Causal Inference with Limited Resources

Abstract

Constraints on treatment resources present problems in many practical settings. For example, during the current coronavirus disease 2019 pandemic, several health care systems have experienced shortages of ventilators, protective equipment and personnel. Similarly, patients with organ failures are allocated to waiting lists because the number of organ transplants are scarce. Investigators are often interested in causal effects of different treatment strategies in these settings, but classical causal inference methods often fail to explicitly consider the resource constraints. This thesis develops a suite of methods for resource limited settings, focusing on the tasks of defining and identifying new causal estimands. In Chapters 1 and 2, this thesis develops a general framework for evaluating the effect of counterfactual treatment allocation and prioritization regimes for limited resource settings. Crucially, this thesis represents the first contribution to consider limited resources with data viewed as a single cluster of causally-connected patients - an observed data structure ubiquitous in limited resource settings. Chapter 2 extends this framework to consider regimes and identification strategies that are specifically tailored for health crises. Chapter 3 departs from the framework of Chapters 1 and 2 and develops an estimand that can be used to define causal effects of treatment strategies that satisfy resource constraints: incremental propensity score interventions for limited resources. Chapter 3 derives a simple class of inverse probability weighted estimators, and applies one such estimator to evaluate the effect of restricting or expanding utilization of ‘increased risk’ liver organs to treat patients with end-stage liver disease.