Research Projects
Research Projects
Optimal Learning From SMARTs With Complex Patient-Centered Outcomes
Project Summary
Some health conditions require treatment adjustments over time based on how a patient responds. Researchers can study these treatment sequences using a study design called a Sequential Multiple Assignment Randomized Trial (SMART). However, it can be difficult to analyze these studies when events such as a disease returning, stopping treatment or death occur before the outcome researchers are studying.
This PCORI-funded research project will develop practical new methods for accounting for these events when comparing treatment strategies that change over time. The methods are intended to help researchers produce more reliable evidence about which treatment sequence may work best for different patients and to support patients and healthcare professionals in making informed treatment decisions.
Statistical Methods
Causal Inference, Cumulative Incidence, Cause-Specific Hazards, Random Survival Forests, Double-Robust Estimators, Q-Learning.
Funding Support
Patient-Centered Outcomes Research Institute
Pragmatic Comparative Effectiveness Using SMARTs
Project Summary
Sequential Multiple Assignment Randomized Trials (SMARTs) provide an avenue for testing the comparative clinical effectiveness of patient-centered treatment strategies. The overall goal of this project is to address several challenges related to the analysis of data from SMARTs, including inflexibility in adjusting treatment allocation, unavailability of interim monitoring methods, and inadequate methods to handle missing data.
We develop pragmatic adaptations to randomization schemes in SMARTs to improve adherence and retention that account for treatment efficacies from earlier stages of the trial. We also develop and evaluate valid interim monitoring approaches to allow early termination of SMARTs for efficacy and/or futility for binary and survival endpoints, and we further develop and test approaches based on machine learning (ML) to handle dropouts and other missing treatment data in multiple stages of SMARTs.
Statistical Methods
Causal Inference, Outcome-adaptive Randomization, Q-Learning, Imputation, Double-Robust Estimators, log-rank tests.