Abstract
BACKGROUND: Several omics methods have been successfully used in hypertension prediction. However, the predictive ability of various multiomics data has not been compared in the same study sample, and it is unknown whether they provide additional predictive value over a good clinical risk factor score. METHODS: Clinical data augmented with modern multiomics methods (systolic blood pressure polygenic risk score, nuclear magnetic resonance metabolite profiling, and gut microbiota) were assessed in 2573 nonhypertensive participants of the FINRISK 2002 cohort. All combinations of these different methods were incorporated into cross-validated machine learning models to predict incident hypertension. Model performance of all combinations of these was assessed using the area under the curve (AUC). Information on incident hypertension was collected using nationwide healthcare register data. RESULTS: Over a mean follow-up of 18.0 years, 393 participants developed hypertension. Models that included the clinical and genetic data resulted in the highest mean AUC (0.735) compared with clinical risk factors alone (AUC=0.725). In the whole study sample, an SD increase in the polygenic risk score was associated with 29% (95% CI, 14%-46%) greater odds of incident hypertension after adjusting for clinical risk factors. Combining metabolome (AUC=0.709) or microbiota (AUC=0.720) data with clinical risk factors did not result in improved risk prediction. CONCLUSIONS: The best prediction combination model for incident hypertension was the clinical model augmented with a polygenic risk score. These data suggest that polygenic risk scores provide limited incremental value over clinical risk factors when assessing risk of incident hypertension.