Background: Proton Pump Inhibitors (PPI) are among the most commonly prescribed medications at the VA and elsewhere. Evidence suggests PPI use is associated with increased risk of adverse health outcomes including cardiovascular disease, pneumonia, Clostridium difficile infections, gastric cancer, and kidney disease. Recent studies suggested that PPI use is associated with increased risk of all-cause-mortality. Significance/Impact: These findings generated widespread concern among VA and outside stakeholders who asked 2 central questions: 1) what are the causes of death associated with long term PPI use, and 2) what are the characteristics of those at highest risk of all-cause and cause-specific mortality associated with PPI use. This proposal falls under several ORD priorities including Quality/Safety, Primary Care Practice, and the results will directly have direct and substantial real-world impact. The proposal will serve an exemplar of transforming data into insightful actionable knowledge to improve the health and wellbeing of veterans. Innovation: The proposal will leverage the power of VA data, state of the art methodologies in pharmacoepidemiology, and cause of death ensemble modeling to address the aims of this proposal. Specific Aims: Specific Aim 1: To estimate the burden of cause-specific mortality among long term users of PPI. In this aim we will build an analytic approach using high-dimensional propensity scores, and instrumental variables in a cause of death ensemble model to estimate the survival probability of PPI users and non-users. This approach will allow the estimation of burden of cause-specific mortality associated with PPI use. Specific Aim 2: a) To identify predictors including (demographic and health characteristics) that modify the relationship between PPI and risk of all-cause and cause-specific mortality among long term PPI users. And b) to build and validate a risk score to stratify patients based on their risk of experiencing PPI-related cause-specific mortality. Methodologies: In aim 1 we will build a cohort of new users of acid suppression therapy including users of PPI and users of H2 Blockers (an active comparator group). To balance both groups, we will build a high- dimensional propensity score and an instrumental variable. We will then apply cause of death ensemble modeling system which incorporates several survival models including proportional hazard models, accelerated failure time models, and additive hazard models. The ensemble approach will allow more accurate estimation of the risk of death and cause-specific mortality associated with PPI use. In aim 2 we will select candidate risk score variables based on a) prior knowledge and biologic plausibility b) prevalence among PPI users or c) strength of effect modification (interaction) between PPI and outcome. Candidate variables will be further selected in logistic regression with backward selection, where a prognostic score representing the probability of...