Abstract
BACKGROUND: Earlier identification of high coronary artery disease (CAD) risk individuals may enable more effective prevention strategies. However, existing 10-year risk frameworks are ineffective at earlier identification. We sought to understand how the variable importance of genomic and clinical factors across life stages may significantly improve lifelong CAD event prediction. METHODS: A longitudinal study was performed using data from 2 cohort studies: the FOS (Framingham Offspring Study) with 3588 participants aged 19 to 57 years and the UKB (UK Biobank) with 327 837 participants aged 40 years to 70 years. A total of 134 765 and 3 831 734 person-time years were observed in FOS and UKB, respectively. Hazard ratios for CAD were calculated for polygenic risk score (PRS) and clinical risk factors at each age of enrollment. The relative importance of PRS and pooled cohort equations in predicting CAD events was also evaluated by age groups. RESULTS: The importance of CAD PRS diminished over the life course, with a hazard ratio of 3.58 (95% CI, 1.39-9.19) at the age of 19 years in FOS and a hazard ratio of 1.51 (95% CI, 1.48-1.54) by the age of 70 years in UKB. Clinical risk factors exhibited similar age-dependent trends. PRS significantly outperformed pooled cohort equations in identifying subsequent CAD events in the 40- to 45-year age group, with 3.2-fold more appropriately identified events. Overall, adding PRS improved the area under the receiving operating curve of the pooled cohort equations by an average of +5.1% (95% CI, 4.9%-5.2%) across all age groups; among individuals <55 years, PRS augmented the area under the receiver operater curve (ROC) of the pooled cohort equations by 6.5% (95% CI, 5.5%-7.5%; P<0.001). CONCLUSIONS: Genomic and clinical risk factors for CAD display time-varying importance across the lifespan. The study underscores the added value of CAD PRS, particularly among individuals younger than 55 years, for enhancing early risk prediction and prevention strategies. All results are available at https://surbut.github.io/dynamicHRpaper/index.html.