in low- and middle-income countries (LMIC) like India is the lack of culturally sensitive screening and evaluation instruments. Concerns about the validity of nosological criteria of dementia in uneducated, multiethnic, multilingual populations have not been adequately raised or addressed. There is a genuine concern that instruments designed for the Western population may not be appropriate and may end up mislabeling and stigmatizing some with only age-related cognitive impairment. This is especially significant given that the treatment gap in dementia in LMICs is more than 90%. 3

Dementia research has also investigated modifiable risk factors for dementia to identify targets for early intervention. Lower levels of educational attainment and decreased physical activity levels have been noted to be associated with Alzheimer’s disease (AD). 4 Hence, the population-level increase in educational status and physical activity levels can potentially reduce the risk of cognitive impairment.

Biomarker Research in Dementia

A major ethical issue in research on biomarkers in dementia has been a disproportionately excessive focus on AD, usually diagnosed as “probable AD” 5 even though other types of dementia commonly exist in clinical practice. Similarly, an emphasis on neuroimaging and biomarkers as indicators of dementia risk has not translated into clinical practice. 6 , 7 For example, it is known that up to 50% decrease in Cerebrospinal Fluid (CSF) Aβ 42 concentrations and twofold to threefold increase in CSF total tau increase in F2-isoprostanes occurs in AD or mild cognitive impairment. Still, these tests are not routinely performed in clinical practice either for diagnosis or prognosis. 8

This leads to a fundamental question on the nature of dementia research, which does not yield any diagnostic or therapeutic benefit to patients with dementia in the short or long term while subjecting cognitively impaired individuals for invasive and painful procedures.

Research on Pharmacological and Nonpharmacological Interventions for Dementia

Both pharmacological and nonpharmacological interventions have been studied for dementia. While pharmacological intervention can be standardized, nonpharmacological interventions may not use a standardized treatment manual, which is ethically questionable. There is also a need to develop standardized methodologies for assessing nonpharmacological interventions to reach a more robust conclusion.

Among pharmacological interventions, there have been extensive trials on acetylcholinesterase inhibitors and memantine. 9 However, since 2003, no new drug has been approved by the FDA to treat AD. More than 200 therapeutic agents have been assessed, and then either they are abandoned or have failed investigational programs. 10

Some Pharmaceutical companies like Pfizer have abandoned dementia research. 11

A pertinent ethical question is about the rights of the participants with dementia who participated in these failed trials in good faith.

Caregiver Support Interventions Research

Caregivers of patients of dementia face many challenges, with studies reporting 30%–55% of caregivers experience anxiety or depressive symptoms, which may adversely impact the quality of care provided to the patient. 12 – 14 Although studies have shown that psychosocial interventions may be effective in dementia caregivers to reduce their burden, they often face logistic barriers, stigma among caregivers, and a lack of structured instructional manuals. 15

The beneficial impact of psychosocial intervention for dementia includes delayed institutionalization of patients, improved symptomatology, and providing services that caregivers highly value. 16 However, despite robust evidence, caregiver interventions have not been translated into policy domain and clinical practice. This is not ethical considering caregivers have a significant burden because of caring for patients with dementia. The lack of translation of evidence-based, low-cost interventions for dementia in policy and clinical care across many nations globally raises important ethical questions for the future research.

Clinical Trials

For developing new drugs for the treatment of PwD, clinical trials are essential, but it requires well written informed consent assent process and rigorous documentation before recruitment. Participants will need more intense monitoring compared to cognitively intact individuals. 17

There is concern about the robustness of the informed consent process for clinical trials for dementia conducted in LMICs with a population with limited baseline educational attainment. In some situations, it may be necessary that an independent clinician may undertake capacity assessments and informed consent to get involved in research instead of those research staff involved in the study.

Genetic Testing

Genetic testing for dementia includes testings for symptomatic individuals (diagnostic testing) and asymptomatic at-risk individuals (predictive testing).

Advance directives may be useful in such cases. As genetic testing results may affect the patient and other family members, familial genetic counseling is essential before testing and for disclosure of results. 18

End of Life/Palliative Care Issues in Dementia Research

It may be difficult to determine the lifespan of patients in advanced stages of dementia, and many patients receive palliative care in hospices in high-income settings. 19 Dementia research may include some participants in palliative care or who enter palliative care during the research. This raises ethical dilemmas of prolonging life at the cost of quality of life to complete research. Transference of a patient in the end stages of life to an unfamiliar setting may also lead to ethical dilemmas. The ability to participate in research may vary based on the clinical condition, and this poses challenges for the researcher, having to reassess this every time. Many exclusions of research participants during research may not serve the purpose, and this poses an ethical question to those who are participating, whether the study serves its purpose.

Ethical Dilemmas in Multicentric International Research

The 10/66 Dementia Research Group found that even though 66% of people with dementia live in developing countries, less than 10% of dementia research is conducted. Underrepresentation of LMIC populations with large absolute numbers of dementia is a fundamental ethical issue in dementia research and must be addressed by policymakers and grant agencies.

Even within LMICs, ethnic, linguistic, and religious minorities are underrepresented, necessitating the development of novel strategies to improve the participation of minority groups. 20 Equitable partnership and participation in research across social groups is an important ethical goal in dementia research pertaining to rights of participation and justice.

Researchers lead most multicentric international dementia research in LMICs from high-income settings. It is often found that many such protocols contain research instruments that have not been culturally and linguistically validated for the target population resulting in underreporting or overreporting of cognitive impairment and dementia. In addition, the informed consent documents in multicentric international trials may not be adequately translated to cater to the informational needs of target populations with low educational attainment, as is often the case in LMICs. They may result in a lack of awareness of rights and compensation to participants.

Ethical Aspects of Data Sharing Agreement to Enhance Research Outputs

For dementia research to reach any breakthrough, it is important to use harmonized protocols and share anonymized data with strict precautions for confidentiality, privacy, and data stewardship for implementing big data analytics. Alzheimer Europe gave a report, “Data Sharing in Dementia Research,” which reviews the recent changes in research policy and gives recommendations to aid data sharing in dementia research. This is ethical utilization of limited resources for dementia research, but such initiatives are lacking in the Asian context. 21

Ethical Issues Specific to Dementia Research

Memory impairment, poor comprehension, hearing, or visual impairment at varying degrees is common in dementia patients. This poses a challenge for the investigators to obtain consent, and more participants with impaired ability to comprehend information will get excluded from the study. On the other hand, vulnerable PwD in institutional care may agree/disagree to participate in research due to possible coercion and fear. Some studies on PwD in advanced stages include PwD with sensory deficits, which are not likely to have competency. This makes such studies difficult due to ethical dilemmas and leaves a gap in the understanding of dementia and probable medical progress for the care of such patients. 22 – 24 The progressive and continuous nature of the illness may necessitate a periodic reassessment of capacity for consent to participate in the study.

If PwD cannot have impaired capacity to give consent, proxy consent from caregivers can be taken. 25

An ethical issue with dementia research is that persons without supportive family or carers may be excluded from the research (and its benefits) because they lack adequate representation. 26

In the case of proxy consent, the patient’s beliefs must be kept in mind while making a decision. 27 Advanced directive may help in this aspect. However, some ethical issues emerge when proxy decision-makers decide as per their beliefs about the right decision for the patient. 28 Some research has shown that the data provided by proxies may differ from data provided by PwD. 29 – 31 In the decision-making in dementia research by caregivers, it is essential for researchers or the ethical committees to ensure that there is no secondary gain by the caregiver considering the late stage of illness and issues around property, will, or any other form.

In dementia research, it is required to obtain both informed consent and assent. 32 Assent may be defined as “the agreement to participate in research-based upon less than full understanding.” 33 In contrast to informed consent, which requires an individual to understand the research protocol, 34 to give assent, an individual must only have a minimal level of understanding to make a meaningful choice. 35 An individual’s level of cooperation may be indicative of assent or dissent. 36 Frustration, discomfort, unhappiness, or passivity may indicate a lack of cooperation in research or dissent. 37 A study which audiotaped informed consent encounters for research concluded that any interpretation regarding assent should be made with caution for PwD. Also, along with the cognitive aspect, the emotional and social dimensions of informed consent warrant attention. 38

One of the main concerns for ethical issues in dementia is related to involving older people in research. Possible barriers may include physical and cognitive impairments, lack of transport, a lower threshold for burden, changes in routine, and negative beliefs about medication. 39 , 40 These factors may affect the risk–benefit ratio for participation in research. Issues of beneficence versus nonmaleficence may need to be addressed.

Recruitment and maintenance of older people in studies may also involve higher costs. 41

On the other hand, failure to include older people in studies could lead to inequities in healthcare and biased results. 42

A databases search (PubMed and CINAHL) has revealed that only 3%–6% of clinical trials were based on older populations. 40 Studies have shown that clinical trials may sometimes involve participants who are not representative of those for whom the medication is most likely to be used. Older individuals are often excluded from the trials. 43 , 44

Respect for autonomy may be difficult in PwDs on account of impaired cognition.

Researchers and Institutional Ethics Committees and Institutional Review Boards can uphold the ethical principles of respect, beneficence, and justice through the informed consent/assent process and objectively assess the risks and benefits of participation in research. 45

Higher involvement of caregivers in dementia research safeguards the rights of PwD.

Patient and caregiver should share a harmonious relationship. However, in some cases, PwD may object to being accompanied by a caregiver for participation in a study. 46 It has also been suggested that the involvement of the caretaker may amount to paternalism. 47 Table 1 highlights the ethical issues in dementia research.

Ethical Issues Specific to Dementia Research.

The progressive nature of the disease in dementia with poor interepisodic recovery, involvement of older age group, high caregiver burden, and issues around consent poses unique ethical challenges around dementia research. Developing decisional capacity assessment tools, advanced care planning, and a standardized approach to research would help in addressing the ethical barriers

• Involving the elderly population: Dementia research should include all age groups and all stages of severity so that results are not biased.
• Research tools: Standardized manuals for nonpharmacological interventions need to be developed for a more structured research approach. Furthermore, developing culturally and linguistically validated instruments and harmonized protocols would help in a uniform approach to research.
• Consent: Supportive decision/assent making should be done to ensure ethical safeguards. It should be assumed that PwD has capacity unless proved otherwise.
• Advanced care planning: In the early stages of dementia, advanced directives should be formed not just for clinical care but also for research participation. This would enhance research participation while upholding ethics.

Along with capacity assessment tools that check decision making in specific areas, knowledge of the patient’s hopes, beliefs, and personal history should also be used to document decision making and withdrawal of cons. 48

Data Safety Management Boards—Serious Adverse Events reporting as per established protocols enhances ethical conduct of dementia research. Scientists should make the study results accessible to the wider scientific community by sharing them in institutional repositories. This allows for secondary data research and avoids unnecessary research in a vulnerable population. At the same time, data sharing and attendant scrutiny increase confidence in research implementation as per ethical and legal regulatory framework.

Recommendations for the ethics committee and regulatory authority are illustrated in Table 2 .

Recommendations for Ethics Committee and Regulatory Authority

Declaration of Conflicting Interests: The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding: The authors received no financial support for the research, authorship, and/or publication of this article.

Skip to Content

'An epidemic of loneliness': How the COVID-19 pandemic changed life for older adults

• Share via Twitter
• Share via Facebook
• Share via LinkedIn
• Share via E-mail

Years after the U.S. began to slowly emerge from mandatory COVID-19 lockdowns, more than half of older adults still spend more time at home and less time socializing in public spaces than they did pre-pandemic, according to new University of Colorado Boulder research. 

Participants cited fear of infection and “more uncomfortable and hostile” social dynamics as key reasons for their retreat from civic life.

“The pandemic is not over for a lot of folks,” said Jessica Finlay, an assistant professor of geography whose findings are revealed in a series of new papers . “Some people feel left behind.”

The study comes amid what the U.S. Surgeon General recently called an ‘ epidemic of loneliness ’ in which older adults— especially those who are immune compromised or have disabilities— are particularly vulnerable.

“We found that the pandemic fundamentally altered neighborhoods, communities and everyday routines among aging Americans and these changes have long-term consequences for their physical, mental, social and cognitive health,” said Finlay.

‘I just can’t go back’

As a health geographer and environmental gerontologist, Finlay studies how social and built environments impact health as we age.

In March 2020 as restaurants, gyms, grocery stores and other gathering places shuttered amid shelter-in-place orders, she immediately wondered what the lasting impacts would be. Shortly thereafter, she launched the COVID-19 Coping Study with University of Michigan epidemiologist Lindsay Kobayashi. They began their research with a baseline and monthly survey. Since then, nearly 7,000 people over age 55 from all 50 states have participated.

The researchers check in annually, asking open-ended questions about how neighborhoods and relationships have changed, how people spend their time, opinions and experiences of the COVID-19 pandemic, and their physical and mental health.

“We’ve been in the field for some incredibly pivotal moments,” said Finlay, noting that surveys went out shortly after George Floyd was murdered in May 2020 and again after the attack on the U.S. Capitol on Jan. 6, 2021.

Collectively, the results paint a troubling picture in which a substantial portion of the older population remains isolated even after others have moved on. 

In one paper published in February in the journal Wellbeing, Space and Society , 60% of respondents said they spend more time in their home while 75% said they dine out less. Some 62% said they visit cultural and arts venues less, and more than half said they attend church or the gym less than before the pandemic.

The most recent survey, taken in spring 2023, showed similar trends, with more than half of respondents still reporting that their socialization and entertainment routines were different than they were pre-pandemic. 

In another paper titled “ I just can’t go back ,” 80% of respondents reported that there are some places they are reluctant to visit in person anymore.

“The thought of going inside a gym with lots of people breathing heavily and sweating is not something I can see myself ever doing again,” said one 72-year-old male.

Those who said they still go to public places like grocery stores reported that they ducked in and out quickly and skipped casual chitchat. 

“It’s been tough,” said one 68-year-old female. “You don’t stop and talk to people anymore.”

Many respondents reported that they were afraid of getting infected with a virus or infecting young or immune-compromised loved ones, and said they felt “irresponsible” for being around a lot of people.

Some reported getting dirty looks or rude comments when wearing masks or asking others to keep their distance— interpersonal exchanges that reinforced their inclination to stay home.

Revitalizing human connection

The news is not all bad, stresses Finlay.

At least 10% of older adults report exercising outdoors more frequently since the pandemic. And a small but vocal minority said that their worlds had actually opened up, as more meetings, concerts and classes became available online. Still, Finlay worries that the loss of spontaneous interactions in what sociologists call “third places” could have serious health consequences.

Previous research shows that a lack of social connection can increase risk of premature death as much as smoking 15 cigarettes a day and exacerbate mental illness and dementia.

“For some older adults who live alone, that brief, unplanned exchange with the butcher or the cashier may be the only friendly smile they see in the day, and they have lost that,” Finlay said.

Societal health is also at risk.

“It is increasingly rare for Americans with differing sociopolitical perspectives to collectively hang out and respectfully converse,” she writes. 

Finlay hopes that her work can encourage policymakers to create spaces more amenable to people of all ages who are now more cautious about getting sick – things like outdoor dining spaces, ventilated concert halls or masked or hybrid events.

She also hopes that people will give those still wearing masks or keeping distance some grace.

“It is a privilege to be able to ‘just get over’ the pandemic and many people, for a multitude of reasons, just don’t have that privilege. The world looks different to them now,” she said.  “How can we make it easier for them to re-engage?”

News Headlines

• Arts & Humanities
• Business & Entrepreneurship
• Climate & Environment
• Education & Outreach
• Health & Society
• Law & Politics
• Science & Technology

Campus Community

• Administration
• Announcements & Deadlines
• Career Development
• Getting Involved
• Mind & Body

Events & Exhibits

• Arts & Culture
• Conferences
• Lectures & Presentations
• Performances & Concerts
• Sports & Recreation
• Workshops & Seminars

Subscribe to CUBT

Sign up for Alerts

Administrative eMemos

Buff Bulletin Board

Events Calendar

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 27 March 2024

The effects of genetic and modifiable risk factors on brain regions vulnerable to ageing and disease

• Jordi Manuello   ORCID: orcid.org/0000-0002-9928-0924 1 , 2 ,
• Joosung Min   ORCID: orcid.org/0000-0002-5541-5014 3 ,
• Paul McCarthy 1 ,
• Fidel Alfaro-Almagro 1 ,
• Soojin Lee 1 , 4 ,
• Stephen Smith 1 ,
• Lloyd T. Elliott 3   na1 ,
• Anderson M. Winkler 5 , 6   na1 &
• Gwenaëlle Douaud   ORCID: orcid.org/0000-0003-1981-391X 1  

Nature Communications volume  15 , Article number:  2576 ( 2024 ) Cite this article

17k Accesses

2213 Altmetric

Metrics details

• Genetics research
• Neuroscience
• Risk factors

We have previously identified a network of higher-order brain regions particularly vulnerable to the ageing process, schizophrenia and Alzheimer’s disease. However, it remains unknown what the genetic influences on this fragile brain network are, and whether it can be altered by the most common modifiable risk factors for dementia. Here, in ~40,000 UK Biobank participants, we first show significant genome-wide associations between this brain network and seven genetic clusters implicated in cardiovascular deaths, schizophrenia, Alzheimer’s and Parkinson’s disease, and with the two antigens of the XG blood group located in the pseudoautosomal region of the sex chromosomes. We further reveal that the most deleterious modifiable risk factors for this vulnerable brain network are diabetes, nitrogen dioxide – a proxy for traffic-related air pollution – and alcohol intake frequency. The extent of these associations was uncovered by examining these modifiable risk factors in a single model to assess the unique contribution of each on the vulnerable brain network, above and beyond the dominating effects of age and sex. These results provide a comprehensive picture of the role played by genetic and modifiable risk factors on these fragile parts of the brain.

Similar content being viewed by others

Exome-wide analysis implicates rare protein-altering variants in human handedness

Dick Schijven, Sourena Soheili-Nezhad, … Clyde Francks

Single-cell multiplex chromatin and RNA interactions in ageing human brain

Xingzhao Wen, Zhifei Luo, … Sheng Zhong

Protein-truncating variants in BSN are associated with severe adult-onset obesity, type 2 diabetes and fatty liver disease

Yajie Zhao, Maria Chukanova, … John R. B. Perry

Introduction

The development of preventative strategies based on modifying risk factors might prove to be a successful approach in ensuring healthy ageing. Factors particularly scrutinised in dementia and unhealthy ageing have included cerebrovascular factors such as high blood pressure, diabetes and obesity, but also lifestyle ones such as alcohol consumption, and protective factors such as exercise 1 . Assessing these modifiable risk factors together makes it possible to identify the unique contribution of each of these factors on the brain or on cognitive decline. A Lancet commission, updated in 2020 to include, e.g., pollution for its possible role in the incidence of dementia 2 , examined the relative impact of 12 modifiable risk factors for dementia, and showed that these 12 factors may account for 40% of the cases worldwide 3 . Conversely, genetic factors are non-modifiable in nature, but can inform us about the mechanisms underlying the phenotypes of interest. These mechanisms sometimes can be shared across these phenotypes. For instance, genetic overlap has been found for Alzheimer’s and Parkinson’s diseases at a locus in the MAPT region 4 . Likewise, one of the most pleiotropic variants, in the SLC39A8 / ZIP8 gene, shows genome-wide associations with both schizophrenia and fluid intelligence, amongst many other phenotypes 5 , 6 .

One way to objectively and robustly assess susceptibility for unhealthy ageing is to look non-invasively at brain imaging markers 7 . Using a data-driven approach on a lifespan cohort, we previously identified an ensemble of higher-order, ‘transmodal’ brain regions that degenerates earlier and faster than the rest of the brain 8 . The very same areas also develop relatively late during adolescence, thus supporting the ‘last in, first out’ (LIFO) hypothesis, which posits that the process of age-related brain decline mirrors developmental maturation. Importantly, this network of brain regions further demonstrated heightened vulnerability to schizophrenia and Alzheimer’s disease, two disorders that impact on brain structure during adolescence and ageing respectively. Accordingly, this LIFO network was strongly associated with cognitive traits whose impairment is specifically related to these two disorders, namely fluid intelligence and long-term memory 8 .

Here, our main objective was to assess both the genetic and modifiable risk factors’ contributions to the vulnerability of these most fragile parts of the brain. We conducted a genome-wide association study on a prospective cohort of nearly 40,000 participants of the UK Biobank study who had received brain imaging, and in total evaluated the association between the LIFO brain network and 161 modifiable risk factors, classified according to 15 broad categories: blood pressure, cholesterol, diabetes, weight, alcohol consumption, smoking, depressive mood, inflammation, pollution, hearing, sleep, socialisation, diet, physical activity and education.

The vulnerable LIFO brain network in UK Biobank

Similar to our previously observed results 8 , the loadings of the LIFO brain network, i.e., the normalised grey matter volume in the network after regressing out the effects of all the other brain maps (see Methods), demonstrated a strong quadratic association with age in the UK Biobank cohort of 39,676 participants ( R 2  = 0.30, P  < 2.23 × 10 −308 , Fig.  1 ). These higher-order regions thus show an accelerated decrease of grey matter volume compared with the rest of the brain. Furthermore, these areas define a network mainly involved in behavioural tasks related to execution, working memory, and attention (Fig.  1 , Supplementary Information ).

Top left, spatial map of the LIFO network (in red-yellow, thresholded at Z  > 4 for visualisation) used to extract the loadings from every scanned participant from UK Biobank ( n  = 39,676). Top right, these LIFO loadings (in arbitrary units) show a strong quadratic association with age in the UK Biobank cohort, i.e. grey matter volume decreases quadratically with older age in these specific regions ( R 2  = 0.30, P  < 2.23 × 10 −308 ; inset: residual scatterplot). Bottom, the vulnerable network appears to encompass areas mainly involved in execution, working memory, and attention (using the BrainMap taxonomy 60 , and with the LIFO brain network thresholded at both Z  = 4 and Z  = 10, see  Supplementary Information ).

Genetic influences over the vulnerable LIFO brain network

Using a minor allele frequency filter of 1% and a –log 10 (P) threshold of 7.5, we found, in the 39,676 participants, genome-wide associations between the LIFO brain network and seven genetic clusters whose top variants were all replicated (Table  1 /Supplementary Data  1 , Fig.  2 ).

Top row, Manhattan plot showing the 7 significant genetic clusters associated with the LIFO brain network (–log 10 ( P ) > 7.5). Second and third rows, regional association plots of the top variants for each of the 5 autosomal genetic clusters: rs6540873 on chromosome (Chr) 1 ( KCNK2 ), rs13107325 on Chr4 ( SLC39A8 ), rs2677109 on Chr6 ( RUNX2 ) (as a proxy in high LD R 2  = 0.86 with indel 6:45442860_TA_T), rs12146713 on Chr12 ( NUAK1 ), and rs2532395 on Chr17 ( MAPT , KANSL1 )(highest variant after tri-allelic rs2693333; see Supplementary Data  4 for a complete list of significant variants in this 5th MAPT genetic cluster). Bottom row, regional association plots of the top variants for the two genetic clusters in the pseudo-autosomal region PAR1 of the X chromosome: rs312238 ( XG , CD99 ) and rs2857316 ( XG )(UK Biobank has no genotyped variants on the 3’ side). Based on Human Genome build hg19. P -values are derived from a two-sided linear association test.

The first autosomal genetic cluster, on chromosome 1, included two variants (lead variant: rs6540873, β  = 0.06, P  = 1.71 × 10 −8 , and rs1452628, with posterior probabilities of inclusion in the causal variant set of 0.56 and 0.45, respectively) close to, and eQTL of, KCNK2 ( TREK1 ). This gene regulates immune-cell trafficking into the central nervous system, controls inflammation, and plays a major role in the neuroprotection against ischemia. Of relevance, these two loci are in particular related in UK Biobank participants with the amount of alcohol consumed, insulin levels, inflammation with interleukin-8 levels, as well as, crucially, with late-onset Alzheimer’s disease (Table  1 /Supplementary Data  1 ).

The second autosomal genetic cluster on chromosome 4 was made of 7 loci, with the lead variant rs13107325 in an exon of SLC39A8/ZIP8 ( β  = 0.14, P  = 2.82 × 10 −13 , posterior probability: 0.99). This locus is one of the most pleiotropic SNPs identified in GWAS, and is, amongst many other associations, related in UK Biobank with cholesterol, blood pressure, weight, inflammation with C-reactive proteins levels, diabetes with insuline-like growth factor 1 levels, alcohol intake, sleep duration, and cognitive performance/impairment, including prospective memory (Table 1 /Supplementary Data  1 ).

The third locus was an indel in chromosome 6 in an intron, and eQTL, of RUNX2 (rs35187443, β  = 0.06, P  = 9.03 × 10 −9 ), which plays a key role in differentiating osteoblasts, and has been very recently shown to limit neurogenesis and oligodendrogenesis in a cellular model of Alzheimer’s disease 9 .

The fourth locus was a SNP in chromosome 12, in an intron of NUAK1 (rs12146713, β  = −0.10, P  = 1.26 × 10 −9 ), and remarkably its top association in UK Biobank was with the contrast between schizophrenia and major depressive disorder 10 , and it was also associated with insulin-like growth factor 1 levels (Table 1 /Supplementary Data  1 ).

The final genetic autosomal genetic cluster was made of 3,906 variants in the MAPT region. Its lead non-triallelic variant, rs2532395 ( β  = −0.09, P  = 3.56 × 10 −15 ) was more specifically <10 kb from KANSL1 and an eQTL of KANSL1 , MAPT and other genes in brain tissues (Table 1 /Supplementary Data  1 , Supplementary Data 4 ). This locus was also associated in UK Biobank with tiredness and alcohol intake. MAPT is in 17q21.31, a chromosomal band involved with a common chromosome 17 inversion 11 . Adding chromosome 17 inversion status as a confounder reduced the significance of the association ( β  = −0.15, P  = 8.45 × 10 −3 ). Since the genotype for rs2532395 was also strongly correlated with chromosome 17 inversion in our dataset (Pearson correlation r  = 0.98, P  < 2 × 10 −16 ), this would suggest that the association between MAPT and the LIFO network is not independent from chromosome 17 inversion. As this extended genetic region is known for its pathological association with many neurodegenerative disorders including Alzheimer’s disease, we investigated whether the LIFO brain regions mediated the effect of the MAPT genetic cluster (using the lead bi-allelic variant rs2532395) on Alzheimer’s disease (see Methods). Despite small average causal mediated effect (ACME) sizes, we found a significant effect for both the dominant model (ACME β  = 1.16 × 10 −4 ; 95% CI = [5.19 × 10 −5 , 1.99 × 10 −4 ]; P  = 4 × 10 −5 ) and the recessive model (ACME β  = 1.55 × 10 −4 ; 95% CI = [3.96 × 10 −5 , 3.74 × 10 −4 ]; P  = 4 × 10 −5 ; full output of the mediation package on the dominant and recessive models in  Supplementary Information ).

The two last genetic clusters of 8 and 9 variants respectively were found on the X chromosome, notably in a pseudo-autosomal region (PAR1), which is interestingly hit at a higher rate than the rest of the genome ( P  = 1.56 × 10 −5 , see  Supplementary Information ). The top variants for these clusters were related to two homologous genes coding for the two antigens of the XG blood group: rs312238 ( β  = −0.05, P  = 1.77 × 10 −10 ) ~ 10 kb from, and an eQTL of, CD99/MIC2 , and rs2857316 ( β  = −0.08, P  = 2.27 × 10 −29 ) in an intron and eQTL of XG  (Table 1 /Supplementary Data  1 ). Since chromosome X has hardly been explored, we carried out our own association analyses between these two top variants and non-imaging variables in UK Biobank. Intriguingly, the first of these two PAR1 loci, rs312238, was found to be significantly associated in the genotyped participants who had not been scanned (out-of-sample analysis in n  = 374,230 UK Biobank participants) with nitrogen dioxide air pollution, our ‘best’ MRF for pollution (see below), and many other environmental, socioeconomic, and early life factors (such as urban or rural setting, distance from the coast, place of birth, number of siblings, breastfed as a baby, maternal smoking around birth), as well as health outcomes (Supplementary Data  2 ). In particular, amongst the more easily interpretable findings of the most associated variables with rs312238, the T allele of this locus was associated with two increased measures of deprivation and/or disability (worse socioeconomic status), the ‘Townsend deprivation index’ and the ‘Health score’, but also with ‘Nitrogen dioxide air pollution’, ‘Maternal smoking around birth’, as well as ‘Number of full brothers’ and ‘Number of full sisters’, thus showing consistent signs of association between this variant and these phenotypes.

We found that the heritability of the LIFO network was significant, with h 2  = 0.15 (se = 0.01). The genetic co-heritability between the LIFO network and Alzheimer’s disease or schizophrenia was not statistically significant (coefficient of co-heritability = −0.12, se = 0.10; P  = 0.23; coefficient of co-heritability = −0.16, se = 0.04, P  = 0.07, respectively).

Modifiable risk factors’ associations with the vulnerable LIFO brain network

Including the modifiable risk factors (MRFs) in a single general linear model allows us to assess the unique contribution of each factor on the LIFO brain network. Not all UK Biobank participants have data available for all of the MRF variables however. An analysis limited to those with complete data for all MRFs would be biased, and based on a relatively small, low-powered sample. We addressed this issue via a two-stage analysis in which: (i) we first identified which variable within each of the 15 MRF categories best represented associations of that category with the LIFO brain network loadings (based on two criteria: significance and <5% missing values), (ii) we investigated the unique contribution of that MRF category, over and above all other categories and the dominating effects of age and sex, to the LIFO loadings.

From the first stage of our analysis, 12 of the 15 categories of MRFs had at least one ‘best’ MRF, i.e., with a significant effect on the LIFO brain network and enough non-missing values across all scanned participants to be investigated further (Table  2 /Supplementary Data  3 ). The contribution of the MRFs on the vulnerable brain network differed vastly depending on whether confounding effects of age, sex and head size were taken into account. The effect size and significance of some MRFs diminished because of some clear collinearity with the confounders. For instance, for the category of blood pressure, the most significant MRF was first “systolic blood pressure, automatic (second) reading” ( r  = −0.20, P  < 2.23 × 10 −308 ), but after regressing out the confounders, the ‘best’ MRF for this category was “medication for blood pressure” ( r  = −0.05, P  = 7.55 × 10 −22 ). Conversely, regressing out the effects of age served to unmask the significant deleterious effects of pollution on the vulnerable brain regions, such as nitrogen dioxide air pollution or particulate matter air pollution (Table  2 /Supplementary Data  3 ).

When considered together in a single model in the second stage of the analysis, 3 best MRFs had an effect on the LIFO brain network that remained significant beyond the dominating effects of age and sex, and of the 9 other best MRFs: diabetes (“diabetes diagnosed by doctor”, r  = −0.05, P  = 1.13 × 10 −24 ), pollution (“nitrogen dioxide air pollution in 2005”, r  = −0.05, P  = 5.39 × 10 −20 ) and alcohol (“alcohol intake frequency”, r  = −0.04, P  = 3.81 × 10 −17 ) (Table  3 ). No MRFs showed any bias in their sub-sampling distribution, i.e., any significant difference between the original sample and the reduced sample of 35,527 participants who had values for all 18 variables considered (the 12 best MRFs and 6 confounders: age, sex, age 2 , age × sex, age 2  × sex, head size; Supplementary Information ). In total, the 12 best MRFs explained 1.5% of the effect on the vulnerable brain network ( F 12;35509  = 43.5).

While 6 out of the 7 genetic clusters associated with the LIFO network were correlated with many variables related to each of the 15 MRF categories, including diabetes, alcohol consumption and traffic pollution (Supplementary Data  1 ), we also found some genetic overlap between the very specific best MRF of “alcohol intake frequency” and the LIFO network in the pleiotropic rs13107325 variant (cluster 2), as well as rs17690703, part of the large genetic cluster 5 in MAPT (Supplementary Data  4 ). No genetic overlap was found for the precise “nitrogen dioxide air pollution in 2005” or “diabetes diagnosed by doctor”, nor for approximate variables.

This study reveals, in a cohort of nearly 40,000 UK Biobank participants, the genetic and modifiable risk factors’ associations with brain regions in a ‘last in, first out’ (LIFO) network that show earlier and accelerated ageing and are particularly vulnerable to disease processes such as that of Alzheimer’s disease 8 . Seven genetic clusters, two of which in the pseudo-autosomal region of the sex chromosomes coding for two antigens of the XG blood system, were found significantly associated and replicated genome-wide. In addition, after accounting for age and sex effects, diabetes, traffic-related pollution and alcohol were the most deleterious modifiable risk factors (MRFs) on these particularly vulnerable brain regions.

Three lead variants for our significant genetic clusters have been previously associated with ageing-related brain imaging measures in recent studies: one, in cluster 1, an eQTL of KCNK2 ( TREK1 ) 12 , 13 , whose increase in expression mediates neuroprotection during ischemia 14 , the ubiquitous rs13107325 (cluster 2), and one, in cluster 4, in an intron of NUAK1 ( ARK5 ) 15 , 16 , 17 , which has been associated with tau pathology 18 (Table  1 /Supplementary Data  1 ). On the other hand, of the seven genetic clusters, three were entirely novel (clusters 3, 6 and 7), and not found in other brain imaging studies, including our most recent work that expanded on our previous GWAS of all of the brain IDPs available in UK Biobank 19 by including more participants—in fact, the same number of participants as analysed in this present work—and, crucially, by also including the X chromosome 20 (Table  1 /Supplementary Data  1 ). This suggests that, beyond the genetic hits that were meaningfully associated with the LIFO brain network and an array of relevant risk factors, lifestyle variables and brain disorders, and found in a few other imaging GWAS, some of the genetic underpinnings of the LIFO network are intrinsically specific to it and to no other pre-existing imaging phenotype.

All five autosomal genetic clusters identified through the GWAS of the LIFO phenotype had relevant associations with risk factors for dementia (Results; Supplementary Data  1 ), including precisely two of the best MRFs (for clusters 2 and 5), and three of them directly related in UK Biobank to the two diseases showing a pattern of brain abnormalities following the LIFO network: schizophrenia (clusters 2 and 4) and Alzheimer’s disease (cluster 1) (Supplementary Data  1 ). In particular, cluster 2 has its lead variant rs13107325 in an exon of one of the most pleiotropic genes ZIP8 , which codes for a zinc and metal transporter. Considering the vulnerability of the LIFO brain network to adolescent-onset schizophrenia and its significant association with fluid intelligence that we previously demonstrated 8 , it is notable that this variant has been associated genome-wide with schizophrenia 6 , as well as intelligence, educational attainment and mathematics ability 5 , 21 . In line with the LIFO brain network being both prone to accelerated ageing and susceptible to Alzheimer’s disease, this genetic locus has also been associated genome-wide with well-known risk factors for dementia. These comprise alcohol—including the exact same variable of “alcohol intake frequency” as identified as one of the best MRFs—cholesterol, weight, sleep—including “sleep duration”—and blood pressure 22 , 23 , 24 , 25 , 26 , all of which significantly contribute to modulating the LIFO brain network when considered separately (Table  2 /Supplementary Data  3 ). Of relevance, this genetic locus is also associated to an increased risk of cardiovascular death 27 . Cluster 5, a large genetic cluster in the MAPT region (Microtubule-Associated Protein Tau), comprised in total 3906 significant variants (Supplementary Data  4 ). This genetic region plays a role in various neurodegenerative disorders related to mutations of the protein tau, such as frontotemporal dementia 28 and progressive supranuclear palsy 29 , but also, of particular pertinence to the LIFO brain network, Alzheimer’s and Parkinson’s disease, with a genetic overlap between these two diseases in a locus included in our significant cluster 5 (rs393152, β  = −0.09, P  = 6.35 × 10 −14 ) 4 . Despite the relatively low number of people with diagnosed Alzheimer’s disease in the genetic discovery cohort, we were able to establish—albeit with small effect sizes—a significant mediation role for the LIFO brain regions between the lead bi-allelic variant for cluster 5 and this Alzheimer’s diagnosis, suggesting once more the importance played by these vulnerable brain areas in unhealthy ageing.

Finally, of the seven clusters, two were located in the pseudo-autosomal region (PAR1) of the sex chromosomes corresponding to the genes XG and CD99 , coding for the two antigens of the XG blood group. This blood group system has been largely neglected, its main contribution related to the mapping of the X chromosome itself, and its clinical role remains elusive 30 . In order to investigate further the possible role of these two variants of the XG blood group, we examined out-of-sample their associations with thousands of non-imaging phenotypes. This analysis revealed that the first of these two loci was significantly and consistently associated with early life factors, environmental factors and health outcomes, including particulate matter and nitrogen dioxide air pollution, the second most deleterious MRF to the LIFO brain network (Supplementary Data  2 ). Whether these associations are due to stratification or genotyping artefacts, or to the fact that this specific variant, which is inherited from a parent, has a parental impact that modulates the effect of early life environment of the UK Biobank participants, the so-called “nature of nurture”, will need further investigation 31 .

Intriguingly, an analysis revealed that the genes involved in the loci associated with the LIFO network (Table  1 /Supplementary Data  1 ) are enriched for the gene ontology terms of leucocyte extravasation, namely “positive regulation of neutrophil extravasation” ( P  = 4.75 × 10 −6 ) and “T cell extravasation” ( P  = 4.75 × 10 −6 ). This result held when removing the genes included in the MAPT extended region (with P  = 2.54 × 10 −6 and P  = 2.54 × 10 −6 , respectively). Leucocyte extravasation facilitates the immune and inflammatory response, and there has been renewed focus on the fact that a breakdown of the blood-brain barrier together with leukocyte extravasation might contribute to both Alzheimer’s disease and schizophrenia 32 , 33 . In line with the enrichment findings, 4 out of the 7 genetic clusters associated with the LIFO network are correlated in UK Biobank blood assays with percentage or count of immune cells (neutrophil, lymphocyte, platelet, monocyte, etc.; Supplementary Data  1 ).

Regarding MRFs’ effects on the LIFO brain network, diabetes and alcohol consumption have been consistently shown to be associated with both cerebral and cognitive decline 34 , 35 . On the other hand, pollution—and notably that of nitrogen oxides—has emerged more recently as a potential MRF for dementia 2 , 36 . In particular, the increase of dementia risk due to nitrogen oxide pollution, a proxy for traffic-related air pollution, seems to be enhanced by cardiovascular disease 37 . In this study, we found that nitrogen dioxide pollution has one of the most deleterious effects onto the fragile LIFO brain regions. This effect could only be unmasked by regressing out the effects of age and sex, as traffic-related air pollution is modestly inversely-correlated with age (Supplementary Data  5 ). It is also worth noting that including age and sex as confounding variables in the first stage of our analysis reduced considerably the contribution of what had appeared at first—before regression—as the most harmful risk factors: blood pressure, cholesterol and weight (Table  2 /Supplementary Data  3 ). Furthermore, the benefit of examining these MRFs in a single model in the second stage of our analysis is that we can assess the unique contribution of each of these factors on the LIFO brain network; in doing so, blood pressure, cholesterol and weight were no longer significant (Table  3 ).

One defining characteristic of the LIFO brain network is how much age explains its variance. Indeed, in the dataset covering most of the lifespan that was initially used to identify the LIFO and spatially define it 8 , age explained 50%. In the UK Biobank imaging project, where imaged participants are over 45 years old, age explained 30% (Fig.  1 ). It is thus perhaps unsurprising that, while the explained variance by each of the MRFs varies widely (Table  2 /Supplementary Data  3 ), it reduces notably once the effect of age and other confounders has been regressed out (without confounders included in the model: maximum 8.4%; with confounders: maximum 0.5%). Combined, the 12 best MRFs explained a significant 1.5% of the effect on the vulnerable brain network after regressing out age, head size and sex effects. Regarding the genetic hits, we found a significant heritability with h 2  = 0.15, in keeping with our results for structural brain phenotypes (except for subcortical and global brain volumes, which demonstrate higher heritability 19 ).

The uniqueness of this study relies on the fact that we combined the strengths of two different cohorts: the first, which revealed the LIFO grey matter network, is lifespan, demonstrating the mirroring of developmental and ageing processes in the LIFO brain areas, something that could never be achieved with UK Biobank because of its limited age range. Of note, for this initial work with the lifespan cohort 8 , we not only included grey matter partial volume images, as done in this current study, but also Freesurfer information of cortical thickness and surface area. The LIFO network showed no contribution from Freesurfer cortical thickness or area. This might hint at processes that only partial volume maps are able to detect due to the LIFO network’s specific localisation, including in the cerebellum and subcortical structures, which are not included in the area and thickness surface methods from Freesurfer.

Limitations of our study pertain to the nature of the data itself and the way each variable is encoded in the UK Biobank (binary, ordinal, categorical, continuous), the number of missing values, what is offered as variables for each modifiable risk factor category (e.g. we chose not to create any compound variables, such as the ratio of cholesterol levels or systolic and diastolic blood pressures), and the curation of each of these variables. Some of the factors might be proxies for another category, but including the ‘best’ ones in a single model alleviate these issues to some extent. Another limitation is the assumption in our models that each risk factor has a linear, additive effect on the vulnerable LIFO brain network. It is also important to note that cross-sectional and longitudinal patterns of brain ageing can differ, as has been shown for instance for adult span trajectories of episodic and semantic memory, especially in younger adults 38 . A recent study has also demonstrated a specific ‘brain age’ imaging measure to be more related to early life influences on brain structure than within-person rates of change in the ageing brain 39 . Further work will be needed to establish how the LIFO network data changes in terms of within-person trends, for instance by investigating the growing UK Biobank longitudinal imaging database. While we took care of assessing the replicability of our genetic results by randomly assigning a third of our dataset for such purposes (all our significant genetic hits were replicated), this was performed within the UK Biobank cohort that exhibits well-documented biases, being well-educated, less deprived, and healthier than the general population, especially for its imaging arm 40 . Independent replications will be needed to confirm the existence of the LIFO-associated genetic loci.

In conclusion, our study reveals the modifiable and non-modifiable factors associated with some of the most fragile parts of the brain particularly vulnerable to ageing and disease process. It shows that, above and beyond the effect of age and sex, the most deleterious modifiable risk factors to this brain network of higher-order regions are diabetes, pollution and alcohol intake. Genetic factors are related to immune and inflammatory response, tau pathology, metal transport and vascular dysfunction, as well as to the XG blood group system from the pseudo-autosomal region of the sex chromosomes, and meaningfully associated with relevant modifiable risk factors for dementia. The unprecedented genome-wide discovery of the two variants on the sex chromosomes in this relatively unexplored blood group opens the way for further investigation into its possible role in underlying unhealthy ageing.

Supplementary Information is available for this paper.

For the present work the imaging cohort of UK Biobank was used and we included 39,676 subjects who had been scanned and for whom the brain scans had been preprocessed at the time of the final set of analyses (M/F 47–53%; 44–82 years, mean age 64 ± 7 years; as of October 2020) 41 , 42 . Structural T1-weighted scans for each participant were processed using the FSL-VBM automated tool to extract their grey matter map 43 , 44 . The ‘last in, first out’ (LIFO) network of mainly higher-order brain regions was initially identified by performing a linked independent component analysis on the grey matter images of another, lifespan observational cohort of 484 subjects 8 , 45 , 46 . This map of interest, along with the other 69 generated by the analysis, was first realigned to the UK Biobank ‘standard’ space defined by the grey matter average across the first 15,000 participants, then regressed into the UK Biobank participants’ grey matter data, to extract weighted average values of grey matter normalised volume inside each of the z-maps, using the z-score as weighting factor. This made it possible to assess the unique contribution of this specific LIFO map, above and beyond all the rest of the brain represented in the other 69 maps. At the end of this process, we obtained a single imaging measure for each of the 39,676 participants, i.e. a ‘loading’ corresponding to their amount of grey matter normalised volume in the LIFO brain network.

Human participants: UK Biobank has approval from the North West Multi-Centre Research Ethics Committee (MREC) to obtain and disseminate data and samples from the participants ( http://www.ukbiobank.ac.uk/ethics/ ), and these ethical regulations cover the work in this study. Written informed consent was obtained from all of the participants.

Modifiable risk factors selection

The following 15 categories of modifiable risk factors (MRFs) for dementia were investigated based on previous literature: blood pressure, diabetes, cholesterol, weight, alcohol, smoking, depression, hearing, inflammation, pollution, sleep, exercise, diet/supplementation, socialisation, and education. These included well-documented cerebrovascular risk factors, and in particular included all of the 12 modifiable risk factors considered in the updated Lancet commission on dementia, with the sole exception of traumatic brain injury 3 . For each category, several MRF variables from UK Biobank were very minimally pre-processed ( Supplementary Information ). In total, 161 MRF variables were obtained. To optimise the interpretability of the results, and to be able to relate them to previous findings, we did not carry out any data reduction, which would have prevented us from identifying exactly which variable—and subsequently, which genetic component for this specific variable—contribute to the effect. For these same reasons, we did not create any compound variable.

Statistical analyses

Genome-wide association study.

We followed the same protocol we had developed for the first genome-wide association study (GWAS) with imaging carried out on UK Biobank 19 . Briefly, we examined imputed UK Biobank genotype data 47 , and restricted the analysis to samples that were unrelated (thereby setting aside only ~450 participants), without aneuploidy and with recent UK ancestry. To account for population stratification, 40 genetic principal components were used in the genetic association tests as is recommended for UK Biobank genetic studies 19 , 20 , 47 . We excluded genetic variants with minor allele frequency <0.01 or INFO score <0.03 or Hardy-Weinberg equilibrium –log 10 ( P ) > 7. We then randomly split the samples into a discovery set with 2/3 of the samples ( n  = 22,128) and a replication set with 1/3 of the samples ( n  = 11,083). We also examined the X chromosome with the same filters, additionally excluding participants with sex chromosome aneuploidy: 12 in non-pseudoautosomal region (PAR) and 9 in PAR for the discovery set, 3 in non-PAR and 6 in PAR for the replication set. Variants were considered significant at –log 10 ( P ) > 7.5, and replicated at P  < 0.05.

Modifiable risk factor study

In the first stage, the general linear model was used to investigate, separately, the association between each of these 161 MRFs and the LIFO network loadings in all the scanned UK Biobank participants ( n  = 39,676). We ran each model twice: once as is, and once adding 6 confounders: age, age 2 , sex, age × sex, age 2 × sex, and head size, to estimate the contribution of these MRFs on the LIFO network above and beyond the dominating effects of age and sex. Sex was based on the population characteristics entry of UK Biobank. This is a mixture of the sex the NHS had recorded for the participant at recruitment, and updated self-reported sex. For the GWAS, both sex and genetic sex were used (the sample was excluded in case of a mismatch). In total, 32 variables tailored to structural imaging had been considered as possible confounders, and we retained those with the strongest association ( R 2  ≥ 0.01; see  Supplementary Information ). Socioeconomic status via the Townsend deprivation index was also considered as a possible confounding variable but explained little variance ( R 2  < 0.001) and thus was not included as a confounder.

MRFs were not considered further if they were not significant—not surviving Bonferroni-correction, i.e., P  > 1.55 × 10 −4 —and if more than 5% of the subjects had their MRF values missing. For each category, a single ‘best’ MRF was then selected as the variable with the highest R 2 among those remaining, after regressing out the confounding effects of age and sex.

In the second stage, all these best MRFs were then included in a single general linear model, together with the same 6 confounders used in the first stage, to assess the unique contribution of each factor on the LIFO brain network loadings. A prerequisite to carry out this single general linear model analysis was to only include participants who would have values for all best MRFs and confounders. This explains the additional criterion of only including MRFs that had no more than 5% of values missing, to ensure that the final sample of participants who had values for all these best and confounding factors would not be biased compared with the original sample—something we formally tested (see  Supplementary Information )—especially as data are not missing at random in UK Biobank, and exhibit some genetic structure 48 . The sample was therefore reduced to a total of 35,527 participants for this second stage analysis (M/F 17,290–18,237; 45–82 years, mean 64 ± 7 years). The effect of these best MRFs taken altogether was considered significant with a very conservative Bonferroni correction for multiple comparisons across all combinations of every possible MRF from each of the initial 15 MRF categories ( P  < 4.62 × 10 −17 , see  Supplementary Information for more details). In addition, both full and partial correlations were computed for the same set of best MRFs and confounders, in order to assess possible relationships between variables.

Post hoc genetic analyses

Chromosome 17 inversion.

We investigated chromosome 17 inversion status of the participants in the discovery cohort by considering their genotype on 32 variants that tag chromosome 17 inversion according to Steinberg et al. 11 . Of these 32 variants, 24 were present in our genetic data. We labelled the participants homozygous inverted, heterozygous, or homozygous direct (not inverted) when all 24 of these alleles indicated the same zygosity. This yielded an unambiguous inversion status for 21,969 participants (99% of the discovery cohort). To examine if the association between the non-triallelic lead variant of the MAPT genetic cluster (rs2532395, Table  1 /Supplementary Data  1 ) and the LIFO network was independent from this common inversion, we determined inversion/direct status of the discovery cohort and: 1. repeated the association test between rs2532395 and the LIFO phenotype, with chromosome 17 inversion status added as a confounder; and 2. correlated the genotype for rs2532395 with chromosome 17 inversion.

Causality within each genetic cluster

We used CAVIAR (Causal Variants Identification in Associated Regions 49 ) to assess causality of variants that passed the genome-wide significance threshold in each of the genetic clusters we report. CAVIAR uses a Bayesian model and the local linkage disequilibrium structure to assign posterior probabilities of causality to each variant in a region, given summary statistics for an association. We did not perform CAVIAR analysis on the genetic cluster on chromosome 17, as its non-triallelic lead variant (rs2532395) was strongly correlated with chromosome 17 inversion, and the LD matrix was large and low rank. We excluded the X chromosome loci from this analysis due to the difficulty in assessing LD in this chromosome.

Enrichment analysis

Based on the genes listed in the ‘Genes’ column of Table  1 /Supplementary Data  1 , we performed an enrichment analysis for the genes associated with the LIFO brain network using PANTHER 50 . PANTHER determines whether a gene function is overrepresented in a set of genes, according to the gene ontology consortium 51 , 52 .

Mediation analysis between MAPT top variant and Alzheimer’s disease, via the LIFO brain network

As the gene MAPT is associated with Alzheimer’s disease, and as we found a significant association between MAPT and the LIFO brain network, we examined to what extent the effect of MAPT is mediated by the LIFO brain regions. We conducted a mediation analysis using the counterfactual framework in which the average indirect effect of the treatment on the outcome through the mediator is nonparametrically identified (version 4.5.0 of the R package ‘mediation' 53 ). This is a general approach that encompasses the classical linear structural equation modelling framework for causal mediation, allowing both linear and non-linear relationships. In this analysis, the genotype for the lead bi-allelic variant of the MAPT association was used as the treatment, the LIFO loadings as the mediator, and Alzheimer’s disease diagnosis as the outcome.

From the ~43 K UK Biobank participants who had been scanned, we searched for those who had been diagnosed with Alzheimer’s disease specifically, regardless of whether this diagnosis occurred before, or after their brain scans. Based on hospital inpatient records (ICD10: F000, F001, F002, F009, G300, G301, G308, and G309 and ICD9: 3310) and primary care (GP) data (Eu00., Eu000, Eu001, Eu002, Eu00z, F110., F1100, F1101, Fyu30, X002x, X002y, X002z, X0030, X0031, X0032, X0033, XaIKB, XaIKC, and XE17j), we identified 65 such cases— UK Biobank being healthier than the general population, and those scanned showing an even stronger healthy bias—of which 34 were included in the discovery set after QC.

We considered two conditions for the effect of the treatment on the outcome. First, a dominant condition in which the minor allele is assumed to be dominant and for which at least one copy of the minor allele is considered treated. Second, a recessive condition in which the minor allele is assumed to be recessive. We considered that either condition was nominally significant if the confidence interval of the average causal mediated effect did not intersect zero, and had an associated P  < 0.05 ÷ 2 (correcting for the two conditions). We assessed confidence intervals and P -values using 50,000 bootstrapped samples.

Associations between the LIFO brain network’s genetic hits and the MRFs

First, we reported in Table  1 / Supplementary Data  1 the significant associations between the LIFO genetic hits and UK Biobank variables related to the 15 categories listed for the MRFs. For this, we used the Open Targets Genetics website, which reports the GWAS carried out in UK Biobank ( https://genetics .opentargets.org/ ). Second, we assessed whether there was any genetic overlap between the known genetic components of the 3 best MRFs and the LIFO phenotype. Again, we used the Open Targets Genetics website outputs for these 3 very specific UK Biobank variables, and compared the significant hits for these 3 best MRFs within ±250 kbp of, or in high LD (>0.8) with, our own LIFO variants. If reported hits were limited, we also searched online for GWAS done on similar variables. Finally, we also included the list of significant hits for diabetes 54 , which focused on a potential genetic overlap between diabetes and Alzheimer’s disease.

Post hoc association for the sex chromosomes variants

The allele counts of each participant for two specific significant variants of the sex chromosomes not—or hardly—available in open databases such as https://genetics.opentargets.org/ 55 were further associated out-of-sample with all non-imaging phenotypes of UK Biobank ( n  = 16,924). This analysis was carried out in the entire genotyped, quality-controlled sample where participants who had been scanned were removed (final sample: 374,230 participants), taking into account the population structure (40 genetic principal components), as well as the confounding effects of age, sex, age x sex, age 2 and age 2 x sex. Results were corrected for multiple comparisons across all non-imaging phenotypes and the two variants.

Heritability

We examined the heritability of the LIFO phenotype, and the coheritability between the LIFO network and Alzheimer’s disease or schizophrenia using LDSC 56 . This method uses regression on summary statistics to determine narrow sense heritability h 2 of a trait, or the shared genetic architecture between two traits. LDSC corrects for bias LD structure using LD calculated from a reference panel (we used LD from the Thousand Genomes Project Phase 1 57 ). We obtained summary statistics for a meta-analysis of Alzheimer’s disease involving 71,880 cases and 383,378 controls 58 . The number of genetic variants in the intersection between the summary statistics was 1,122,435. For schizophrenia, the summary statistics were obtained from a meta-analysis involving 53,386 cases and 77,258 controls 59 . A total of 1,171,319 genetic variants were in the intersection with the summary statistics for LIFO. For both Alzheimer’s and schizophrenia, the X chromosome was not included in the heritability calculation, as it was excluded from the meta-analysis that we sourced the summary statistics from.

Reproducibility

No data was excluded for the MRF analyses. For the genetic analyses, these were restricted to samples that were unrelated, without aneuploidy and with recent UK ancestry (see above).

No statistical method was used to predetermine sample size. The experiments were not randomised. The Investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

Data availability

All the FLICA decomposition maps − including the LIFO grey matter network − in UK Biobank standard space, the UK Biobank grey matter template, scripts, and the LIFO loadings for all of the participants are freely available on a dedicated webpage: open.win.ox.ac.uk/pages/douaud/ukb-lifo-flica/ .

Baumgart, M. et al. Summary of the evidence on modifiable risk factors for cognitive decline and dementia: a population-based perspective. Alzheimers Dement. 11 , 718–726 (2015).

Article   PubMed   Google Scholar  

Chen, H. et al. Exposure to ambient air pollution and the incidence of dementia: a population-based cohort study. Environ. Int. 108 , 271–277 (2017).

Article   CAS   PubMed   Google Scholar  

Livingston, G. et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet 396 , 413–446 (2020).

Article   PubMed   PubMed Central   Google Scholar  

Desikan, R. S. et al. Genetic overlap between Alzheimer’s disease and Parkinson’s disease at the MAPT locus. Mol. Psychiatry 20 , 1588–1595 (2015).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Hill, W. D. et al. A combined analysis of genetically correlated traits identifies 187 loci and a role for neurogenesis and myelination in intelligence. Mol. Psychiatry 24 , 169–181 (2019).

Pardinas, A. F. et al. Common schizophrenia alleles are enriched in mutation-intolerant genes and in regions under strong background selection. Nat. Genet. 50 , 381–389 (2018).

Douaud, G. et al. Preventing Alzheimer’s disease-related gray matter atrophy by B-vitamin treatment. Proc. Natl Acad. Sci. USA 110 , 9523–9528 (2013).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Douaud, G. et al. A common brain network links development, aging, and vulnerability to disease. Proc. Natl Acad. Sci. USA 111 , 17648–17653 (2014).

Nakatsu, D. et al. BMP4-SMAD1/5/9-RUNX2 pathway activation inhibits neurogenesis and oligodendrogenesis in Alzheimer’s patients’ iPSCs in senescence-related conditions. Stem Cell Rep. 18 , 1246 (2023).

Article   CAS   Google Scholar  

Peyrot, W. J. & Price, A. L. Identifying loci with different allele frequencies among cases of eight psychiatric disorders using CC-GWAS. Nat. Genet. 53 , 445–454 (2021).

Steinberg, K. M. et al. Structural diversity and African origin of the 17q21.31 inversion polymorphism. Nat. Genet. 44 , 872–880 (2012).

Le Guen, Y. et al. eQTL of KCNK2 regionally influences the brain sulcal widening: evidence from 15,597 UK Biobank participants with neuroimaging data. Brain Struct. Funct. 224 , 847–857 (2019).

Jonsson, B. A. et al. Brain age prediction using deep learning uncovers associated sequence variants. Nat. Commun. 10 , 5409 (2019).

Heurteaux, C. et al. TREK-1, a K+ channel involved in neuroprotection and general anesthesia. EMBO J. 23 , 2684–2695 (2004).

Vojinovic, D. et al. Genome-wide association study of 23,500 individuals identifies 7 loci associated with brain ventricular volume. Nat. Commun. 9 , 3945 (2018).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Zhao, B. et al. Genome-wide association analysis of 19,629 individuals identifies variants influencing regional brain volumes and refines their genetic co-architecture with cognitive and mental health traits. Nat. Genet. 51 , 1637–1644 (2019).

Zhao, B. et al. Large-scale GWAS reveals genetic architecture of brain white matter microstructure and genetic overlap with cognitive and mental health traits ( n = 17,706). Mol. Psychiatry https://doi.org/10.1038/s41380-019-0569-z (2019).

Lasagna-Reeves, C. A. et al. Reduction of Nuak1 decreases tau and reverses phenotypes in a tauopathy mouse model. Neuron 92 , 407–418 (2016).

Elliott, L. T. et al. Genome-wide association studies of brain imaging phenotypes in UK Biobank. Nature 562 , 210–216 (2018).

Smith, S. M. et al. An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank. Nat. Neurosci. 24 , 737–745 (2021).

Lee, J. J. et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nat. Genet. 50 , 1112–1121 (2018).

Teslovich, T. M. et al. Biological, clinical and population relevance of 95 loci for blood lipids. Nature 466 , 707–713 (2010).

Sanchez-Roige, S. et al. Genome-Wide Association Study Meta-Analysis of the Alcohol Use Disorders Identification Test (AUDIT) in Two Population-Based Cohorts. Am. J. Psychiatry 176 , 107–118 (2019).

Locke, A. E. et al. Genetic studies of body mass index yield new insights for obesity biology. Nature 518 , 197–206 (2015).

Dashti, H. S. et al. Genome-wide association study identifies genetic loci for self-reported habitual sleep duration supported by accelerometer-derived estimates. Nat. Commun. 10 , 1100 (2019).

International Consortium for Blood Pressure Genome-Wide Association, S. et al. Genetic variants in novel pathways influence blood pressure and cardiovascular disease risk. Nature 478 , 103–109 (2011).

Article   ADS   Google Scholar  

Johansson, A. et al. Genome-wide association and Mendelian randomization study of NT-proBNP in patients with acute coronary syndrome. Hum. Mol. Genet. 25 , 1447–1456 (2016).

Poorkaj, P. et al. Tau is a candidate gene for chromosome 17 frontotemporal dementia. Ann. Neurol. 43 , 815–825 (1998).

Baker, M. et al. Association of an extended haplotype in the tau gene with progressive supranuclear palsy. Hum. Mol. Genet. 8 , 711–715 (1999).

Johnson, N. C. XG: the forgotten blood group system. Immunohematology 27 , 68–71 (2011).

Kong, A. et al. The nature of nurture: Effects of parental genotypes. Science 359 , 424–428 (2018).

Zenaro, E., Piacentino, G. & Constantin, G. The blood-brain barrier in Alzheimer’s disease. Neurobiol. Dis. 107 , 41–56 (2017).

Pong, S., Karmacharya, R., Sofman, M., Bishop, J. R. & Lizano, P. The role of brain microvascular endothelial cell and blood-brain barrier dysfunction in schizophrenia. Complex Psychiatry 6 , 30–46 (2020).

Chatterjee, S. et al. Type 2 diabetes as a risk factor for dementia in women compared with men: a pooled analysis of 2.3 million people comprising more than 100,000 cases of dementia. Diabetes Care 39 , 300–307 (2016).

Veldsman, M. et al. Cerebrovascular risk factors impact frontoparietal network integrity and executive function in healthy ageing. Nat. Commun. 11 , 4340 (2020).

Power, M. C., Adar, S. D., Yanosky, J. D. & Weuve, J. Exposure to air pollution as a potential contributor to cognitive function, cognitive decline, brain imaging, and dementia: a systematic review of epidemiologic research. Neurotoxicology 56 , 235–253 (2016).

Grande, G., Ljungman, P. L. S., Eneroth, K., Bellander, T. & Rizzuto, D. Association between cardiovascular disease and long-term exposure to air pollution with the risk of dementia. JAMA Neurol. 77 , 801–809, (2020).

Ronnlund, M., Nyberg, L., Backman, L. & Nilsson, L. G. Stability, growth, and decline in adult life span development of declarative memory: cross-sectional and longitudinal data from a population-based study. Psychol. Aging 20 , 3–18 (2005).

Vidal-Pineiro, D. et al. Individual variations in ‘brain age’ relate to early-life factors more than to longitudinal brain change. Elife https://doi.org/10.7554/eLife.69995 (2021).

Fry, A. et al. Comparison of sociodemographic and health-related characteristics of UK Biobank participants with those of the general population. Am. J. Epidemiol. 186 , 1026–1034 (2017).

Alfaro-Almagro, F. et al. Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. NeuroImage 166 , 400–424 (2018).

Miller, K. L. et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nat. Neurosci. 19 , 1523–1536 (2016).

Good, C. D. et al. A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage 14 , 21–36 (2001).

Douaud, G. et al. Anatomically related grey and white matter abnormalities in adolescent-onset schizophrenia. Brain 130 , 2375–2386 (2007).

Groves, A. R. et al. Benefits of multi-modal fusion analysis on a large-scale dataset: life-span patterns of inter-subject variability in cortical morphometry and white matter microstructure. NeuroImage 63 , 365–380 (2012).

Smith, S. et al. Structural variability in the human brain reflects fine-grained functional architecture at the population level. J. Neurosci. 39 , 6136–6149 (2019).

Bycroft, C. et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 562 , 203–209 (2018).

Mignogna, G. et al. Patterns of item nonresponse behaviour to survey questionnaires are systematic and associated with genetic loci. Nat. Hum. Behav . https://doi.org/10.1038/s41562-023-01632-7 (2023).

Hormozdiari, F., Kostem, E., Kang, E. Y., Pasaniuc, B. & Eskin, E. Identifying causal variants at loci with multiple signals of association. Genetics 198 , 497–508 (2014).

Thomas, P. D. et al. PANTHER: Making genome-scale phylogenetics accessible to all. Protein Sci. 31 , 8–22 (2022).

Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25 , 25–29 (2000).

Gene Ontology, C. et al. The Gene Ontology knowledgebase in 2023. Genetics https://doi.org/10.1093/genetics/iyad031 (2023).

Imai, K., Keele, L. & Tingley, D. A general approach to causal mediation analysis. Psychol. Methods 15 , 309–334 (2010).

Hardy, J., de Strooper, B. & Escott-Price, V. Diabetes and Alzheimer’s disease: shared genetic susceptibility? Lancet Neurol. 21 , 962–964 (2022).

Carvalho-Silva, D. et al. Open Targets Platform: new developments and updates two years on. Nucleic Acids Res. 47 , D1056–D1065 (2019).

Gazal, S., Marquez-Luna, C., Finucane, H. K. & Price, A. L. Reconciling S-LDSC and LDAK functional enrichment estimates. Nat. Genet. 51 , 1202–1204 (2019).

Genomes Project, C. et al. A global reference for human genetic variation. Nature 526 , 68–74 (2015).

Jansen, I. E. et al. Author Correction: Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk. Nat. Genet. 52 , 354 (2020).

Trubetskoy, V. et al. Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature 604 , 502–508 (2022).

Laird, A. R. et al. ALE meta-analysis workflows via the brainmap database: progress towards a probabilistic functional brain atlas. Front. Neuroinform. 3 , 23 (2009).

Download references

Acknowledgements

We are grateful to Profs Christian K. Tamnes, Lars T. Westlye, Kristine B. Walhovd and Anders M. Fjell, and Dr Andreas Engvig for providing the lifespan cohort which was used to initially derive the original ‘last in, first out’ brain network map, and to Prof Augustine Kong for helpful discussion on the associations between the PAR hit and early life and environmental factors. G.D. was supported by a UK MRC Career Development Fellowship (MR/K006673/1) and a Wellcome Collaborative Award (215573/Z/19/Z). S.S. was supported by Wellcome (203139/Z/16/Z; 215573/Z/19/Z). L.E. was funded by NSERC grants (RGPIN/05484-2019; DGECR/00118-2019) and a Michael Smith Health Research BC Scholar Award. A.M.W. received support through the NIH Intramural Research Program (ZIA-MH002781; ZIA-MH002782). This research was funded in whole, or in part, by the Wellcome Trust (215573/Z/19/Z; 203139/Z/16/Z; 203139/A/16/Z). For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. This research was also supported by the NIHR Oxford Health Biomedical Research Centre (NIHR203316). The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. The Wellcome Centre for Integrative Neuroimaging is supported by core funding from the Wellcome Trust (203139/Z/16/Z and 203139/A/16/Z).

Author information

These authors contributed equally: Lloyd T. Elliott, Anderson M. Winkler.

Authors and Affiliations

FMRIB Centre, Wellcome Centre for Integrative Neuroimaging (WIN), Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK

Jordi Manuello, Paul McCarthy, Fidel Alfaro-Almagro, Soojin Lee, Stephen Smith & Gwenaëlle Douaud

FOCUS Lab, Department of Psychology, University of Turin, Turin, Italy

Jordi Manuello

Department of Statistics and Actuarial Science, Simon Fraser University, Burnaby, Canada

Joosung Min & Lloyd T. Elliott

Pacific Parkinson’s Research Centre, The University of British Columbia, Vancouver, BC, Canada

National Institutes of Mental Health, National Institutes of Health, Bethesda, MD, USA

Anderson M. Winkler

Department of Human Genetics, University of Texas Rio Grande Valley, Brownsville, TX, USA

You can also search for this author in PubMed   Google Scholar

Contributions

G.D. conceived and supervised the work, and carried out some of the genetic and modifiable risk factors analyses. J.Ma. carried out most of the genetic and modifiable risk factors analyses. J.Mi., S.L., A.M.W., and L.T.E. carried out additional genetics analyses. G.D., P. McC., F.A.-A., S.S., and L.T.E. created/extracted the imaging and genetics data, and organised the non-imaging data and confound variables. L.T.E. co-supervised the genetic analyses. A.M.W. co-supervised the modifiable risk factor analyses. G.D. interpreted the results and wrote the paper. J.Ma., S.S., L.T.E., and A.M.W. revised the paper.

Corresponding author

Correspondence to Gwenaëlle Douaud .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature Communications thanks Xavier Caseras and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information, peer review file, description of additional supplementary files, supplementary data 1-5, reporting summary, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Manuello, J., Min, J., McCarthy, P. et al. The effects of genetic and modifiable risk factors on brain regions vulnerable to ageing and disease. Nat Commun 15 , 2576 (2024). https://doi.org/10.1038/s41467-024-46344-2

Download citation

Received : 15 February 2023

Accepted : 22 February 2024

Published : 27 March 2024

DOI : https://doi.org/10.1038/s41467-024-46344-2

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

• Entertainment
• Celebrities
• Ticket Sales
• Promoted: What to Watch on Prime Video

Jay Leno officially granted conservatorship of wife Mavis amid her dementia diagnosis and ‘difficult period’

• View Author Archive
• Get author RSS feed

Thanks for contacting us. We've received your submission.

Jay Leno has been officially granted conservatorship of his wife, Mavis Leno, who was diagnosed with dementia.

Mavis’ lawyer said she was “in agreement” with the conservatorship during a hearing at Los Angeles County Superior Court on Tuesday, People reports. Mavis added that she’s “receiving excellent care with her husband, Mr. Leno.”

“I think she’s in the least restrictive environment,” the judge agreed. “I think she’s in very good care with Mr. Leno.”

The judge then told Leno, 73, “everything you’re doing is right” and “I totally understand this is a difficult period.” He added that the TV personality is “fit” for the role.

The former “Tonight Show” host went on to note that he is in the process of estate planning and looking at his wife’s will.

Earlier this month, court documents revealed that Mavis, 77, “sometimes does not know her husband nor her date of birth” amid her health struggles. The statement, which was made by Mavis’ court-appointed attorney, Robert Ostrin, was filed in Los Angeles as an update to an earlier petition to allow Leno to be her conservator.

Additionally, the report explained that Mavis “has a lot of disorientation” and “will ruminate about her parents who have both passed and her mother who died about 20 years ago.”

“Based on my interaction with Mr. Leno, plus some research I have done, he seems to be a standup guy,” Ostrin said. “And his private persona matches the public persona he projects.”

In January, Leno told Daily Mail he had plans to set up her will. “She’s fine. Everybody’s good, we’re doing well,” the comedian said at the time. “[I] set up a will, in case something happens.”

The couple married in 1980 after meeting at the Comedy Store in the ’70s. They do not have any children.

Share this article:

Advertisement