Universal Sensitivity Analysis for Unmeasured Confounding in Drug-Related Public Policy Evaluation - Project Summary Unmeasured confounding is a major source of bias in causal inference for drug-related public policy evaluation, and a sensitivity analysis is typically needed to examine how sensitive a related causal conclusion is to unmeasured confounding. Existing sensitivity analysis methods are underdeveloped in drug-related policy evaluation and can severely harm the evidence for (or against) causal claims. For example, in matched observational studies, one of the most widely used causal inference methods in policy research, existing sensitivity analysis methods typically focus on the case when the treatment is binary or there is a single outcome; also, they often ignore possible subgroup-specific effects. However, many drug-related policy measures are non-binary (e.g., ordinal or continuous), such as alcohol or tobacco tax rates, minimum legal purchase ages, alcohol policy scores, tobacco control index, and mobility scores. Also, drug-related policies are often evaluated using several outcomes, either those related to different types of drug use, or those related to different aspects of society such as health, justice, and economics. Finally, due to existing disparities in drug-related outcomes, there is an intense focus on accurately measuring the effects of drug-related policies among subgroups defined by race and socioeconomic status. The broad objective of this project is to develop a universal sensitivity analysis framework for unmeasured confounding in matched observational studies that can work with binary or non-binary treatments, single or multiple outcomes, and overall or subgroup-specific effects. There are three specific aims. Aim 1 will develop a universal sensitivity analysis framework for matched observational studies with general (binary or non-binary) treatments. Aim 2 will further develop the sensitivity analysis for multiple outcomes and subgroup-specific effects. Aim 3 will illustrate the proposed sensitivity analysis by studying the causal influences of mobility policies, such as social distancing policies and transportation policies where the measures (e.g., mobility scores) are continuous in nature, on drug-related outcomes (e.g., drug overdose deaths, tobacco use, and excessive drinking). We will evaluate the effect on the overall population and those among different racial groups. Aim 3 will also develop a publicly available and user-friendly R package to implement our universal sensitivity analysis framework.
