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

Effective care for chronic conditions with high rates of non-response or relapse requires personalized and adaptive treatment guidelines known as dynamic treatment regimens (DTRs). Sequential, multiple assignment, randomized trials (SMARTs) are the gold standard for estimating DTRs, but SMARTs, like any trial, may struggle with recruitment and retention due to patient treatment preferences. A partially randomized, patient preference SMART (PRPP-SMART) design overcomes these issues by assigning participants with a preference to their preferred treatment and randomizing indifferent participants at each stage of the SMART. The PRPP-SMART design and methods were motivated in part by STAR*D (NCT00021528), which produced second- and third-line treatment recommendations for patients with major depressive disorder (MDD) who tend to have comorbidities, prior treatment history, and treatment preferences. Traditional, randomized clinical trials typically fail to recruit generalizable study populations, and traditional methods are not appropriate to combine data from randomized and non-randomized participants. We have previously shown that weighted and replicated regression models (WRRMs) combining data from all participants in a PRPP-SMART estimate DTR effects given binary outcomes with minimal bias. Here, we evaluate WRRMs to estimate PRPP-SMART DTRs with continuous outcomes and find that the performance of our method is robust to different preference rates and outcome distributions. We illustrate our method using data adapted from STAR*D which considered preferences in treatment assignment but did not compare DTRs. The PRPP-SMART design and methods presented here would have overcome many shortcomings of STAR*D and establish a framework for meaningful DTR estimates in future pragmatic SMARTs.