A generalized outcome-adaptive sequential multiple assignment randomized trial design

Area of Science:

• Biostatistics
• Clinical Trial Design
• Health Services Research

Background:

• Dynamic treatment regimes (DTRs) guide sequential treatment decisions for chronic diseases.
• Sequential Multiple Assignment Randomized Trials (SMARTs) construct DTRs but ignore historical patient data.
• Ignoring past data in SMARTs can lead to suboptimal treatment assignments and reduced patient adherence.

Purpose of the Study:

• To propose a Generalized Outcome-adaptive (GO) SMART design to improve DTR construction.
• To address the limitation of standard SMARTs by incorporating historical patient data.
• To develop statistical methods for unbiased estimation of DTR effects within the GO-SMART framework.

Main Methods:

• Introduced a GO-SMART design that adaptively adjusts randomization probabilities based on previous treatment effectiveness.
• Proposed G-estimators and inverse-probability-weighted estimators to correct for bias in adaptive randomization.
• Conducted analytical derivations and simulations to assess the performance of the GO-SMART design.

Main Results:

• The GO-SMART design significantly increased the proportion of patients receiving the optimal DTR.
• GO-SMART achieved a higher number of total patient responses compared to standard SMART designs.
• The proposed methods demonstrated consistency and maintained statistical power comparable to or better than existing adaptive randomization strategies.

Conclusions:

• The GO-SMART design offers an effective strategy for optimizing sequential treatment selection in clinical practice.
• This adaptive approach enhances treatment efficacy and patient adherence by leveraging historical treatment outcome data.
• The developed statistical estimators provide reliable methods for evaluating DTRs within adaptive trial frameworks.