Magnesium supplement and vascular health: Machine learning from the longitudinal medical record

点击翻译按钮获取中文摘要

Project Summary/Abstract Over half of adult Americans use dietary supplements. However, little is known about their safety and effectiveness as these products are not approved by the US Food and Drug Administration (FDA) and post-marketing surveillance is limited to adverse events. The NIH Office of Dietary Supplements (ODS) seeks to fill in that gap and has identified electronic health record (EHR) data as a potential tool to advance that goal. Preliminary data from our pilot study sponsored by the NIH ODS that used advanced machine/deep learning techniques suggest that magnesium supplements may lower the risk of heart failure (HF) in patient with diabetes mellitus (DM) and may improve outcomes in those with HF. Both HF and DM affect the health and outcomes of millions of Americans. DM is a risk factor for HF and adversely affects outcomes in those with HF. Magnesium is an integral part of over 300 human enzyme systems, which are impaired in magnesium deficiency. Findings from our study suggest that a low dietary magnesium intake is associated with a higher risk of incident HF, especially among those with DM. However, less is known about this relationship in patients with HF. The Specific Aims 1 and 2 of the proposed projects are to test the hypotheses that a new prescription for oral magnesium supplement is associated with a lower risk of incident HF in those with DM and of mortality and hospitalization in patients with HF. Although magnesium is inexpensive and relatively safe, its long-term effects may vary for individual patients. Thus, instead of recommending it to millions of patients, it would be ideal to recommend to individuals who are most likely to benefit. Thus, our Specific Aim 3 is to develop and validate a novel explainable deep learning-based risk prediction model to determine with precision the optimal clinical setting under which an individual may derive clinical benefits from magnesium supplementation given their individual characteristics including multimorbidity and polypharmacy. These aims will be achieved by interrogating the Veterans Affairs (VA) national EHR data that includes over 2 million Veterans with DM and 1 million with HF with ~20 years of longitudinal data on magnesium supplements, serum magnesium, and outcomes. We will use a new-user design, marginal structural model (propensity score weighting) with machine- learning-based estimation and stability analyses to minimize confounding and account for potential biases. The prediction model for individual risk/benefit will be validated using the Cerner Health Facts® data for generalizability in non-Veteran populations. The findings of proposed study will generate new evidence that will have direct clinical implications and those of Aim 3 specifically will provide a novel precision medicine tool to individualize magnesium supplement use.