SMARTs with treatment preference: Pragmatic SMART design and methods motivated by STAR*D

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Personalized Medicine

Background:

• Chronic conditions require adaptive treatment guidelines called dynamic treatment regimens (DTRs).
• Sequential, Multiple Assignment, Randomized Trials (SMARTs) are ideal for estimating DTRs but face recruitment challenges due to patient preferences.
• Existing methods struggle to integrate data from both randomized and non-randomized participants in trials that consider preferences.

Purpose of the Study:

• To evaluate Weighted and Replicated Regression Models (WRRMs) for estimating DTRs in Partially Randomized, Patient Preference SMART (PRPP-SMART) designs with continuous outcomes.
• To assess the robustness of WRRMs across varying patient preference rates and outcome distributions.
• To demonstrate the application of PRPP-SMART and WRRMs using data adapted from the STAR*D trial for major depressive disorder.

Main Methods:

• Utilized a Partially Randomized, Patient Preference SMART (PRPP-SMART) design, which randomizes only indifferent participants.
• Employed Weighted and Replicated Regression Models (WRRMs) to combine data from both randomized and non-randomized participants.
• Evaluated WRRMs for continuous outcomes, testing performance with diverse preference rates and outcome distributions.

Main Results:

• WRRMs demonstrated robust performance in estimating DTR effects for continuous outcomes within the PRPP-SMART framework.
• The method's accuracy was consistent across different patient preference levels and outcome distributions.
• Illustrative analysis using STAR*D-adapted data showcased the potential of PRPP-SMART for pragmatic clinical trials.

Conclusions:

• The PRPP-SMART design, coupled with WRRMs, offers a superior framework for estimating DTRs compared to traditional randomized clinical trials.
• This approach effectively addresses recruitment and retention issues by accommodating patient preferences.
• The proposed methods provide a foundation for developing more personalized and effective DTRs in future pragmatic clinical research.