Models and algorithms to solve robust causal inference problems from large-scale observational data

Abstract

Causal inference enables us to discover key insights from data by analyzing how an action or intervention impacts the desired outcome. Identifying cause-effect relations is one of the most important steps in making policy decisions. While conducting a randomized experiment is the gold standard approach for making causal inference, randomized experiments are often infeasible due to legal, ethical, time, and budget constraints. In addition, in today's data-rich world, most of the data are collected as natural observation over time and are called observational data. In this dissertation, we present scalable algorithms and techniques to make robust causal inference from large-scale observational data. First, we consider causal inference testing with binary outcomes and propose computationally efficient algorithms that are adaptable for large-scale observational studies. By leveraging the structure of the optimization model, we propose a robustness condition that further reduces the computational burden. We validate the efficiency of the proposed algorithms by testing the causal relation between the Hospital Readmission Reduction Program (HRRP) and readmissions to different hospital from the State of California Patient Discharge Database. Second, we develop a unique computational framework to solve the robust causal inference test instances from observational data with continuous outcomes. We propose a unique framework to reformulate the nonlinear binary optimization problems as feasibility problems. By leveraging the structure of the feasibility formulation, we develop greedy schemes that are efficient in solving robust test problems. In many cases, the proposed algorithms achieve global optimal solution. Finally, we propose an Outcome Adaptive Elastic-Net (ONet) method specifically designed for causal inference to select the confounders and outcome predictors for inclusion in the propensity score model or in the matching mechanism. ONet provides two major advantages over existing methods: it performs superiorly on correlated data, and it can be applied to any matching method and any estimates. We apply the proposed ONet on National Survey of Drug Use and Health (NSDUH) data to identify the effect of opioid use disorder on suicidal behavior.--Author's abstract