Application of novel technologies to cardiovascular prevention

Abstract

This dissertation aimed to employ advanced methodologies in machine learning (ML), causal inference, and metabolomics in epidemiological research to contribute to the field of CVD prevention. First, quantifying sodium intake has been a persistent challenge in epidemiological research due to the measurement errors, limiting the explorations of sodium exposure in relation to disease outcomes, such as early-onset hypertension. Chapter 1 describes the development of ML algorithms that predict dietary sodium intake—measured via multiple 24-hour urinary sodium excretion—based on questionnaire data in the subcohorts of the Nurses' Health Study (NHS), NHS II, and Health Professional Follow-up Study (HPFS). These algorithms well predicted absolute sodium intake, albeit with a modest improvement in mitigating measurement error bias. In Chapter 2, using the developed algorithms, we evaluated the population-attributable risks (PARs) of modifiable lifestyle factors concerning the incidence of overall and early-onset hypertension across the full cohorts of NHS, NHS II, and HPFS. The findings suggest that maintaining healthy weight may substantially lower the risk of hypertension, and that adopting a combination of lifestyle modifications could further reduce the risk across all ages, particularly stronger in younger populations. Second, studies exploring heterogeneous treatment effect (HTE) gain traction in epidemiology, potentially offering advancements in precision medicine. Chapter 3 delineates the methodological frameworks for HTE research using a single randomized controlled trial. In an application example using the Preventing Overweight Using Novel Strategies (POUNDS LOST) trial, we developed and validated an ITR to optimize the efficacy of high- and low-fat diet interventions for weight loss over two years. Third, Chapter 4 delves into the development of a novel biomarker identified through metabolomics technology. We investigated the associations between circulating branched-chain amino acids (BCAAs)— metabolites indicative of diabetes risk—and established cardiometabolic biomarkers in the Women’s Health Study, thus characterizing BCAAs as biomarkers representing CVD risk independent of glucose metabolism.