Traditional clinical trials evaluate interventions as a single, static package. A participant is randomized to either a treatment group or a control group at the beginning of a study, and researchers measure the outcome weeks or months later. However, as Omer T. Inan et al. note in Digitizing clinical trials, the rise of mobile health (mHealth) and wearables allows for continuous, real-world data collection and dynamic support.
When a digital intervention fails—such as the massive drop-offs in user engagement highlighted by Jessica M. Lipschitz et al. in The Engagement Problem: a Review of Engagement with Digital Mental Health Interventions and Recommendations for a Path Forward—a traditional randomized controlled trial (RCT) cannot pinpoint why. Did the user get bored of the notifications? Was the timing wrong? Or was the content itself ineffective? To optimize these dynamic systems, researchers require a fundamentally different experimental paradigm.
The Mechanism of Micro-Randomization
To optimize Just-in-Time Adaptive Interventions (JITAIs), researchers use Micro-Randomized Trials (MRTs). As Tianchen Qian et al. explain in The microrandomized trial for developing digital interventions: Experimental design and data analysis considerations., instead of randomizing a participant once at the beginning of a study, an MRT randomizes participants hundreds or thousands of times at discrete "decision points" throughout the trial.
A decision point is any moment where an intervention could realistically be delivered—such as every morning, or whenever a wearable detects a prolonged period of inactivity. At each point, the system makes a probabilistic decision to deliver a specific intervention component or to remain passive. This high-frequency sequential randomization allows researchers to leverage within-person longitudinal data over time, a statistical framework originally formalized by Peng Liao et al. in Sample size calculations for micro‐randomized trials in mHealth.
Traditional RCT:
[Participant] ─── Randomization ───► [Treatment Group] ───────► End Outcome
└───► [Control Group] ─────────► End Outcome
Micro-Randomized Trial (MRT):
[Participant] ───► Decision Point 1 ───► Probabilistic Nudge? (Yes/No) ───► Proximal Outcome 1
───► Decision Point 2 ───► Probabilistic Nudge? (Yes/No) ───► Proximal Outcome 2
───► Decision Point N ───► Probabilistic Nudge? (Yes/No) ───► Proximal Outcome N
Why It Works: Isolating Proximal Causal Effects
The core statistical power of an MRT lies in its ability to isolate proximal causal effects—the immediate, near-term impact of a specific intervention component. In Designing m-Health interventions for precision mental health support, Niranjan Bidargaddi et al. emphasize that frequent randomization allows researchers to establish direct causal links between a single push notification and a user's immediate behavior, rather than waiting to evaluate a finished product as a single, unchangeable unit.
Furthermore, this probabilistic delivery solves a major confounding variable in digital health: the observer effect. As Noelle E. Carlozzi et al. demonstrate in Improving outcomes for care partners of persons with traumatic brain injury: Protocol for a randomized control trial of a just-in-time-adaptive self-management intervention, by randomly choosing whether or not to send a prompt at any given decision point, the experimental design statistically separates the impact of the actual content from the general effect of a user simply knowing they are being monitored by an app.
Mapping Context and Time-Varying Moderators
Human behavior is highly contextual. A nudge that works on Monday morning might fail or cause frustration on Friday night. MRTs excel at identifying time-varying moderators—the shifting environmental, temporal, or psychological states that influence whether an intervention succeeds.
For instance, in a study tracking medical interns, Timothy NeCamp et al. found in Assessing Real-Time Moderation for Developing Adaptive Mobile Health Interventions for Medical Interns: Micro-Randomized Trial that push notifications regarding mood and sleep were highly beneficial when interns were actively struggling, but had a negative impact when they were already performing well. As Emily B. Kroska et al. outline in Optimizing an Acceptance and Commitment Therapy Microintervention Via a Mobile App With Two Cohorts: Protocol for Micro-Randomized Trials, repeating these assignments over short intervals reveals how an intervention's efficacy shifts alongside a participant's immediate environment and psychological state.
By analyzing these interactions, MRTs provide the empirical foundation needed to build the decision-making logic for AI-driven health behavior systems, a goal highlighted by Chao Zhang 0071 et al. in Preface to Using AI Methods for Health Behavior Change. These dynamics can even be mapped across social structures. In Effectiveness of gamified team competition as mHealth intervention for medical interns: a cluster micro-randomized trial, Jitao Wang et al. extended this framework to cluster micro-randomized trials, randomizing entire teams over discrete intervals to measure how social contexts and temporal fatigue interact with gamified interventions.
From Micro-Interventions to Macro-Pathways
While MRTs focus on the granular, moment-to-moment optimization of digital triggers—such as the engagement strategies explored by Mashfiqui Rabbi et al. in Toward Increasing Engagement in Substance Use Data Collection: Development of the Substance Abuse Research Assistant App and Protocol for a Microrandomized Trial Using Adolescents and Emerging Adults and Alexandra M. Psihogios et al. in Understanding Adolescent and Young Adult 6-Mercaptopurine Adherence and mHealth Engagement During Cancer Treatment: Protocol for Ecological Momentary Assessment—they represent part of a broader shift toward adaptive trial architectures.
On a macro scale, Sequential Multiple Assignment Randomized Trials (SMARTs) apply a similar logic over longer time horizons. As Julien Rouvere et al. describe in Personalization Strategies for Increasing Engagement With Digital Mental Health Resources: Sequential Multiple Assignment Randomized Trial, SMART designs re-randomize participants at broader clinical decision points based on their ongoing response or characteristics. According to a scoping review by Nikki L. B. Freeman et al. in Design characteristics of Sequential Multiple Assignment Randomized Trials (SMARTs) for human health: a scoping review of studies between 2009-2024, this multi-stage adaptive randomization allows clinical researchers to systematically compare entire sequences of treatments to map out optimal, personalized clinical pathways.
By shifting randomization from a single upfront choice to a continuous, contextual process, both MRTs and SMARTs provide the statistical machinery required to turn static digital tools into precise, responsive health interventions.
Key Foundational Work
Protocol: Adaptive Implementation of Effective Programs Trial (ADEPT): cluster randomized SMART trial comparing a standard versus enhanced implementation strategy to improve outcomes of a mood disorders program (2014) — contributed 2 concept elements to this direction.
Steptacular: An incentive mechanism for promoting wellness (2012) — contributed 1 concept elements to this direction.