This project will develop and validate wrist accelerometer-derived digital biomarkers to predict major chronic diseases, physical function decline (gait speed, grip strength, ADLs), and cognitive impairment/dementia.
Research questions:
(1) Do high-resolution digital biomarkers improve prediction beyond demographics, lifestyle factors (sleep, physical activity, diet), and polygenic risk scores (PRS)?
(2) What minimal and optimal combinations of objectively measured sleep, activity, and diet confer the lowest chronic disease and dementia risks?
(3) How do 24-hour activity rhythms, exercise timing (morning/afternoon/evening), and physical activity modality (self-reported vs device-measured; leisure vs occupational) influence disease risk and interact with genetic susceptibility?
Objectives:
(i) Extract multiscale time-series and circadian features (RA, IS, IV, M10, L5) from UK Biobank accelerometry and derive interpretable digital biomarkers via feature selection and machine learning.
(ii) Quantify the added predictive value of digital biomarkers, lifestyle behaviours, and PRS using sequential Cox and mixed-effects models (C-index, NRI, calibration) with internal and external validation.
(iii) Identify minimal and optimal behaviour profiles using compositional data analysis, isotemporal substitution, and response-surface modelling.
(iv) Explore PRS×behaviour interactions and proteomic/metabolomic mediation to elucidate biological pathways.
(v) Translate findings into actionable targets (e.g. additional minutes of MVPA or improved sleep regularity) yielding clinically meaningful risk reduction.
Scientific rationale:
Objective 24-hour accelerometry overcomes self-report bias and captures intensity, timing, and rhythm patterns critical to disease mechanisms. Integrating accelerometry with genomics and molecular profiling enhances risk prediction and reveals pathways linking lifestyle and disease, supporting precision prevention and healthy ageing.