Master protocol with adaptive design is a new complex innovative trial design that combines an adaptive treatment strategy and master protocol. It is more flexible and adjustable. In the complex clinical trial environment, the dynamics emphasized in this design are consistent with the idea of traditional Chinese medicine (TCM) syndrome differentiation and treatment. In this study, we summarized its concept, characteristics and advantages, and we also discussed its application in TCM clinical research. We hope this paper can provide more thinking and suggestions for TCM clinical trials.
Objective This study aims to systematically review the current application status of sequential multiple assignment randomized trials (SMART) in the past decade. The goal is to clarify the research fields, research objectives, design elements, and data analysis methods of SMART clinical reports, and to provide evidence-based references for the subsequent standardized design and reporting of SMART. Methods The PubMed, Embase, Web of Science, APA PsychInfo, Scopus, CNKI, WanFang Data, VIP databases were electronically searched to collect studies on SMART-related clinical studies published from 2015 to 2025. Descriptive statistics and inductive thematic analysis methods were used to summarize and analyse the extracted data. Results A total of 153 articles were included. The results showed that the number of publications has been increasing year by year; the research was mainly concentrated in the United States (n=133), followed by China (n=11). The research fields were mainly in psychology and psychiatry (42%), endocrinology (12%), and cancer (11%). The research goals were diverse, with the comparison of dynamic treatment strategies (14%) being the most common. In terms of trial design, the initial grouping was mostly two groups, with a 1: 1 ratio between groups being the most common; two-stage multiple randomizations were mostly used, ultimately forming 4-8 subgroups; sample sizes were mostly between 100 and 500 cases (48%). Data analysis methods were diverse, depending on the research purpose, data characteristics, and design type. Longitudinal data analysis mainly used linear mixed-effects models (66 times) and generalized estimating equations (31 times), and Q-learning (16 times) was the mainstream method for constructing optimal decision rules. Additionally, the study found that the detail related to data processing was generally underreported. Conclusion As a primary method for evaluating clinical dynamic treatment strategies, SMART has issues such as imbalanced geographical and disciplinary distribution, incomplete reporting of design elements, and insufficient standardization of data processing. In the future, it is necessary to promote the expansion of international reporting standards, strengthen methodological research, and encourage the validation of its extrapolation and clinical translation value in a wider range of disease fields and regions.
