Imperial College London

ID-333-1

Evaluates genetic and metabolomic data to improve disease risk prediction in South Asian vs White populations, expanding LOLIPOP linkage in WSIC and integrating multi-omic data to enhance modelling of cardiovascular, diabetes and cardiometabolic outcomes.

We will undertake a linked-cohort analysis using the LOLIPOP baseline research dataset (expanded to ~80,000 participants) linked via NHS number to longitudinal routine healthcare data within WSIC/NWL. Baseline variables will include demographics (age, sex, self-reported ethnicity), socioeconomic status (IMD), anthropometry (BMI, waist/hip measures where available), clinical risk factors (blood pressure, lipids, glycaemia/HbA1c), lifestyle factors (eg smoking, if available), comorbidities and medications. We will additionally incorporate multi-omic exposures, including pre-computed genetic risk scores (GRSs) for relevant traits and diseases derived from LOLIPOP genotyping, and metabolomic profiling measures. Outcomes will be defined using validated phenotyping code lists within routine data and will include incident type 2 diabetes, cardiovascular disease (eg coronary heart disease, stroke, heart failure), chronic kidney disease, and other long-term conditions; where feasible we will also evaluate cancer outcomes. Follow-up Version 1.5 – October 2025 Page 8 of 29 will run from baseline assessment date to first outcome event, death, deregistration, or end of data availability. Analytically, we will first describe baseline characteristics and outcome incidence by ethnic group, and quantify data completeness and linkage rates. For disease prediction, we will develop and compare risk models in South Asian and White European participants using a staged approach: (i) clinical variables only, (ii) clinical plus genetic risk scores, (iii) clinical plus metabolomics, and (iv) combined clinical plus genetic plus metabolomics. Modelling will use time-to-event methods (eg Cox proportional hazards models) for incident outcomes, with consideration of competing risks where relevant, and regularised regression (eg elastic net) and/or machine learning approaches (eg gradient boosting) for high-dimensional metabolomic predictors. We will assess incremental predictive value of omic data using discrimination (C-statistic/AUC), calibration (calibration slope/intercept and plots), and reclassification metrics (eg net reclassification improvement) alongside decision-curve analysis where appropriate. Internal validation will use crossvalidation or bootstrapping, and we will undertake external validation in an independent cohort (eg UK Biobank) where comparable variables and outcomes are available. We will also evaluate model performance separately by ethnic group to ensure equitable calibration and discrimination, and perform sensitivity analyses to assess robustness to missing data (eg multiple imputation), alternative outcome definitions, and differential healthcare utilisation. All analyses will be conducted entirely within the OneLondon SDE secure environment. No individual-level data will be extracted from the environment at any stage. All outputs for publication or dissemination will contain only aggregated summary statistics, model coefficients, and performance metrics, with small numbers suppressed in accordance with the SDE disclosure control policy. Data minimisation principles are applied throughout: only the minimum variables required for the specified analyses will be accessed, and access will be limited to the named users listed in Section 8.2 of this application under the authorisation of the Principal Investigator.

20/05/2026

21/11/2026

21/05/2026

De-Personalised

25/05/2026

TRE