Project Summary

Contemporary comparative clinical effectiveness research is almost universally conducted across diverse health systems, and many studies are cluster-randomized trials (CRTs). In a CRT, the clusters could be any higher-level unit that engages with a patient population, e.g., surgeons, therapists or primary care clinics. While the total causal effect has historically been the focus of CRTs, there is also enormous interest in understanding the mechanism by which a cluster-level intervention improves patient-centered outcomes, so that the intervention pathways for improving treatment benefits to patients can be identified and targeted. 

Causal mediation methods are well-suited for this objective, and findings from causal mediation analyses hold great promise to advance process evaluation, inform healthcare policy and optimize system-level interventions. While causal mediation methods for nonclustered data are well developed, guidance for assessing causal mediation in CRTs within CRTs is relatively scarce, prohibiting robust knowledge generation and evidence translation through embedded healthcare delivery and implementation science research. This study will fill in this gap in methodology for causal mediation analysis by developing new estimands, model-robust estimators and sensitivity methods in CRTs, along with accessible software that integrates these connected components. 

First, researchers will decompose the cluster-level total effect parameter into indirect and direct effects. The indirect effect summarizes the extent to which the total effect is channeled through the intermediate outcome. The study team will further decompose the indirect effect into individual and spillover mediation effects to study the extent to which the indirect effect works through the mediators of other patients in the same cluster. The team will characterize the causal assumptions and formulas for estimating each of these target values, and further motivate model-robust methods that protect against biases from model misspecification. Results will also be developed based on the individual-level total effect parameter. 

Second, this study will expand the causal mediation framework to accommodate two or more mediators with an unknown structure in CRTs. Researchers will develop new mediation effect measures that can account for the separate and interactive effects among the mediators. For each mediation measure, the research team will characterize the formulas for point estimation and develop model-robust strategies for inference, and will carry out simulations under plausible data structures to compare the new methods to standard alternatives. 

Third, the study team will develop a framework to study causal mediation effects under departures from key assumptions in CRTs. For example, we will derive interpretable confounding functions to assess the magnitude and direction of violating the no unmeasured confounding assumption. Researchers will also consider specific forms of the confounding functions to represent several common scenarios of interest and develop methods to bias-correct the effect estimates accordingly. 

Finally, researchers will apply these new methods to completed CRTs and create open-access statistical software to facilitate their implementation by other research teams. To maximize the tools' value and dissemination to the broader research community, the team will work with stakeholders, including principal investigators of completed CRTs and leaders in the field of pragmatic cluster randomized trials. Collectively, this study introduces new causal inference tools to enhance the rigor of CRTs and will enable researchers to better operationalize several current PCORI Methodology Standards.

Project Information

Fan Li, Ph.D.
Yale University
$985,945 *

Key Dates

36 months *
November 2023
2023

*All proposed projects, including requested budgets and project periods, are approved subject to a programmatic and budget review by PCORI staff and the negotiation of a formal award contract.

Tags

Award Type
State State The state where the project originates, or where the primary institution or organization is located. View Glossary
Last updated: January 24, 2024