Dementia Risk Prediction Research Articles

Genetic risk scores and dementia risk across different ethnic groups in UK Biobank.

Genetic Risk Scores (GRS) for predicting dementia risk have mostly been used in people of European ancestry with limited testing in other ancestry groups. We conducted a logistic regression with all-cause dementia as the outcome and z-standardised GRS as the exposure across diverse ethnic groups. There was variation in frequency of APOE alleles across ethnic groups. Per standard deviation (SD) increase in z-GRS including APOE, the odds ratio (OR) for dementia was 1.73 (95%CI 1.69-1.77). Z-GRS excluding APOE also increased dementia risk (OR 1.21 per SD increase, 95% CI 1.18-1.24) and there was no evidence that ethnicity modified this association. Prediction of secondary outcomes was less robust in those not of European ancestry when APOE was excluded from the GRS. z-GRS derived from studies in people of European ancestry can be used to quantify genetic risk in people from more diverse ancestry groups. Urgent work is needed to include people from diverse ancestries in future genetic risk studies to make this field more inclusive.

Validity of three risk prediction models for dementia or cognitive impairment in Australia.

no studies have compared the predictive validity of different dementia risk prediction models in Australia. (i) to investigate the predictive validity of the Australian National University-Alzheimer's Disease Risk Index (ANU-ADRI), LIfestyle for BRAin Health (LIBRA) Index and cardiovascular risk factors, ageing and dementia study (CAIDE) models for predicting probable dementia/cognitive impairment in an Australian cohort. (ii) To develop and assess the predictive validity of a new hybrid model combining variables from the three models. the Hunter Community Study (HCS) included 3,306 adults aged 55-85years with a median follow-up of 7.1years. Probable dementia/cognitive impairment was defined using Admitted Patient Data Collection, dispensing of cholinesterase inhibitors or memantine, or a cognitive test. Model validity was assessed by calibration and discrimination. A hybrid model was developed using deep neural network analysis, a machine learning method. 120 (3.6%) participants developed probable dementia/cognitive impairment. Mean calibration by ANU-ADRI, LIBRA, CAIDE and the hybrid model was 19, 0.5, 4.7 and 3.4%, respectively. The discrimination of the models was 0.65 (95% CI 0.60-0.70), 0.65 (95% CI 0.60-0.71), 0.54 (95% CI 0.49-0.58) and 0.80 (95% CI 0.78-0.83), respectively. ANU-ADRI and LIBRA were better dementia prediction tools than CAIDE for identification of high-risk individuals in this cohort. ANU-ADRI overestimated and LIBRA underestimated the risk. The new hybrid model had a higher predictive performance than the other models but it needs to be validated independently in longitudinal studies.

Improved risk prediction of all‐cause dementia, Alzheimer’s disease, and vascular dementia by inflammation‐related biomarkers

AbstractBackgroundDementia risk prediction models, such as the widely used Cardiovascular Risk Factors, Aging and Dementia (CAIDE) model, combining factors from the fields of demographic, cognition, physical and health, have proven themselves. However, it is of interest whether inflammatory biomarkers additionally improve such dementia risk prediction models.MethodsThe Olink Target 96 Inflammation panel was assessed in serum samples of a large German, population‐based cohort study. In a nested case‐cohort design (n=1688), the CAIDE model was used for dementia risk prediction and aimed to improve by adding 72 inflammation‐related biomarkers of the Olink panel. Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression and bootstrapping were utilized to select relevant biomarkers and determine areas under the curve (AUCs) for all‐cause dementia, Alzheimer’s disease (AD) and vascular dementia (VD) in the total cohort, as well as in mid‐life and late‐life sub‐samples.ResultsOver 17 years of follow‐up, 440 participants developed dementia. Of those, 147 were diagnosed with AD and 195 with VD. The prediction model including APOE ε4 carrier status (CAIDE model 2) had a better discriminative ability for all‐cause dementia, AD, and VD than the model without APOE ε4 (CAIDE model 1) with AUCs of 0.751, 0.767, and 0.725, respectively. Adding 16 inflammation‐related biomarkers selected by LASSO regression to the CAIDE model significantly increased the AUC for all‐cause dementia risk prediction by 0.025 (95%CI: 0.001–0.045). The AUC increases for AD (0.024) and VD (0.019) were of similar magnitude but not statistically significant due to lower case numbers. Moreover, the CAIDE model generally performed better in mid‐life (50‐64 years) than in late‐life (65‐75 years) sub‐samples, and all increases in the AUC caused by inflammation‐related biomarkers were larger in the mid‐life sub‐sample.ConclusionInflammation‐related biomarkers significantly improved the CAIDE model for all‐cause dementia prediction. Especially the predictive abilities of these blood‐based biomarkers in mid‐life are of clinical relevance to guide early preventive measures.

Algorithmic Fairness in Biomarker‐Based Machine Learning Models to Predict Alzheimer’s Dementia in Individuals with Mild Cognitive Impairment

AbstractBackgroundWith increasing availability of large‐scale data sets in biomedical research, machine learning (ML) algorithms rapidly gain importance for research and clinical practice. However, ML models bear the risk of increasing health disparities, whenever trained with imbalanced or biased data. Based on these ethical concerns the WHO guidelines on Artificial Intelligence for Healthcare recommend incorporating algorithmic fairness in ML models by design. Although research on biomarker detection and dementia risk prediction in Alzheimer’s disease increasingly employs ML algorithms, the fairness of these models has not been in focus to date.MethodUsing data from the Alzheimer’s Disease Neuroimaging Initiative database, we generated ML models to predict conversion to Alzheimer’s dementia in 1017 individuals with mild cognitive impairment. Two models were trained based on (1) biomarkers and (2) biomarkers and demographic features. Subsequently, fairness metrics were calculated to investigate prediction disparities in the trained models regarding sensitive attributes sex, education, race and ethnicity.ResultThe models did not show substantial differences in predictive performance. Compared to Non‐Hispanic‐White individuals, true positive rates for Hispanics and for Non‐White individuals were lower in the biomarker model. Inclusion of demographic features to the model led to a further drop of true positive rates and positive predictive values for both Hispanics and Non‐White individuals. None of the models showed any predictive disparities regarding sex and education, with both models having a slightly higher accuracy, positive predictive value, and negative predictive value for women and for individuals with lower educational attainment.ConclusionThis study empirically demonstrates the potential ethical hazards of ML models for the prediction of conversion to Alzheimer’s dementia. While ML models exhibit fair prognostication when based on balanced data regarding sensitive attributes, underrepresentation of specific groups may lead to violations of fairness criteria with important implications for the clinical translation of these models. Such fairness disparities can be present in models based on only biological data and the inclusion of demographic information might exacerbate the predictive disparities in ML‐models.

A Population‐Based Meta‐Analysis of Plasma Glial Fibrillary Acidic Protein for Cognitive Decline and Incident Dementia

AbstractBackgroundExpression of glial fibrillary acidic protein (GFAP), a marker of reactive astrocytosis, co‐localizes with neuropathology in the brain. Blood‐derived levels of GFAP have been associated with cognitive decline and dementia status. However, further examinations at a population‐based level are necessary to broaden generalizability and assess the biomarker’s potential efficacy in community settings.MethodPlasma GFAP levels were assayed using a Simoa HD‐1 analyzer in 4,338 adults without prevalent dementia from four longitudinal cohort studies, the Framingham Heart Study (FHS), the Cardiovascular Health Study (CHS), the Age, Gene/Environment Susceptibility – Reykjavik Study (AGES), and the Coronary Artery Risk Development in Young Adults Study (CARDIA). The associations between plasma GFAP levels with a general cognition score derived from multiple cognitive tests, total brain volume, and hippocampal volume were evaluated with separate linear regression models in each cohort with adjustment for age, sex, education, race, diabetes, systolic blood pressure, antihypertensive medication, body mass index, apolipoprotein ε4 status, site, and time between GFAP blood draw and the outcome. Associations with incident all‐cause and Alzheimer’s disease dementia were evaluated with adjusted Cox proportional hazard models. Meta‐analysis was performed on the estimates derived from each cohort using random effects models with inverse variance weighting. The Sidik‐Jonkman estimator was used to report overall associations across studies.ResultMeta‐analysis indicated that higher plasma GFAP associated with lower general cognition (ß=‐0.09, [95% confidence interval [CI]: ‐0.15 to ‐0.03], p=0.005), but not with total brain or hippocampal volume (p>0.05 for all). However, higher GFAP was significantly associated with a 2.5‐fold higher risk of incident all‐cause (Hazard Ratio [HR]: 2.47 (95% CI: 1.52‐4.01)) and Alzheimer’s disease dementia (HR: 2.54 (95% CI: 1.42‐4.53) per standard deviation increase over up to 15‐years of follow‐up.ConclusionThe findings suggest that elevations in plasma GFAP may be detectable several years prior to a clinical diagnosis, highlighting the potential value of the biomarker for aiding dementia risk prediction at a population‐based level

The Split-Half Reliability and Construct Validity of the Virtual Reality-Based Path Integration Task in the Healthy Population.

The virtual reality (VR)-based path integration task shows substantial promise in predicting dementia risk. However, the reliability and validity in healthy populations need further exploration. The present study investigates the relationship between task indicators and brain structures in a healthy population using a VR-based navigation task, particularly the entorhinal cortex (EC) and hippocampus. Sixty healthy adults were randomly recruited to perform a VR-based path integration task, the digit span task (DST), and an MRI scan. The indicators of the VR-based path integration task were calculated, including the absolute distance error (ADE), degree of angle deviation (DAD), degree of path deviation (DPD), and return time (Time). The reliability of the above indicators was then estimated using the split-half method and Cronbach's alpha. Correlation and regression analyses were then performed to examine the associations between these indicators and age, general cognitive ability (DST), and brain structural measures. ADE, DAD, and DPD showed reasonable split-half reliability estimates (0.84, 0.81, and 0.72) and nice Cronbach's alpha estimates (0.90, 0.86, and 0.96). All indicators correlated with age and DST. ADE and DAD were sensitive predictors of hippocampal volume, and return time was a predictor of EC thickness. Our findings demonstrate that the VR-based path integration task exhibits good reliability and validity in the healthy population. The task indicators are age-sensitive, can capture working memory capacity, and are closely related to the integrity of individual EC and hippocampal structures.

Dementia risk prediction in individuals with mild cognitive impairment: a comparison of Cox regression and machine learning models

BackgroundCox proportional hazards regression models and machine learning models are widely used for predicting the risk of dementia. Existing comparisons of these models have mostly been based on empirical datasets and have yielded mixed results. This study examines the accuracy of various machine learning and of the Cox regression models for predicting time-to-event outcomes using Monte Carlo simulation in people with mild cognitive impairment (MCI).MethodsThe predictive accuracy of nine time-to-event regression and machine learning models were investigated. These models include Cox regression, penalized Cox regression (with Ridge, LASSO, and elastic net penalties), survival trees, random survival forests, survival support vector machines, artificial neural networks, and extreme gradient boosting. Simulation data were generated using study design and data characteristics of a clinical registry and a large community-based registry of patients with MCI. The predictive performance of these models was evaluated based on three-fold cross-validation via Harrell’s concordance index (c-index), integrated calibration index (ICI), and integrated brier score (IBS).ResultsCox regression and machine learning model had comparable predictive accuracy across three different performance metrics and data-analytic conditions. The estimated c-index values for Cox regression, random survival forests, and extreme gradient boosting were 0.70, 0.69 and 0.70, respectively, when the data were generated from a Cox regression model in a large sample-size conditions. In contrast, the estimated c-index values for these models were 0.64, 0.64, and 0.65 when the data were generated from a random survival forest in a large sample size conditions. Both Cox regression and random survival forest had the lowest ICI values (0.12 for a large sample size and 0.18 for a small sample size) among all the investigated models regardless of sample size and data generating model.ConclusionCox regression models have comparable, and sometimes better predictive performance, than more complex machine learning models. We recommend that the choice among these models should be guided by important considerations for research hypotheses, model interpretability, and type of data.

Validity and Performance of Blood Biomarkers for Alzheimer Disease to Predict Dementia Risk in a Large Clinic-Based Cohort.

Blood biomarkers for Alzheimer disease (AD) have consistently proven to be associated with CSF or PET biomarkers and effectively discriminate AD from other neurodegenerative diseases. Our aim was to test their utility in clinical practice, from a multicentric unselected prospective cohort where patients presented with a large spectrum of cognitive deficits or complaints. The MEMENTO cohort enrolled 2,323 outpatients with subjective cognitive complaint (SCC) or mild cognitive impairment (MCI) consulting in 26 French memory clinics. Participants had neuropsychological assessments, MRI, and blood sampling at baseline. CSF sampling and amyloid PET were optional. Baseline blood Aβ42/40 ratio, total tau, p181-tau, and neurofilament light chain (NfL) were measured using a Simoa HD-X analyzer. An expert committee validated incident dementia cases during a 5-year follow-up period. Overall, 2,277 individuals had at least 1 baseline blood biomarker available (n = 357 for CSF subsample, n = 649 for PET subsample), among whom 257 were diagnosed with clinical AD/mixed dementia during follow-up. All blood biomarkers but total tau were mildly correlated with their equivalence in the CSF (r = 0.33 to 0.46, p < 0.0001) and were associated with amyloid-PET status (p < 0.0001). Blood p181-tau was the best blood biomarker to identify amyloid-PET positivity (area under the curve = 0.74 [95% CI = 0.69; 0.79]). Higher blood and CSF p181-tau and NfL concentrations were associated with accelerated time to AD dementia onset with similar incidence rates, whereas blood Aβ42/40 was less efficient than CSF Aβ42/40. Blood p181-tau alone was the best blood predictor of 5-year AD/mixed dementia risk (c-index = 0.73 [95% CI = 0.69; 0.77]); its accuracy was higher in patients with clinical dementia rating (CDR) = 0 (c-index = 0.83 [95% CI = 0.69; 0.97]) than in patients with CDR = 0.5 (c-index = 0.70 [95% CI = 0.66; 0.74]). A "clinical" reference model (combining demographics and neuropsychological assessment) predicted AD/mixed dementia risk with a c-index = 0.88 [95% CI = 0.86-0.91] and performance increased to 0.90 [95% CI = 0.88; 0.92] when adding blood p181-tau + Aβ42/40. A "research" reference model (clinical model + apolipoprotein E genotype and AD signature on MRI) had a c-index = 0.91 [95% CI = 0.89-0.93] increasing to 0.92 [95% CI = 0.90; 0.93] when adding blood p181-tau + Aβ42/40. Chronic kidney disease and vascular comorbidities did not affect predictive performances. In a clinic-based cohort of patients with SCC or MCI, blood biomarkers may be good hallmarks of underlying pathology but add little to 5-year dementia risk prediction models including traditional predictors.

Development of a novel dementia risk prediction model in the general population: A large, longitudinal, population-based machine-learning study.

SummaryBackgroundThe existing dementia risk models are limited to known risk factors and traditional statistical methods. We aimed to employ machine learning (ML) to develop a novel dementia prediction model by leveraging a rich-phenotypic variable space of 366 features covering multiple domains of health-related data.MethodsIn this longitudinal population-based cohort of the UK Biobank (UKB), 425,159 non-demented participants were enrolled from 22 recruitment centres across the UK between March 1, 2006 and October 31, 2010. We implemented a data-driven strategy to identify predictors from 366 candidate variables covering a comprehensive range of genetic and environmental factors and developed the ML model to predict incident dementia and Alzheimer's Disease (AD) within five, ten, and much longer years (median 11.9 [Interquartile range 11.2–12.5] years).FindingsDuring a follow-up of 5,023,337 person-years, 5287 and 2416 participants developed dementia and AD, respectively. A novel UKB dementia risk prediction (UKB-DRP) model comprising ten predictors including age, ApoE ε4, pairs matching time, leg fat percentage, number of medications taken, reaction time, peak expiratory flow, mother's age at death, long-standing illness, and mean corpuscular volume was established. Our prediction model was internally evaluated based on five-fold cross-validation on discrimination and calibration, and it was further compared with existing prediction scales. The UKB-DRP model can achieve high discriminative accuracy in dementia (AUC 0.848 ± 0.007) and even better in AD (AUC 0.862 ± 0.015). The model was well-calibrated (Hosmer-Lemeshow goodness-of-fit p-value = 0.92), and the predictive power was solid in different incidence time groups. More importantly, our model presented an apparent superiority over existing models like Cardiovascular Risk Factors, Aging, and Incidence of Dementia Risk Score (AUC 0.705 ± 0.008), the Dementia Risk Score (AUC 0.752 ± 0.007), and the Australian National University Alzheimer's Disease Risk Index (AUC 0.584 ± 0.017). The model was internally validated in the general population of European ancestry and White ethnicity; thus, further validation with independent datasets is necessary to confirm these findings.InterpretationOur ML-based UKB-DRP model incorporated ten easily accessible predictors with solid predictive power for incident dementia and AD within five, ten, and much longer years, which can be used to identify individuals at high risk of dementia and AD in the general population.FundingThis study was funded by grants from the Science and Technology Innovation 2030 Major Projects (2022ZD0211600), National Key R&D Program of China (2018YFC1312904, 2019YFA070950), National Natural Science Foundation of China (282071201, 81971032, 82071997), Shanghai Municipal Science and Technology Major Project (2018SHZDZX01), Research Start-up Fund of Huashan Hospital (2022QD002), Excellence 2025 Talent Cultivation Program at Fudan University (3030277001), Shanghai Rising-Star Program (21QA1408700), Medical Engineering Fund of Fudan University (yg2021-013), and the 111 Project (No. B18015).

A population-based meta-analysis of circulating GFAP for cognition and dementia risk.

ObjectiveExpression of glial fibrillary acidic protein (GFAP), a marker of reactive astrocytosis, colocalizes with neuropathology in the brain. Blood levels of GFAP have been associated with cognitive decline and dementia status. However, further examinations at a population‐based level are necessary to broaden generalizability to community settings.MethodsCirculating GFAP levels were assayed using a Simoa HD‐1 analyzer in 4338 adults without prevalent dementia from four longitudinal community‐based cohort studies. The associations between GFAP levels with general cognition, total brain volume, and hippocampal volume were evaluated with separate linear regression models in each cohort with adjustment for age, sex, education, race, diabetes, systolic blood pressure, antihypertensive medication, body mass index, apolipoprotein E ε4 status, site, and time between GFAP blood draw and the outcome. Associations with incident all‐cause and Alzheimer's disease dementia were evaluated with adjusted Cox proportional hazard models. Meta‐analysis was performed on the estimates derived from each cohort using random‐effects models.ResultsMeta‐analyses indicated that higher circulating GFAP associated with lower general cognition (ß = −0.09, [95% confidence interval [CI]: −0.15 to −0.03], p = 0.005), but not with total brain or hippocampal volume (p > 0.05). However, each standard deviation unit increase in log‐transformed GFAP levels was significantly associated with a 2.5‐fold higher risk of incident all‐cause dementia (Hazard Ratio [HR]: 2.47 (95% CI: 1.52–4.01)) and Alzheimer's disease dementia (HR: 2.54 [95% CI: 1.42–4.53]) over up to 15‐years of follow‐up.InterpretationResults support the potential role of circulating GFAP levels for aiding dementia risk prediction and improving clinical trial stratification in community settings.

Plasma metabolomic profiles of dementia: a prospective study of 110,655 participants in the UK Biobank

BackgroundPlasma metabolomic profile is disturbed in dementia patients, but previous studies have discordant conclusions.MethodsCirculating metabolomic data of 110,655 people in the UK Biobank study were measured with nuclear magnetic resonance technique, and incident dementia records were obtained from national health registers. The associations between plasma metabolites and dementia were estimated using Cox proportional hazard models. The 10-fold cross-validation elastic net regression models selected metabolites that predicted incident dementia, and a 10-year prediction model for dementia was constructed by multivariable logistic regression. The predictive values of the conventional risk model, the metabolites model, and the combined model were discriminated by comparison of area under the receiver operating characteristic curves (AUCs). Net reclassification improvement (NRI) was used to estimate the change of reclassification ability when adding metabolites into the conventional prediction model.ResultsAmongst 110,655 participants, the mean (standard deviation) age was 56.5 (8.1) years, and 51 186 (46.3%) were male. A total of 1439 (13.0%) developed dementia during a median follow-up of 12.2 years (interquartile range: 11.5–12.9 years). A total of 38 metabolites, including lipids and lipoproteins, ketone bodies, glycolysis-related metabolites, and amino acids, were found to be significantly associated with incident dementia. Adding selected metabolites (n=24) to the conventional dementia risk prediction model significantly improved the prediction for incident dementia (AUC: 0.824 versus 0.817, p =0.042) and reclassification ability (NRI = 4.97%, P = 0.009) for identifying high risk groups.ConclusionsOur analysis identified various metabolomic biomarkers which were significantly associated with incident dementia. Metabolomic profiles also provided opportunities for dementia risk reclassification. These findings may help explain the biological mechanisms underlying dementia and improve dementia prediction.

Subjective short-term memory difficulties at ages 50-75 predict dementia risk in a community-based cohort followed over 17years.

Subjective cognitive decline (SCD) is an established precursor of dementia. However, the relationship between SCD and dementia has been mostly studied among people aged 65+. We aimed to assess the association between subjective memory difficulties at ages 50-75 with all-cause dementia and dementia-subtypes in a community-based cohort with long-term follow-up. 6,190 individuals (51% female) aged 50-75years (median age, 62) attending a general health examination (by a total of 684 general practitioners) in Saarland, Germany, in 2000-2002 were recruited for a community-based cohort study. Subjective difficulties regarding short-term and long-term memory were assessed at baseline with two simple yes/no questions. Associations with dementia (-subtypes) diagnoses during 17years of follow-up were estimated by Cox proportional hazards models. 492 participants were diagnosed with dementia during 17years of follow-up. Participants with short-term memory difficulties were at higher risk to receive incident all-cause dementia and vascular dementia diagnoses both within 0-9years (age and sex adjusted hazard ratios (aHR), 1.80 and 2.00, respectively) and within 0-17years (aHR 1.55 and 1.78, respectively) from recruitment (P < 0.05 in all cases). For clinical Alzheimer's disease, a significant association was only seen within the initial 6years. There were no associations of long-term memory difficulties with any type of dementia. Subjective difficulties in short-term memory predict both intermediate and long-term risk of vascular and all-cause dementia even among late middle-age adults. These results underline the importance of cardiovascular disease prevention efforts well before old age for maintaining cognitive health.

Determining Vascular Risk Factors for Dementia and Dementia Risk Prediction Across Mid- to Later Life: The Framingham Heart Study.

The association between vascular risk factors and dementia varies with age, making generalizability of dementia risk prediction rules to individuals of different ages challenging. We determined the most important vascular risk factors for inclusion in age-specific dementia risk scores. Framingham Heart Study Original and Offspring cohort participants with available data on the Framingham Stroke Risk Profile (FSRP) at midlife (age 55; n = 4,899, 57% women), late life (ages 65 or 70), or later life (ages 75 or 80 [n = 2,386, 62% women]) were followed for 10-year incident dementia risk from ages 65, 70, 75, and 80. Age- and sex-adjusted midlife risk factors associated with 10-year risk of dementia from age 65 included FSRP (hazard ratio [HR] 1.16, 95% CI 1.06-1.26, per 1 SD increment in log-transformed score), diabetes mellitus (DM; HR 4.31, 95% CI 1.97-9.43), and systolic blood pressure (SBP; HR 1.12, 95% CI 1.02-1.24, per 10 mm Hg increment). Late-life risk factors associated with 10-year incident dementia from ages 65 or 70 included FSRP (age 65 only: HR 1.06, 95% CI 1.02-1.10), antihypertensive use (age 65 reported: HR 1.66, 95% CI 1.12-2.46), DM (age 65 reported: HR 1.96, 95% CI 1.09-3.52), atrial fibrillation (age 65 reported: HR 2.30, 95% CI 1.00-5.27), nonstroke cardiovascular disease (nsCVD; age 65 reported: HR 1.95, 95% CI 1.24-3.07), and stroke (age 70 only: HR 3.61, 95% CI 2.21-5.92). Later-life risk factors associated with 10-year incident dementia from ages 75 or 80 included antihypertensive use (age 80 only: HR 0.74, 95% CI 0.62-0.89), DM (age 80 reported: HR 1.40, 95% CI 1.04-1.89), atrial fibrillation (age 80 reported: HR 1.43, 95% CI 1.07-1.92), and stroke (age 80 reported: HR 1.63, 95% CI 1.13-2.35). In stepwise models, SBP and DM at age 55, nsCVD at age 65, DM and stroke at ages 70 and 75, and DM, stroke, and use of antihypertensives (protective) at age 80 were the most important vascular risk factors for dementia. Our findings support the use of age-specific dementia risk scores, which should prioritize including, at age 55, SBP and DM; at age 65, nsCVD; at ages 70 and 75, DM and stroke; and at age 80, DM, stroke, and antihypertensive use.

Dementia risk predictions from German claims data using methods of machine learning.

We examined whether German claims data are suitable for dementia risk prediction, how machine learning (ML) compares to classical regression, and what the important predictors for dementia risk are. We analyzed data from the largest German health insurance company, including 117,895 dementia-free people age 65+. Follow-up was 10 years. Predictors were: 23 age-related diseases, 212 medical prescriptions, 87 surgery codes, as well as age and sex. Statistical methods included logistic regression (LR), gradient boosting (GBM), and random forests (RFs). Discriminatory power was moderate for LR (C-statistic=0.714; 95% confidence interval [CI]=0.708-0.720) and GBM (C-statistic=0.707; 95% CI =0.700-0.713) and lower for RF (C-statistic=0.636; 95% CI =0.628-0.643). GBM had the best model calibration. We identified antipsychotic medications and cerebrovascular disease but also a less-established specific antibacterial medical prescription as important predictors. Our models from German claims data have acceptable accuracy and may provide cost-effective decision support for early dementia screening.

Derivation and validation of the Rapid Assessment of Dementia Risk (RADaR) for older adults.

BackgroundThere is no practical dementia risk score in the clinical setting.ObjectiveTo derive and validate a score obtained by a rapid and simple assessment, which guides primary care providers in predicting the risk of dementia among older adults.DesignA total of 4178 participants from three longitudinal cohorts (mean age at baseline = 76.8 [SD = 7.6] years), without baseline dementia, followed annually for a median of 10 years (IQR: 5 to16 years, Reverse Kaplan-Meier).ParticipantsTo derive the score, we used data from 1,780 participants from the Rush Memory and Aging Project (93% White). To validate the score, we used data from 1,299 participants from the Religious Order Study (92% White), and to assess generalizability, 679 participants from the Minority Aging Research Study (100% Black).MeasurementsClinician-based dementia diagnosis at any time after baseline and predictive variables associated with dementia risk that can be collected in a primary care setting: demographics, clinical indicators, medical history, memory complaints, cognitive and motor tests, and questions to assess functional disability, depressive symptoms, sleep, social isolation, and genetics (APOE e4 and AD polygenic risk score).ResultsAt baseline, age, memory complaint, the ability to handle finances, the recall of the month, recall of the room, and recall of three words, were associated with the cumulative incidence of dementia, in the derivation cohort. The discrimination of the RADaR (Rapid Risk Assessment of Dementia) was good for the derivation and external-validation cohorts (AUC3 years = 0.82–0.86), compared to the overall discrimination of age alone (AUC3 years = 0.73), a major risk factor for dementia. Adding genetic data did not increase discrimination.LimitationsParticipants were volunteers, may not represent the general population.ConclusionsThe RADaR, derived from community-dwelling older persons, is a brief and valid tool to predict dementia risk at 3 years in older White and Black persons.

IoT-Based Unobtrusive Physical Activity Monitoring System for Predicting Dementia

Mental health-related disorders are common in elderly populations. Among the various mental health disorders, one most significant threat is dementia, and prediction of dementia has become an important issue related to well-being in old age, because the disease progression of dementia can be slowed by early diagnosis and disease control. In this paper, we propose an unobtrusive dementia-prediction system for monitoring physical activities of elderly persons either living alone or as a couple in different house structures, achieved through passive infrared (PIR) motion sensors combined with data processing. The proposed feature extraction algorithm extracts feature values related to physical activities from simple passive infrared sensors located in each room space. We then apply a variety of common popular classification models, including Deep Neural Networks (DNNs), to predict the risk of dementia in a sensor-enabled home. We implemented and validated algorithms on data collected for over a month from 18 participants who were engaged with a variety of living conditions. The proposed system was effective in predicting dementia risk, with up to an 0.99 area under the curve (AUC) using DNN with principal component analysis (PCA) and a quantile transformer scaler. In terms of the result based on leave-one-subject-out (LOSO) analysis, an accuracy of 63.38% was achieved using DNN with PCA and a standard scaler. The proposed methodology is non-invasive and cost-effective, and can be used for a variety of long-term monitoring and early symptom detection systems, helping caregivers provide optimal interventions to elderly individuals at risk for dementia.

Performance of Machine Learning Algorithms for Predicting Progression to Dementia in Memory Clinic Patients

Machine learning algorithms could be used as the basis for clinical decision-making aids to enhance clinical practice. To assess the ability of machine learning algorithms to predict dementia incidence within 2 years compared with existing models and determine the optimal analytic approach and number of variables required. This prognostic study used data from a prospective cohort of 15 307 participants without dementia at baseline to perform a secondary analysis of factors that could be used to predict dementia incidence. Participants attended National Alzheimer Coordinating Center memory clinics across the United States between 2005 and 2015. Analyses were conducted from March to May 2021. 258 variables spanning domains of dementia-related clinical measures and risk factors. The main outcome was incident all-cause dementia diagnosed within 2 years of baseline assessment. In a sample of 15 307 participants (mean [SD] age, 72.3 [9.8] years; 9129 [60%] women and 6178 [40%] men) without dementia at baseline, 1568 (10%) received a diagnosis of dementia within 2 years of their initial assessment. Compared with 2 existing models for dementia risk prediction (ie, Cardiovascular Risk Factors, Aging, and Incidence of Dementia Risk Score, and the Brief Dementia Screening Indicator), machine learning algorithms were superior in predicting incident all-cause dementia within 2 years. The gradient-boosted trees algorithm had a mean (SD) overall accuracy of 92% (1%), sensitivity of 0.45 (0.05), specificity of 0.97 (0.01), and area under the curve of 0.92 (0.01) using all 258 variables. Analysis of variable importance showed that only 6 variables were required for machine learning algorithms to achieve an accuracy of 91% and area under the curve of at least 0.89. Machine learning algorithms also identified up to 84% of participants who received an initial dementia diagnosis that was subsequently reversed to mild cognitive impairment or cognitively unimpaired, suggesting possible misdiagnosis. These findings suggest that machine learning algorithms could accurately predict incident dementia within 2 years in patients receiving care at memory clinics using only 6 variables. These findings could be used to inform the development and validation of decision-making aids in memory clinics.

Development and external validation of a novel dementia risk prediction score in the UK Biobank cohort

AbstractBackgroundGlobally, up to 40% of dementia cases may be prevented if several key risk factors, such as low education and obesity, are targeted. This has motivated interest into the development of risk scores that aim to quantify an individual’s risk of developing dementia within a given time frame. However, translation to a clinical setting has been hampered by either a lack of external validation, poor out‐of‐sample performance, or the integration of measures (e.g., MRI, cognitive testing) that are costly or time intensive to administer.This study aimed to develop a novel dementia risk score suitable for a primary care setting and compare its discrimination accuracy with other risk scores.Method208,108 UK Biobank (UKB) participants were examined, with cases of incident dementia (mean time to diagnosis ± SD = 5.91 years ± 2.02) identified through self‐report, hospital inpatient records and death certificates (n = 1,270, 0.6%). The dataset was randomly split into a training (80%) and test set (20%). Potential predictors (n = 30) included Apolipoprotein E4 (APOE4) status and cardiovascular, medical and lifestyle factors collected at baseline. Logistic LASSO regression was employed for variable selection and logistic regression was then used to derive beta coefficients for each predictor included in the UKB dementia risk score (UKB‐DRS). The area under the curve (i.e., AUC) was used to compare the discriminative ability of the UKB‐DRS to the CAIDE, ANU‐ADRI, Dementia Risk Score (i.e., DRS) and Framingham risk score, in both the test set and the Whitehall II study (WHIII, n = 2,974), which served as an external dataset.ResultThe UKB‐DRS consisted of age, sex, education, APOE4 status, a medical history of diabetes, stroke and depression and a family history of dementia. The UKB‐DRS demonstrated good discrimination accuracy in both the test set (AUC [95% ] = 0.79 [0.77, 0.82]) and external dataset (AUC [95% CI] = 0.83 [0.79,0.87]). Additionally, the UKB‐DRS outperformed all other risk scores (the DRS had the next best AUC [95% CI]: UKB = 0.77 [0.74, 0.8]; WHII = 0.77 [0.73, 0.81]).ConclusionThis study contributes a novel dementia risk score that may be used to guide primary prevention strategies.

Self‐reported and informant‐reported depressive symptoms are independent and divergent predictors of new onset dementia in older adults

AbstractBackgroundDepressive symptoms in older adults are important predictors of incident dementia. Symptoms can be assessed in several ways, including self‐reported or informant‐reported measures. It is unclear whether one is superior in predicting dementia onset or how to interpret discordant reports. Here, in a large sample of non‐dementia older adults for whom depressive symptoms were assessed via both self‐report and informant‐report, we test independent and divergent prognostic utility for dementia.MethodData from 10,684 non‐dementia subjects from the National Alzheimer’s Coordinating Centre were analysed. This included individuals with normal cognition (NC; CDR=0) and Mild Cognitive Impairment (MCI; CDR=0.5). Geriatric Depression Scale‐15 (GDS) was used for self‐report and the Neuropsychiatric Inventory Questionnaire depression item (NPI‐Q‐D) was used for informant‐report. Cox regression and Kaplan Meier (KM) survival analyses were used to explore the measures as independent predictors of dementia, and the putative interaction between them. Outcome was defined as change to CDR≥1 at follow‐up. The cox model was adjusted for age, sex, and education. CDR category was included to determine applicability of findings across cognitive categories. GDS‐15 and NPI‐Q‐D scores were added as continuous predictors. For the KM analysis, depressive symptoms were dichotomized as GDS+/GDS‐ and NPI‐Q‐D+/NPI‐Q‐D‐ using cut‐off scores of ≥5 for GDS‐15 and ≥2 for NPI‐Q‐Depression.ResultGDS (HR: 1.096; CI: 1.070–1.122) and NPI‐Q‐D (HR: 1.428; CI: 1.291–1.580) independently predicted dementia, with a significant interaction between them (p&lt;0.000) even after controlling for CDR. KM analyses showed that in the self‐reported non‐depressed group (GDS‐), there was a significant difference in the hazard of incident dementia based on the informant report. The discordant group i.e., GDS‐/NPI‐Q‐D+ progressed to dementia significantly faster (9.6yrs; SD=0.292) compared to the GDS‐/NPIQ‐D‐ group (12.0yrs; SD=0.055) (p&lt;0.000; Log‐rank test). No significant difference was found based on informant reports in the GDS+ group (p=0.114).ConclusionSelf‐reported and informant‐reported depressive symptoms are not duplicative, and independently predict dementia risk across cognitive categories. Informant report is particularly important when self‐report is negative, to detect individuals with compromised insight at higher risk for progression. Discordance of self‐informant reports require greater clinical and research consideration in NC and MCI populations.

Development and validation of new screening tool for predicting dementia risk in community-dwelling older Japanese adults

BackgroundEstablished clinical assessments for detecting dementia risk often require time, cost, and face-to-face meetings. We aimed to develop a Simplified Telephone Assessment for Dementia risk (STAD) (a new screening tool utilizing telephonic interviews to predict dementia risk) and examine the predictive validity of the STAD for the incidence of dementia.MethodsWe developed STAD based on a combination of literature review, statistical analysis, and expert opinion. We selected 12 binary questions on subjective cognitive complaints, depressive symptoms, and lifestyle activities. In the validation study, we used STAD for 4298 community-dwelling older adults and observed the incidence of dementia during the 24-month follow-up period. The total score of STAD ranging from 0 to 12 was calculated, and the cut-off point for dementia incidence was determined using the Youden index. The survival rate of dementia incidence according to the cut-off points was determined. Furthermore, we used a decision-tree model (classification and regression tree, CART) to enhance the predictive ability of STAD for dementia risk screening.ResultsThe cut-off point of STAD was set at 4/5. Participants scoring ≥ 5 points showed a significantly higher risk of dementia than those scoring ≤ 4 points, even after adjusting for covariates (hazard ratio [95% confidence interval], 2.67 [1.40–5.08]). A decision tree model using the CART algorithm was constructed using 12 nodes with three STAD items. It showed better performance for dementia prediction in terms of accuracy and specificity as compared to the logistic regression model, although its sensitivity was worse than the logistic regression model.ConclusionsWe developed a 12-item questionnaire, STAD, as a screening tool to predict dementia risk utilizing telephonic interviews and confirmed its predictive validity. Our findings might provide useful information for early screening of dementia risk and enable bridging between community and clinical settings. Additionally, STAD could be employed without face-to-face meetings in a short time; therefore, it may be a suitable screening tool for community-dwelling older adults who have negative attitudes toward clinical examination or are non-adherent to follow-up assessments in clinical trials.