Causal inference in the absence of positivity: The role of overlap weights.

How to analyze data when there is violation of the positivity assumption? Several possible solutions exist in the literature. In this paper, we consider propensity score (PS) methods that are commonly used in observational studies to assess causal treatment effects in the context where the positivity assumption is violated. We focus on and examine four specific alternative solutions to the inverse probability weighting (IPW) trimming and truncation: matching weight (MW), Shannon's entropy weight (EW), overlap weight (OW), and beta weight (BW) estimators. We first specify their target population, the population of patients for whom clinical equipoise, that is, where we have sufficient PS overlap. Then, we establish the nexus among the different corresponding weights (and estimators); this allows us to highlight the shared properties and theoretical implications of these estimators. Finally, we introduce their augmented estimators that take advantage of estimating both the propensity score and outcome regression models to enhance the treatment effect estimators in terms of bias and efficiency. We also elucidate the role of the OW estimator as the flagship of all these methods that target the overlap population. Our analytic results demonstrate that OW, MW, and EW are preferable to IPW and some cases of BW when there is a moderate or extreme (stochastic or structural) violation of the positivity assumption. We then evaluate, compare, and confirm the finite-sample performance of the aforementioned estimators via Monte Carlo simulations. Finally, we illustrate these methods using two real-world data examples marked by violations of the positivity assumption.

Characteristics of High-Performing Hospitals in Cardiogenic Shock Following Acute Myocardial Infarction.

Cardiogenic shock after acute myocardial infarction (AMI-CS) carries significant mortality despite advances in revascularization and mechanical circulatory support. We sought to identify the process-based and structural characteristics of centers with lower mortality in AMI-CS. We analyzed 16,337 AMI-CS cases across 440 centers enrolled in the National Cardiovascular Data Registry's Chest Pain-MI Registry, a retrospective cohort database, between January 1, 2015, and December 31, 2018. Centers were stratified across tertiles of risk-adjusted in-hospital mortality rate (RAMR) for comparison. Risk-adjusted multivariable logistic regression was also performed to identify hospital-level characteristics associated with decreased mortality. The median participant age was 66 (interquartile range 57 to 75) years, and 33.0% (n = 5,390) were women. The median RAMR was 33.4% (interquartile range 26.0% to 40.0%) and ranged from 26.9% to 50.2% across tertiles. Even after risk adjustment, lower-RAMR centers saw patients with fewer co-morbidities. Lower-RAMR centers performed more revascularization (92.8% vs 90.6% vs 85.9%, p <0.001) and demonstrated better adherence to associated process measures. Left ventricular assist device capability (odds ratio [OR] 0.78 [0.67 to 0.92], p = 0.002), more frequent revascularization (OR 0.93 [0.88 to 0.98], p = 0.006), and higher AMI-CS volume (OR 0.95 [0.91 to 0.99], p = 0.009) were associated with lower in-hospital mortality. However, several such characteristics were not more frequently observed at low-RAMR centers, despite potentially reflecting greater institutional experience or resources. This may reflect the heterogeneity of AMI-CS even after risk adjustment. In conclusion, low-RAMR centers do not necessarily exhibit factors associated with decreased mortality in AMI-CS, which may reflect the challenges in performing outcomes research in this complex population.

Defining the Need for Causal Inference to Understand the Impact of Social Determinants of Health: A Primer on Behalf of the Consortium for the Holistic Assessment of Risk in Transplantation (CHART).

Objective: This study aims to introduce key concepts and methods that inform the design of studies that seek to quantify the causal effect of social determinants of health (SDOH) on access to and outcomes following organ transplant. Background: The causal pathways between SDOH and transplant outcomes are poorly understood. This is partially due to the unstandardized and incomplete capture of the complex interactions between patients, their neighborhood environments, the tertiary care system, and structural factors that impact access and outcomes. Designing studies to quantify the causal impact of these factors on transplant access and outcomes requires an understanding of the fundamental concepts of causal inference. Methods: We present an overview of fundamental concepts in causal inference, including the potential outcomes framework and direct acyclic graphs. We discuss how to conceptualize SDOH in a causal framework and provide applied examples to illustrate how bias is introduced. Results: There is a need for direct measures of SDOH, increased measurement of latent and mediating variables, and multi-level frameworks for research that examine health inequities across multiple health systems to generalize results. We illustrate that biases can arise due to socioeconomic status, race/ethnicity, and incongruencies in language between the patient and clinician. Conclusions: Progress towards an equitable transplant system requires establishing causal pathways between psychosocial risk factors, access, and outcomes. This is predicated on accurate and precise quantification of social risk, best facilitated by improved organization of health system data and multicenter efforts to collect and learn from it in ways relevant to specialties and service lines.