Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Canada Postdoctoral Research Award (CPRA) program

McGill University

$140,000.00

An interpretable Digital Twin model for personalized cognitive decline prediction and sample size reduction in Alzheimer's disease clinical trials

EF 2025-26

Résumé
Cognitive decline in Alzheimer's disease (AD) is caused by accumulated brain pathology combined with health-related risk factors. Computer software can combine this information to predict decline, helping doctors choose treatments and patients plan their future. However, current tools are often complex artificial intelligence "black boxes" that doctors find hard to trust. To address this issue, in my PhD I developed an interpretable Digital Twin (DT) model. For a patient, the model identifies medically similar people (Twins) from large AD databases, then uses the Twins' past cognitive test scores to predict decline. This method is transparent - it directly uses real patient data. My DT model can also improve clinical trials of new treatments. By comparing each participant's observed decline to their predicted decline, trials can detect treatment effects more precisely, reducing the number of participants needed. In simulated Phase 3 AD trials, I combined my DT predictions with a new analysis method called Prediction-Powered Inference (PPI) and showed a 24% reduction in required participants compared to standard analysis. In my new project, I will first improve prediction accuracy for a 24-month window by adding key AD-related factors to Twin selection: sex, education, and three MRI markers of brain damage. I will validate the model in AD datasets with 9,000+ patients. Next, I will apply the improved model to two ongoing Canadian AD trials testing (1) whether a sleep-improving drug can slow decline, and (2) whether physical and cognitive training can prevent decline. I will also apply it to a past anti-amyloid drug trial. With an improved longitudinal PPI method that analyzes all trial time points, I aim to show the DT model can reduce required trial size by over 40%. No other DT method uses MRI biomarkers, patient-specific matching, or longitudinal PPI. This project will be the first to test these techniques in real Canadian trials, paving the way for clinical use.
Programme
Canada Postdoctoral Research Award (CPRA) program
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
CIHR Funding Decisions Database

Tel que publié

Id
536891
Name
Andrews Daniel
Region
Québec
Country
Canada
Orgtype
Institution Paid; Research Institution
Instname2
Aging
Theme2
Biomedical
Prcname2
Canada Postdoctoral Research Award - A
Projecttitle
An interpretable Digital Twin model for personalized cognitive decline prediction and sample size reduction in Alzheimer's disease clinical trials
Programtype2
Fellowship Programs
Orgnameinp2
McGill University
Cihramount2
$140,000
Cihrequipment2
$0
Programname2
Canada Postdoctoral Research Award (CPRA) program
Keyworddelim
Alzheimer'S Disease; Clinical Trial; Cognitive Decline; Data Analysis; Data Simulation; Dementia; Digital Twins; Power Analysis; Prediction; Statistics
Abstract
Cognitive decline in Alzheimer's disease (AD) is caused by accumulated brain pathology combined with health-related risk factors. Computer software can combine this information to predict decline, helping doctors choose treatments and patients plan their future. However, current tools are often complex artificial intelligence "black boxes" that doctors find hard to trust. To address this issue, in my PhD I developed an interpretable Digital Twin (DT) model. For a patient, the model identifies medically similar people (Twins) from large AD databases, then uses the Twins' past cognitive test scores to predict decline. This method is transparent - it directly uses real patient data. My DT model can also improve clinical trials of new treatments. By comparing each participant's observed decline to their predicted decline, trials can detect treatment effects more precisely, reducing the number of participants needed. In simulated Phase 3 AD trials, I combined my DT predictions with a new analysis method called Prediction-Powered Inference (PPI) and showed a 24% reduction in required participants compared to standard analysis. In my new project, I will first improve prediction accuracy for a 24-month window by adding key AD-related factors to Twin selection: sex, education, and three MRI markers of brain damage. I will validate the model in AD datasets with 9,000+ patients. Next, I will apply the improved model to two ongoing Canadian AD trials testing (1) whether a sleep-improving drug can slow decline, and (2) whether physical and cognitive training can prevent decline. I will also apply it to a past anti-amyloid drug trial. With an improved longitudinal PPI method that analyzes all trial time points, I aim to show the DT model can reduce required trial size by over 40%. No other DT method uses MRI biomarkers, patient-specific matching, or longitudinal PPI. This project will be the first to test these techniques in real Canadian trials, paving the way for clinical use.
Competitiondate
202509
Approvedterm2
2 yrs 0 mth
Pinamesdelim
Andrews, Daniel
Approvedterm
2 ans 0 mois
Cihrcontribution2
~ $140,000~ $0~