CAUSAL MODELING SOFTWARE FOR LONGITUDINAL STUDIES

DESCRIPTION (provided by applicant): This SBIR proposes to develop software for validly inferring causality from longitudinal observational data. Such data arise for example in epidemiologic studies, where an important goal is to determine how a treatment affects disease. Often treatments occur over time. Treatment decisions are based on other variables which also affect disease. Standard analysis procedures -- even with well intentioned adaptations -- cannot validly infer causal relationships from such data. The ultimate goal is to create software that (1) uses causal graphs to assist in defining causal models, (2) implements algorithms for causal models that estimate and test for causal effects, and (3) provides analytic tools to assess sensitivity to assumptions. In Phase I, we will implement a prototype, which can be completed and extended to further applications in Phase II. We will also address computational issues that need to be resolved for a practical implementation. Using the prototype on real data will motivate a preliminary software design. Our proposed software will provide tools to help researchers validly infer causality from longitudinal observational data, even when it is too expensive, not timely, or not ethical to conduct randomizedtrials. This software wilt impact the type of conclusions that can be reached from data collected to assess the effect of interventions on public health.