Abstract
Despite the increased adoption of SMART designs, substantial reporting variability persists. Most primary analyses underutilize the capability of SMARTs to generate data for developing dynamic treatment regimes. SMART-specific standardized reporting guidelines can help accelerate the scientific and clinical impact of SMARTs.
Introduction
In clinical practice, healthcare is rarely a single event; it is a sequence of "if-then" decisions. For example, if a patient does not respond to an initial medication after four weeks, a clinician must decide whether to increase the dosage, switch to a different drug, or add a behavioral therapy. Traditional Randomized Controlled Trials (RCTs) are often too static to evaluate these adaptive sequences. Sequential Multiple Assignment Randomized Trials (SMARTs) were developed to address this gap, allowing researchers to randomize participants at multiple decision points based on their ongoing response to treatment.
While the adoption of SMART designs has accelerated, their complexity—involving multiple stages, re-randomization, and embedded treatment paths—presents significant reporting challenges. This scoping review examines 88 SMART studies published between 2009 and 2024 to identify trends, reporting gaps, and the degree to which these trials fulfill their promise of informing Dynamic Treatment Regimes (DTRs).
Methodology
The researchers conducted a comprehensive scoping review following the PRISMA-ScR (Scoping Reviews) extension. The team systematically searched five major databases, including PubMed and Embase, for SMART-related protocol and primary analysis papers published between January 2009 and January 2024.
To ensure methodological rigor, the study employed dual independent screening of 5,486 citations. Data extraction focused on granular operational details unique to SMARTs, such as re-randomization criteria, second-stage allocation ratios, and the specific statistical methods used to analyze DTRs. Unlike previous reviews that focused on specific niches like oncology, this study encompassed all human health domains.
The Landscape of SMART Research
The final sample included 88 studies, consisting of 59 protocol papers and 29 primary analysis papers. The review highlighted several key demographic and geographic trends:
- Growth Trend: SMART adoption is increasing rapidly, with over 80% of the included papers published in 2019 or later.
- Domain Distribution: Behavioral and mental health emerged as the most frequent therapeutic areas utilizing this design.
- Geographic Concentration: The majority of studies (75–83%) were conducted in the United States, suggesting a need for broader global adoption.
- Sample Sizes: Most protocols targeted 500 or fewer participants, while completed primary analyses typically reported on trials with 200 or fewer participants.
Key Insights and Reporting Gaps
The "DTR Underutilization" Trap
The primary purpose of a SMART is to develop and evaluate a Dynamic Treatment Regime (DTR). However, the review found a significant "translation gap": only 48.3% of primary analysis papers actually evaluated these embedded regimes. Many researchers essentially treated the SMART as a series of disconnected single-stage trials, failing to leverage the design's full potential for precision medicine.
Missing Operational Blueprints
Critical design elements are frequently omitted from publications. While most protocols reported sample size calculations, only 74.6% of those calculations actually accounted for the complexity of multiple randomizations. Furthermore, only 37.9% of primary analyses reported allocation ratios for the second stage of the trial, hindering replicability and evidence synthesis.
Blinding and Analytic Rigor
Blinding is notoriously difficult in behavioral SMARTs, yet only 65% of papers discussed whether blinding was attempted or why it was not feasible. Regarding statistical analysis, researchers tended to favor simpler methods like "weighting and replication" (31%). Advanced causal inference methods, such as G-computation or Targeted Maximum Likelihood Estimation (TMLE), were used in only a single paper each. Notably, zero primary analysis papers explored "tailored" DTRs learned through machine learning or reinforcement learning.
Limitations
The review is subject to several limitations, including a language bias (only English-language publications were included) and a potential publication gap, as SMARTs described only in trial registries without published protocols or results were not captured. Additionally, the study assessed reporting consistency rather than the underlying clinical quality or risk of bias of the trials themselves.
Conclusion and Future Directions
The findings reveal that while the use of SMARTs is growing, the field lacks a standardized framework for reporting. To bridge the gap between complex trial design and clinical application, the authors advocate for the development of a "CONSORT for SMARTs"—a standardized reporting checklist. Such guidelines would ensure that researchers document the "if-then" logic and statistical nuances necessary to move personalized, adaptive interventions from research papers into clinical practice.
Summary updated