ABSTRACT
While polygenic risk scores (PRSs) have been shown to identify a small number of individuals with increased clinical risk for several common diseases, non-genetic factors that change during a lifetime, such as lifestyle, employment, diet, and pollution, have a larger role in clinical prediction. We analyzed data from 459,613 participants of the UK Biobank to investigate the independent and combined roles of demographics (e.g., sex and age), 96 environmental exposures, and common genetic variants in atrial fibrillation, coronary artery disease, inflammatory bowel disease, and type 2 diabetes. We develop an additive modelling approach to estimate and validate a poly-exposure score (PXS) that goes beyond consideration of a handful of factors such as smoking and pollution. PXS is able to identify groups with high prevalence of the four common disease comparable to, if not better, than the PRS. Type 2 diabetes has the largest discrepancy in PXS and PRS performance, defined as the maximum area under the receiver-operator curve (AUC) (PXS AUC of 0.828 [0.821-0.836], PRS AUC of 0.711 [0.702-0.720]). Most importantly, we show that PXS identifies individuals that have low genetic risk but high overall risk for disease. While PRS is useful for screening genetically exceptional individuals in a time-invariant way, broader consideration of multiple non-genetic and modifiable factors is required to fully translate risk scores to the bedside for precision medicine. All results and the PXS calculator can be found in our web application http://apps.chiragjpgroup.org/pxs/.
Footnotes
