Experimental designs with repeated randomizations for optimizing adaptive interventions: guidelines and practical considerations

In clinical practice, as well as in other areas where interventions are provided, a sequential individualized approach to treatment is often necessary, whereby each treatment is adapted based on the object's response. An adaptive intervention is a sequence of decision rules which formalizes the provision of treatment at critical decision points in the care of an individual. In order to inform the development of an adaptive intervention, scientists are increasingly interested in the use of sequential multiple assignment randomized trials (SMART), which is a type of multi-stage randomized trial where individuals are randomized repeatedly at critical decision points to a set treatment options. While there is great interest in the use of SMART and in the development of adaptive interventions, both are relatively new to the medical and behavioral sciences. As a result, many clinical researchers will first implement a SMART pilot study (i.e., a small-scale version of a SMART) to examine feasibility and acceptability considerations prior to conducting a full-scale SMART study. A primary aim of this paper is to introduce a new methodology to calculate minimal sample size necessary for conducting a SMART pilot.

The logic of the SMART (Sequential Multiple Assignment Randomization Trial) design was applied to assess the replicability of original-replicate study pairs for Open Science Collaboration (OSC) intervention studies. Within SMART, we utilized both subtests of the correspondence test (CT) to assess study pair comparability. First, we implemented a CT difference test to determine if an effect size difference between the original study and its replicate pair was close to zero; second, we implemented a CT equivalence test to determine if the effect size difference of that study pair was within a designated threshold. In Stage 1 of SMART, each study pair was randomly assigned to one of two alphas (.01 and .05), thereby creating two, probabilistically similar subsets of study pairs. Within each alpha subset, successful difference tests (test of significance was not significantly different than zero) and unsuccessful difference tests were then determined. In Stage 2 of SMART, study pairs in each combination of alpha level and successful or unsuccessful difference tests were randomly assigned to one of two thresholds (±.25 SD, ±.50 SD). Equivalence tests were then conducted for all study pairs in each of these four subsets. Successful equivalence occurred when the distance between an original and its replicate pair was statistically significantly less than a given threshold. Thus, initial randomization followed by a second randomization was used to gauge comparability of each OSC original study and its replicate, for two alpha levels and two thresholds. In the first set of results, to mirror the common replicability assessment case in which only difference tests are conducted, 16 of 96 difference tests (16.7%) conducted in Stage 1 were successful. In the second set of results, for initially successful difference tests, two thresholds were used to determine the percent of study pairs that also passed the equivalence test. Depending on α and threshold, 8.0%-13.8% of studies successfully passed both difference and equivalence CT subtests. In the third set of results, using SMART, after randomization to two α-values and contingent on success or lack of success of a difference test, study pairs were randomized to two thresholds and a statistical test of equivalence conducted. Using meta-analysis methods within SMART-based subsets of study pairs, original-replicate average effect size differences were compared to differences in the second set of results. We found a similarly-sized 10.3% of study pairs passed both CT subtests (nine of 87 study pairs successfully passed the difference test at either alpha and successfully passed the equivalence test at either threshold). Reflecting the importance of incorporating both CT subtests, of 16 study pairs that initially passed the difference test, nearly half (43.7%) failed the equivalence test. Thus, for CT success, we found that α choice had little impact, while threshold choice was an important determinant. In all three sets of results, the percent of successful replications was substantially smaller than the 36% of OSC replicates that were statistically significant. To confirm this study’s replicability, we found very similar patterns of CT success and lack of success for two, SMART-based tables, one for alpha = .01 and one for alpha = .05. The current research extends the utility of CT established by Steiner and Wong (2018) in which results were based on simulation data.

Sequential Multiple Assignment Randomized Trials: An Opportunity for Improved Design of Stroke Reperfusion Trials

There are limited data on the effects of adaptive social communication interventions with a speech-generating device in autism. This study is the first to compare growth in communications outcomes among three adaptive interventions in school-age children with autism spectrum disorder (ASD) who are minimally verbal. Sixty-one children, ages 5–8 years, participated in a sequential, multiple-assignment randomized trial (SMART). All children received a developmental behavioral communication intervention: joint attention, symbolic play, engagement and regulation (JASP) with enhanced milieu teaching (EMT). The SMART included three 2-stage, 24-week adaptive interventions with different provisions of a speech-generating device (SGD) in the context of JASP+EMT. The first adaptive intervention, with no SGD, initially assigned JASP+EMT alone, then intensified JASP+EMT for slow responders. In the second adaptive intervention, slow responders to JASP+EMT were assigned JASP+EMT+SGD. The third adaptive intervention initially assigned JASP+EMT+SGD; then intensified JASP+EMT+SGD for slow responders. Analyses examined between-group differences in change in outcomes from baseline to Week 36. Verbal outcomes included spontaneous communicative utterances and novel words. Nonlinguistic communication outcomes included initiating joint attention and behavior regulation, and play. The adaptive intervention beginning with JASP+EMT+SGD was estimated as superior. There were significant (p < .05) between-group differences in change in spontaneous communicative utterances and initiating joint attention. School-age children with ASD who are minimally verbal make significant gains in communication outcomes with an adaptive intervention beginning with JASP+EMT+SGD. Future research should explore mediators and moderators of the adaptive intervention effects and second-stage intervention options that further capitalize on early gains in treatment.

Sequential, multiple assignment, randomized trials (SMARTs) compare sequences of treatment decision rules called dynamic treatment regimes (DTRs). In particular, the Adaptive Treatment for Alcohol and Cocaine Dependence (ENGAGE) SMART aimed to determine the best DTRs for patients with a substance use disorder. While many authors have focused on a single pairwise comparison, addressing the main goal involves comparisons of >2 DTRs. For complex comparisons, there is a paucity of methods for binary outcomes. We fill this gap by extending the multiple comparisons with the best (MCB) methodology to the Bayesian binary outcome setting. The set of best is constructed based on simultaneous credible intervals. A substantial challenge for power analysis is the correlation between outcome estimators for distinct DTRs embedded in SMARTs due to overlapping subjects. We address this using Robins' G-computation formula to take a weighted average of parameter draws obtained via simulation from the parameter posteriors. We use non-informative priors and work with the exact distribution of parameters avoiding unnecessary normality assumptions and specification of the correlation matrix of DTR outcome summary statistics. We conduct simulation studies for both the construction of a set of optimal DTRs using the Bayesian MCB procedure and the sample size calculation for two common SMART designs. We illustrate our method on the ENGAGE SMART. The R package SMARTbayesR for power calculations is freely available on the Comprehensive R Archive Network (CRAN) repository. An RShiny app is available at https://wilart.shinyapps.io/shinysmartbayesr/.

Minority adolescents are at highest risk for obesity and extreme obesity; yet, there are few clinical trials targeting African American adolescents with obesity. The purpose of the study was to develop an adaptive family-based behavioral obesity treatment for African American adolescents using a sequential multiple assignment randomized trial (SMART) design. Fit Families was a SMART where 181 African American adolescents (67% female) aged 12-17 were first randomized to office-based versus home-based behavioral skills treatment delivered from a Motivational Interviewing foundation. After 3 months, nonresponders to first phase treatment were rerandomized to continued home-based behavioral skills treatment or contingency management with voucher-based reinforcement for adolescent weight loss and for caregiver adherence to the program. All interventions were delivered by community health workers. The primary outcome was treatment retention and percent overweight. All adolescents reduced percent overweight by -3.20%; there were no significant differences in percent overweight based on treatment sequence. Adolescents receiving home-based delivery in Phase 1 and contingency management in Phase 2 completed significantly more sessions than those receiving office-based treatment and continued skills without CM (M = 8.03, SD = 3.24 and M = 6.62, SD = 2.95, respectively). The effect of contingency management was strongest among older and those with lower baseline confidence. Younger adolescents experienced greater weight reductions when receiving continued skills (-4.90% compared with -.02%). Behavioral skills training can be successfully delivered to African American adolescents with obesity and their caregivers by community health workers when using a home-based service model with incentives. More potent interventions are needed to increase reductions in percent overweight and may need to be developmentally tailored for younger and older adolescents.

Due to the cost and complexity of conducting a sequential multiple assignment randomized trial (SMART), it is desirable to pre-define a small number of personalized regimes to study. We proposed a simulation-based approach to studying personalized dosing strategies in contexts for which a therapeutic agent's pharmacokinetic and pharmacodynamics properties are well understood. We take dosing of warfarin as a case study, as its properties are well understood. We consider a SMART in which there are five intervention points in which dosing may be modified, following a loading phase of treatment. Realistic SMARTs are simulated, and two methods of analysis, G-estimation and Q-learning, are used to assess potential personalized dosing strategies. In settings where outcome modelling may be complex due to the highly non-linear nature of the pharmacokinetic and pharmacodynamics mechanisms of the therapeutic agent, G-estimation provides for which the more promising method of estimating an optimal dosing strategy. Used in combination with the simulated SMARTs, we were able to improve simulated patient outcomes and suggest which patient characteristics were needed to best individually tailor dosing. In particular, our simulations suggest that current dosing should be determined by an individual's current coagulation time as measured by the international normalized ratio (INR), their last measured INR, and their last dose. Tailoring treatment only based on current INR and last warfarin dose provided inferior control of INR over the course of the trial. The ability of the simulated SMARTs to suggest optimal personalized dosing strategies relies on the pharmacokinetic and pharmacodynamic models used to generate the hypothetical patient profiles. This approach is best suited to therapeutic agents whose effects are well studied. Prior to investing in a complex randomized trial that involves sequential treatment allocations, simulations should be used where possible in order to guide which dosing strategies to evaluate.

Interventions often require multiple decisions to improve outcomes for every student. Whether the decision to implement a practice, tailor an existing protocol, or change approaches, these decisions should be based on individual variables and outcomes via a sequence of treatment. To develop adaptive interventions that have sufficient evidence to support decisions, components, and sequences, they must be evaluated as they operate. The sequential multiple-assignment randomized trial is a design that experimentally assesses the efficacy of the decisions, components, and sequence of an adaptive intervention. The purpose of this article is to provide an overview of this novel methodology and describe how this design can provide meaningful information about components and sequence based on individual differences and response to maximize educational outcomes in special education.

A sequential multiple assignment randomized trial (SMART) protocol for empirically developing an adaptive preventive intervention for college student drinking reduction

M-bridge was a sequential multiple assignment randomized trial (SMART) that aimed to develop a resource-efficient adaptive preventive intervention (API) to reduce binge drinking in first-year college students. The main results of M-bridge suggested no difference, on average, in binge drinking between students randomized to APIs versus assessment-only control, but certain elements of the API were beneficial for at-risk subgroups. This paper extends the main results of M-bridge through an exploratory analysis using Q-learning, a novel algorithm from the computer science literature. Specifically, wesought to further tailorthe two aspects of the M-bridge APIs to an individual and test whether deep tailoring offers a benefit over assessment-only control. Q-learning is a method to estimate decision rules that assign optimal treatment(i.e., to minimize binge drinking) based on student characteristics. For the first aspect of the M-bridge API (when to offer), we identified the optimal tailoring characteristic post hoc from a set of 20 candidate variables. For the second (how to bridge), we used a known effect modifier from the trial. The results of our analysis are two rules that optimize (1) the timing of universal intervention for each student based on their motives for drinking and (2) the bridging strategy to indicated interventions (i.e., among those who continue to drink heavily mid-semester) based on mid-semester binge drinking frequency.We estimate that this newly tailored API, if offered to all first-year students, would reduce binge drinking by 1 occasion per 2.5 months (95% CI: decrease of 1.45 to 0.28 occasions, p < 0.01) on average. Our analyses demonstrate a real-world implementation of Q-learning for a substantive purpose, and, if replicable in future trials, our results have practical implications for college campuses aiming to reduce student binge drinking.

I-SPY2 is a long-running phase 2 platform trial that evaluates neoadjuvant treatments for locally advanced breast cancer to identify those with high efficacy that are likely to be successful in phase 3 trials, assigning patients to novel agents using response-adaptive randomization (RAR). Recently, I-SPY2 was reconfigured as a sequential multiple assignment randomized trial (SMART), with up to three stages of therapy. At the first stage, a patient is assigned to a tumor-subtype-specific therapy. If the patient fails to show a satisfactory response, the patient is assigned to a second subtype-specific therapy, and receives a third, rescue therapy if response is still not achieved. The I-SPY2 SMART thus supports identification of highly efficacious entire treatment regimes. The transition of I-SPY2 to a SMART required development of a RAR scheme that updates randomization probabilities at each stage, aligned with the goal of maximizing the number of patients who achieve a pathological complete response (pCR). We present our Bayesian RAR approach, which updates randomization probabilities based on the posterior probability that treatments are part of the optimal regime. Empirical studies demonstrate that it results in more patients having treatment experience consistent with highly efficacious regimes, improves overall within-trial pCR rates, and identifies optimal regimes post trial at rates similar to or exceeding those under simple, uniform, nonadaptive randomization.

Adherence to medications for chronic conditions is often very low, limiting the benefit to patients, even when the medications are effective and have favorable side effect profiles. This article reviews some of the prior work on treatment adherence, introduces novel concepts from behavioral economics that can be used to design interventions to improve adherence, and proposes new approaches for clinical trials. Relevant experience of the authors and insights from the literature were combined to identify key issues and propose methodological improvements. Specific examples regarding adherence to warfarin are provided. Several new approaches to trial design can be effectively applied in the context of medication adherence. These include tailored intervention strategies and sequential multiple assignment randomized trial (SMART) designs. While we have proposed to use these new approaches for ongoing studies of adherence in behavioral health, practical experience with their application is still somewhat limited. Behavioral economics offer a variety of concepts that, when used in the design of interventions to improve adherence, may be more successful than traditional approaches. New clinical trial designs can also be adopted to improve the efficiency of studies that assess these approaches.

Advances in digital technologies have created unprecedented opportunities to deliver effective and scalable behavior change interventions. Many digital interventions include multiple components, namely several aspects of the intervention that can be differentiated for systematic investigation. Various types of experimental approaches have been developed in recent years to enable researchers to obtain the empirical evidence necessary for the development of effective multiple-component interventions. These include factorial designs, Sequential Multiple Assignment Randomized Trials (SMARTs), and Micro-Randomized Trials (MRTs). An important challenge facing researchers concerns selecting the right type of design to match their scientific questions. Here, we propose MCMTC – a pragmatic framework that can be used to guide investigators interested in developing digital interventions in deciding which experimental approach to select. This framework includes five questions that investigators are encouraged to answer in the process of selecting the most suitable design: (1) Multiple-component intervention: Is the goal to develop an intervention that includes multiple components; (2) Component selection: Are there open scientific questions about the selection of specific components for inclusion in the intervention; (3) More than a single component: Are there open scientific questions about the inclusion of more than a single component in the intervention; (4) Timing: Are there open scientific questions about the timing of component delivery, that is when to deliver specific components; and (5) Change: Are the components in question designed to address conditions that change relatively slowly (e.g., over months or weeks) or rapidly (e.g., every day, hours, minutes). Throughout we use examples of tobacco cessation digital interventions to illustrate the process of selecting a design by answering these questions. For simplicity we focus exclusively on four experimental approaches—standard two- or multi-arm randomized trials, classic factorial designs, SMARTs, and MRTs—acknowledging that the array of possible experimental approaches for developing digital interventions is not limited to these designs.

This pilot study was conducted to assess the feasibility and acceptability of 4 adaptive treatment strategies (ATSs) for adolescent depression to plan for a subsequent full-scale clinical trial. The ATSs aim to address 2 questions that arise when personalizing treatment: (a) For adolescents treated with Interpersonal Psychotherapy for depressed adolescents (IPT-A; Mufson et al., 2004), at what time point should therapists make the determination that the adolescent is not likely to respond if the initial treatment plan is continued (week 4 or week 8)? (b) For adolescents who are judged to need their treatment augmented, should the therapist increase the number of IPT-A sessions or add pharmacotherapy (fluoxetine)? A 16-week pilot sequential multiple assignment randomized trial (SMART) was conducted with 32 adolescents (M age = 14.9) who had a diagnosis of major depressive disorder, dysthymic disorder, or depressive disorder not otherwise specified. Adolescents were primarily female (75%) and Caucasian (84.4%). Data regarding the feasibility and acceptability of the study and treatment procedures and treatment response rates were collected. Week 4 was the more feasible and acceptable decision point for assessing need for a change to treatment. Adolescents, parents, and therapists reported a range of attitudes about medication and more intensive therapy as treatment options. Results from the pilot study have yielded additional research questions for the full-scale SMART and will improve our ability to successfully conduct the trial.

Results from randomized control trials (RCTs) may not be representative when individuals refuse to be randomized or are excluded for having a preference for which treatment they receive. If trial designs do not allow for participant treatment preferences, trials can suffer in accrual, adherence, retention, and external validity of results. Thus, there is interest surrounding clinical trial designs that incorporate participant treatment preferences. We propose a Partially Randomized, Patient Preference, Sequential, Multiple Assignment, Randomized Trial (PRPP-SMART) which combines a Partially Randomized, Patient Preference (PRPP) design with a Sequential, Multiple Assignment, Randomized Trial (SMART) design. This novel PRPP-SMART design is a multi-stage clinical trial design where, at each stage, participants either receive their preferred treatment, or if they do not have a preferred treatment, they are randomized. This paper focuses on the clinical trial design for PRPP-SMARTs and the development of Bayesian and frequentist weighted and replicated regression models (WRRMs) to analyze data from such trials. We propose a two-stage PRPP-SMART with binary end of stage outcomes and estimate the embedded dynamic treatment regimes (DTRs). Our WRRMs use data from both randomized and non-randomized participants for efficient estimation of the DTR effects. We compare our method to a more traditional PRPP analysis which only considers participants randomized to treatment. Our Bayesian and frequentist methods produce more efficient DTR estimates with negligible bias despite the inclusion of non-randomized participants in the analysis. The proposed PRPP-SMART design and analytic method is a promising approach to incorporate participant treatment preferences into clinical trial design.