One of the prominent impacts of UK Biobank (UKB) on human health is to reveal etiological factors using its vast amount of data. Such results enable the prediction of disease risk (DR) and prophylaxis with corresponding risk countermeasures. It is however notable that although a large body of studies have been conducted in this regard, few of them addressed a related and significant subject matter, i.e., can we use UKB data to predict health risk (HR)? Here we define HR as the likelihood of a significant, longitudinal change of a physiological trait within the normal range, such as the rise of fasting blood glucose below 100 mg/dL (above which pre-diabetes is diagnosed). By this definition, HR reflects subtle shift from optimal to suboptimal health, which is distinct from DR (switch from health to disease). HR research is relevant for preventive medicine, since it enables the studies of risk countermeasures that stop the progression of disease at the very beginning. In essence, an HR weakens the homeostatic resilience against stress, whereas an HR countermeasure does the opposite. Unfortunately, due to the priority placed on disease by the research community hitherto, HR, as a health measure, is yet to receive attention.
Our objective is to develop a library of HR (as defined above) panels of the clinically-relevant traits (e.g., blood pressure, lipids, and glucose) using the multi-modal parameters documented at UKB database.
Tier 2 UKB data will be organized, curated, quality-controlled on UKB RAP, and used for training and validating our HR-predicting models. Given the large volume and the multi-modality of the data, we plan to employ AI, such as Large Language Model, for probability analytics. Specifically, for a trait of interest, the AI model identifies a signature panel of covariates (e.g., anthrometrics, lab tests, and omics) that differentiates at-risk individuals from the study population based on the longitudinal change of the trait.