COIOS
Weighing data and models Item refines preprint

Polygenic scores add most to prediction where clinical risk is already low

Across 150 diseases in 900,000 UK Biobank and FinnGen participants, polygenic risk scores added most to clinical prediction in metabolic, cardiovascular, autoimmune and neurological disease, and their gains were consistently larger among those at lower clinical risk.

Models combining clinical history and polygenic risk scores were developed and validated for 150 diseases and all-cause mortality in 900,000 participants from UK Biobank and FinnGen. Genetic information added most in metabolic, cardiovascular, autoimmune and neurological conditions and little in genitourinary and respiratory disease, with all significant score-by-clinical-risk interactions negative. For atherosclerotic cardiovascular disease, the polygenic score expanded the clinically identified high-risk group by 40%, and the reclassified group had 4.9-fold higher incidence during follow-up than those remaining low-risk.

Why it is interesting: It places the marginal value of genetic prediction precisely where clinical records are least informative, and quantifies the reclassification for cardiovascular disease.

Source
medRxiv, 27 September 2026
DOI
10.64898/2026.09.25.26364029
Type
Preprint
Design
Prediction model development and validation across 150 diseases and all-cause mortality, 900,000 participants, UK Biobank and FinnGen, with external validation in three biobanks
Verdict
Refines prior evidence
Driver
Forecasts and their record
Driver
Heart and stroke