H. Yang, J. Guo, X. Shi, K. Ma, Y. Jiang, D. Li, X. Zhou, Y. Yu, D. Huang
BACKGROUND Contemporary cardiovascular risk equations guide prevention, yet cardiovascular risk remains distributed beyond the highest-risk groups. Whether broader baseline phenotyping can improve risk stratification and whether this information can be reduced to clinically accessible measures is uncertain. OBJECTIVES To assess whether baseline phenotyping improves cardiovascular risk stratification beyond PREVENT and CHARGE-AF and whether this information can be compressed into clinically accessible measures. METHODS We studied 266,410 UK Biobank participants without prevalent CHD, HF, AF, or stroke, divided into training, validation, and held-out test samples. PREVENT equations were used to estimate 10-year risk for CHD, HF, and stroke, and CHARGE-AF to estimate 5-year AF risk. Models considered 3,838 baseline phenotypes. Geographic replication was assessed in Scotland and Wales. RESULTS The highest-risk 20% identified by the clinical equations accounted for 44.8% to 55.2% of subsequent events across the 4 outcomes. Broad phenotyping refined risk stratification beyond these equations. After compression from 3,838 baseline phenotypes, disease-specific sets of 13 to 19 clinically accessible measures retained part of the incremental discrimination provided by broader phenotyping, yielding held-out IPCW AUC improvements of 0.013 to 0.030 relative to the corresponding clinical equations, with the largest gain for HF. At the same 20% prioritization level, the clinically accessible measures increased the proportion of subsequent events captured by 2.7 to 4.9 percentage points. Findings were directionally consistent in geographic replication. CONCLUSIONS Disease-specific sets of clinically accessible measures retained part of the information identified from broad phenotyping and refined cardiovascular risk stratification beyond contemporary clinical equations. Independent external validation is required before clinical use.