Research Question:
We propose to leverage UK Biobank’s multimodal datasets to build personalized predictive models that can produce patient-level risk scores for diseases, project disease trajectories, and forecast comorbidities. For the first stage of this study, we will focus on chronic kidney disease (CKD).
Objectives:
We plan to use clinical records and genetic data to conduct the following analyses.
1. Identify indicators of CKD progression:
Identify key predictors of CKD onset and progression among demographic, lifestyle, clinical, and genetic variables. Candidate variables include BMI, townsend deprivation index, diet, preceeding dieseases, and APOE gene variants.
2. Develop personalized CKD risk prediction models:
Use SOTA artificial intelligence methods to build predictive models that estimate patient-level risk for CKD onset and progression with the indicators identified in Objective 1.
3. Forecast CKD trajectories and cormorbidities:
Build longitudinal models to forecast the trajectories of common progression patterns of CKD stages and identify co-occurrent diseases.
Scientific Rationale:
Recent studies suggest that the transformer architecture used in SOTA artificial intelligence models can be borrowed to predict patient-level disease risk and prognosis [1,2]. Another study also indicates that leveraging multimodal data may improve the predictive accuracy of disease prediction models [3]. However, these studies are generic instead of disease-specific. We will build disease-specific models with highly accurate predictions..
References:
1. Nature (2025). https://doi.org/10.1038/s41586-025-09529-3
2. arXiv preprint (2025) arXiv:2508.12104.
3. at Commun 16, 3767 (2025). https://doi.org/10.1038/s41467-025-58724-3