Using propensity scores to estimate effects of treatment initiation decisions: State of the science

Abstract

Confounding can cause substantial bias in nonexperimental studies that aim to estimate causal effects. Propensity score methods allow researchers to reduce bias from measured confounding by summarizing the distributions of many measured confounders in a single score based on the probability of receiving treatment. This score can then be used to mitigate imbalances in the distributions of these measured confounders between those who received the treatment of interest and those in the comparator population, resulting in less biased treatment effect estimates. This methodology was formalized by Rosenbaum and Rubin in 1983 and, since then, has been used increasingly often across a wide variety of scientific disciplines. In this review article, we provide an overview of propensity scores in the context of real-world evidence generation with a focus on their use in the setting of single treatment decisions, that is, choosing between two therapeutic options. We describe five aspects of propensity score analysis: alignment with the potential outcomes framework, implications for study design, estimation procedures, implementation options, and reporting. We add context to these concepts by highlighting how the types of comparator used, the implementation method, and balance assessment techniques have changed over time. Finally, we discuss evolving applications of propensity scores.

Assessment of Obesity and Cardiometabolic Status by Integrase Inhibitor Use in REPRIEVE: A Propensity-Weighted Analysis of a Multinational Primary Cardiovascular Prevention Cohort of People With Human Immunodeficiency Virus

“…This study was cross-sectional and was not a randomized trial of INSTI versus non-INSTI-based regimens. To account for this, we performed a rigorous analysis using IPT weighting methodology that successfully balanced measured confounders between INSTI users and nonusers in a weighted analysis population, reducing potential bias of the treatment effect estimates [ 27 ]. We performed sensitivity analyses to account for differences in ART duration and NRTI regimens—specifically, TAF—between groups.…”

Introduction to computational causal inference using reproducible Stata, R, and Python code: A tutorial

“…However, there are other estimators based on matching strategies that we did not cover here 19 . Readers can find a more detailed overview of the propensity score and matching methods in a recently published article 48 …”

Optimal radiotherapy dose in cervical esophageal squamous cell carcinoma patients treated with definitive concurrent chemoradiotherapy: A population based study

“…We adopted a propensity score (PS) approach and used PS weighting (PSW) as the framework for analyses as advocated in the literature. 19 , 20 , 21 , 22 We estimated the probability of receiving a high radiotherapy dose (vs. standard dose) with a logistic regression model based on all the above covariates, and then assessed the balance of covariates between groups after PSW using overlap weight 23 via the standardized difference (SDif) rather than the chi‐square or t ‐test. 19 , 20 We compared the hazard ratio (HR) of death between the group A and group B groups during the entire follow‐up period via Cox proportional hazards model in the weighted sample for point estimation and used the bootstrap method to estimate the 95% confidence interval (95% CI).…”

“… 19 , 20 , 21 , 22 We estimated the probability of receiving a high radiotherapy dose (vs. standard dose) with a logistic regression model based on all the above covariates, and then assessed the balance of covariates between groups after PSW using overlap weight 23 via the standardized difference (SDif) rather than the chi‐square or t ‐test. 19 , 20 We compared the hazard ratio (HR) of death between the group A and group B groups during the entire follow‐up period via Cox proportional hazards model in the weighted sample for point estimation and used the bootstrap method to estimate the 95% confidence interval (95% CI). 24 , 25 , 26 We used E‐value to assess the robustness of our finding regarding potential unmeasured confounder(s) as suggested in the literature 27 because a PS approach is only be valid under the assumption of no unmeasured confounder(s).…”