This study tested whether machine-learning-selected plasma proteins predict myocardial infarction (MI) beyond conventional factors and polygenic scores. It measured over 9,500 proteins in 4,000 Chinese adults, 1,937 with MI, using a case-cohort design. The Boruta-selected 23-protein score gave a 3.3-fold gap in 10-year MI incidence between top and bottom fifths. Conventional risk factors gave a 2.5-fold gap and polygenic scores a 1.5-fold gap. Adding the proteins to conventional factors raised C-statistics from 0.846 to 0.864, and 0.710 to 0.733 in UK Biobank. UK Biobank validation used 37,187 Europeans with only 630 MI cases.
Why it is interesting: Proteomic assays measure thousands of proteins, yet just 23 improved MI prediction in both Chinese and European adults. In the Chinese cohort the protein score also outperformed polygenic scores.
Cite this item
BibTeX
Free sign-up. A digest of the new papers twice a week, grouped by domain.