Abstract 4119941: Identification of individuals who benefit from omega-3 fatty acid supplementation to prevent coronary heart disease: A machine-learning analysis of the VITAL

Backgrounds: Randomized controlled trials (RCTs) have demonstrated benefits of marine omega-3 polyunsaturated fatty acids (omega-3 FA) supplementation for the prevention of coronary heart disease (CHD). However, it has not been clear which individuals would benefit the most from the supplementation. We sought to predict the individual treatment effect of omega-3 FA supplementation on CHD prevention and to develop an omega-3 effect score to stratify individuals according to their expected benefit from the supplementation. Methods: Among the 25,871 randomized participants without history of CVD in the VITamin D and OmegA-3 TriaL (VITAL), we applied machine-learning (ML) approaches to predict individual treatment effect of omega-3 FA supplementation on 5-year CHD risk (a composite of myocardial infarction, coronary revascularization, and CHD death) using 11 covariates pre-specified in the VITAL trial protocol. A 10-fold cross-validation was used and held-out test dataset was used for the evaluation. An omega-3 effect score was developed such that each covariate contributed linearly, and utility of the score was further evaluated by transportability analysis using the National Health and Nutrition Examination Survey (NHANES) data as the target population. Results: Omega-3 FA intervention led to absolute 0.48% [SE: 0.20] reduction in CHD in the total population. ML algorithms effectively stratified participants by their expected benefit according to individual factors; decreased CHD risk was observed in those with quintile 1 and 2 of the expected benefit (absolute CHD risk reduction %: 1.30 % [0.55] and 1.32 % [0.51] in quintile 1 and 2, respectively). Race, diabetes, and fish intake most contributed to the omega-3 effect score. CHD incidence rates per 1000 person-year were 5.5 [0.44] if treated and 8.5 [0.55] if not treated (35.3% reduction) in individuals with the score ≥11 (upper 40th percentile), and 3.9% [0.31] if treated and 3.4% [0.55] if not treated (14.7% increase) in those with the score <11, respectively. The transportability of the utility of the score to baseline NHANES data was confirmed. Conclusion: In VITAL, ML approaches identified individuals who experienced greater CHD reduction from the omega-3 FA intervention, and an omega-3 effect score stratified the expected benefit. Although it warrants testing in a new RCT, the omega-3 effect score holds promise for guiding the indication of omega-3 FA supplementation in US primary prevention population.