Just-in-Time Adaptive Interventions for Dietary Adherence: A Research and Clinical Application Framework

Abstract

This article synthesizes current evidence and methodologies for Just-in-Time Adaptive Interventions (JITAIs) in promoting dietary adherence. Targeting researchers and clinical professionals, it explores the foundational principles of JITAIs that deliver personalized, context-aware support via mobile technology. The content covers the design frameworks, including the core components outlined by Nahum-Shani et al., and their application in conditions like hypertension, obesity, and type 2 diabetes. It critically examines implementation challenges, such as user engagement and the balance between personalization and burden, and reviews empirical evidence from recent trials. Finally, it discusses future directions for optimizing these dynamic interventions to improve long-term health outcomes in biomedical research and clinical practice.

The Science of JITAIs: Foundations and Core Concepts for Dietary Behavior Change

Defining JITAIs and Their Role in Modern Digital Health

Just-in-Time Adaptive Interventions (JITAIs) represent a transformative approach in digital health, designed to deliver personalized support to individuals at moments of heightened vulnerability or opportunity. By leveraging mobile and sensing technologies, JITAIs dynamically adapt intervention type, timing, and intensity based on a user's changing internal state and context. This article examines the core components, theoretical foundations, and practical implementation of JITAIs, with a specific focus on their application in dietary adherence research. We present structured protocols, quantitative outcomes, and visualization tools to equip researchers with methodologies for developing and evaluating effective JITAIs for health behavior change.

Just-in-Time Adaptive Interventions (JITAIs) represent a paradigm shift in digital health, moving from static, one-size-fits-all approaches to dynamic, personalized support systems. JITAIs are behavioral interventions designed to provide the right type and amount of support, at the right time, by adapting to an individual's changing internal and contextual state [1]. The scientific motivation for JITAIs stems from recognition that health behaviors are dynamic, context-dependent processes that cannot be effectively addressed through static intervention designs [1].

In modern digital health, JITAIs leverage increasingly powerful mobile and sensing technologies to monitor individuals in their natural environments and deliver ecologically sound support when it is most needed [1]. This approach is particularly valuable for addressing health behaviors that fluctuate throughout the day, such as dietary choices, physical activity, and medication adherence. The appeal of JITAIs lies in their potential to provide support during critical moments of vulnerability or opportunity, while minimizing unnecessary interruptions that might lead to intervention fatigue [1].

Within the specific context of dietary adherence research, JITAIs offer promising solutions for addressing the challenge of dietary lapses - specific instances of nonadherence to dietary goals that frequently undermine behavioral obesity treatment [2]. By proactively monitoring lapse risk and delivering contextually appropriate support, JITAIs can help individuals maintain adherence to prescribed dietary guidelines, thereby improving weight loss outcomes and reducing cardiovascular disease risk [3].

Core Components and Theoretical Framework

Fundamental Elements of JITAIs

According to the established conceptual framework set forth by Nahum-Shani et al., JITAIs are characterized by six core components that work in concert to provide real-time adaptive intervention [2] [1]:

• Distal Outcome: The ultimate long-term goal of the intervention, typically a primary clinical outcome such as weight loss or improved blood pressure control [2] [4].
• Proximal Outcome: Short-term goals the intervention aims to achieve, often serving as mediators or intermediate measures of the distal outcome [2]. In dietary adherence, this might be the reduction of dietary lapses.
• Tailoring Variables: Individual-specific information used to determine when and how to intervene, which can include states of vulnerability/opportunity and receptivity [2] [5].
• Decision Points: Specific moments in time when an intervention decision is made, such as following the completion of an ecological momentary assessment (EMA) survey [2].
• Intervention Options: The array of possible support actions that might be employed at any decision point [2].
• Decision Rules: Algorithms that specify which intervention option to offer, to whom, and when, systematically linking tailoring variables and intervention options [2] [5].

The JITAI Decision-Making Process

The following diagram illustrates how these components interact in a dynamic feedback system for dietary adherence:

This systematic framework ensures that JITAIs are not merely reactive but constitute a sophisticated, theory-driven approach to behavior change that dynamically adapts to the individual's evolving needs and circumstances [1].

JITAI Implementation Protocols and Methodologies

Microrandomized Trial Design for Dietary Adherence

The microrandomized trial (MRT) has emerged as a preferred methodological approach for developing and optimizing JITAIs. This design enables researchers to empirically test the effectiveness of various intervention components in real-time [2]. The following protocol outlines a comprehensive MRT for dietary adherence:

Table 1: Key Elements of an MRT Protocol for Dietary Adherence JITAI

Protocol Component	Specification	Rationale
Study Population	Adults with overweight/obesity and ≥1 CVD risk factor (n=159) [2]	Targets population most likely to benefit from improved dietary adherence
Intervention Duration	6-month web-based BOT with JITAI, plus 3-month JITAI-only follow-up [2]	Allows examination of both initial efficacy and maintenance
Decision Points	6 times daily (8:30 AM, 11:00 AM, 1:30 PM, 4:00 PM, 6:30 PM, 9:00 PM) [2]	Captures critical moments throughout the day when lapse risk may vary
Tailoring Variables	Behavioral, psychological, and environmental triggers assessed via EMA [2]	Provides multidimensional assessment of lapse risk factors
Intervention Conditions	No intervention, generic risk alert, or 1 of 4 theory-driven interventions [2]	Enables component testing to determine most effective approach
Primary Outcome	Occurrence of dietary lapse in 2.5 hours following randomization [2]	Measures proximal effect of intervention on target behavior

Experimental Workflow for JITAI Optimization

The experimental workflow for optimizing a dietary adherence JITAI involves multiple stages of development and testing:

This systematic approach to JITAI optimization emphasizes iterative refinement based on empirical data, ensuring the final intervention is both effective and efficient [2]. The MRT design is particularly valuable because it allows for testing intervention components across numerous decision points (potentially over 100 per participant), providing robust data on what works, for whom, and under what circumstances [2].

Quantitative Outcomes and Efficacy Data

Clinical and Behavioral Outcomes Across Health Domains

Recent empirical studies have generated mixed but promising evidence regarding the effectiveness of JITAIs across various health domains. The table below summarizes key quantitative findings from recent JITAI trials:

Table 2: Efficacy Data from Recent JITAI Implementation Studies

Health Domain	Study Design	Primary Outcome	Key Efficacy Findings	Reference
Hypertension Management	RCT (n=602) over 6 months	Change in systolic BP	-5.2 mmHg (intervention) vs -5.7 mmHg (control), p=0.76 (NS) [6]	[6]
Physical Activity Promotion	System ID study (n=48) over 270 days	Daily step count	Intervention group: +170 steps vs control: -319 steps, p=0.040 [6]	[6] [7]
Dietary Sodium Reduction	RCT (n=602) over 6 months	Daily sodium intake	Intervention: -1145 mg vs control: -860 mg, p=0.002 [6]	[6]
Behavioral Obesity Treatment	Meta-analysis	Distal health outcomes	Large effect size vs waitlist (Hedges' g=1.653) [8]	[8]

Engagement and Usability Metrics

Beyond clinical and behavioral outcomes, engagement and usability metrics are critical for evaluating JITAI implementation:

• Application Engagement: In the myBPmyLife hypertension trial, participants used the mobile application on a median of 81 days (45.76% of eligible days), with a mean of 1.96 interactions per active day [6].
• Usability Assessment: The System Usability Scale (SUS) score for the myBPmyLife JITAI was 73.6 (SD 19), considered above average with good usability [6].
• User Acceptance: A qualitative study of an EMA-driven JITAI for type 2 diabetes found generally positive feedback, with participants reporting motivating and enjoyable messages, though notable individual differences emerged regarding preferred intervention intensity and personalization [4].

The mixed results from these studies highlight both the potential and challenges of JITAIs. While the hypertension trial failed to demonstrate significant between-group differences in the primary outcome (blood pressure reduction), despite significant improvements in secondary behavioral outcomes (step count and sodium reduction) [6], other domains have shown more consistently positive effects [8].

Successful JITAI development and implementation requires a diverse methodological toolkit. The following table outlines essential "research reagents" and resources for dietary adherence JITAI studies:

Table 3: Essential Methodological Resources for Dietary Adherence JITAI Research

Resource Category	Specific Tools/Methods	Function/Application	Key Considerations
Assessment Tools	Ecological Momentary Assessment (EMA) [2]	Repeated sampling of behaviors, cognitions, emotions in natural environment	Should be brief (1-2 min), administered 5-6 times daily [2]
Analytical Algorithms	Machine learning for risk detection [2]	Calculates real-time lapse risk based on EMA responses	Requires training data from pilot studies [2]
Experimental Designs	Microrandomized Trials (MRT) [2]	Tests intervention component effects at decision points	Enables optimization before efficacy testing [2]
Theoretical Frameworks	Self-regulation theory [2]	Informs intervention content for self-control	One of four theory-driven approaches tested [2]
Outcome Measures	Dietary lapse frequency [2]	Primary proximal outcome in dietary JITAIs	Defined as specific instances of nonadherence to dietary goals [2]
Mobile Platforms	Smartphone applications with push notifications [6]	Delivery mechanism for intervention components	Should include both active and passive data collection [4]

These methodological resources provide the essential infrastructure for developing, testing, and implementing effective JITAIs for dietary adherence. The combination of real-time assessment, adaptive algorithms, and theory-driven interventions creates a powerful toolkit for addressing the dynamic challenge of dietary lapses in behavioral obesity treatment [2].

JITAIs represent a significant advancement in digital health, offering a sophisticated methodology for addressing the dynamic nature of health behaviors like dietary adherence. By leveraging mobile technology and adaptive algorithms, JITAIs can provide personalized, contextually relevant support at critical moments of need. The structured frameworks, experimental protocols, and efficacy data presented in this article provide researchers with practical tools for developing and evaluating JITAIs across health domains.

Future research should focus on enhancing the personalization of JITAIs through more sophisticated analytical approaches, including machine learning and passive sensing technologies [5]. Additionally, greater attention should be paid to long-term engagement and intervention fatigue, potentially through dynamic adaptation of intervention intensity and type [1]. As the field evolves, JITAIs hold tremendous promise for creating more effective, efficient, and scalable interventions for promoting health behavior change and improving clinical outcomes in diverse populations.

The Just-in-Time Adaptive Intervention (JITAI) represents a transformative approach in digital health, designed to provide tailored support that adapts to an individual's changing internal states and external contexts [1]. Unlike traditional static interventions, JITAIs dynamically deliver the right type of support at the right time, thereby addressing the critical challenge of sustaining health behavior changes in real-world settings [9]. The conceptual framework established by Nahum-Shani and colleagues offers a systematic structure for developing these sophisticated interventions, emphasizing the need to translate empirical, theoretical, and practical evidence into dynamic models that account for the temporal nature of health behaviors [9] [10]. For researchers investigating dietary adherence, this framework provides an essential roadmap for constructing interventions that can effectively respond to momentary risks and opportunities in participants' daily lives, ultimately aiming to improve long-term health outcomes through precisely timed support.

Core Components of the JITAI Framework

Fundamental Elements

The Nahum-Shani framework defines six core components that collectively form the architecture of any JITAI [5] [1]. These components work in concert to create an intervention system capable of adapting to individual needs over time. Table 1 summarizes these fundamental elements and their roles in intervention design.

Table 1: Core Components of the JITAI Framework

Component	Definition	Role in JITAI Design	Dietary Adherence Example
Distal Outcome	The ultimate long-term goal of the intervention	Primary clinical outcome that the JITAI aims to achieve	Weight loss, reduced blood pressure, improved glycemic control
Proximal Outcome	Short-term goals indicating progress toward distal outcomes	Mediators or intermediate measures of the distal outcome	Daily adherence to dietary goals, reduced dietary lapses
Tailoring Variables	Individual-specific information used for personalization	Determines when and how to intervene based on dynamic factors	Location (grocery store), mood, cravings, time since last meal
Decision Points	Points in time when intervention decisions are made	Opportunities to deliver support when it may be most effective	When entering a restaurant, after reporting cravings, when sedentary behavior is detected
Intervention Options	Array of possible treatments or actions	Suite of support strategies that can be deployed	Alternative food suggestions, coping strategies, motivational messages
Decision Rules	Algorithms specifying which intervention to offer, to whom, and when	Systematic linkage between tailoring variables and intervention options	IF at restaurant AND high lapse risk THEN suggest pre-selected healthy menu items

Interrelationship of Components

The power of the JITAI framework lies in the dynamic relationships between these components. Decision rules operationalize the adaptation process by systematically connecting tailoring variables to intervention options at specific decision points [1]. This systematic linkage enables the intervention to respond to both states of vulnerability (periods of susceptibility to negative health outcomes) and states of opportunity (periods conducive to positive behavior change) [1]. For dietary adherence research, this means constructing decision rules that can detect imminent risk of dietary lapses while also identifying teachable moments when participants may be most receptive to support for healthy eating behaviors [3].

Application to Dietary Adherence Research

Theoretical Foundations for Dietary JITAIs

The application of the Nahum-Shani framework to dietary adherence addresses a fundamental challenge in behavioral obesity treatment: the frequent dietary lapses that undermine weight loss efforts [3]. Research indicates that dietary lapses occur 3-4 times per week in behavioral obesity treatment and are strongly associated with poorer weight outcomes [3]. The JITAI approach reconceptualizes dietary adherence as a dynamic process influenced by momentary factors rather than a stable trait, requiring interventions that can adapt to rapidly changing contexts, internal states, and environmental triggers [3] [1].

Empirical Evidence and Clinical Outcomes

Recent randomized controlled trials have demonstrated both the promise and challenges of JITAIs for dietary behaviors. The myBPmyLife trial, a 6-month RCT with 602 hypertension patients, tested a JITAI targeting physical activity and lower-sodium food choices [6]. While the intervention significantly reduced sodium intake by 1145 mg compared to 860 mg in controls (p=0.002), it did not produce significant differences in systolic blood pressure reduction between groups (-5.2 mmHg vs. -5.7 mmHg, p=0.76) [6]. This dissociation between behavioral changes and clinical outcomes highlights the complexity of translating momentary dietary improvements into sustained physiological benefits. Table 2 summarizes key quantitative findings from recent dietary-focused JITAI trials.

Table 2: Efficacy Data from Dietary JITAI Randomized Controlled Trials

Trial/Study	Population	Sample Size	Primary Outcome	Behavioral Outcome	Clinical Outcome
myBPmyLife [6]	Hypertension patients	602	SBP change at 6 months	Sodium intake ↓1145 mg (int) vs ↓860 mg (ctrl), p=0.002	SBP: -5.2 mmHg (int) vs -5.7 mmHg (ctrl), p=0.76
LowSalt4Life 2 [11]	Hypertension patients	410	SBP change at 2 months	Results pending (2025)	Results pending (2025)
Goldstein MRT [3]	Overweight/obesity with CVD risk	159	Dietary lapse occurrence in 2.5h post-intervention	Study ongoing	Study ongoing

Experimental Protocols for Dietary JITAI Development

Microrandomized Trial Design for Intervention Optimization

The Microrandomized Trial (MRT) has emerged as a critical methodological innovation for developing and optimizing JITAIs [3]. This experimental design enables researchers to test the proximal effects of intervention components repeatedly over time within individuals. The protocol for dietary adherence JITAIs typically involves:

• Participant Recruitment: Adults with overweight or obesity and at least one cardiovascular disease risk factor (e.g., hypertension, hypercholesterolemia, type 2 diabetes) [3].

• Baseline Assessment: Comprehensive evaluation including weight, cardiovascular metrics, dietary patterns, and psychological measures.

• EMA Configuration: Implementation of 6 daily EMA surveys at anchor times (8:30 AM, 11:00 AM, 1:30 PM, 4:00 PM, 6:30 PM, 9:00 PM) with 90-minute response windows [3].

• Risk Algorithm Training: Machine learning algorithms analyze EMA responses in real-time to calculate ongoing lapse risk based on behavioral, psychological, and environmental triggers.

• Microrandomization Procedure: Each time elevated lapse risk is detected, participants are randomized to either no intervention, a generic risk alert, or one of four theory-driven interventions (enhanced education, building self-efficacy, fostering motivation, improving self-regulation) [3].

• Proximal Outcome Measurement: Occurrence of dietary lapse assessed via EMA in the 2.5 hours following randomization [3].

This MRT approach generates intensive longitudinal data that informs the optimization of decision rules by identifying which intervention modalities work best for specific individuals under particular contextual circumstances.

Ecological Momentary Assessment Protocol

The assessment foundation for dietary JITAIs relies on carefully constructed EMA protocols that balance comprehensiveness with participant burden:

EMA Content Domains for Dietary Lapse Risk Assessment:

• Behavioral Triggers: Time since last meal, proximity to food, food availability
• Environmental Context: Location (home, work, restaurant), social setting, food cues
• Internal States: Hunger, cravings, mood, stress, self-control capacity
• Psychological Factors: Self-efficacy, motivation, outcome expectancies

This comprehensive assessment strategy enables the JITAI to detect vulnerability states with greater precision and deliver appropriately tailored interventions [3].

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Research Reagents for Dietary JITAI Development

Tool Category	Specific Solution	Function in JITAI Research	Implementation Example
Mobile Assessment Platforms	ilumivu ivu System [12]	Rapid prototyping of EMI/JITAI interventions; implements decision rules without custom software	Lowers development costs; enables iterative refinement of intervention components
Passive Sensing Technologies	GPS, accelerometry, Bluetooth beacons [5]	Context detection without active user input; identifies risk locations and situations	Detects proximity to restaurants or grocery stores; monitors sedentary behavior
Active Assessment Tools	Ecological Momentary Assessment (EMA) [3]	Repeated sampling of behaviors, cognitions, emotions in natural environment	6 daily surveys assessing lapse triggers, mood, context
Analytical Algorithms	Machine learning lapse prediction [3]	Real-time analysis of EMA responses to calculate ongoing lapse risk	Classifies risk level based on combination of endorsed triggers
Intervention Delivery Systems	SMS text messaging, push notifications, in-app alerts [4]	Timely delivery of tailored support when lapse risk is elevated	Sends coping strategy when high risk is detected
Evaluation Frameworks	Microrandomized Trial (MRT) design [3]	Tests proximal effects of intervention components repeatedly over time	Randomizes intervention type each time high risk is detected

Implementation Considerations and Current Challenges

Addressing Heterogeneity in Intervention Response

Recent research highlights significant individual differences in response to JITAIs, necessitating more sophisticated personalization approaches. A qualitative acceptability study of an EMA-driven JITAI for type 2 diabetes found notable variations in user experiences, particularly regarding intervention intensity and perceived personalization [4]. Participants reported that while EMA was generally low burden and easy to use, it provided "too much of a snapshot and too little context," reducing the perceived tailoring of intervention options [4]. This suggests the need for more nuanced assessment strategies that balance active and passive data collection to better capture individual contexts.

Integration with Comprehensive Treatment Approaches

The most effective dietary JITAIs appear to be those integrated within broader behavioral treatment frameworks rather than operating as standalone interventions [5]. Current evidence suggests that JITAIs combining algorithmic tailoring with guidance from human clinicians yield superior outcomes [5]. Furthermore, systematic review evidence indicates that JITAIs for health behavior are still in early developmental stages, with opportunities for improvement in both development and testing methodologies [5]. Future directions should focus on enhancing the sophistication of decision rules through more complex analytical techniques, including machine learning applications that can handle real-time data streams from multiple sources [5].

The Critical Link Between Dietary Lapses and Long-Term Health Outcomes

Dietary adherence is a cornerstone of successful behavioral obesity treatment (BOT) and long-term health maintenance. Dietary lapses, defined as specific instances of nonadherence to one or more prescribed dietary goals, represent a critical challenge in nutritional science and weight management interventions [13]. These lapses occur frequently during lifestyle interventions, with research indicating they happen approximately 3-4 times per week in standard BOT programs [3]. The systematic study of these lapses is essential because they directly impact energy balance and ultimately influence the risk for chronic diseases, including cardiovascular disease, diabetes, and neurodegenerative conditions [13] [14].

The emerging field of just-in-time adaptive interventions (JITAIs) offers a promising framework for addressing dietary lapses by providing the right type and amount of support at the exact moment of need [1]. JITAIs represent a dynamic intervention design that adapts support over time to an individual's changing internal and contextual state, with the goal of delivering support when the person needs it most and is most likely to be receptive [1]. This approach is particularly suited to dietary adherence because it can proactively monitor lapse risk and provide timely interventions to prevent lapses in an adaptive manner.

Quantifying the Impact of Dietary Lapses

Direct Effects on Nutritional Intake

Understanding the quantitative impact of dietary lapses is essential for appreciating their significance in long-term health outcomes. Research utilizing ecological momentary assessment (EMA) and 24-hour dietary recalls has provided robust evidence of how lapses influence daily nutritional intake.

Table 1: Impact of Dietary Lapses on Daily Nutritional Intake

Nutritional Parameter	Effect of Dietary Lapse	Statistical Significance	Study Reference
Total Daily Caloric Intake	Increase of 139.20 kcal	B = 139.20, p < 0.05	[13]
Daily Grams of Added Sugar	Increase of 16.24 grams	B = 16.24, p < 0.001	[13]
Likelihood of Exceeding Daily Calorie Goal	Significant increase	B = 0.89, p < 0.05	[13]
Estimated Weekly Caloric Impact	Additional 600-750 kcal per week	Based on participant food records	[13]

The data demonstrate that dietary lapses have meaningful effects on energy balance, primarily through increased consumption of empty calories from added sugars rather than through fundamental shifts in macronutrient composition [13]. This pattern of lapse-induced consumption contributes directly to positive energy balance, potentially explaining the association between lapse frequency and suboptimal weight loss outcomes in lifestyle interventions.

Long-Term Health Implications

The cumulative effect of dietary lapses extends beyond immediate weight management challenges to influence broader health outcomes, particularly in aging populations and those with preexisting conditions.

Table 2: Dietary Interventions and Cognitive Health Outcomes in Adults with Preexisting Cognitive Impairment

Dietary Intervention	Study Duration	Cognitive Domains Improved	Study Population
Pomegranate Seed Oil	1 year	Executive function, Working memory, Planning, Inhibition	Mild Cognitive Impairment (MCI)	[14]
Probiotic Mixture	12 weeks	Executive function, Learning and memory, Attention, Psychomotor speed	Older adults with MCI	[14]
Microalgae Extract	12 weeks	Executive function, Learning and memory, Attention	Older adults with memory complaints	[14]
Folic Acid and DHA	12 weeks	Learning and memory, Attention, Psychomotor speed	Older adults with MCI	[14]
MIND Diet	36 weeks	No significant improvement in global cognition	Older adults predisposed to dementia	[14]

The mechanisms through which dietary components influence long-term health are multifaceted. Emerging evidence suggests that specific dietary interventions may exert neuroprotective effects primarily by reducing oxidative stress and neuroinflammation, and by enhancing brain vascular function [14]. These mechanisms may promote neuroplasticity through the modulation of neurotrophic signaling pathways, thereby preserving cognitive function despite aging-related decline.

Methodological Framework for Studying Dietary Lapses

Assessment Protocols for Dietary Adherence

Accurate measurement of dietary intake and adherence is fundamental to understanding the relationship between lapses and health outcomes. Multiple assessment methods exist, each with distinct strengths and limitations.

Table 3: Dietary Assessment Methods for Research and Clinical Applications

Assessment Method	Time Frame	Primary Use Cases	Strengths	Limitations
24-Hour Dietary Recall	Short-term (previous 24 hours)	Total diet assessment, Intervention studies	High accuracy for recent intake, Does not require literacy	Relies on memory, Interviewer training required	[15]
Food Record	Short-term (typically 3-4 days)	Total diet assessment, Metabolic studies	Comprehensive recording, Weighed portions possible	Reactivity (changing behavior), Participant burden	[15]
Food Frequency Questionnaire	Long-term (months to years)	Epidemiological studies, Habitual intake	Cost-effective for large samples, Ranks individuals by intake	Less precise for absolute intakes, Limited food list	[15]
Ecological Momentary Assessment	Real-time, multiple times daily	Behavioral interventions, Trigger identification	Ecological validity, Minimal recall bias	Participant burden, Requires technology	[13]

The choice of assessment method depends on the research question, study design, sample characteristics, and available resources [15]. For studying dietary lapses specifically, EMA has emerged as a particularly valuable tool because it captures data in near real-time, improving ecological validity and reliability compared to retrospective recalls [13].

Experimental Protocol for JITAI Implementation

The following protocol outlines a systematic approach for implementing and evaluating a JITAI targeting dietary lapses, based on established methodologies [3].

Protocol Title: Microrandomized Trial for Optimizing a Just-in-Time Adaptive Intervention to Improve Dietary Adherence

Objective: To optimize a smartphone-based JITAI that uses daily surveys to assess triggers for dietary lapses and deliver interventions when the risk of lapse is high.

Population: Adults with overweight or obesity (BMI 25-50 kg/m²) and at least one diagnosed cardiovascular disease risk factor (e.g., hypertension, hypercholesterolemia, type 2 diabetes).

Exclusion Criteria: Medical conditions contraindicating weight loss, pregnancy or breastfeeding, enrollment in another weight loss program, significant recent weight loss (>5% in past 6 months), use of weight loss medications, history of bariatric surgery, or diagnosed eating disorders (excluding Binge Eating Disorder).

Study Design:

• Duration: 6-month intervention comprising 3 months of active BOT with weekly sessions and 3 months of monthly booster sessions.
• JITAI Components:
  • Ecological Momentary Assessment: 6 daily surveys at semi-random times (anchor times at 8:30 AM, 11:00 AM, 1:30 PM, 4:00 PM, 6:30 PM, and 9:00 PM) with 90-minute response windows.
  • Risk Algorithm: Machine learning algorithm analyzing EMA responses in real-time to calculate ongoing lapse risk.
  • Intervention Delivery: Randomized intervention delivery when elevated lapse risk is detected.

Microrandomization Procedure: When the JITAI algorithm detects elevated lapse risk, participants are randomized to one of six conditions:

• No intervention
• Generic risk alert
• Theory-driven intervention targeting education on dietary goals
• Theory-driven intervention building self-efficacy
• Theory-driven intervention enhancing motivation
• Theory-driven intervention improving self-regulation

Primary Outcome: Occurrence of dietary lapse in the 2.5 hours following randomization, measured via EMA.

Secondary Outcomes: Eating characteristics (duration, rate, bite count) via wrist-based monitoring; weight change at 3 and 6 months; intervention engagement and satisfaction.

Analytical Plan: Use of mixed effects models to evaluate intervention effects on proximal outcomes, accounting for within-person clustering of repeated measurements.

Visualization of JITAI Framework and Mechanisms

JITAI Decision Pathway for Dietary Lapse Prevention

Mechanisms Linking Dietary Lapses to Health Outcomes

Research Reagent Solutions Toolkit

Table 4: Essential Research Tools and Methodologies for Dietary Adherence Research

Tool Category	Specific Tool/Platform	Primary Function	Application Notes
EMA Platforms	LifeData App	Delivery of EMA surveys, HIPAA-compliant data collection	Configurable survey schedules, Real-time data capture	[13]
Dietary Assessment	Automated Self-Administered 24-hour Recall (ASA-24)	24-hour dietary recall administration	Reduces interviewer burden, Free for researchers	[15]
Nutritional Analysis	Nutrition Data System for Research (NDSR)	Comprehensive nutritional analysis of dietary intake	Detailed nutrient composition, Research-grade output	[13]
JITAI Development	Custom smartphone applications with machine learning algorithms	Real-time risk assessment and intervention delivery	Requires multidisciplinary development team	[3]
Adherence Biomarkers	Doubly labeled water, Recovery biomarkers for energy and protein	Objective validation of self-reported dietary intake	Considered gold standard but costly and complex	[15]
Behavioral Assessment	Theory-driven intervention content libraries	Provision of evidence-based behavioral strategies	Content targeting education, self-efficacy, motivation, self-regulation	[3]

Dietary lapses represent a critical mediating factor between behavioral interventions and long-term health outcomes. The evidence consistently demonstrates that these lapses directly impact daily caloric intake, particularly through increased consumption of added sugars, and consequently influence weight management success and chronic disease risk [13]. The emergence of JITAIs as a methodological framework offers a promising approach to addressing the dynamic nature of dietary adherence challenges by providing timely, personalized support at moments of heightened vulnerability [1] [3].

Future research should focus on optimizing intervention components within JITAIs through microrandomized trials and similar experimental designs that can isolate active ingredients of behavior change. Additionally, greater integration of objective biomarkers and passive sensing technologies may enhance the precision of lapse detection and intervention timing. As the field advances, the systematic application of JITAIs holds significant potential for improving dietary adherence and thereby actualizing the long-term health benefits of behavioral obesity treatments.

Ecological Momentary Assessment (EMA) has emerged as a transformative methodology for capturing real-time data on health behaviors, emotions, and contextual factors as they occur naturally in daily life. Unlike traditional retrospective assessments prone to recall bias and aggregation errors, EMA provides rich, context-laden data through repeated sampling in participants' natural environments [16]. This approach is particularly valuable for understanding complex, fluctuating behaviors such as dietary intake, where momentary triggers and contextual influences significantly impact adherence patterns.

Within the framework of Just-in-Time Adaptive Interventions (JITAIs) for dietary adherence, EMA serves as the critical data collection backbone that enables real-time intervention tailoring. By identifying dynamic predictors of dietary lapses and capturing within-person fluctuations in eating behaviors, EMA provides the essential temporal density and resolution needed to inform adaptive decision rules [17]. The integration of EMA with JITAI architectures represents a paradigm shift from one-size-fits-all dietary interventions toward personalized, context-sensitive approaches that can prevent lapses at critical decision points [17].

Key Methodological Approaches in Dietary EMA

EMA Sampling Protocols for Dietary Research

Dietary EMA studies employ various sampling approaches, each with distinct advantages for capturing different aspects of eating behavior. The choice of protocol significantly influences data quality, participant burden, and the types of research questions that can be addressed.

Table 1: Comparison of EMA Sampling Protocols in Dietary Research

Protocol Type	Description	Key Applications	Advantages	Limitations
Signal-Contingent	Participants respond to random or fixed prompts	Assessing eating episodes, moods, contexts	Captures non-eating moments; reduces selection bias	May miss unplanned eating events
Event-Contingent	Participants initiate reports after specific events	Capturing specific eating episodes, binge events	Ensures capture of target behaviors; reduces burden	Underreporting if participants forget to initiate
Interval-Contingent	Participants report at predetermined intervals	End-of-day summaries, scheduled check-ins	Predictable for participants; consistent data	May increase recall bias for distant events
Hybrid Approaches	Combination of multiple protocols	Comprehensive dietary assessment	Balances comprehensiveness with feasibility	Increased complexity and participant burden

Recent research has compared the relative effectiveness of these approaches for specific dietary behaviors. A 2023 study on online food delivery use found that event-contingent sampling was 3.53 times more likely to capture target behaviors compared to signal-contingent sampling, though both approaches showed similar compliance rates (72.5% vs 73.2%) [18]. However, a 2025 feasibility study comparing personalized versus fixed-interval signal-contingent EMA found no significant difference in adherence (65.7% vs 66.3%) or agreement with validation measures, suggesting that advanced personalization of assessment timing may require more sophisticated approaches [19].

Technological Implementation and Compliance

Modern EMA implementations leverage mobile technologies to reduce participant burden and improve data quality. Smartphone applications with programmed response options, photograph capabilities, and flexible sampling frameworks have become standard in dietary EMA research [16] [18].

Compliance rates across studies generally range from 65% to 75%, with specific factors influencing adherence:

• Survey length: Brief assessments (1-2 minutes) demonstrate higher compliance [16]
• Recording duration: Shorter monitoring periods (7-14 days) maintain engagement [19]
• Prompt frequency: 3-5 prompts daily balances data density with burden [16] [18]
• Ease of use: Intuitive interfaces significantly impact acceptability [18]

The eTRIP V.1 smartphone application used in a 2025 Singapore study achieved a remarkable 97.7% completion rate through careful design, including predetermined mealtimes, photograph integration, and brief 1-2 minute check-ins [16]. This demonstrates how technological optimization can maximize data quality in dietary EMA.

EMA Evidence Base for Dietary Adherence and Obesity

Identifying Real-Time Dietary Triggers

EMA research has substantially advanced our understanding of the dynamic, contextual factors that influence dietary adherence and lapses. A 2025 systematic review synthesizing findings from 89 EMA studies on obesity and overweight identified consistent patterns in dietary triggers across diverse populations [20].

Table 2: Key Dietary Triggers and Adherence Factors Identified Through EMA

Factor Category	Specific Triggers	Association with Dietary Adherence	Evidence Source
Environmental Context	Food accessibility; Travelling; Home meal preparation	Self-prepared meals: 5% increased adherence; Travelling: 7% decreased adherence	[16]
Emotional States	Negative emotions (stress, nervousness, sadness); Premenstrual syndrome	Significant predictors of dietary lapses; Variations in individual susceptibility	[16] [20]
Social Context	Eating alone vs. with others; Family meals; Peer interactions	Alone vs. not alone: lower overeating; Family meals: higher overeating but lower loss of control	[21]
Food Types	Sweet foods; Salty/fried foods; Pizza/fast food; Sweetened beverages	Consistent associations with binge eating symptoms and overeating	[21]
Temporal Factors	Time of day; Day of week; Meal timing	Variations in eating patterns and lapse susceptibility across temporal cycles	[20]

A large-scale EMA study with 250 participants and 4,708 assessments found that 76.4% of responses indicated adherence to dietary plans, with non-adherence primarily driven by food accessibility and negative emotions [16]. This research highlights the individual variability in dietary triggers, underscoring the need for personalized approaches rather than one-size-fits-all interventions.

EMA in Special Populations and Contexts

EMA methodologies have been successfully adapted for specific populations and eating behaviors:

• Binge Eating in Adolescents: A 2025 study identified distinct factor structures for overeating and loss of control eating, with significant associations between social contexts (family, peers) and food types (sweet, salty/fried) [21].

• Online Food Delivery (OFD) Use: EMA research has revealed that pizza (18.5%) and fried chicken (14.5%) comprise the bulk of OFD orders, with most orders placed at home (79%) for one person (54.8%) [18].

• Childhood Dietary Assessment: EMA-assisted 24-hour recalls are being validated as more accurate alternatives to traditional dietary assessment in young children, leveraging photograph uploads and real-time reporting [22].

These applications demonstrate EMA's flexibility in capturing context-specific eating behaviors across diverse populations and settings, providing essential data for targeted interventions.

Experimental Protocols and Workflows

Core Protocol: EMA for Dietary Trigger Identification

Objective: To identify real-time dietary triggers and adherence patterns among adults with overweight or obesity.

Participants: 250 adults with BMI ≥23 kg/m² recruited from public and specialist obesity management centers [16].

Duration: 7-day monitoring period with assessments at three user-specified mealtimes.

Materials:

• eTRIP V.1 smartphone application
• Eating Behaviour Lapse Inventory Survey Singapore (eBLISS)
• Automated prompting system with photograph capability

Procedure:

• Baseline Assessment: Demographic data collection, BMI measurement, mealtime specification
• EMA Training: Instruction on application use, photograph standards, and response protocols
• Data Collection Phase:
  • Participants receive prompts at three predetermined mealtimes daily
  • At each prompt: photograph meal, complete eBLISS survey on dietary triggers and adherence
  • Optional event-contingent reports for snacks and unplanned eating episodes
  • Average completion time: 1-2 minutes per assessment
• Data Validation: Cross-checking of photograph data with self-reported food intake
• Debriefing: Participant feedback on acceptability and burden

Measures:

• Primary outcome: Adherence to dietary plan (dichotomous yes/no)
• Predictor variables: Seven trigger domains (place, emotions, physiological state, eating company, food provider, activity before eating, sleep hours)
• Covariates: Time of day, day of week, participant characteristics

Analysis:

• Generalized Estimating Equations (GEE) for correlated data
• Bonferroni adjustment for multiple comparisons (p<0.001)
• Relative risk calculations for trigger-adherence associations

This protocol successfully generated 4,708 assessments with 97.7% completion rate, identifying key modifiable triggers for dietary interventions [16].

Comparative Protocol: Personalized vs. Fixed EMA Scheduling

Objective: To compare feasibility and accuracy of personalized versus fixed-interval EMA for capturing dietary data in young adults with irregular eating patterns.

Design: Double-blinded crossover feasibility study with random assignment to initial condition [19].

Participants: 24 young adults (18-30 years), 13 female, mean age 26.0 (SD 2.1).

Conditions:

• Personalized Schedule: Assessment timing based on individual eating patterns from baseline food images
• Fixed Interval Schedule: Six predetermined prompts daily between 7:00-22:00

Materials:

• Mobile EMA application (ExpiWell or equivalent)
• Time-stamped food image collection system
• 24-hour dietary recall instruments

Procedure:

• Baseline Phase: 3-day food image collection to establish individual eating patterns
• Randomization: Participants assigned to personalized or fixed schedule start condition
• First Monitoring Phase: 7-day EMA with assigned schedule condition
• Washout Period: 7-14 days without data collection
• Second Monitoring Phase: 7-day EMA with alternate schedule condition
• Validation: Simultaneous 24-hour recalls and food image collection

Outcomes:

• Feasibility: Adherence rates (response completion)
• Accuracy: Agreement with food images and 24-hour recalls
• Acceptability: Participant ratings of burden, usability, and interference

This protocol revealed similar adherence (65.7% fixed vs 66.3% personalized) and agreement (52.0% vs 47.7%) between approaches, highlighting challenges in personalization for irregular eaters [19].

Integration with Just-in-Time Adaptive Interventions

The translation of EMA findings into effective JITAIs requires systematic frameworks that link assessment to intervention. The following workflow illustrates this integration:

EMA-JITAI Integration Workflow

This integration framework enables dynamic intervention tailoring based on real-time risk assessment. For example, when EMA detects negative emotions (a known dietary trigger [16]), the system can deliver emotion-regulation strategies precisely when needed. Similarly, location-based triggers (e.g., proximity to fast food outlets) can prompt alternative action plans to prevent automatic, habitual responses.

The critical decision points in this workflow include:

• Risk State Identification: Classifying current context as high-risk based on EMA-derived triggers
• Intervention Selection: Matching specific intervention components to identified triggers
• Delivery Timing: Ensuring support arrives during the critical window of vulnerability
• Dosage Adjustment: Modifying intervention intensity based on cumulative risk exposure

This systematic approach transforms EMA from a passive assessment tool into an active intervention component that drives personalized support.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Tools and Platforms for Dietary EMA Research

Tool Category	Specific Solutions	Key Functionality	Research Applications
EMA Platforms	ExpiWell; mEMASense (ilumivu); REDCap	Survey delivery; Scheduling; Multi-mode sampling; Compliance monitoring	General dietary assessment; Trigger identification; Adolescent eating behaviors [19] [17] [18]
Mobile Applications	eTRIP V.1; DevilSPARC	Dietary logging; Photograph integration; Real-time prompting	Dietary adherence monitoring; Obesity research; College student nutrition [16]
Data Collection Instruments	Eating Behaviour Lapse Inventory Survey Singapore (eBLISS); Chrono-Nutrition Behavior Questionnaire (CNBQ)	Standardized trigger assessment; Temporal eating patterns; Contextual factors	Dietary lapse prediction; Cultural eating patterns; Meal timing research [16]
Analytical Frameworks	TRIPOD-AI; PROBAST-AI; DECIDE-AI	Prediction model reporting; Risk of bias assessment; Early clinical evaluation	Model validation; Intervention development; Clinical translation [23]
Integration Frameworks	JITAI Design Principles; Multiphase Optimization Strategy (MOST)	Intervention optimization; Adaptive algorithm design; Resource allocation	JITAI development; Component screening; Efficacy optimization [17]

These research reagents provide the methodological infrastructure necessary for implementing rigorous dietary EMA studies and translating findings into effective JITAIs. Platform selection should align with specific research questions, with considerations for sampling flexibility, integration capabilities, and compliance monitoring features.

The integration of EMA with JITAIs represents a promising frontier in dietary adherence research, with several emerging trends shaping future applications:

Technological Innovations: The convergence of EMA with wearable sensors, geolocation tracking, and passive physiological monitoring will enable more comprehensive context assessment with reduced participant burden [17]. Advanced personalization algorithms that adapt assessment frequency based on individual patterns and states will improve the efficiency of data collection [19].

Methodological Advances: Hybrid approaches combining signal-contingent, event-contingent, and sensor-triggered assessments will provide more complete coverage of eating episodes while minimizing participant burden [19] [18]. The development of standardized reporting guidelines (TRIPOD-AI, CONSORT-AI) will enhance methodological rigor and comparability across studies [23].

Clinical Translation: Future research must bridge the gap between EMA identification of dietary triggers and the development of effective JITAIs that leverage these insights in real-time. This requires closed-loop systems that continuously adapt intervention content, timing, and intensity based on EMA-derived risk assessments [17]. The growing emphasis on ethical AI frameworks and responsible implementation will ensure these technologies benefit diverse populations without exacerbating health disparities [23].

In conclusion, EMA provides the essential data infrastructure for understanding the dynamic, contextual nature of dietary behavior. By capturing real-time triggers and within-person fluctuations, EMA enables the development of JITAIs that can deliver personalized support at moments of heightened vulnerability. As methodological and technological innovations continue to advance, EMA-guided JITAIs hold significant promise for transforming dietary adherence from a static intervention approach to a dynamic, responsive, and ultimately more effective support system.

Just-in-Time Adaptive Interventions (JITAIs) represent an innovative approach in digital health, designed to deliver personalized support at moments of heightened vulnerability and opportunity. Within dietary adherence research, JITAIs dynamically adapt intervention timing, content, and intensity based on real-time data from smartphones and wearable sensors [5]. This application note synthesizes current evidence and methodologies regarding JITAI efficacy for two critical dietary components: sodium reduction and calorie control. The integration of ecological momentary assessment (EMA), sensor technology, and machine learning algorithms enables these interventions to address the complex, context-dependent nature of eating behaviors, offering promising avenues for managing chronic conditions like hypertension and obesity [24] [3].

Quantitative Evidence of Efficacy

Recent clinical trials and pilot studies provide quantitative evidence supporting the potential of JITAIs to modify dietary behaviors. The data, summarized in the table below, demonstrate effects on sodium intake, caloric control, and secondary health outcomes.

Table 1: Efficacy Outcomes from JITAI Dietary Studies

Study & Intervention	Primary Target	Study Design & Duration	Key Efficacy Findings	Statistical Significance
myBPmyLife Trial [6]	Sodium & Physical Activity	RCT, 602 hypertension patients, 6 months	Sodium intake: -1145 mg (Intervention) vs -860 mg (Control); SBP: -5.2 mmHg (Intervention) vs -5.7 mmHg (Control)	Sodium: p=0.002; SBP: p=0.76 (NS)
LowSalt4Life Pilot [25]	Sodium	RCT, 50 hypertension patients, 8 weeks	Sodium (Spot Urine): -462 mg (App) vs +381 mg (No App); Sodium (FFQ): -1553 mg (App) vs -515 mg (No App)	Spot Urine: p=0.03; FFQ: p=0.01
AGILE Trial Protocol [26]	Calorie Control & Weight	Factorial RCT, 608 young adults, 6 months	Protocol; tests adaptive goals, messaging, and self-monitoring for weight loss.	Results Pending
Dietary Lapse JITAI [3]	Calorie Control (Lapses)	Microrandomized Trial, 159 adults, 6 months	Protocol; uses ML to predict lapse risk and deliver theory-driven interventions.	Results Pending

Abbreviations: RCT: Randomized Controlled Trial; SBP: Systolic Blood Pressure; FFQ: Food Frequency Questionnaire; NS: Not Significant; ML: Machine Learning.

The findings indicate that JITAIs can effectively promote significant reductions in sodium intake, as evidenced by the myBPmyLife and LowSalt4Life trials [6] [25]. However, the dissociation between sodium reduction and corresponding blood pressure changes in the myBPmyLife trial highlights the complex, multifactorial nature of hypertension and underscores the need for JITAIs to be part of a broader management strategy [6]. Research on JITAIs for direct caloric control and weight loss, such as the AGILE trial and the dietary lapse JITAI, is actively evolving, with rigorous methodologies being applied to optimize these interventions [26] [3].

Detailed Experimental Protocols

A critical component for advancing JITAI research is the transparent reporting of methodological frameworks. Below are detailed protocols for two seminal studies in sodium reduction and calorie control.

Protocol 1: The LowSalt4Life JITAI for Sodium Reduction

The LowSalt4Life intervention is a contextual JITAI designed to reduce dietary sodium intake in adults with hypertension [25].

Table 2: Key Components of the LowSalt4Life Protocol

Component	Description
Objective	To evaluate the effect of a contextual just-in-time mobile app intervention on reducing sodium intake in adults with hypertension.
Study Design	Single-center, prospective, open-label RCT for 8 weeks.
Participants	Adults ≥18 years, on antihypertensive therapy, owning a smartphone. Exclusions: CKD, heart failure, SBP >180 mmHg.
Intervention Group	Received the LowSalt4Life mobile application.
Control Group	Received usual dietary advice (standard of care).
Baseline Assessment	1. Sodium Screener: Identified participant's top 5 high-sodium foods.2. Alternative Selection: Participants selected lower-sodium alternatives.3. Geotagging: Participants tagged locations (home, restaurant, grocery store) associated with these foods.
JITAI Delivery	1. Context Sensing: A cloud-based service used smartphone sensors to detect when a participant entered a geotagged location or a predicted restaurant/grocery store.2. Just-in-Time Messaging: Upon entry, the app sent a push notification with a tailored message promoting the pre-selected low-sodium alternative.3. In-App Support: The app provided curated low-sodium restaurant menus and a barcode scanner for finding low-sodium grocery items.
Outcome Measures	Primary: Change in estimated 24-hour urinary sodium from spot urine.Secondary: Change in sodium intake (24-hour dietary recall, FFQ), blood pressure, self-confidence.

Figure 1: The LowSalt4Life JITAI Workflow for Sodium Reduction. The diagram outlines the process from baseline assessment to the delivery of context-aware interventions.

Protocol 2: A JITAI for Dietary Lapses in Calorie Control

This protocol details a microrandomized trial (MRT) designed to optimize a JITAI that prevents dietary lapses (instances of non-adherence to caloric goals) during behavioral obesity treatment [3].

Table 3: Key Components of the Dietary Lapse JITAI Protocol

Component	Description
Objective	To optimize a JITAI by evaluating the proximal efficacy of theory-driven interventions for preventing dietary lapses.
Study Design	Microrandomized Trial (MRT) embedded within a 6-month behavioral obesity treatment.
Participants	Adults with overweight/obesity and at least one cardiovascular disease risk factor.
JITAI Framework	Based on the conceptual model by Nahum-Shani et al.
Decision Points	Immediately after each of 6 daily Ecological Momentary Assessment (EMA) surveys.
Tailoring Variables	Real-time EMA data on behavioral, psychological, and environmental lapse triggers (e.g., mood, location, cravings).
Proximal Outcome	Occurrence of a dietary lapse within 2.5 hours of a decision point.
Distal Outcome	Body weight change at 3 and 6 months.
Intervention Options	At each high-risk decision point, participants are microrandomized to: 1. No intervention; 2. Generic risk alert; 3. One of four theory-driven interventions (Education, Self-Efficacy, Motivation, Self-Regulation).
Data Analysis	A machine learning algorithm analyzes EMA data to calculate real-time lapse risk, triggering the randomization procedure. MRT data will inform an optimized decision rule for a future RCT.

Figure 2: Decision Workflow for the Dietary Lapse JITAI. This MRT design tests intervention efficacy at moments of high lapse risk, identified via machine learning analysis of EMA data.

The Scientist's Toolkit: Research Reagent Solutions

Implementing and evaluating dietary JITAIs requires a suite of technological and methodological "reagents." The following table catalogues essential components for researchers in this field.

Table 4: Essential Research Reagents for Dietary JITAI Development

Tool Category	Specific Examples	Function in JITAI Research
Sensing & Data Acquisition	Smartphone Sensors (GPS, Accelerometer) [25]	Passively monitors context (location, movement) to identify decision points (e.g., entering a restaurant).
	Wearable Activity Monitors (e.g., Fitbit) [27]	Provides objective data on physical activity and sedentary behavior, used for tailoring or as an outcome.
	Wrist-Based Monitors (for ingestive behavior) [3]	Passively detects eating episodes (duration, bites) for objective proximal outcome measurement.
Behavioral & Outcome Assessment	Ecological Momentary Assessment (EMA) [4] [3]	The core method for active data collection on real-time triggers, behaviors, and psychological states.
	24-Hour Dietary Recalls (e.g., ASA24) [25]	A standardized tool for subjective assessment of dietary intake as a secondary outcome.
	Food Frequency Questionnaire (FFQ) [25]	Assesses habitual dietary patterns and screens participants at baseline.
	Biomarkers (24-hour Urinary Sodium) [25]	The gold-standard objective measure for validating changes in sodium intake.
Intervention Delivery & Platform	Cloud Platforms (e.g., Google Cloud Functions) [27]	Provides the backend infrastructure for serverless computation, data processing, and triggering interventions.
	Application Programming Interfaces (APIs) [25]	Enables integration of external services (e.g., nutrition databases for food content).
Analytical Frameworks	Microrandomized Trial (MRT) Design [3]	An experimental design optimized for constructing JITAIs by testing intervention effects at numerous decision points over time.
	Machine Learning Algorithms [3]	Used to develop predictive models (e.g., for lapse risk) that form the basis of sophisticated decision rules.
	Conceptual Frameworks (e.g., Nahum-Shani et al.) [5] [3]	Provides a systematic structure for designing, reporting, and evaluating JITAIs.

JITAIs represent a paradigm shift in dietary adherence research, moving from static, one-size-fits-all interventions to dynamic, personalized, and context-aware support. Current evidence robustly demonstrates the efficacy of JITAIs for reducing sodium intake, while the field for caloric control is rapidly advancing with methodologically rigorous trials [6] [25] [3]. The future optimization of JITAIs hinges on several key factors: the adoption of sophisticated analytical techniques like machine learning for improved tailoring [5] [3], a deeper investigation into the core components of JITAIs such as receptivity and optimal decision rules [26] [5], and the implementation of novel experimental designs like MRTs to efficiently build empirical foundations [3]. As these elements coalesce, JITAIs hold immense potential to deliver on the promise of precision health in dietary behavior modification.

Building Effective JITAIs: Design Frameworks and Practical Implementation

Establishing Distal and Proximal Outcomes for Dietary Interventions

In the development of Just-in-Time Adaptive Interventions (JITAIs) for dietary adherence, the precise definition of proximal and distal outcomes is fundamental to intervention design and evaluation. JITAIs are defined as interventions that provide "behavioral support that directly corresponds to a need in real-time" with content "adapted or tailored according to input collected by the system" [28]. Within this framework, proximal outcomes represent the immediate, short-term behaviors directly targeted by intervention components, while distal outcomes reflect the ultimate health goals that are expected to improve as a result of sustained changes in proximal outcomes [2]. This distinction creates a causal pathway where successful manipulation of proximal outcomes through timely intervention is hypothesized to lead to the achievement of distal health outcomes.

Defining Proximal and Distal Outcomes

Conceptual Framework

The conceptual model for a JITAI links momentary intervention decisions to long-term health goals through a series of mediating variables. As outlined in the framework by Nahum-Shani et al. [2], this includes:

• Decision Points: Times when intervention decisions are made (e.g., following Ecological Momentary Assessment surveys).
• Tailoring Variables: Information used to decide when and how to intervene (e.g., current lapse risk).
• Intervention Options: The specific support provided at decision points.
• Proximal Outcomes: The immediate behaviors targeted (e.g., preventing a single dietary lapse).
• Distal Outcomes: The ultimate health goals (e.g., weight loss, improved clinical markers).

Classification of Outcomes for Dietary JITAIs

Table 1: Outcome Classification for Dietary Adherence JITAIs

Category	Definition	Measurement Timeline	Specific Examples
Proximal Outcomes	Immediate behaviors directly targeted by JITAI components	Minutes to hours after intervention	- Occurrence of dietary lapse in 2.5-hour window [2]- Passive eating characteristics (duration, bite count) [2]- Successful resistance to a specific temptation
Distal Outcomes	Long-term health changes resulting from sustained proximal change	Weeks to months	- Weight loss (e.g., 5-10% of body weight) [2]- Reduction in HbA1c levels [29]- Regression to normoglycaemia [29] [30]

Quantitative Data and Evidence Base

Empirical Support for Outcome Selection

Research indicates that dietary lapses occur 3-4 times per week in behavioral obesity treatment and are significantly associated with poorer weight loss outcomes [2]. This evidence supports the selection of lapse occurrence as a key proximal outcome. In prediabetes interventions, achieving normoglycaemia represents a clinically meaningful distal outcome, with studies showing intensive lifestyle interventions can reduce diabetes incidence by 58% over three years [29] [30].

Table 2: Quantitative Evidence for Outcome Selection

Outcome	Measurement Method	Evidence of Clinical Significance	Source
Dietary Lapses	Ecological Momentary Assessment (EMA) via smartphone [2]	Direct association with suboptimal weight loss; occurs 3-4 times/week [2]	JMIR (2021)
Weight Loss	Body weight measurement	5-10% reduction reduces cardiovascular disease risk [2]	JMIR (2021)
Normoglycaemia	HbA1c measurement	58% reduction in diabetes incidence with lifestyle change [29]	Public Health Nutrition (2021)

Experimental Protocols for Outcome Assessment

Protocol: Microrandomized Trial for Proximal Outcome Evaluation

Purpose: To optimize a JITAI by evaluating the effects of intervention options on proximal outcomes [2].

Design Elements:

• Participants: Adults with overweight/obesity and ≥1 cardiovascular disease risk factor (target N=159) [2].
• Duration: 6-month intervention with 3-month follow-up.
• JITAI Components:
  • EMA Surveys: 6 prompts daily (8:30 AM, 11:00 AM, 1:30 PM, 4:00 PM, 6:30 PM, 9:00 PM) with 90-minute response windows.
  • Risk Algorithm: Machine learning algorithm calculates real-time lapse risk from EMA responses.
  • Intervention Randomization: When elevated risk is detected, random assignment to: no intervention, generic risk alert, or 1 of 4 theory-driven interventions (education, self-efficacy, motivation, self-regulation).

Primary Proximal Outcome Measurement:

• Occurrence of dietary lapse in the 2.5 hours following randomization, measured via subsequent EMA [2].

Secondary Proximal Outcomes:

• Passive measurement of eating characteristics via wrist-based monitoring during the first 14 days of treatment and subsequent 14-day periods at 3 and 6 months [2].

Protocol: Assessing Distal Outcomes in Prediabetes Interventions

Purpose: To evaluate long-term health outcomes following a dietary intervention [29] [30].

Design Elements:

• Participants: Adults with prediabetes (HbA1c 41-49 mmol/mol) and BMI ≥ 25 kg/m² [29] [30].
• Intervention Structure: Pragmatic primary care nurse-led intervention with individualised dietary assessment, goal setting, and advice considering socio-economic and cultural influences [29] [30].
• Assessment Schedule: Baseline, 6 months, 12 months, and 24 months [29] [30].

Distal Outcome Measurements:

• Primary: Glycaemic status (normoglycaemia, persistent prediabetes, or progression to type 2 diabetes) [29] [30].
• Secondary: Body weight, cardiovascular risk factors, health-related quality of life [29] [30].

Visualizing JITAI Workflows and Outcome Pathways

JITAI Conceptual Model for Dietary Interventions

Diagram 1: JITAI conceptual model showing the pathway from assessment to outcomes.

Outcome Assessment Workflow

Diagram 2: Workflow showing the relationship between proximal and distal outcome assessment.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Research Materials and Tools for Dietary JITAI Studies

Tool Category	Specific Solution	Function in Research	Implementation Example
Assessment Platforms	Ecological Momentary Assessment (EMA) [2]	Repeated real-time data collection on behaviors, triggers, and contexts	6 daily surveys via smartphone app assessing lapse triggers [2]
Analytical Algorithms	Machine Learning Risk Algorithm [2]	Calculates real-time lapse risk from EMA data to trigger interventions	Algorithm analyzes EMA responses to determine elevated lapse risk for microrandomization [2]
Intervention Delivery Systems	Smartphone JITAI Platform [2]	Delivers tailored interventions at moments of elevated risk	System-triggered notifications with theory-driven content when lapse risk is high [2]
Passive Monitoring Tools	Wrist-Based Sensors [2]	Objective measurement of eating behaviors without user input	Captures eating duration, rate, and bite count in 2.5-hour post-intervention windows [2]
Clinical Outcome Measures	HbA1c Assays [29]	Standardized measurement of glycaemic control	Assesses progression to normoglycaemia, persistent prediabetes, or diabetes at 6-month intervals [29]
Behavioral Frameworks	Socio-Ecological Model (SEM) [29] [30]	Guides understanding of multi-level influences on dietary behavior	Informs intervention design addressing personal, interpersonal, and environmental factors [29] [30]

Just-in-Time Adaptive Interventions (JITAIs) represent a transformative approach in digital health, designed to provide the right type and amount of support at the right time by adapting to an individual's changing internal and contextual state [1]. The efficacy of JITAIs hinges on the identification and deployment of tailoring variables—individual-specific information used for personalization that determines when and how to intervene [1]. In the specific context of dietary adherence for chronic conditions like type 2 diabetes (T2D), key dynamic tailoring variables include mood, location, and cravings [31]. These variables are crucial because dietary choices are not made in a vacuum but are profoundly influenced by momentary emotional states, environmental contexts, and visceral physiological signals. This document provides detailed application notes and experimental protocols for identifying, measuring, and leveraging these variables to optimize JITAIs for dietary adherence research.

Theoretical and Empirical Foundations

The Role of Tailoring Variables in JITAI Architecture

Within the JITAI framework, tailoring variables are dynamic components that feed into decision rules to select appropriate intervention options at specific decision points [1]. Their purpose is to operationalize the adaptation mechanism, ensuring that support is ecologically sound and delivered during states of vulnerability or opportunity [1]. For dietary behavior, which is highly episodic and context-dependent, static factors like demographic information or baseline preferences are insufficient for real-time adaptation. Dynamic factors such as mood, location, and cravings provide the necessary granularity to intervene proactively against lapses in dietary adherence.

A survey identified these factors as significant facilitators and barriers for a healthy lifestyle among individuals with lifestyle-related chronic conditions [31]. Specifically:

• Mood is a critical factor for both physical activity and dietary choices.
• Location (e.g., eating out in a restaurant) directly influences the type and quality of food available and consumed.
• Cravings present a momentary, powerful internal state that can override planned dietary intentions.

Evidence from Feasibility Studies

A recent acceptability study on an EMA-driven JITAI for T2D management provides preliminary evidence for the feasibility of using these variables [31]. In this study, participants completed daily Ecological Momentary Assessments (EMAs) about their activity, location, mood, overall condition, and cravings. This data was used to deliver tailored support via SMS text messaging. The study found that while the EMA was perceived as easy to use and low in burden, participants felt it provided too much of a snapshot and too little context, thereby reducing the perceived tailoring of the intervention options [31]. This highlights a critical challenge: the need for assessment methods that balance active (EMA) and passive data collection to minimize user burden while maximizing contextual understanding.

Defining and Measuring Core Tailoring Variables

Mood

Conceptual Definition: Mood in the context of dietary adherence refers to a transient, subjective emotional state that can influence impulsive decision-making and self-regulatory capacity. Negative affect, such as stress or sadness, is particularly linked to unhealthy food choices.

Measurement Protocols:

• EMA Prompt Design: Utilize a 5-item bipolar scale (e.g., "Stressed - Relaxed," "Sad - Happy") presented via smartphone prompts at random intervals 3-5 times per day or event-contingently following self-identified stressful events.
• Objective Correlates (Passive Sensing):
  • Social Rhythm: Measure variability in phone usage patterns or social connectivity (via call/text logs) as a proxy for routine disruption, which correlates with mood instability.
  • Voice Analytics: Use smartphone microphones to analyze speech patterns (rate, tone, energy) during consented phone calls for markers of negative affect.

Data Integration: Fuse active EMA ratings with passive social rhythm metrics to create a composite mood instability score. A deviation from the individual's baseline (e.g., increased stress + disrupted social rhythm) triggers a state of vulnerability.

Table 1: Mood Measurement Methods and Metrics

Method Type	Specific Tool/Data Source	Key Metrics	Frequency
Active (EMA)	Smartphone-delivered visual analog scales	Arousal, Valence (e.g., 1-7 Likert scales)	3-5 random prompts/day + event-contingent
Passive (Sensing)	Smartphone usage logs (calls, texts, app use)	Social rhythm metrics, keystroke dynamics	Continuous
Passive (Sensing)	Audio analysis (with user consent)	Speech prosody, pitch, energy	During phone calls

Location

Conceptual Definition: Location extends beyond geographic coordinates to encompass the food environment characteristics of a place, which can enable or constrain healthy dietary choices.

Measurement Protocols:

• Geofencing around Pre-Identified Zones: Use GPS or Wi-Fi positioning to create virtual boundaries (geofences) around locations the individual has previously identified as high-risk (e.g., favorite fast-food restaurant, workplace cafeteria) or low-risk (e.g., gym, home).
• Crowdsourced Food Environment Data: Integrate with third-party APIs (e.g., Google Places) to classify the type of food outlets in the individual's immediate vicinity (e.g., "fast food," "grocery store," "healthy restaurant").
• Travel Trajectory Analysis: Model GPS traces in real-time to determine if the individual is on a route that commonly leads to a high-risk food environment.

Data Integration: Location is not used in isolation. A JITAI is most potent when location is combined with other states. For instance, entering a geofence around a fast-food restaurant may not trigger an intervention alone. However, if this location is coupled with a self-reported or predicted negative mood state, it constitutes a high-risk context warranting JITAI support.

Cravings

Conceptual Definition: A craving is a strong, conscious, and specific desire to consume a particular type of food (often energy-dense, sweet, or salty). It is a subjective experience with both psychological and physiological underpinnings.

Measurement Protocols:

• EMA for Self-Report: Implement a 2-item craving EMA that is user-initiated or triggered by time (e.g., post-prandial). The items should assess: a) intensity of the craving ("How strong is your craving right now?" on a 1-10 scale), and b) target food ("What food are you craving?").
• Passive Physiological Proxies:
  • Heart Rate Variability (HRV): Use optical sensors on wearable devices (e.g., smartwatches) to capture a drop in HRV, which is associated with reduced self-control and may precede craving episodes.
  • Skin Conductance: Monitor for subtle increases in electrodermal activity indicating heightened arousal related to craving.

Data Integration: A machine learning classifier can be trained on an individual's baseline data to predict craving states from passive physiological streams. A prediction of high-craving probability can trigger a low-burden EMA for confirmation, thus balancing accuracy with user burden.

Table 2: Cravings and Location Measurement Matrix

Tailoring Variable	Primary Measurement	Supplementary Data	Key Derived Metric
Cravings	User-initiated EMA (Intensity, Food Type)	Heart Rate Variability (HRV) from wearables	Craving risk score (combination of self-report and physiological arousal)
Location	GPS / Wi-Fi positioning	Geofencing, Points-of-Interest API data	Classification of current location as High/Low Risk

Experimental Workflow for JITAI Development

The following diagram illustrates the integrated experimental workflow for developing a JITAI using mood, location, and cravings.

Diagram 1: JITAI experimental workflow for dietary adherence.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Digital Tools for JITAI Research

Item/Tool Category	Specific Examples	Function in JITAI Research
Mobile Sensing Platforms	AWARE Framework (Android/iOS), Beiwe	Open-source platforms for collecting passive sensor data (GPS, accelerometer, usage logs) from smartphones.
EMA Delivery Systems	PACO (Personal Analytics Companion), mEMA (Ilumivu)	Configurable systems to design and deliver smartphone-based EMA surveys assessing mood, cravings, and context.
Wearable Physio Monitors	Empatica E4, ActiGraph GT9X, Consumer Smartwatches (Fitbit, Apple Watch)	Capture physiological data streams (HRV, EDA, activity) that serve as passive proxies for states like stress and cravings.
Geospatial Analysis Tools	Google Places API, OpenStreetMap, GIS software (QGIS)	Characterize the food environment of a participant's location by identifying types of nearby food outlets.
JITAI Orchestration Engine	RADAR-base, Mobile Health Platform (mHP)	Backend systems that host decision rules, integrate data from multiple sources, and automate the delivery of intervention options.
Data Analysis Environments	R (with specific packages for mmASD), Python (Pandas, Scikit-learn)	Statistical environments for conducting microrandomized trial analyses and developing machine learning models for state prediction.

Detailed Experimental Protocol for a 14-Day JITAI Feasibility Study

This protocol is adapted from a referenced acceptability study [31] and is designed for researchers to implement and test a JITAI for dietary adherence.

Objective: To evaluate the feasibility, acceptability, and preliminary efficacy of an EMA-driven JITAI that uses mood, location, and cravings to support dietary adherence in adults with T2D.

Participants:

• Recruitment: N=20-30 adults with clinically diagnosed T2D.
• Inclusion Criteria: Age 18-75, owns a smartphone, willing to use a study-provided wearable sensor.
• Ethics: Obtain approval from an Institutional Review Board (IRB). Acquire informed consent detailing data collection from phone sensors, wearables, and EMAs.

Materials & Setup:

• Smartphone App: Configure a JITAI-enabled app (e.g., using AWARE or a custom build) with the following:
  • Scheduled EMAs (3-5 random prompts per day).
  • User-initiated craving EMA button.
  • Permissions for continuous GPS and accelerometer data.
  • SMS integration for message delivery.
• Wearable Sensor: Provide each participant with a device capable of measuring heart rate (e.g., Empatica E4 or Fitbit).
• Backend Server: Implement decision rules on a server (e.g., using mHP) that ingests sensor and EMA data in near-real-time.

Procedure:

• Baseline Phase (Day 1):
  • Collect demographic and clinical data.
  • Have participants identify and geo-tag 3-5 personal high-risk and low-risk food locations via the app.
  • Conduct a 5-minute baseline resting heart rate measurement.
• Intervention Phase (Days 2-15):
  • Data Collection:
    • Passive: Continuous GPS, accelerometer, and heart rate data.
    • Active: Random EMAs (mood, current activity), event-triggered EMAs (upon entering a high-risk geofence), and user-initiated craving EMAs.
  • Decision Rules: Implement the following exemplar logic:
    • IF (participant enters High-Risk Location geofence) AND (mood EMA score indicates "stressed" or "sad") THEN send intervention option A (distraction + alternative action).
    • IF (craving intensity >7/10) AND (passive HRV drops below personal baseline) THEN send intervention option B (urge-surfing meditation).
    • IF (location is "home") AND (evening time) AND (no craving reported) THEN send intervention option C (planning for the next day).
  • Intervention Options: Deliver via SMS or in-app notification. Options should be grounded in Cognitive Behavioral Therapy (CBT) and Mindfulness-Based Eating.
• Post-Study (Day 16):
  • Conduct a semi-structured qualitative interview to assess acceptability, perceived usefulness, and burden of the JITAI system and its components.

Data Analysis Plan:

• Feasibility: Recruitment rate, retention rate, adherence to EMA prompts (>80% target), sensor data completeness.
• Acceptability: Thematic analysis of interview transcripts, focusing on perceptions of personalization, timing, and relevance of messages.
• Preliminary Efficacy: Use microrandomized trial analysis techniques to test the marginal effect of intervention messages on the proximal outcome of next-meal healthfulness (also assessed via EMA).

Mood, location, and cravings are not merely correlates of dietary behavior but are dynamic, interwoven components of the decision-making architecture that JITAIs are uniquely positioned to target. Successfully leveraging these variables requires a multi-method assessment strategy that blends active EMAs with passive sensing to create a rich, contextualized understanding of the individual's state. The protocols and tools outlined here provide a foundational roadmap for researchers.

Future development must focus on refining the decision rules through experimental designs like Microrandomized Trials (MRTs) and improving the contextual granularity of assessments to move beyond "snapshots." Furthermore, as noted in prior research, participants express a need for even more personalized support [31], suggesting that future work should explore idiographic (N-of-1) models for tailoring variable selection and decision rule optimization. The ultimate goal is to create JITAIs that are not only scientifically rigorous but also deeply resonant with the complex, lived experience of individuals managing their dietary health.

The integration of artificial intelligence (AI) into dietary adherence research represents a paradigm shift from generalized nutritional guidance to highly personalized, dynamic intervention strategies. This evolution is particularly critical for just-in-time adaptive interventions (JITAIs), which aim to provide tailored support to individuals at moments of greatest need. The foundation of these systems lies in their decision rules—the logical frameworks that process incoming data to determine when and how to intervene. The sophistication of these rules ranges from simple, human-programmed if-then statements to complex machine learning (ML) models that predict individual risk states from multimodal data streams. Research demonstrates that AI-generated dietary interventions can lead to substantial health improvements, including a 39% reduction in IBS symptom severity and a 72.7% diabetes remission rate in clinical studies, significantly outperforming traditional one-size-fits-all approaches [32].

This progression of decision logic is not merely a technical exercise but a fundamental requirement for addressing the complex, context-dependent challenge of dietary adherence. Traditional rule-based systems offer transparency and control but lack the adaptability to individual metabolic variability and changing circumstances. In contrast, ML-based approaches can model complex, non-linear relationships between contextual cues, physiological states, and behavioral outcomes, enabling more precise and effective intervention timing. The power of modern dietary JITAIs stems from their ability to leverage this advanced decision logic in conjunction with continuous data from wearable sensors and mobile self-reports, creating a closed-loop system that adapts to the user's evolving needs [33] [6].

The Evolution of Decision Logic in Dietary Interventions

Foundational Rule-Based Systems

Rule-based systems form the historical and conceptual foundation for most digital interventions. In the context of dietary adherence, these systems operate on a fixed set of logical conditions programmed by researchers or clinicians. The IF-THEN statement is the fundamental building block, evaluating a condition and triggering a predetermined action if the condition is met. For example, a simple rule might state: IF the system time is 12:30 PM (typical lunch hour) AND the user has not logged a meal, THEN send a reminder to eat a healthy lunch. The AND logical operator is crucial, as it allows multiple conditions to be combined, creating more specific and context-aware rules [34].

More complex rule structures, such as Sequential Rules and Rule Chains, allow for multi-stage decision processes that are particularly useful in dietary questionnaires or structured counseling protocols. For instance, a Rule Chain might first determine a user's current hunger level, then assess available food options, and finally provide a specific recommendation based on the combined assessment. Tree Rules offer another sophisticated approach, using a branching structure to represent a series of conditional decisions, visually mapping out the possible paths from a user's initial state to a final recommendation [34] [35]. While highly interpretable and easy to implement, the primary limitation of these systems is their static nature; they cannot learn from new data or adapt their logic to patterns observed across a user population.

The Shift to Machine Learning-Driven Rules

Machine learning represents a fundamental shift from programmer-defined rules to data-driven, algorithmically derived decision models. Instead of being explicitly told which conditions to check, ML models are trained on historical data to identify complex patterns and relationships that predict dietary lapses or adherence. For example, a study utilizing continuous glucose monitors (CGM), accelerometers, and food diaries achieved an F1-score of 0.88 in distinguishing between fasting and non-fasting states using a Support Vector Machine (SVM) model. This high level of accuracy in classifying a behavioral state from sensor data demonstrates the predictive power of ML that rule-based systems struggle to match [33].

These models excel at integrating and interpreting multimodal data streams. In dietary JITAIs, relevant features can include time-series glucose data, physical activity levels from accelerometers, self-reported mood and hunger, and temporal patterns. The model synthesizes these inputs to generate a moment-by-moment probability of a dietary lapse. When this probability exceeds a certain threshold, the system can trigger an intervention. This approach was central to the "myBPmyLife" JITAI trial, which aimed to promote physical activity and lower-sodium food choices. The system used real-time data to tailor push notifications to users at critical moments when they were more likely to succeed, moving beyond a fixed, time-based schedule to a contextually responsive one [36] [6].

Table 1: Comparison of Rule-Based and Machine Learning Approaches for Dietary JITAIs

Feature	Rule-Based Systems	Machine Learning Systems
Logic Source	Expert-defined, pre-programmed	Data-derived, model-generated
Adaptability	Static; requires manual updates	Dynamic; improves with more data
Complexity Handling	Limited to pre-specified conditions	Can model complex, non-linear interactions
Transparency	High; easily interpretable	Low; "black box" models can be opaque
Data Utilization	Uses direct inputs for conditional checks	Can integrate and weight multimodal data streams
Primary Strength	Simplicity, control, and interpretability	Predictive accuracy and personalization
Example Use Case	Reminder at a fixed meal time	Predicting lapse risk based on glucose variability and location

Hybrid Approaches

The dichotomy between rule-based and ML systems is not always absolute. Hybrid approaches are increasingly common, leveraging the strengths of both paradigms. In a typical hybrid JITAI, an ML model might act as the "risk detector," processing sensor and self-report data to identify high-risk moments for a dietary lapse. Once a high-risk state is identified, a separate, rule-based system could determine the most appropriate type of intervention from a library of options. This second step might be based on simpler rules that consider immediate context, such as the user's current location (e.g., home vs. restaurant) or time of day [36]. This architecture combines the predictive power of ML with the transparency and control of rule-based systems for action selection.

Quantitative Evidence and Outcomes

Recent clinical trials and studies provide compelling evidence for the efficacy of AI-driven dietary interventions. A 2025 systematic review of 11 studies found that among nine studies with comparison groups, six reported statistically significant improvements in the AI groups, while two found outcomes that were comparable or better than the control. The interventions reviewed leveraged a variety of AI methods, including conventional ML algorithms, deep learning (DL), and hybrid approaches integrating ML with Internet of Things (IoT)-based systems. The outcomes measured were clinically significant, spanning improved glycemic control, enhanced metabolic health, and even psychological well-being [32].

Another 2025 randomized controlled trial, the "myBPmyLife" study, investigated a JITAI for patients with hypertension. While the primary outcome of systolic blood pressure reduction did not differ significantly from the control group, the intervention successfully improved two key behavioral secondary outcomes: it increased daily step count by 170 steps (compared to a decrease of 319 steps in the control) and reduced estimated daily sodium intake by 1145 mg (compared to a reduction of 860 mg in the control). These results highlight the potential of JITAIs to positively influence the very behaviors that traditional interventions struggle to modify [6].

Table 2: Key Outcomes from Recent AI-Driven Dietary Intervention Studies

Study / Trial	AI Method	Population	Key Outcome(s)
Systematic Review (2025) [32]	Conventional ML, Deep Learning, Hybrids	Adults with chronic conditions (e.g., Diabetes, IBS)	39% reduction in IBS symptom severity; 72.7% diabetes remission rate; significant improvements in glycemic control.
Sensor-based Fasting Evaluation (2025) [33]	Support Vector Machine, Random Forest	Individuals on intermittent fasting regimens	F1-score of 0.88 for classifying fasting/non-fasting states using CGM and accelerometer data.
myBPmyLife JITAI RCT (2025) [6]	Just-in-Time Adaptive Intervention	Patients with hypertension	Increased daily steps by 170 (+489 vs control); Reduced sodium intake by 1145 mg (-285 mg vs control).
JITAI for Obesity Protocol (2021) [36]	Microrandomized Trial Design	Adults with overweight/obesity	Protocol to optimize JITAI for dietary lapses; uses theory-driven interventions (e.g., self-efficacy, motivation).

Experimental Protocols for JITAI Development and Evaluation

Protocol: Microrandomized Trial for JITAI Optimization

Objective: To optimize the composition of a JITAI by empirically testing the acute effect of different intervention components on proximal outcomes (e.g., dietary lapse prevention) in real-world contexts [36].

Background: This protocol is designed for the development phase of a JITAI, prior to a definitive RCT on distal outcomes like weight loss.

Methodology:

• Participant Recruitment: Enroll adults (e.g., n=159) with the target condition (e.g., overweight or obesity) and cardiovascular disease risk into a base intervention (e.g., a 6-month web-based behavioral obesity treatment).
• Risk Detection & Randomization: The JITAI system is programmed to assess elevated lapse risk periodically via daily surveys or sensor data. Each time elevated risk is detected, the participant is microrandomized to one of the following:
  • No intervention.
  • A generic lapse risk alert.
  • One of several theory-driven intervention components (e.g., enhanced education, building self-efficacy, fostering motivation, improving self-regulation).
• Primary Outcome Measurement: The primary proximal outcome is the occurrence of a dietary lapse within a short window (e.g., 2.5 hours) following randomization.
• Data Analysis: Data is analyzed to determine:
  • The overall effect of receiving any intervention versus none.
  • The comparative effectiveness of different theory-driven components.
  • Contextual moderators (e.g., time of day, location) of intervention efficacy.

Outcome: The data informs an optimized decision rule that selects the most effective intervention type for a given individual and context [36].

Protocol: Developing an ML Model for Behavioral State Classification

Objective: To develop and validate a supervised machine learning model capable of classifying a specific dietary behavior state (e.g., fasting vs. non-fasting) using time-series sensor data [33].

Background: This protocol enables the passive, objective assessment of dietary adherence, a core requirement for an automated JITAI.

Methodology:

• Data Collection:
  • Sensors: Collect time-series data from continuous glucose monitors (CGM) and tri-axial accelerometers.
  • Labels: Collect ground-truth data on dietary behavior (e.g., start/end times of meals and fasting) using food diaries or mobile app logs.
• Data Preparation & Harmonization:
  • Import all data into a unified framework (e.g., Python pandas).
  • Harmonize sampling rates and align all data sources by timestamp.
  • Label sensor data periods with the corresponding behavioral state (e.g., "fasting" or "non-fasting").
• Feature Engineering:
  • Glucose Data: Compute established metrics of glycemic control and variability (e.g., using the cgmquantify Python package) over a sliding window (e.g., 45 minutes).
  • Acceleration Data: Compute features related to activity recognition (e.g., signal peaks, magnitude) over the same window.
  • Feature Selection: Use an algorithm like Recursive Feature Elimination (RFE) to identify the most predictive features for the model.
• Model Training & Evaluation:
  • Address class imbalance in the dataset using techniques like SMOTETomek.
  • Scale features using a Min-Max scaler.
  • Train and compare multiple supervised ML models (e.g., Support Vector Machine, Random Forest, Multi-layer Perceptron).
  • Evaluate model performance using appropriate metrics (e.g., F1-score, accuracy) on a hold-out test set and external validation cohorts.

Outcome: A validated binary classification model that can be deployed within a JITAI to detect behavioral states from sensor data alone [33].

Visualization of Decision Logic Workflows

Rule-Based Decision Tree for Meal Initiation

Machine Learning Workflow for Dietary Lapse Prediction

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Tools for Dietary JITAI Research

Tool / Reagent	Type	Primary Function in Research
Continuous Glucose Monitor (CGM) [33]	Sensor	Provides passive, high-frequency time-series data on interstitial glucose levels, a key biomarker for metabolic state and meal timing.
Tri-axial Accelerometer [33]	Sensor	Captures objective physical activity and movement data, used for activity recognition and as a contextual feature for intervention delivery.
Mobile Food Diary App [33]	Software	Collects self-reported dietary intake and meal timing data, serving as ground truth for model training and adherence validation.
scikit-learn Python Package [33]	Software Library	Provides a comprehensive suite of tools for machine learning, including feature selection algorithms (RFE), model training, and evaluation.
Imbalanced-learn Python Package [33]	Software Library	Offers specialized techniques for handling class imbalance in behavioral datasets (e.g., using SMOTETomek).
JITAI Platform (e.g., HeartSteps) [36] [6]	Software Framework	Provides the backbone for developing and deploying adaptive interventions, including randomization engines and notification systems.
Dash/Plotly Libraries [33]	Software Library	Enables the creation of interactive dashboards for researchers to visualize model outputs, sensor data, and participant adherence.

The Microrandomized Trial (MRT) Design for JITAI Optimization

Just-in-Time Adaptive Interventions (JITAIs) are mobile health (mHealth) interventions designed to deliver the right support component at the right time by adapting to an individual's changing internal state and external context [37]. The micro-randomized trial (MRT) has emerged as a specialized experimental design specifically developed to optimize these dynamic intervention systems [37] [38]. Unlike traditional randomized controlled trials (RCTs) that evaluate intervention packages as a whole, MRTs enable researchers to assess the causal effects of individual intervention components and investigate how these effects vary over time and context [37]. This design is particularly valuable for dietary adherence research, where factors influencing lapses are dynamic and context-dependent [2] [36].

MRTs accomplish this by randomly assigning participants to different intervention options at numerous decision points throughout the study, sometimes numbering in the hundreds per participant [39]. This high-frequency randomization generates data to understand whether intervention components have intended proximal effects, when they are most effective, and what contextual factors moderate their efficacy [37]. For dietary adherence research, this methodology enables precise optimization of JITAIs to prevent lapses during behavioral obesity treatment [2] [36].

Core Components of the MRT Design

Theoretical Foundations and Key Elements

The MRT design operates on the principle that mobile interventions require understanding momentary, time-varying causal effects rather than static, overall effects [37] [38]. In an MRT, each participant serves as their own control across multiple randomization instances, allowing for highly efficient within-person comparisons while also enabling between-person contrasts [37]. This is particularly advantageous for dietary research, where within-person variability in lapse triggers is substantial [2].

Key elements of the MRT design include decision points (times when intervention decisions are made), intervention options (the different components being tested), tailoring variables (information used to decide when and how to intervene), proximal outcomes (immediate effects of interventions), and distal outcomes (long-term health goals) [2] [5]. For dietary adherence, a typical proximal outcome might be the occurrence of a dietary lapse in the hours following randomization, while distal outcomes include weight loss or improved cardiovascular risk factors [2] [36].

Comparison with Traditional Experimental Designs

Table 1: Comparison of MRTs with Traditional Experimental Designs

Design Feature	Standard RCT	Factorial Design	Micro-Randomized Trial
Unit of Randomization	Participant	Participant	Decision point
Primary Question	Does the intervention package work?	Which components are effective?	When and under what conditions are components effective?
Time-Varying Effect Moderation	Limited assessment	Limited assessment	Explicitly modeled
Optimization Target	Intervention package	Component selection	Timing and context for delivery
Data Efficiency	Lower	Moderate	Higher for proximal effects

Traditional RCTs are designed to assess whether an intervention package as a whole has an effect on behavior but are not designed to investigate which components are efficacious, when they are efficacious, or what psychosocial or contextual factors influence their efficacy [37]. Factorial designs within the Multiphase Optimization Strategy (MOST) framework can assess effects of individual components but do not adequately address when it is most effective to deliver each component [37]. MRTs overcome these limitations by specifically enabling research into time-varying effect moderation, which is essential for developing effective decision rules in JITAIs [37] [38].

Application to Dietary Adherence Research

Protocol for Dietary Lapse Prevention

A specific application of MRTs in dietary research is exemplified by a protocol to optimize a JITAI for dietary adherence in behavioral obesity treatment [2] [36]. This study aims to address the critical problem of dietary lapses, which occur 3-4 times per week on average during behavioral obesity treatment and are associated with poorer weight loss outcomes [2]. The protocol employs an MRT to evaluate the efficacy of various theory-driven interventions delivered at moments of high lapse risk, with the ultimate goal of developing an optimized JITAI that selects the most appropriate intervention strategy for a given individual and context [2].

Adults with overweight or obesity and cardiovascular disease risk (n=159) participate in a 6-month web-based behavioral obesity treatment while using the JITAI to prevent dietary lapses [2]. Each time the JITAI's machine learning algorithm detects elevated lapse risk based on ecological momentary assessment (EMA) data, the system randomizes the participant to one of six conditions: no intervention, a generic risk alert, or one of four theory-driven interventions targeting education, self-efficacy, motivation, or self-regulation [2]. The primary proximal outcome is the occurrence of a dietary lapse in the 2.5 hours following randomization, assessed via EMA [2].

Quantitative Parameters and Outcomes

Table 2: Key Parameters from a Dietary Adherence MRT

Parameter	Specification	Rationale
Sample Size	159 participants	Powered for primary aims
Study Duration	6 months (3 months BOT + 3 months follow-up)	Allows examination of time-varying effects
Decision Points	~6 per day via EMA	Captures varying contexts throughout day
Expected Randomizations	>100 per participant	Provides statistical power for proximal effects
Primary Outcome	Dietary lapse in 2.5 hours post-randomization	Proximal effect of intervention
Intervention Options	6 conditions (including control)	Tests multiple theory-driven approaches

The dietary adherence MRT includes several key assessment strategies. Ecological momentary assessment occurs at six predetermined times daily (8:30 AM, 11:00 AM, 1:30 PM, 4:00 PM, 6:30 PM, and 9:00 PM), with participants given 90 minutes to respond to each survey [2]. Secondary outcomes include passive measurement of eating characteristics via wrist-based monitoring during specified periods throughout the study [2]. Contextual moderators such as location, time of day, and whether the participant is in active treatment or follow-up are explored to determine circumstances under which interventions are more or less effective [2].

Methodological Workflow and Decision Framework

MRT Workflow for Dietary Intervention

JITAI Decision Framework in MRT Context

Research Reagents and Methodological Tools

Essential Research Components for Dietary MRTs

Table 3: Research Reagent Solutions for Dietary Adherence MRTs

Research Component	Function in MRT	Example Implementation
Ecological Momentary Assessment (EMA)	Captures time-varying tailoring variables and proximal outcomes	6 daily surveys assessing lapse triggers, mood, context [2]
Machine Learning Algorithm	Determines real-time lapse risk based on EMA data	Risk classification using behavioral, psychological, environmental triggers [2]
Randomization Engine	Assigns intervention conditions at decision points	Algorithm randomizing to 6 conditions when risk elevated [2] [39]
Passive Sensing Technology	Provides contextual data without user burden	Wrist-based monitoring of eating characteristics [2]
Theory-Driven Intervention Content	Specific strategies targeting hypothesized mechanisms	Education, self-efficacy, motivation, self-regulation components [2] [36]
Data Integration Platform	Combines active and passive data streams for decision rules	Mobile platform integrating EMA, sensor data, and intervention delivery [2]

The research reagents table highlights essential methodological components for implementing a dietary adherence MRT. Ecological momentary assessment serves as the foundation for identifying decision points and measuring proximal outcomes [2]. Machine learning algorithms transform this EMA data into real-time risk assessments that trigger the micro-randomization process [2]. The randomization engine represents the core MRT component, assigning conditions repeatedly throughout the study [39]. Passive sensing technologies complement self-report data by providing objective measures of behavior and context [2]. Theory-driven intervention content ensures that the tested components are grounded in established behavioral mechanisms [2] [36]. Finally, data integration platforms bring these elements together into a functional JITAI system [2].

Analysis Considerations and Future Directions

Statistical Approaches and Methodological Challenges

Analyzing MRT data requires specialized statistical methods that account for the complex temporal structure of the measurements and interventions [37] [38]. Primary analyses focus on estimating causal excursion effects, which represent the causal effect of an intervention option on a proximal outcome when the decision to intervene is made according to a particular regime [37]. These analyses must properly account for the potential time-varying confounding that can occur despite the sequential randomization, as well as the correlation within individuals due to repeated measurements [37].

Moderator analyses are particularly important in MRTs, as they help identify under what circumstances interventions are most effective [37] [2]. In dietary adherence research, potential moderators include time-varying factors such as location, time of day, presence of others, and emotional states [2]. The intensive longitudinal nature of MRT data also enables investigation of how intervention effects change over the course of the study, addressing questions about habituation or learning effects [37] [38].

Methodological challenges in MRT implementation include participant burden, technology reliability, and developing appropriate statistical power calculations for the numerous randomizations [37] [2]. Additionally, there are ethical considerations around the frequency of interventions and the use of control conditions at moments of identified risk [37]. Future directions for MRT methodology in dietary research include incorporating more passive sensing data to reduce participant burden, developing adaptive randomization strategies that evolve based on accumulating data, and integrating MRTs with single-case experimental designs to further personalize intervention approaches [37] [40].

Just-in-time adaptive interventions (JITAIs) represent a transformative approach in digital health, designed to deliver personalized support at moments of heightened vulnerability or opportunity [4]. By leveraging data from smartphones, wearables, and ecological momentary assessments (EMAs), JITAIs dynamically adapt intervention type and timing to individual contexts, addressing the critical challenge of sustaining adherence to dietary recommendations [3] [41]. This article presents three detailed case studies applying JITAI methodology to hypertension, obesity, and type 2 diabetes management, providing researchers with structured protocols, outcome data, and implementation frameworks to advance dietary adherence research.

Case Study 1: Hypertension Management via Sodium Reduction

Application Note: The myBPmyLife Trial

The myBPmyLife study was a randomized controlled trial evaluating a JITAI to promote lower-sodium food choices and physical activity in 602 patients with hypertension over six months [42] [43]. The intervention adapted components from two previously developed JITAIs: HeartSteps (for physical activity) and LowSalt4Life (for sodium reduction) [43]. Participants had a mean age of 59.6 years, with 25.6% self-identifying as non-White and 48% as women [43]. The intervention group used the myBPmyLife application which delivered context-aware push notifications when users were near restaurants or grocery stores, suggesting lower-sodium alternatives [43].

Table 1: Primary and Secondary Outcomes from myBPmyLife Trial

Outcome Measure	Intervention Group	Control Group	P-value
Systolic BP Change (mmHg)	-5.2	-5.7	0.76
Daily Step Change	+170 steps	-319 steps	0.040
Sodium Intake Change (mg/day)	-1145 mg	-860 mg	0.002

Despite significant improvements in behavioral outcomes (step count and sodium reduction), the intervention did not yield significantly greater systolic blood pressure reduction compared to the control group [42] [43]. This highlights the complex relationship between behavioral changes and clinical endpoints in hypertension management.

Experimental Protocol: LowSalt4Life 2 Trial

The LowSalt4Life 2 trial builds upon this approach with an enhanced design [11]:

Study Design: 6-month randomized controlled trial with adaptive re-randomization.

Participants: 410 adults with hypertension.

Intervention Arms:

• Phase 1: 1:1 randomization to LowSalt4Life app with JITAI vs. app alone
• Phase 2 (Month 2): JITAI group re-randomized to standard JITAI vs. personalized JITAI (pJITAI) using reinforcement learning based on prior engagement

Primary Outcome: Change in systolic blood pressure at 2 months.

Secondary Outcomes: Changes in BP medication, dietary sodium intake, and engagement metrics.

JITAI Decision Rules:

• Tailoring Variables: GPS location (restaurants/grocery stores), prior engagement, self-reported dietary habits
• Intervention Options: Push notifications suggesting lower-sodium alternatives, recipe modifications, situational advice
• Decision Points: When user enters geofenced food-purchasing locations or during typical meal planning times

Case Study 2: Obesity Management via Dietary Lapse Prevention

Application Note: The Nudge Pilot Microrandomized Trial

The Nudge study was a 12-week pilot microrandomized trial (MRT) testing seven types of behavior change technique (BCT)-based messages on daily goal achievement among 52 young adults (BMI 25-40 kg/m²) [44]. Participants received a comprehensive weight loss intervention including lessons, tailored feedback, self-monitoring tools, and daily behavioral goals (weighing, active minutes, red foods limit) [44].

Table 2: Effects of Message Viewing on Daily Goal Achievement in Nudge Trial

Daily Goal	Odds Ratio	95% Confidence Interval
Weighing	1.91	1.06-3.43
Active Minutes	1.63	1.20-2.20
Red Foods Limit	2.09	1.46-3.01

Randomization to receive any message versus none did not impact overall goal achievement, but receiving and viewing any message significantly increased odds of achieving all three daily goals [44]. Social comparison messages about red foods were specifically associated with fewer red foods tracked (OR=0.92, 95% CI: 0.85-0.99) [44].

Experimental Protocol: Dietary Lapse Prevention JITAI

An ongoing MRT aims to optimize a JITAI for dietary lapses during behavioral obesity treatment [3]:

Study Design: 6-month microrandomized trial with 159 adults with overweight/obesity and cardiovascular risk factors.

Intervention Components:

• Web-based behavioral obesity treatment
• Smartphone JITAI with EMA-measured lapse triggers
• Machine learning algorithm for real-time lapse risk prediction

Randomization Scheme: When elevated lapse risk is detected, participants are randomized to:

• No intervention
• Generic risk alert
• Theory-driven interventions (enhanced education, building self-efficacy, fostering motivation, improving self-regulation)

Primary Proximal Outcome: Occurrence of dietary lapse in the 2.5 hours following randomization.

Decision Points: Following completion of each of 6 daily EMA surveys (8:30 AM, 11:00 AM, 1:30 PM, 4:00 PM, 6:30 PM, 9:00 PM) [3].

Figure 1: Dietary Lapse Prevention JITAI Workflow

Case Study 3: Type 2 Diabetes Management via Lifestyle Support

Application Note: EMA-Driven JITAI for T2D

A qualitative acceptability study evaluated an EMA-driven JITAI for lifestyle support in 8 individuals with type 2 diabetes [4]. Participants had a mean age of 70.5 years, BMI of 32.1 kg/m², and T2D duration of 15.6 years [4]. The intervention involved:

• Daily EMAs about activity, location, mood, overall condition, weather, and cravings
• Tailored support delivered via SMS text messaging
• Two-week intervention period

Participants provided positive feedback on motivating and enjoyable messages, but noted challenges with timing, frequency, and perceived personalization [4]. The EMA was perceived as easy to use with low burden, but participants felt it provided "too much of a snapshot and too little context," reducing perceived tailoring of intervention options [4].

Experimental Protocol: E-Supporter 1.0 JITAI Development

The JITAI was developed based on the E-Supporter 1.0 digital coach, designed according to the Nahum-Shani framework [4]:

JITAI Components:

• Distal Outcomes: Improvement in light to moderate vigorous physical activities; improved adherence to national dietary guidelines
• Proximal Outcome: Daily goal achievement
• Tailoring Variables: Weather, mood, daily activity, location, cravings, overall condition (collected via EMA)
• Intervention Options: Tailored motivational messages via SMS
• Decision Points: Multiple times daily based on EMA completion

Theoretical Foundation: Health Action Process Approach (HAPA) and behavior maintenance theories [4].

Assessment Methodology: Semistructured interviews analyzed via hybrid thematic analysis to assess acceptability of EMA content, intervention options, and overall JITAI use.

Cross-Cutting Methodological Framework

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Research Materials for JITAI Implementation

Tool Category	Specific Examples	Research Function
Mobile Platforms	Native iOS/Android apps (Nudge, myBPmyLife)	Primary intervention delivery mechanism
Wearable Sensors	Fitbit activity trackers, wireless scales (Fitbit Aria)	Passive data collection (steps, weight)
EMA Platforms	Custom smartphone EMA systems	Real-time assessment of behaviors, cognitions, emotions, environmental factors
Analytical Algorithms	Machine learning lapse prediction, reinforcement learning personalization	Tailoring intervention timing and content
Geolocation Services	GPS-based restaurant/grocery store detection	Context-aware intervention triggering

JITAI Conceptual Architecture

Figure 2: JITAI Conceptual Architecture for Dietary Adherence

These case studies demonstrate both the promise and challenges of JITAIs for dietary adherence across chronic conditions. While JITAIs show consistent improvements in proximal behavioral outcomes (step counts, sodium reduction, daily goal achievement), effects on distal clinical outcomes (blood pressure, weight) remain inconsistent [42] [44] [43]. Critical success factors include: message viewing (not just delivery) [44], perceived personalization [4], and balancing assessment burden with contextual richness [4]. Future research should focus on advanced personalization algorithms, integration of passive sensing data, and longer-term trials to establish clinical efficacy. The protocols and frameworks presented provide researchers with methodological foundations for advancing JITAI science in dietary adherence research.

Navigating Real-World Challenges and Enhancing JITAI Effectiveness

Adherence, defined as the extent to which a person's behavior corresponds with agreed recommendations from a healthcare provider, is a pivotal factor in achieving optimal therapeutic outcomes [45]. In chronic disease management, non-adherence remains a pervasive challenge, contributing to suboptimal treatment results, higher complication rates, increased hospitalizations, and substantial healthcare expenditures [45]. Understanding the multidimensional nature of adherence barriers is essential for developing effective interventions, particularly for emerging approaches such as just-in-time adaptive interventions (JITAIs) that aim to provide support in response to fluctuating states and contexts [6] [11].

This application note examines three critical barrier domains relevant to dietary adherence research: social and environmental contexts, user burden and system demands, and alert fatigue. We synthesize evidence from multiple chronic conditions and healthcare settings to provide researchers and drug development professionals with methodological insights for designing, implementing, and evaluating JITAIs targeting dietary behaviors.

Social and Environmental Context Barriers

Social and environmental factors present substantial challenges to adherence by creating conflicts between recommended behaviors and situational demands. These barriers are particularly relevant to dietary adherence, where food choices are often influenced by external circumstances.

Table 1: Social and Environmental Barrier Typology

Barrier Category	Specific Manifestations	Relevant Contexts
Competing Priorities	• Children's needs taking precedence [46] • Family responsibilities requiring deviation from dietary plans [47]	• Home environment • Family gatherings
Social Interactions	• Socializing to avoid loneliness [47] • Stigma or embarrassment about adherence behaviors [48]	• Social events • Public settings • Workplace environments
Environmental Constraints	• Numerous people in one's area making adherence difficult [47] • Limited access to recommended foods [6]	• Restaurants • Grocery stores • Travel situations

Experimental Protocol for Assessing Social Context Barriers

Objective: To identify and characterize social and environmental barriers to dietary adherence in real-world settings using ecological momentary assessment (EMA).

Procedure:

• Participant Recruitment: Recruit adults with chronic conditions requiring dietary management (e.g., hypertension, diabetes). Sample size calculations should account for multiple measures per participant.
• Baseline Assessment: Collect demographic data, medical history, and psychosocial measures including health literacy, social support, and self-efficacy.
• EMA Data Collection: Implement a 14-day EMA protocol with 5-7 random prompts per day assessing:
  • Current location and social context
  • Recent dietary choices
  • Conflict between recommended and actual behaviors
  • Emotional state and decision factors
• Data Analysis: Use multilevel modeling to identify contextual predictors of non-adherence, accounting for within-person and between-person variance.

Materials:

• Mobile devices with custom EMA application
• Backend database for real-time data capture
• Analysis software (R, SPSS, or similar) with multilevel modeling capabilities

User Burden and System Demands

The complexity of treatment regimens and system demands creates substantial barriers to adherence, particularly for vulnerable populations. These factors increase cognitive load, time commitment, and physical effort required for adherence.

Key Dimensions of User Burden

Cognitive Load: Patients with chronic conditions frequently report forgetfulness as a primary barrier, especially for regimens requiring multiple daily doses [49] [48]. This challenge is exacerbated by declining cognitive abilities in vulnerable elderly populations and low health literacy that impedes understanding of treatment rationales [49].

Physical Challenges: Physical difficulties and conditions such as impaired dexterity or visual limitations create substantial barriers, particularly for medication delivery devices like inhalers [49] [48]. Elderly patients often experience reduced general fitness that complicates adherence behaviors [49].

Financial and System Navigation: Financial burden represents one of the most consistently reported barriers across populations [50]. Patients often develop complex coping strategies, including altering medication dosage to extend prescriptions or sourcing medications from other countries to reduce costs [48]. System-level barriers include disorganized hospital discharge processes, inaccessible personal physicians, and communication challenges between providers and patients [49].

Table 2: Experimental Measures for Quantifying User Burden

Burden Dimension	Assessment Method	Metrics
Cognitive Load	• Medication Adherence Questionnaire (MAQ) • Prospective memory tasks • Health Literacy Scale	• Recall accuracy • Adherence self-reports • Comprehension scores
Physical Demand	• Dexterity assessment • Visual acuity testing • Usability testing	• Task completion time • Error rates • Success/failure rates
Financial Burden	• Adherence Barriers Questionnaire (ABQ) • Out-of-pocket cost tracking	• Cost-related non-adherence • Financial toxicity scores
System Navigation	• Healthcare System Distrust Scale • Patient satisfaction surveys • Continuity of care measures	• Appointment adherence • Trust ratings • Care coordination scores

Experimental Protocol for Evaluating Intervention Burden

Objective: To quantify and compare user burden across different dietary adherence intervention approaches.

Procedure:

• Intervention Design: Develop three versions of a dietary tracking intervention with varying burden levels:
  • Low burden: Passive monitoring only
  • Medium burden: Daily check-ins with simplified tracking
  • High burden: Comprehensive food logging with portion estimation
• Participant Assignment: Randomly assign participants to one of the three conditions for a 30-day period.
• Burden Assessment: Collect daily burden measures using:
  • NASA Task Load Index (TLX)
  • System Usability Scale (SUS)
  • Time spent on adherence tasks
• Adherence Monitoring: Track intervention engagement and dietary adherence through:
  • Application usage metrics
  • Dietary recall interviews
  • Biomarker monitoring (where applicable)
• Data Analysis: Examine relationships between burden measures and adherence outcomes using correlation and regression analyses.

Alert Fatigue in Clinical Systems

Alert fatigue represents a critical barrier in digital health interventions, occurring when users become desensitized to notifications due to excessive volume, poor timing, or limited relevance. In clinical decision support systems, an estimated 7,000 passive alerts may be delivered daily to a single critical care practitioner, leading to inappropriate overrides of important warnings [51].

Facilitators and Barriers to Alert Effectiveness

Qualitative research with clinicians has identified key factors influencing alert response [51] [52]:

Facilitators:

• Patient safety relevance: Alerts perceived as critical to preventing harm
• Ease of response: Minimal effort required to address the alert
• Specificity: Well-defined, actionable recommendations
• Prioritization: Clear indication of urgency
• Feedback: Understanding the rationale for the alert

Barriers:

• Excessive quantity: Too many alerts, particularly low-value notifications
• Work environment: High cognitive load and time pressure
• Difficulty in response: Complex or time-consuming resolution requirements
• Irrelevance: Alerts not applicable to the specific patient context

Experimental Protocol for Mitigating Alert Fatigue in JITAIs

Objective: To develop and test alert personalization strategies that reduce fatigue while maintaining effectiveness in dietary JITAIs.

Procedure:

• Alert System Development: Create a JITAI for dietary sodium reduction with multiple alert types:
  • Context-aware restaurant recommendations
  • Grocery store selection prompts
  • Meal planning reminders
• Personalization Algorithm: Implement a reinforcement learning system that adapts alert timing, frequency, and content based on:
  • Historical user response patterns
  • Current context (location, time, social setting)
  • User-defined preferences
  • Recent adherence behavior
• Randomized Comparison: Assign participants to one of three conditions:
  • Standard JITAI (fixed alert schedule)
  • Personalized JITAI (adapts based on engagement)
  • Control (self-monitoring only)
• Outcome Measures:
  • Alert responsiveness rates
  • User satisfaction (System Usability Scale)
  • Perceived burden (NASA-TLX)
  • Dietary behavior change (sodium intake)
  • Systolic blood pressure change
• Analysis Plan: Compare conditions on primary outcomes using ANOVA or mixed effects models, with mediation analyses to examine whether personalization effects operate through reduced fatigue.

The Scientist's Toolkit

Table 3: Essential Research Reagents and Materials for Adherence Research

Item	Function/Application	Example Implementation
Adherence Barriers Questionnaire (ABQ)	Multidimensional assessment of adherence barriers	Identifies financial, therapy-related, and health system barriers [50]
Patient Health Questionnaire (PHQ-9)	Depression screening	Assesses psychological barriers to adherence [50]
System Usability Scale (SUS)	Standardized usability assessment	Quantifies perceived burden of digital interventions [6]
NASA Task Load Index (TLX)	Multidimensional workload assessment	Measures cognitive, temporal, and frustration dimensions of burden [51]
Ecological Momentary Assessment (EMA)	Real-time data collection in natural environments	Captures context-behavior interactions [6]
Just-in-Time Adaptive Intervention Platform	Mobile delivery of context-sensitive support	Enables personalized timing of adherence supports [11]

Conceptual Framework and Visualizations

Addressing adherence barriers requires a multidimensional approach that recognizes the complex interplay between social contexts, user burden, and intervention design factors. Social and environmental barriers create contextual conflicts that pull individuals away from recommended behaviors, while user burden creates internal resistance through cognitive, physical, and financial demands. Alert fatigue represents a critical design failure that undermines even well-conceived interventions.

For JITAIs targeting dietary adherence, our analysis suggests several strategic priorities: (1) implementing context-aware personalization that adapts to social and environmental constraints, (2) minimizing user burden through intelligent defaults and passive monitoring where possible, and (3) employing reinforcement learning to optimize alert timing and content based on individual response patterns. The experimental protocols outlined provide a methodological foundation for advancing this research agenda.

Future work should focus on developing more sophisticated context sensing algorithms, establishing thresholds for alert fatigue across different populations, and examining cross-domain interactions between different barrier types. By systematically addressing these common adherence barriers, researchers can enhance the effectiveness of JITAIs and ultimately improve dietary adherence in real-world settings.

Balancing Personalization with Simplication in Intervention Design

Application Notes: Core Principles for JITAI Design in Dietary Adherence

Theoretical Foundation of Adaptive Interventions

Adaptive interventions (AIs) provide a systematic framework for operationalizing personalized treatment sequences through a series of decision rules that specify whether, how, when, and based on which measures to alter treatment type, dosage, or delivery. These interventions are characterized by four core elements: decision stages, treatment options, tailoring variables, and decision rules [53]. Within this broader category, Just-in-Time Adaptive Interventions (JITAIs) represent a specialized form that leverages mobile technology to deliver dynamically tailored support in real-time based on an individual's changing internal states and external context [5]. The fundamental structure of a JITAI is built upon six core components as defined by Nahum-Shani et al.: distal outcomes, proximal outcomes, tailoring variables, decision points, decision rules, and intervention options [4] [5].

The development of high-quality adaptive interventions requires addressing critical sequencing questions that cannot be adequately answered through traditional randomized controlled trial designs. The Sequential Multiple Assignment Randomized Trial (SMART) has been developed explicitly to build optimal adaptive interventions by experimentally evaluating the timing, sequencing, and individualization of treatment decisions [53]. This design is particularly valuable for dietary adherence research, where individual heterogeneity in response to interventions is substantial and the optimal intervention strategy may vary based on both baseline characteristics and early response indicators.

Personalization-Simplicity Tradeoffs in Practice

A central challenge in JITAI implementation involves balancing the depth of personalization against practical constraints of usability and scalability. Research on JITAIs for type 2 diabetes management has demonstrated that while participants value highly tailored support, excessive complexity in assessment or delivery can reduce engagement and adherence [4]. Users report that ecological momentary assessment (EMA) driven interventions provide motivating and contextually relevant support, but also express the need for the right balance between assessment burden and personalization precision [4].

The personalization-localization framework emerging from digital health intervention research highlights that effective tailoring operates at two levels: individual-specific factors (biological, behavioral, and psychological) and environmental-contextual factors (social, physical, and cultural) [54]. This dual approach ensures that interventions are not only biologically appropriate but also practically feasible within an individual's daily life and local environment. For dietary adherence specifically, this might involve adapting recommendations not only to genetic predispositions or metabolic responses but also to local food availability, cultural preferences, and socioeconomic constraints.

Table 1: Key Considerations for Balancing Personalization and Simplicity in Dietary JITAIs

Design Aspect	Personalization Emphasis	Simplicity Emphasis	Balanced Approach
Data Collection	Multi-modal data (genetic, metabolic, behavioral, contextual) [55] [56]	Minimal user burden, passive sensing where possible [4]	Strategic combination of passive sensing with low-burden active assessment at critical decision points [4] [5]
Intervention Delivery	Fully adaptive content, timing, and intensity based on real-time analytics [5]	Standardized, fixed intervention schedule	Adaptive intervention type with standardized timing or vice versa [4]
Decision Rules	Complex machine learning algorithms incorporating multiple tailoring variables [56] [5]	Simple if-then rules based on single indicators	Staged decision trees with limited branching factors [53]
User Interface	Highly customizable display and notification preferences [54]	One-size-fits-all interface	Modular interface with preset customization options [4]

Experimental Protocols

Protocol 1: Developing an EMA-Driven JITAI for Dietary Adherence

Background and Objectives This protocol outlines the methodology for developing and testing an ecological momentary assessment (EMA)-driven JITAI to support dietary adherence in individuals with type 2 diabetes, based on established frameworks for JITAI development [4] [5]. The primary objective is to create a tailored intervention that dynamically adapts to individuals' daily experiences, contexts, and barriers to healthy eating while maintaining sufficient simplicity for sustainable engagement.

Materials and Equipment

• Mobile application platform with EMA capability
• Backend system for real-time data processing and intervention delivery
• SMS integration for message delivery (as an alternative to push notifications)
• Dietary adherence assessment tools (e.g., brief food frequency questionnaires, photo-based food records)
• Data management system for storing and processing individual profiles

Procedure

• Define Core JITAI Components [5]:
  • Distal Outcome: Improved adherence to national dietary guidelines [4]
  • Proximal Outcome: Daily achievement of personalized dietary goals [4]
  • Tailoring Variables: Activity level, location, mood, overall condition, weather, cravings [4]
  • Decision Points: Mealtimes and predefined vulnerable periods for dietary lapses
  • Intervention Options: Goal setting, reminder messages, alternative food suggestions, motivational messages, stress management techniques
  • Decision Rules: If-then algorithms linking tailoring variables to intervention options

• System Development:

  • Implement daily EMA surveys assessing tailoring variables
  • Develop algorithm processing EMA data to select appropriate interventions
  • Create SMS-based intervention delivery system
  • Build provider dashboard for monitoring participant engagement

• Testing and Refinement:

  • Conduct acceptability testing with 8-15 target users for 2 weeks [4]
  • Use semi-structured interviews to assess perceived personalization, burden, and usefulness
  • Iteratively refine system based on user feedback

• Evaluation:

  • Measure dietary adherence through self-report and biometric indicators
  • Assess engagement with EMA and intervention acceptance rates
  • Evaluate perceived personalization and simplicity through validated scales

Analysis and Interpretation Qualitative data from interviews should be analyzed using thematic analysis to identify key themes related to the personalization-simplicity balance. Quantitative metrics should include EMA completion rates, intervention acceptance rates, and pre-post changes in dietary adherence measures. Particular attention should be paid to individual differences in preferences for personalization level and tolerance for assessment burden.

Diagram 1: JITAI Development Workflow

Protocol 2: AI-Driven Personalized Nutrition Intervention

Background and Objectives This protocol describes the implementation of an artificial intelligence (AI)-driven personalized nutrition intervention that integrates genetic, metabolic, and behavioral data to provide tailored dietary recommendations. The objective is to leverage machine learning approaches to enhance the precision of dietary advice while maintaining interpretability and practical applicability for end users [56].

Materials and Equipment

• Genetic profiling capability (e.g., DNA microarray for nutrition-relevant SNPs)
• Continuous glucose monitors (CGMs) or other metabolic tracking devices
• Mobile application for food tracking and recommendation delivery
• Machine learning infrastructure for data integration and model training
• Secure data storage compliant with relevant privacy regulations

Procedure

• Participant Profiling:
  • Collect genetic data focusing on variants relevant to nutrient metabolism (e.g., MTHFR for folate, BCMO1 for beta-carotene, APOE for lipid response) [55]
  • Establish baseline metabolic parameters through CGMs and standard blood tests
  • Assess dietary patterns, preferences, and barriers through validated questionnaires
  • Document relevant medical history, medications, and supplements

• AI Model Development:

  • Implement supervised learning models (e.g., multilayer perceptrons, LSTM networks) to predict postprandial glycemic responses to different foods [56]
  • Apply unsupervised methods (e.g., k-means clustering, PCA) for phenotype-driven stratification [56]
  • Incorporate reinforcement learning algorithms for continuous personalization based on user feedback and outcomes [56]
  • Integrate explainable AI techniques to maintain interpretability of recommendations

• Intervention Delivery:

  • Provide initial dietary recommendations based on integrated genetic and metabolic profiling
  • Offer real-time feedback through mobile platform based on logged food intake and metabolic response
  • Adapt recommendations dynamically based on ongoing data collection and user feedback
  • Include educational components explaining the rationale behind recommendations

• Evaluation:

  • Assess short-term outcomes (glycemic variability, dietary adherence)
  • Monitor long-term outcomes (HbA1c, weight, cardiovascular risk factors)
  • Evaluate user engagement with the platform and satisfaction with recommendations
  • Assess understanding and perceived usefulness of the personalized advice

Analysis and Interpretation Primary analysis should focus on the efficacy of AI-driven personalization compared to standard dietary advice. Secondary analyses should explore whether certain genetic or metabolic profiles respond more favorably to the personalized approach. Implementation outcomes should include adherence rates, user satisfaction, and qualitative feedback on the balance between personalization and complexity.

Table 2: Key Genetic Variants for Personalized Nutrition Interventions [55]

Gene	Function	Nutritional Influence	Personalization Approach
MTHFR	Methylenetetrahydrofolate reductase	Folate metabolism	Individuals with TT genotype may require higher folate intake than RDA [55]
APOA1	Apolipoprotein A1	HDL cholesterol metabolism	A allele carriers show increased HDL with omega-3 PUFA supplementation [55]
BCMO1	Beta-carotene oxygenase 1	Beta-carotene to vitamin A conversion	Genetic variations affect conversion efficiency and plasma carotenoid levels [55]
FTO	Fat mass and obesity-associated protein	Energy balance and obesity risk	Specific genotypes may respond differently to dietary fat composition [55]
SGK1	Serum- and glucocorticoid-inducible kinase 1	Sodium retention	Risk allele carriers may show increased blood pressure on high-salt diets [55]

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Research Reagents and Platforms for JITAI Development

Tool Category	Specific Examples	Function in Research	Implementation Considerations
EMA Platforms	Mobile apps with customizable survey scheduling, SMS-based assessment systems [4]	Capture dynamic changes in internal states and context for tailoring	Balance between assessment frequency and participant burden; typically 3-5 prompts daily [4]
Passive Sensing Technologies	Activity trackers, GPS, smartphone sensors (accelerometer, microphone) [5]	Objective measurement of behavior and context without user burden	Battery life concerns, data processing complexity, privacy implications [4]
AI/ML Frameworks	Random forests, LSTM networks, reinforcement learning algorithms [56]	Develop predictive models of individual responses and optimize decision rules	Trade-off between model complexity and interpretability; need for explainable AI in clinical applications [56]
Biometric Sensors	Continuous glucose monitors, sleep trackers, heart rate variability monitors [56]	Objective physiological data for personalization and outcome assessment	Cost, user comfort, data integration challenges, clinical validation requirements [56]
Intervention Delivery Platforms	Mobile messaging systems, adaptive content delivery frameworks, push notification engines [4] [5]	Deliver tailored interventions at optimal moments	Timing precision, message personalization capabilities, cross-platform compatibility [4]
Data Integration Systems	FHIR-based health data platforms, custom API architectures, secure cloud storage	Combine diverse data sources for comprehensive personalization	Data privacy compliance (HIPAA, GDPR), interoperability standards, real-time processing needs [56]

Diagram 2: JITAI Personalization Framework

This application note details advanced methodologies for enhancing dietary adherence through Just-in-Time Adaptive Interventions (JITAIs). It provides a structured framework for integrating the Fixed-Quality Variable-Type (FQVT) dietary paradigm with Reinforcement Learning (RL) models. The document includes a conceptual overview, quantitative performance comparisons, detailed experimental protocols, and a catalog of essential research tools. This guide is intended to assist researchers and drug development professionals in building robust, personalized nutrition intervention systems.

Dietary adherence remains a significant challenge in clinical nutrition research and chronic disease management. Traditional one-size-fits-all dietary interventions often fail to account for individual preferences, cultural backgrounds, and dynamic behavioral states, leading to suboptimal adherence and variable health outcomes [57] [58]. To address these limitations, Just-in-Time Adaptive Interventions (JITAIs) have emerged as a promising framework. JITAIs leverage mobile technology to provide timely, context-aware, and personalized support to individuals at moments of heightened vulnerability or opportunity [6] [2].

This document explores the synergy of two advanced concepts to optimize JITAIs:

• The Fixed-Quality Variable-Type (FQVT) Dietary Intervention: A novel approach that standardizes overall diet quality while allowing flexibility in diet type to accommodate personal and cultural preferences [57].
• Reinforcement Learning (RL): A branch of machine learning ideal for making sequential, personalized decisions that adapt to an individual's changing state and context [59] [60] [61].

By combining the structural rigor of FQVT with the adaptive intelligence of RL, researchers can develop next-generation dietary interventions that are both scientifically valid and highly adherent.

Conceptual Framework Integration

The integration of FQVT and RL within a JITAI creates a powerful, closed-loop system for dietary management. The FQVT paradigm establishes the nutritional "rules," ensuring that all dietary recommendations meet a fixed, high-quality standard as measured by tools like the Healthy Eating Index (HEI) [57]. Simultaneously, the RL agent operates within these rules to make dynamic, micro-level decisions about what, when, and how to recommend specific foods or actions to the user. The following diagram illustrates the logical workflow and relationship between these components.

Performance Data and Comparative Analysis

Evaluations of standalone and integrated systems demonstrate the potential of these advanced methods. The table below summarizes key quantitative findings from recent studies.

Table 1: Performance Metrics of RL and FQVT in Dietary Interventions

Model / Approach	Key Performance Metrics	Comparative Outcome	Source
CFRL (RL + Collaborative Filtering)	User satisfaction, Nutritional adequacy, User acceptance score	Superior performance vs. four baseline methods; enhances user acceptance and adherence.	[59]
DeepEN (Deep RL for Enteral Nutrition)	Estimated mortality, Nutritional biomarkers	3.7 ± 0.17 percentage-point reduction in mortality (18.8% vs. 22.5%) vs. clinician-led care.	[60]
DDPG-based Food Recommendation	Compliance with healthy guidelines, High-calorie food intake	Improved compliance by 19.8%; reduced high-calorie intake by 16.3%.	[61]
FQVT Dietary Intervention	Participant adherence, Satisfaction, External validity	Potential to enhance adherence, satisfaction, and generalizability of study findings.	[57]
myBPmyLife JITAI (for Hypertension)	Step count, Sodium intake, Systolic Blood Pressure (SBP)	Increased steps by 489/day (p=0.04); reduced sodium by 285 mg/day (p=0.002); no significant SBP change.	[6]

Detailed Experimental Protocols

Protocol 1: Implementing an FQVT Dietary Framework

This protocol establishes the foundational dietary structure for an intervention [57].

1. Define and Standardize Diet Quality:

• Tool Selection: Adopt a validated diet quality index as the fixed standard. The Healthy Eating Index (HEI)-2020 is a recommended choice, as it aligns with the Dietary Guidelines for Americans and has established robustness [57].
• Quality Threshold: Set a minimum HEI score that all dietary patterns in the study must achieve (e.g., a score of 80 out of 100 to signify a high-quality diet).

2. Develop Variable Diet Types:

• Pattern Identification: Identify distinct dietary patterns that are culturally relevant and appealing to the target population (e.g., Mediterranean, vegetarian, low-carbohydrate, or Asian dietary patterns).
• Menu Standardization: Create standardized menus for each diet type, ensuring they are isocaloric and all meet the pre-defined HEI score threshold. This ensures that the only major variable is the type of diet, not its overall nutritional quality.

3. Assign and Monitor:

• Participant Assignment: Randomly assign participants to one of the variable diet types that aligns with their stated preferences.
• Quality Assurance: Use digital dietary assessment tools (e.g., photo-based apps, 24-hour recalls) to periodically verify that participants' actual food intake maintains the target HEI score [57].

Protocol 2: Developing a Reinforcement Learning Agent for JITAIs

This protocol outlines the process of building an RL model to deliver personalized support within an FQVT-structured intervention, drawing from established RL applications in nutrition [59] [60] [2].

1. Formulate the Markov Decision Process (MDP):

• State (s): Define the state space to include user-specific variables. These can be static (e.g., age, sex, BMI, food preference profile) and dynamic (e.g., current hunger, location, time of day, recent physical activity, past dietary adherence, predicted lapse risk) [60] [2].
• Action (a): Define the action space as the set of interventions the JITAI can deliver. Examples include: sending a motivational message, suggesting a specific FQVT-compliant meal, providing a recipe, or delivering a cognitive behavioral tip to counter lapse risk [2].
• Reward (r): Design a reward function that balances short-term and long-term goals. The function can include: R = w1 * (User Engagement) + w2 * (Short-term Health Biomarker Improvement) + w3 * (Long-term Adherence) + w4 * (Mortality Reduction) [60] Here, w1-w4 are weights assigned to each component based on clinical importance.

2. Select and Train the RL Algorithm:

• Algorithm Choice: For discrete action spaces (e.g., sending one of five message types), Dueling Double Deep Q-Networks (D3QN) with Conservative Q-Learning (CQL) regularization is recommended for offline training, as it mitigates overestimation bias and enhances safety [60]. For continuous action spaces (e.g., fine-tuning calorie targets), the Deep Deterministic Policy Gradient (DDPG) algorithm is suitable [61].
• Offline Training: Train the model on a large-scale retrospective dataset (e.g., from previous intervention studies or electronic health records) using offline RL techniques to learn an initial policy without interacting with real users [60].

3. Deploy in a Microrandomized Trial (MRT):

• Trial Design: Before a full-scale efficacy trial, deploy the RL agent within an MRT framework [2].
• Proximal Outcomes: At each decision point (e.g., when the system predicts high lapse risk), randomize the delivery of actions and measure proximal outcomes (e.g., dietary lapse occurrence in the following 2.5 hours). This provides high-quality data to further refine the RL policy and understand contextual moderators [2].

The following diagram illustrates the architecture of a specific RL model, CFRL, which combines collaborative filtering with reinforcement learning for meal recommendation.

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Tools and Resources for JITAI Development

Tool / Resource	Function in Research	Application Example
Healthy Eating Index (HEI)	A validated metric for assessing diet quality based on adherence to dietary guidelines.	Serves as the objective, fixed-quality standard in the FQVT paradigm [57].
Conservative Q-Learning (CQL)	An offline reinforcement learning algorithm.	Prevents overestimation of Q-values for unseen actions, ensuring safer policy learning from historical datasets in healthcare [60].
Microrandomized Trial (MRT) Design	An experimental design for optimizing JITAIs.	Used to test the proximal effect of intervention components at numerous decision points over time [2].
Food Preference Questionnaire (FPQ)	A tool to classify an individual's food liking profile.	Input for personalizing dietary recommendations within an FQVT framework to improve adherence [62].
Dueling Double Deep Q-Network (D3QN)	A deep reinforcement learning algorithm.	Used to estimate the value of actions in a complex state space while reducing overestimation bias [60].
Behavior Change Wheel (BCW) / COM-B Model	A framework for systematically designing behavior change interventions.	Guides the selection of intervention functions and behavior change techniques for the JITAI's content [62].

Strategies for Improving User Engagement and Long-Term Retention

Just-in-Time Adaptive Interventions (JITAIs) represent a transformative approach in digital health, designed to provide tailored support to individuals at moments of heightened vulnerability or opportunity [1]. For researchers and drug development professionals focusing on dietary adherence, the challenge is not only to initiate healthy behaviors but to ensure that patients remain engaged with these interventions over the long term. This document synthesizes current evidence and protocols to outline effective strategies for boosting user engagement and retention within JITAI frameworks for dietary research, providing a scientific toolkit for implementation and evaluation.

Quantitative Evidence from Recent Trials

Recent large-scale trials provide critical quantitative data on engagement patterns and behavioral outcomes associated with JITAIs. The table below summarizes key findings from a randomized controlled trial of the myBPmyLife JITAI application, which targeted physical activity and lower-sodium food choices in patients with hypertension [6].

Table 1: Six-Month Outcomes from the myBPmyLife JITAI Trial

Metric	Intervention Group	Control Group	P-value	Effect Size
Change in Systolic BP (mmHg)	-5.2 (SD 15)	-5.7 (SD 15)	0.76	0.029
Change in Daily Step Count	+170 (SD 2690)	-319 (SD 2612)	0.040	0.184
Change in Sodium Intake (mg)	-1145 (SD 1023)	-860 (SD 1001)	0.002	0.282
Application Engagement	Median of 81 days (IQR 44, 126)45.76% of eligible days	Not Applicable	Not Applicable	Not Applicable
System Usability Scale (SUS) Score	73.6 (SD 19) - "Good"	Not Applicable	Not Applicable	Not Applicable

Despite significant improvements in the secondary behavioral outcomes of step count and sodium reduction, the lack of a corresponding significant effect on the primary clinical endpoint (systolic blood pressure) highlights a critical challenge in JITAI research: translating engagement and intermediate behavioral changes into long-term clinical benefits [6].

Core JITAI Components and Engagement Protocol

The architecture of an effective JITAI for dietary adherence is built upon six fundamental components, as defined by Nahum-Shani et al. [1]. The following protocol provides a methodology for implementing these components with a focus on engagement and retention.

Table 2: Core JITAI Components and Implementation Protocol

JITAI Component	Definition	Implementation Protocol for Dietary Adherence
Distal Outcome	The ultimate long-term goal of the intervention [1].	Improvement in adherence to national dietary guidelines (e.g., reduced sodium intake, increased fruit/vegetable consumption). Assess via 24-hour dietary recalls or validated food frequency questionnaires at baseline, 3, 6, and 12 months [4].
Proximal Outcome	Short-term goals indicating progress toward the distal outcome [1].	Daily goal achievement (e.g., "Did the participant stay under their daily sodium target?"). Measure via daily Ecological Momentary Assessment (EMA) or sensor-based data (e.g., passive sodium sensor if available) [4].
Decision Points	Points in time when an intervention decision is made [1].	Protocol: Program 3-5 decision points per day, triggered by a combination of fixed schedules (e.g., meal times) and dynamic contexts (e.g., entering a grocery store via GPS geofencing) [4].
Tailoring Variables	Dynamic information used to personalize the intervention [1].	Active Data: Collect via low-burden EMA prompts about current mood, cravings, location, and social context [4].Passive Data: Leverage GPS, smartphone usage, and wearable device data (e.g., step count, sleep).
Intervention Options	The array of possible support actions [1].	- Passive Content: Push educational messages about low-sodium alternatives.- Active Support: Prompt user to log their next meal or commit to a healthy choice.- Motivational Feedback: Provide positive reinforcement for achieving daily goals.
Decision Rules	The algorithm specifying which intervention to offer, and when [1].	Example Rule: IF (User is near a grocery store) AND (EMA reports high craving for salty snacks) THEN (Push notification with a healthy, low-sodium alternative and a nearby store location).

Visualizing the JITAI Engagement Workflow

The logical flow of data and intervention in a dietary JITAI, as outlined in the protocol above, can be visualized as a continuous feedback cycle.

The Scientist's Toolkit: Research Reagent Solutions

Implementing and evaluating a JITAI requires a suite of methodological "reagents." The following table details essential tools and their functions for research in this domain.

Table 3: Essential Research Reagents for JITAI Development and Evaluation

Research Reagent	Function/Definition	Application in Dietary JITAI Research
Ecological Momentary Assessment (EMA)	A low-effort self-report method involving repeated sampling of behaviors and states in a user's natural environment [4].	Captures dynamic tailoring variables such as mood, cravings, and context that sensors cannot measure, enabling deeper personalization [4].
System Usability Scale (SUS)	A ten-item, Likert-scale questionnaire for measuring the subjective usability of a system [63].	Provides a standardized metric (score 0-100) to quantitatively assess user satisfaction and interface problems with the JITAI application [6] [63].
User-Centered Design (UCD)	An iterative design process that involves users throughout the design and development cycle to ensure the product meets their needs [63].	Critical for pre-trial development to identify user needs, preferences, and potential barriers to engagement, thereby improving initial adoption and long-term retention.
Randomized Controlled Trial (RCT)	The gold-standard study design for evaluating the efficacy of an intervention by randomly assigning participants to treatment or control groups [6].	Measures the causal effect of the JITAI on both primary clinical outcomes (e.g., blood pressure) and secondary engagement metrics (e.g., step count, sodium intake) [6].
iDISK Knowledge Base	An integrated Dietary Supplement Knowledge base with a standardized ontological structure, curating evidence-based information [63].	Can serve as a backend knowledge graph to inform and validate dietary advice and supplement-related information provided within the JITAI, ensuring scientific accuracy.

Visualization of the Retention Challenge and Strategy

Long-term retention is a primary challenge in JITAI research. The following diagram conceptualizes the key factors influencing engagement and the strategies to address them, based on user feedback from qualitative studies [4].

Enhancing user engagement and long-term retention in JITAIs for dietary adherence requires a meticulously designed, multi-faceted approach. As evidenced by recent trials and qualitative studies, success hinges on the seamless integration of dynamic personalization, low-burden data collection, and contextually relevant support. The protocols, visualizations, and research reagents detailed in this document provide a scientific foundation for researchers and drug development professionals to design, implement, and rigorously evaluate JITAI systems that are not only effective in changing short-term behaviors but are also engaging enough to be sustained by users over time, ultimately leading to meaningful and lasting health outcomes.

Integrating Passive Sensing and Active EMA to Reduce Participant Burden

Ecological Momentary Assessment (EMA) is a powerful tool for capturing real-time, in-the-moment data on behaviors, cognitions, and emotions in naturalistic settings, thereby minimizing recall bias and increasing ecological validity [20] [64]. However, a significant challenge in EMA research is participant burden, which can lead to reduced compliance, missing data, and potential biases that threaten the validity of study findings [65] [66]. This burden stems from repeated interruptions requiring active self-report, which can be intrusive and demanding for participants [64].

Just-in-Time Adaptive Interventions (JITAIs) represent a promising framework for managing this burden. JITAIs aim to provide the right type and amount of support at the right time by adapting to the individual's changing state and context [11] [3]. A key strategy for reducing burden within JITAIs is the strategic integration of passive sensing with active EMA. Passive sensing uses smartphone and wearable sensors to continuously collect objective behavioral and contextual data without requiring active participant input [65] [67]. This integration allows researchers to leverage the rich, objective data from passive sensing to minimize the frequency of active EMA prompts or to trigger them at more contextually appropriate moments, thereby creating a more efficient and less burdensome data collection ecosystem [65] [3]. This Application Note details protocols for achieving this integration, with a specific focus on dietary adherence research.

Key Concepts and Evidence Base

The rationale for combining passive and active sensing is supported by empirical evidence demonstrating its benefits for prediction accuracy and burden reduction.

Comparative Performance of Sensing Modalities

Research by the SenseWhy study, which monitored individuals with obesity in free-living settings, provides compelling quantitative evidence for the superior predictive power of a combined sensing approach. The study compared machine learning models (XGBoost) built on different data streams for predicting overeating episodes, with results summarized in the table below [68].

Table 1: Predictive Performance for Overeating Episodes (SenseWhy Study)

Data Modality	AUROC (Mean)	AUPRC (Mean)	Key Predictive Features
EMA-Only	0.83	0.81	Perceived overeating, pre-meal biological hunger, evening eating
Passive Sensing-Only	0.69	0.69	Number of chews and bites, chew interval
Combined (Feature-Complete)	0.86	0.84	Perceived overeating, number of chews, loss of control

The findings clearly show that while EMA data is a strong predictor on its own, the highest performance is achieved by integrating both data streams [68]. This synergy allows for the detection of complex behavioral phenotypes, such as "Stress-driven Evening Nibbling" or "Uncontrolled Pleasure Eating," which are critical for personalizing interventions [68].

The Burden of EMA and Mitigation Strategies

Participant compliance is a critical indicator of burden. A systematic review of EMA in movement behavior research found that compliance is influenced by multiple design factors [66]. The table below synthesizes key findings on how protocol choices can impact participant response rates.

Table 2: EMA Design Factors and Their Impact on Participant Burden and Compliance

Design Factor	Impact on Burden & Compliance	Evidence from Literature
Prompt Frequency	Higher frequency increases burden and can lower compliance. Studies with one prompt/day achieved 91% compliance vs. 77% with more frequent prompts [66].	[66]
Number of Questions	A higher number of questions per prompt is negatively correlated with response rate (r = -0.433, p < .001) [64].	[64]
Timing of Prompts	Responsiveness varies by context. Participants are most responsive in the evening and on weekdays, but this is influenced by demographics [64].	[64]
Study Duration	Study length itself is not strongly correlated with compliance, but response quality (e.g., carefulness, variance) can decline over time [66] [64].	[66] [64]
Platform	Smartphone-based apps can lessen burden compared to web-based formats or paper diaries [66].	[66]

These findings highlight that burden is not inevitable but can be managed through careful study design. Passive sensing can directly address several of these factors; for instance, by using sensor data to identify optimal moments for prompting (e.g., during behavioral transitions or when the user is at home) to improve response rates [65] [64].

Integrated Protocol for Dietary Adherence Research

This section provides a detailed, actionable protocol for implementing an integrated passive and active sensing system in a JITAI aimed at improving dietary adherence, based on established frameworks [3] and recent research [68] [67].

Conceptual Workflow

The following diagram illustrates the logical workflow and decision points for a JITAI that uses passive sensing to minimize active EMA burden.

Detailed Methodology and Reagent Toolkit

Implementation of the above workflow requires a suite of research tools and technologies. The following table catalogs the essential "research reagents" and their functions.

Table 3: Research Reagent Solutions for Integrated Sensing JITAIs

Item Category	Specific Examples	Function & Rationale
Passive Sensing Hardware	Wrist-worn accelerometer (e.g., research-grade actigraph), smartphone sensors (GPS, microphone), wearable camera (e.g., SenseWhy camera) [68] [67].	Continuously captures objective data on activity, location, audio environment, and eating behaviors (bites, chews) without participant effort. Provides input for behavioral inference.
Active EMA Platform	Smartphone application with push notification capabilities for EMA surveys [66] [3].	Delivers active self-report surveys. Smartphone-based apps are less burdensome and facilitate real-time data capture and intervention delivery.
Behavioral Inference Algorithms	Validated algorithms for eating episode detection (from wrist motion), bite count, chew count, and eating rate [68] [67]. Machine learning models (e.g., XGBoost) for lapse risk prediction [68] [3].	Translates raw passive data into meaningful behavioral constructs. Critical for determining when to trigger an EMA or intervention, thereby reducing unnecessary prompts.
JITAI Decision Engine	Microrandomized trial (MRT) software architecture [3]. Reinforcement learning models for personalization [11].	The core "brain" of the intervention. Uses inference engine output and/or EMA responses to make real-time decisions about whether and what type of intervention to deliver.

Step-by-Step Experimental Protocol

Protocol Title: Integrated Passive and Active Sensing for Dietary Lapse Prevention in Obesity Treatment.

Primary Objective: To optimize a JITAI that reduces participant burden by using passive sensing to minimize the frequency of active EMA prompts while effectively preventing dietary lapses.

Study Population: Adults (age 18-65) with overweight or obesity participating in a behavioral obesity treatment program [3].

Duration: 6 months [3].

Procedures:

• Baseline Assessment & Device Provisioning:

  • Collect demographic and clinical data.
  • Provide participants with a wrist-worn sensor (e.g., ActiGraph) and install the study smartphone app configured for both passive sensing and EMA [67] [3].
  • Train participants on device use, EMA completion, and the concept of dietary lapses.

• Continuous Passive Data Collection:

  • The wrist sensor continuously collects tri-axial accelerometry data.
  • The smartphone passively collects GPS data to infer location (home, work, restaurant) [68] [3].
  • Inference: Pre-validated algorithms process the wrist motion data in near-real-time to detect eating episodes and extract features like bite count, eating duration, and eating rate [67].

• Risk Triggered & Scheduled EMA:

  • Fixed Schedule: Participants receive a maximum of 4-6 EMA prompts per day at semi-random times within broad windows (e.g., 11:00, 13:30, 16:00, 18:30) to assess baseline states and lapse triggers [3]. Adherence to this schedule is the worst-case burden scenario.
  • Context-Triggered Suppression: The JITAI engine uses passive data to suppress scheduled prompts in high-burden contexts. For example, if the participant is driving (inferred from GPS speed and accelerometry) or engaged in high-intensity physical activity, the prompt is withheld and may be re-issued at a later time [64].
  • Event-Contingent Triggering: A scheduled prompt is replaced if the passive inference engine detects the onset of an eating episode with characteristics associated with lapse (e.g., a high eating rate in the evening) [67]. This shifts assessment to the most critical moments.

• Intervention Delivery (Microrandomized Trial):

  • At each decision point (after an EMA is completed or a high-risk passive context is detected), the participant is randomized to receive either no intervention, a generic alert, or a theory-driven intervention (e.g., building self-efficacy, fostering motivation) [3].
  • The proximal outcome (e.g., lapse occurrence in the next 2.5 hours) is measured via a subsequent, brief EMA or passive sensing of meal characteristics [67] [3].

• Data Integration & Analysis:

  • Time-sync passive sensor data, EMA responses, and intervention deliveries.
  • Use machine learning models (e.g., XGBoost) to identify the most parsimonious set of passive and active features that predict lapse risk with high accuracy (AUROC > 0.80) [68].
  • Analyze compliance (response rate) and compare it to studies using fixed-interval EMA to quantify burden reduction [66] [64].

The integration of passive sensing and active EMA represents a paradigm shift in dietary adherence research, moving away from fixed, high-frequency prompting towards a dynamic, context-aware, and participant-friendly model. By leveraging commercially available sensors and robust machine learning algorithms, researchers can construct JITAIs that are not only more effective through personalization but also more respectful of participants' time and cognitive resources. The protocols outlined here provide a concrete roadmap for achieving this integration, with the ultimate goal of developing more engaging, sustainable, and efficacious digital health interventions for obesity and related chronic conditions.

Evaluating Efficacy and Comparing JITAI Approaches in Clinical Research

In the development of just-in-time adaptive interventions (JITAIs) for dietary adherence, a critical challenge lies in selecting and analyzing the most appropriate trial endpoints. JITAIs are mobile health (mHealth) interventions designed to provide the right support at the right time by adapting to an individual's changing internal and contextual state [4]. The core tension in evaluating their efficacy often centers on the relationship between behavioral change endpoints (e.g., reduced sodium intake, increased step count) and clinical endpoints (e.g., reduction in systolic blood pressure) [6]. While behavioral changes are often the direct target of JITAIs, clinical outcomes represent the ultimate therapeutic goals. This document provides a structured framework for researchers to analyze these endpoints within dietary adherence trials, ensuring that trial designs are both scientifically rigorous and aligned with the dynamic nature of JITAIs.

Theoretical Foundations: The Behavior-Clinical Outcome Pathway

Understanding the pathway from intervention to behavior change to clinical outcome is fundamental. The intention-action gap, a well-documented phenomenon where conscious intentions do not reliably translate into behavior, is a primary target for JITAIs [69]. Theoretical models like the COM-B framework posit that successful behavior change requires a synergistic combination of Capability, Opportunity, and Motivation [69]. JITAIs are uniquely positioned to address these factors dynamically by leveraging real-time data from wearables, smartphone sensors, and Ecological Momentary Assessments (EMAs) to deliver contextually relevant support [4].

However, effecting behavioral change does not guarantee a corresponding clinical improvement. A recent randomized controlled trial of the myBPmyLife JITAI for hypertension demonstrated this disconnect: while the intervention significantly increased daily step counts and reduced sodium intake, it did not produce a greater reduction in systolic blood pressure compared to the control group over six months [6]. This underscores the necessity of measuring both behavioral and clinical endpoints to fully understand an intervention's mechanism of action and overall effectiveness.

The following diagram illustrates the theoretical pathway from JITAI delivery to clinical outcomes, highlighting key moderating factors and feedback loops.

Diagram 1: JITAI Impact Pathway. This diagram visualizes the theorized causal pathway through which a Just-in-Time Adaptive Intervention (JITAI) influences clinical endpoints, primarily through the mediation of behavioral change. Moderating factors can influence the relationship between behavioral determinants and the resulting behavioral endpoint. A feedback loop allows clinical progress to inform future JITAI personalization.

Quantitative Outcomes: Comparing Behavioral and Clinical Endpoints

Data from recent clinical trials provide critical insights into the relationship between behavioral and clinical outcomes. The table below summarizes primary and secondary outcomes from two key JITAI trials focused on dietary sodium reduction and blood pressure management.

Table 1: Endpoint Comparison from Recent Sodium-Reduction JITAIs

Trial / Outcome	Intervention Group Change	Control Group Change	Between-Group Difference (95% CI)	P-value
myBPmyLife Trial (6-month) [6]
Systolic BP (Primary)	-5.2 mmHg	-5.7 mmHg	0.5 mmHg (-2.4 to 3.3)	0.76
Diastolic BP	-3.0 mmHg	-3.6 mmHg	0.6 mmHg (-1.2 to 2.3)	0.52
Daily Step Count	+170 steps	-319 steps	489 steps (22 to 956)	0.040
Sodium Intake	-1145 mg	-860 mg	-285 mg (-462 to -108)	0.002
LowSalt4Life 2 (2-month) [11]
Systolic BP (Primary)	Results Awaited	Results Awaited	Results Awaited	Results Awaited

The data from the myBPmyLife trial is particularly revealing. It shows a clear dissociation between behavioral and clinical efficacy: the intervention successfully modified the target behaviors (physical activity and sodium intake) but did not translate into a superior blood pressure reduction compared to an active control [6]. This highlights the potential for unmeasured moderating variables or the possibility that the magnitude of behavioral change, while statistically significant, was insufficient to drive a clinical difference in this population.

Experimental Protocols for JITAI Trials

Core JITAI Trial Design Protocol

The following protocol outlines a standard methodology for a randomized controlled trial (RCT) evaluating a dietary JITAI, based on the designs of published and ongoing studies [6] [11].

• Objective: To evaluate the efficacy of a JITAI for reducing dietary sodium intake and improving blood pressure control in adults with hypertension.
• Design: Two-arm, randomized controlled trial with a 6-month intervention period.
• Participants:
  • Inclusion: Adults (age ≥18) with hypertension, owning a smartphone.
  • Exclusion: Estimated sodium intake <1500 mg/day, conditions prohibiting dietary changes.
• Randomization: 1:1 allocation to Intervention (JITAI) or Active Control group.
• Intervention Group:
  • Receives a smartphone application with JITAI functionality.
  • The JITAI uses a combination of wearable sensor data (step count) and EMAs (Ecological Momentary Assessments) to collect tailoring variables like location, mood, and cravings [4].
  • Decision rules trigger tailored support messages (SMS) at critical decision points (e.g., near meal times, when inactivity is detected) to encourage low-sodium choices and physical activity.
• Active Control Group:
  • Uses the same smartphone application for self-monitoring of blood pressure and receives a smartwatch.
  • Does not receive the proactive, context-aware messaging from the JITAI.
• Data Collection Points: Baseline, 2 months, 4 months, and 6 months.

Just-in-Time Adaptive Intervention Workflow

The operational core of a JITAI is a continuous feedback loop that personalizes support in real-time. The workflow below details this automated process.

Table 2: JITAI Workflow Components [4]

Step	Component	Action	Example in Dietary Trial
1	Data Input	Passive and active data collection.	Wearable tracks location; EMA prompt: "Are you planning a meal?"
2	Tailoring Variable Assessment	Algorithm processes input against decision rules.	If location == restaurant AND time == lunch, then risk is high.
3	Decision Point	The specific moment an intervention decision is made.	User confirms they are at a restaurant menu.
4	Intervention Option Selection	A support message is selected from a library.	Select message: "Try asking for dressings/sauces on the side."
5	Delivery	The message is sent to the user.	SMS is delivered to the user's smartphone.

The following diagram maps this dynamic process, showing how data flows through the JITAI system to deliver a personalized intervention.

Diagram 2: JITAI Operational Workflow. This diagram details the continuous operational cycle of a Just-in-Time Adaptive Intervention (JITAI). The process begins with data collection from passive sensors and active surveys, which informs the assessment of tailoring variables. When a predefined decision point is reached, a tailored intervention is selected and delivered. User engagement and outcomes are then recorded, creating a feedback loop for ongoing system learning and personalization.

The Scientist's Toolkit: Essential Reagents & Digital Solutions

Successfully implementing a JITAI trial requires a suite of digital tools and methodological approaches.

Table 3: Key Research Reagent Solutions for JITAI Trials

Item / Solution	Function in JITAI Research
Mobile Application Platform	Core delivery vehicle for interventions, assessments, and educational content. Often requires custom development for JITAI logic.
Ecological Momentary Assessment (EMA)	A low-burden self-report method using short, repeated surveys on a mobile device to capture dynamic states (mood, cravings, context) that sensors cannot measure [4].
Wearable Activity Tracker	Provides passive, objective data on physical activity (step count) and sleep, serving as both a behavioral endpoint and a tailoring variable.
Clinical Outcome Assessment (COA)	A measurement tool used to capture clinical endpoints that are meaningful to patients, such as quality of life, physical function, and symptom burden [70].
Automated BP Monitor	Provides objective, patient-measured clinical endpoint data for cardiovascular trials. Essential for hypertension studies.
Cloud-Based Data Visualization Platform	Enables real-time aggregation, analysis, and visualization of complex, multi-source trial data for study teams, supporting faster decision-making [71] [72].
Behavior Change Technique (BCT) Taxonomy	A standardized taxonomy (e.g., the COM-B model or Behavior Change Wheel) to define, code, and replicate the active ingredients of the intervention [69].

The analysis of clinical trial outcomes for JITAIs demands a dual-focused approach that rigorously measures both behavioral change and clinical endpoints. As evidenced by recent trials, success in one domain does not automatically guarantee success in the other. Future research must prioritize deeper personalization, explore the use of multi-source data and AI [73], and integrate patient-centered digital health technologies (DHTs) and clinical outcome assessments (COAs) to capture the full impact of interventions [70]. By adopting the structured protocols and analytical frameworks outlined in this document, researchers can robustly evaluate the complex pathway from JITAI-driven behavioral adherence to meaningful clinical improvement.

Application Notes: Efficacy and Implementation

Table 1: Comparative Effects of JITAIs and Standard Counseling on Health Outcomes

Outcome Measure	JITAI Effectiveness	Standard Counseling Effectiveness	Comparative Notes
Mental Health (e.g., Depression, Anxiety)	Small between-group effect size (Hedges' g = 0.15); significant effects sustained at 1, 3, and 6-month follow-ups. Shorter interventions (<6 weeks) showed greater longevity of effects (g=0.71) [74].		JITAIs provide sustained benefits beyond the intervention period, often complementing therapeutic measures [74].
Physical Activity	Significantly increased daily step count (+489 steps compared to control) and reduced sedentary behavior [6].		Effects on systolic blood pressure (SBP) were not significant despite positive behavior change [6].
Dietary Sodium Intake	Significant reduction in daily sodium intake (-285 mg more than control) [6].
User Engagement & Acceptability	High perceived usability (mean score 73.6 on SUS); participants found support motivating and well-tailored [6] [4].		Challenges include sensor reliability, notification timing, and perceived personalization [4] [28].

Key Differentiating Characteristics

Table 2: Core Characteristics of JITAIs versus Standard Behavioral Counseling

Feature	Just-in-Time Adaptive Interventions (JITAIs)	Standard Behavioral Counseling
Intervention Timing	Dynamic, real-time support during "teachable moments" or states of vulnerability/opportunity [5].	Typically pre-scheduled sessions (e.g., weekly, monthly).
Adaptation Mechanism	Continuous, automated tailoring based on real-time data from EMAs and/or passive sensors [75] [5].	Manual, static tailoring based on baseline assessments or periodic check-ins.
Contextual Sensitivity	High; leverages individual's current internal state (mood, stress) and external context (location, weather) [4].	Moderate to low; relies on recalled experiences and generalized strategies.
Delivery Modality	Smartphone apps with push notifications, often integrated with wearable sensors [6] [7].	In-person, phone calls, or telehealth sessions; static digital programs.
Primary Goal	Momentary behavior change and habit formation through repeated, timely support [74] [5].	Knowledge transfer and long-term skill building through reflection.

Experimental Protocols

Protocol 1: Randomized Controlled Trial for JITAI Efficacy

This protocol outlines a method to evaluate the efficacy of a JITAI against an active control group, as demonstrated in the myBPmyLife study [6].

a. Study Design

• Type: Two-arm Randomized Controlled Trial (RCT).
• Duration: 6-month intervention with baseline, midpoint (e.g., 3-month), and post-intervention (6-month) assessments. Including a follow-up period (e.g., 3 or 6 months post-intervention) is recommended to assess longevity of effects [74].
• Participants: Recruit adults from the target population. The myBPmyLife study, for example, enrolled 602 participants with hypertension [6].

b. Intervention Groups

• JITAI Group: Receives the just-in-time adaptive intervention.
  • Components: A smartphone application that delivers support based on a defined algorithm.
  • Tailoring Variables: These can include active Ecological Momentary Assessments (EMAs) on mood, cravings, and activity [4], and passive sensor data (e.g., step count from a wearable, GPS).
  • Decision Rules: Pre-defined "if-then" rules for intervention delivery. For example, "IF the user's daily step count is below their goal AND they are in a location previously associated with walking, THEN send an activity suggestion notification." [7] [5].
• Active Control Group: Controls for the non-specific effects of attention and self-monitoring.
  • Components: Participants self-monitor their blood pressure and are given a smartwatch but do not receive the adaptive, context-sensitive notifications [6].

c. Primary and Secondary Outcomes

• Primary Outcome: Distal clinical outcome relevant to the target behavior (e.g., change in systolic blood pressure for hypertension management [6]; change in depression/anxiety symptom scores for mental health [74]).
• Secondary Outcomes:
  • Proximal behavioral outcomes (e.g., mean daily step count, estimated daily dietary sodium intake [6]).
  • Usability and acceptability measures (e.g., System Usability Scale, qualitative interviews [6] [4]).
  • Intervention engagement (e.g., number of application logins, responses to EMAs).

d. Data Analysis

• Analyze the primary outcome using an intention-to-treat approach, comparing the change in the primary outcome from baseline to 6 months between groups using analysis of covariance (ANCOVA) or mixed-effects models for repeated measures [6].

Protocol 2: System ID for JITAI Optimization

This protocol, based on Project JustWalk, describes an N-of-1 experimental method to empirically identify the optimal components of a JITAI, specifically the "just-in-time" state [7] [76].

a. Study Design

• Type: Micro-randomized trial or system identification study within a single-subject design.
• Duration: Extended period to capture sufficient data per participant (e.g., 270 days [7]).
• Participants: A smaller sample of participants from the target population (e.g., N=48 [7]).

b. Experimental Manipulation

• Intervention Components: Systematically vary two components:
  • Walking Notifications: Deliver prompts to walk up to 4 times per day.
  • Daily Step Goals: Vary suggested goals within a range related to the participant's baseline [7].
• JIT State Operationalization: Notifications are randomly or semi-randomly delivered across different combinations of theoretically-defined states:
  • Need: Whether the participant has met their daily step goal.
  • Opportunity: Whether the next 3-hour window is a time they have previously been active.
  • Receptivity: Whether the participant has previously responded to notifications by walking [7] [76].

c. Primary Outcome

• Proximal Outcome: Short-term behavior change following the intervention, such as step count within the 3-hour window after a notification is delivered [7].

d. Data Analysis

• Use system identification techniques to estimate idiographic (N-of-1) computational models for each participant.
• These models predict how an individual responds to different types of support under varying conditions, informing the optimization of the JITAI's decision rules [7].

Conceptual Framework and Workflow

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Tools for JITAI Development and Testing

Tool / Reagent	Function / Purpose	Example Use Case
Mobile Application Platform	Core delivery mechanism for interventions and EMAs.	Custom-built (e.g., myBPmyLife app [6]) or platforms like RADAR-base, Beiwe.
Wearable Activity Sensor	Passively collects physical activity and physiological data.	Fitbit Versa 3 [7] or similar devices to track step count, heart rate, and sleep.
Ecological Momentary Assessment (EMA)	Actively collects self-reported data on internal states and context.	Short, smartphone-prompted surveys on mood, stress, cravings, and environment [4].
JITAI Conceptual Framework	Guides the systematic design of the intervention components.	Nahum-Shani et al. framework (defines distal/proximal outcomes, tailoring variables, decision rules) [5].
System ID / Modeling Software	Analyzes intensive longitudinal data to build predictive models of behavior.	R, Python with time-series and control system libraries to model N-of-1 data for optimization [7] [76].

Usability and Acceptability Findings from Recent Feasibility Studies

Just-in-Time Adaptive Interventions (JITAIs) represent a transformative approach in digital health, designed to provide personalized support to individuals at moments when it is most likely to be effective. These interventions are particularly relevant for dietary adherence research, where maintaining long-term behavior change is challenging. JITAIs leverage mobile technology to deliver tailored interventions based on real-time data about an individual's state and context. The development and refinement of JITAIs depend heavily on feasibility and pilot studies that establish the usability and acceptability of these digital platforms among target populations. This review synthesizes recent findings from feasibility studies to guide researchers and intervention developers in optimizing digital tools for dietary adherence research.

Key Feasibility Findings from Recent Digital Health Studies

The table below summarizes usability and acceptability metrics from recent feasibility studies of digital health interventions, providing comparative benchmarks for researchers.

Table 1: Usability and Acceptability Metrics from Recent Digital Health Feasibility Studies

Study & Intervention	Target Population	Sample Size	Usability Metrics	Acceptability Findings	Key Outcomes
STAC-T App [77]	Middle school students (rural, low-income)	229 students	Not specified	82.1%-90.0% found it acceptable; 78.6%-83.0% found it relevant	88.8% of witnesses used ≥1 STAC strategy; program relevance predicted strategy use (p=.016)
myBPmyLife JITAI [6]	Adults with hypertension	235 (usability subset)	System Usability Scale (SUS): 73.6 (SD 19)	Good usability; 45.76% daily engagement over 6 months	Increased daily steps (+170 vs -319, p=0.04); reduced sodium intake (-1145mg vs -860mg, p=0.002)
P-STEP App [78]	Adults with long-term cardiorespiratory conditions	61 completers	SUS: 61.68 (SD 22.9); Bespoke score: 66.82 (SD 14.75)	Moderate usability; User Engagement Score: 3.08 (SD 0.79)	Feasible for planning exercise routes to minimize pollution exposure
AI-Mediterranean Diet App [79]	Adults assessing Mediterranean Diet adherence	24 participants	61.8% mean Average Precision in food recognition	Positive user and dietitian feedback; automated feedback reports valued	Mean difference in adherence score vs. dietitian: 3.5% (not significant)

Experimental Protocols for Assessing Usability and Acceptability

Protocol: Evaluating a Bystander Intervention App (STAC-T Model)

Objective: To evaluate the acceptability, relevance, and short-term effectiveness of a technology-based bullying bystander intervention among middle school students [77].

• Study Design: Part of a larger randomized controlled trial at six middle schools in rural, low-income U.S. communities.
• Participants: 229 middle school students (grades 6-8, mean age 12.1) randomized to the intervention group who completed the 30-day follow-up survey.
• Intervention: STAC-T app, a technology-based version of an evidence-based bullying bystander intervention. The app includes a 40-minute training and a 15-minute booster session.
• Data Collection:
  • Acceptability & Relevance: Assessed via a post-intervention questionnaire.
  • Strategy Use: Measured at 30-day follow-up by asking students who witnessed bullying whether they used any of the four STAC strategies (Stealing the Show, Turning it Over, Accompanying Others, Coaching Compassion).
• Analysis: Descriptive statistics for acceptability, relevance, and strategy use. Linear regression to assess the relationship between program acceptability/relevance and use of STAC strategies.

Protocol: Feasibility of a JITAI for Hypertension Management (myBPmyLife Model)

Objective: To evaluate the efficacy and usability of a JITAI (myBPmyLife app) in promoting physical activity and lower-sodium food choices to reduce blood pressure over six months [6].

• Study Design: Randomized controlled trial.
• Participants: 602 adults with hypertension.
• Intervention: The myBPmyLife JITAI app, which provided:
  • Context-aware physical activity suggestions.
  • Support for selecting lower-sodium foods at grocery stores and restaurants.
  • BP self-monitoring support.
• Control Group: Received a smartwatch and performed BP self-monitoring without the JITAI.
• Data Collection:
  • Clinical Outcomes: Systolic and diastolic BP measured at baseline and 6 months.
  • Behavioral Outcomes: Daily step counts (via smartwatch) and estimated daily dietary sodium intake (via dietary assessment).
  • Usability: System Usability Scale (SUS) completed by the intervention group at the end of the study.
  • Engagement: Application usage data (days used, interactions per day) collected automatically.
• Analysis: Primary outcome: change in SBP. Secondary outcomes: changes in daily steps and sodium intake. SUS scores interpreted via standard benchmarks (mean of 73.6 is considered "good").

Protocol: Usability of an AI-Powered Dietary Assessment System

Objective: To develop and evaluate the feasibility of an AI-powered smartphone application that automatically calculates Mediterranean Diet adherence (MDA) score from meal images [79].

• Study Design: Feasibility study.
• Participants: 24 participants.
• Intervention: An end-to-end system integrated into a smartphone app that:
  • Recognizes food and drink items from a single meal photo using a Convolutional Neural Network (CNN).
  • Estimates their respective serving sizes.
  • Automatically calculates a weekly MDA score.
  • Generates and presents an automated feedback report to the user.
• Data Collection:
  • System Performance: Mean Average Precision (mAP) for food recognition on a testing set and on images from the feasibility study.
  • Accuracy: Comparison of the system-calculated MDA score versus the score calculated by an expert dietitian.
  • User Feedback: Qualitative feedback from participants and dietitians on the system's feasibility and usability.
• Analysis: Quantitative analysis of mAP and MDA score difference. Qualitative summary of user feedback.

Visualization of JITAI Workflow and Assessment Framework

The following diagram illustrates the core operational logic of a JITAI for dietary adherence, integrating the assessment of usability and acceptability as critical feedback mechanisms.

Diagram 1: JITAI Operational Feedback Loop. This workflow shows how continuous data monitoring triggers intervention decisions, with usability and acceptability assessments providing essential feedback for system refinement.

The diagram below outlines a multi-modal framework for evaluating digital health interventions, combining quantitative and qualitative methods as employed in the reviewed studies.

Diagram 2: Multi-Modal Feasibility Assessment. This framework illustrates the key dimensions for evaluating intervention feasibility and their relationship to ultimate health outcomes.

The Scientist's Toolkit: Key Reagents and Measures

Table 2: Essential Tools for Digital Intervention Feasibility Research

Tool or Measure	Type	Primary Function	Application Example
System Usability Scale (SUS)	Standardized Scale	A 10-item questionnaire providing a global view of subjective usability assessments.	Used to evaluate the myBPmyLife app, yielding a score of 73.6, indicating good usability [6].
User Engagement Scale (UES)	Standardized Scale	Measures user engagement through dimensions like focused attention and perceived usability.	The P-STEP study reported a UES Short Form score of 3.08 to gauge participant engagement [78].
Application Usage Analytics	Behavioral Metric	Passively collected data on frequency and type of application use (e.g., logins, feature access).	myBPmyLife intervention group used the app on a median of 81 days (45.76% of eligible days) [6].
Acceptability & Relevance Questionnaire	Custom Survey	Study-specific measures of how appropriate and meaningful the intervention is perceived by the target population.	82.1-90.0% of students found the STAC-T app acceptable, and 78.6-83.0% found it relevant [77].
Convolutional Neural Network (CNN)	AI Model	Recognizes and classifies multiple food items from a single meal image for automated dietary assessment.	Used in the AI-Mediterranean Diet app to recognize 31 food categories and estimate serving sizes [79].
Focus Groups / Interviews	Qualitative Method	Gathers in-depth feedback on user experience, barriers, and facilitators to intervention use.	Employed in the DePEC-Nutrition study to explore participants' ability to make dietary changes [80].

The Gap Between Short-Term Behavior Change and Sustained Health Improvements

Application Notes: Efficacy and Limitations of JITAIs in Dietary Research

Just-in-Time Adaptive Interventions (JITAIs) represent a promising frontier in digital health, designed to provide personalized support for healthy behaviors at moments when individuals are most receptive [6]. While these interventions demonstrate significant efficacy in promoting short-term behavior change, a critical gap persists in translating these gains into sustained health improvements, particularly in the domain of dietary adherence [4].

Key Efficacy Findings from Recent Trials

Table 1: Outcomes from the myBPmyLife JITAI Randomized Controlled Trial (2025) [6]

Outcome Measure	Intervention Group Change	Control Group Change	P-value	Clinical Significance
Primary Outcome: Systolic BP	-5.2 mmHg	-5.7 mmHg	0.76	Not Significant
Secondary: Daily Step Count	+170 steps	-319 steps	0.040	Significant, modest effect
Secondary: Sodium Intake	-1145 mg	-860 mg	0.002	Significant, substantial effect

The data in Table 1 illustrates a central challenge: the myBPmyLife JITAI successfully modified two key target behaviors—physical activity and sodium intake—over a six-month period, yet this did not yield a statistically significant reduction in the primary clinical endpoint, systolic blood pressure, beyond the change observed in the control group [6]. This dissociation underscores the complex pathway between behavioral modification and physiological outcomes.

The Sustainability Challenge in Behavioral Outcomes

The phenomenon of short-term success without long-term maintenance is not unique to JITAIs. Health and Wellness Coaching (HWC), another patient-centered approach, shows positive impacts on health behaviors like exercise and nutrition for up to six months post-intervention [81]. However, evidence for the long-term stability of these behavior changes remains limited and unclear [81]. A four-year follow-up study of coached heart failure patients found only small, sustained benefits for behaviors like diet, with no long-term changes in quality of life [81]. This suggests that the strategies effective for initiating change may differ from those required for maintenance.

Experimental Protocols for JITAI Research

To systematically investigate and bridge the sustainability gap, researchers require robust and replicable experimental methodologies. The following protocols detail key approaches.

Protocol 1: Randomized Controlled Trial of a Multi-Domain JITAI

This protocol is adapted from the myBPmyLife study, a large-scale RCT evaluating a JITAI targeting physical activity and dietary sodium intake [6].

• Objective: To assess the efficacy of a dynamic mobile application-based JITAI in promoting physical activity and lower-sodium food choices for patients with hypertension over 6 months.
• Population: Adults with hypertension. Key exclusion criteria include estimated sodium intake <1500 mg/day.
• Study Arms:
  • Intervention Group: Receives the JITAI application (e.g., myBPmyLife), which includes BP self-monitoring, activity suggestions, and prompts for lower-sodium food choices based on user context.
  • Control Group: Performs BP self-monitoring and is given a smartwatch but does not receive the adaptive intervention components.
• Primary Outcome: Change in systolic blood pressure from baseline to 6 months.
• Secondary Outcomes:
  • Mean daily step count, measured via wearable device or smartphone sensor.
  • Estimated daily dietary sodium intake, measured via a validated short FFQ (e.g., the Dutch Healthy Diet FFQ) [82].
  • Application usability, measured via the System Usability Scale (SUS).
• Analysis: Comparison of change scores between groups using appropriate statistical tests (e.g., t-test for continuous outcomes).

Protocol 2: Qualitative Acceptability Assessment of an EMA-Driven JITAI

This protocol is based on a study exploring the acceptance of just-in-time adaptive lifestyle support for people with Type 2 Diabetes [4]. Understanding user experience is critical for designing engaging and sustainable interventions.

• Objective: To assess the acceptability and perceived personalization of an Ecological Momentary Assessment (EMA)-driven JITAI.
• Population: Individuals with Type 2 Diabetes.
• Intervention: Participants use the JITAI for a period of 2 weeks. The system involves:
  • EMA Component: Daily short questionnaires on activity, location, mood, overall condition, and cravings.
  • JITAI Component: Tailored support messages (e.g., via SMS) triggered by EMA responses and other sensor data.
• Data Collection: Semi-structured interviews conducted by telephone after the intervention period. Topics should cover acceptability of EMA prompts, perceived relevance and tailoring of messages, and overall user experience.
• Data Analysis: Hybrid approach of thematic analysis to identify key themes related to intervention design, decision points, and mechanisms underlying adherence.

Protocol 3: Mixed-Methods Exploration of Dietary Determinants

This protocol leverages mixed-methods to gain a deeper, contextual understanding of the factors driving dietary adherence, which can inform more effective JITAI tailoring variables [82].

• Objective: To identify how quantitative and qualitative data on dietary guideline adherence correspond and complement each other, and to explore the interdependence of determinants.
• Design: Convergent parallel mixed-methods study.
• Quantitative Component:
  • Cross-sectional survey administered to a large sample (n > 1000).
  • Measures: Adherence to dietary guidelines (using a validated index like the DHD15-index), socio-economic position, and behavioral determinants (e.g., cognitive restraint, habit strength, cooking skills).
  • Analysis: Linear regression to assess associations between determinants and total guideline adherence.
• Qualitative Component:
  • Semi-structured telephone interviews conducted with a sub-sample of participants (n ≈ 24), stratified by socio-economic background.
  • Analysis: Directed content analysis to explore perceptions of dietary behaviors, barriers, and facilitators.
• Integration: Quantitative and qualitative results are analyzed independently and then compared and interpreted in an iterative, reflexive manner to provide a multi-level perspective.

Visualizing the JITAI Framework and Sustainability Model

The following diagrams illustrate the core structure of a JITAI and a theoretical model for sustaining nutritional behavior change.

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials and Tools for Dietary JITAI Research

Item	Function/Description	Example Use in Research
Validated Short FFQ	A food frequency questionnaire designed to estimate habitual nutrient intake and adherence to dietary guidelines.	The Dutch Healthy Diet FFQ (DHD15-index) provides a total score (0-150) for adherence to national guidelines and was used to measure sodium intake [6] [82].
Ecological Momentary Assessment (EMA)	A method for collecting real-time data on behaviors, cognitions, and environmental factors in a participant's natural environment via short, repeated questionnaires.	Used in JITAIs to gather self-reported data on mood, cravings, and location as tailoring variables for support messages [4].
Three-Factor Eating Questionnaire	A validated instrument measuring uncontrolled eating, emotional eating, and cognitive restraint.	Employed in mixed-methods studies to quantitatively assess psychological determinants of dietary guideline adherence [82].
System Usability Scale (SUS)	A simple, ten-item scale giving a global view of subjective assessments of usability and acceptability.	Administered at the end of intervention trials to gauge user satisfaction and perceived ease of use of the JITAI application [6].
Board-Certified Health Coach	A professional trained in behavior change theory, motivational strategies, and communication techniques to enhance intrinsic motivation.	Can be integrated into a hybrid JITAI model to provide human guidance, which may improve long-term effectiveness [81] [4].

Just-in-time adaptive interventions (JITAIs) represent a transformative approach in digital health, designed to deliver personalized support at moments of vulnerability and opportunity. For dietary adherence research, JITAIs dynamically adapt intervention type, timing, and intensity based on continuously collected individual data [17]. Despite promising foundations, critical knowledge gaps remain in scaling these interventions for diverse populations and optimizing the integration of artificial intelligence with human clinical support. This application note delineates key future research priorities and provides detailed experimental protocols to advance the field of JITAIs for dietary adherence, with a specific focus on scalability, diversity, and hybrid models.

Current Evidence and Identified Gaps

Recent systematic reviews and clinical trials highlight both the potential and limitations of current JITAI methodologies. The table below summarizes key findings from recent studies investigating JITAIs for lifestyle behaviors and chronic disease management.

Table 1: Summary of Recent JITAI Evidence and Identified Research Gaps

Study Focus	Key Findings	Identified Gaps
myBPmyLife Trial (n=602) [6]	- No significant SBP reduction vs. control (-5.2 vs. -5.7 mmHg, p=0.76)- Increased daily steps (+170 vs. -319, p=0.04)- Reduced sodium intake (-1145 mg vs. -860 mg, p=0.002)	Disconnect between behavioral changes and clinical outcomes; Limited effectiveness in diverse populations
JITAI/EMI Meta-Analysis (23 studies) [74]	Small between-group effect on mental health (g=0.15); Effects sustained up to 6 months; Shorter interventions (<6 weeks) showed greater effect longevity (g=0.71)	Need for clearer decision rules; Long-term effect uncertainty; Control group selection impacts effect sizes
Dynamically Tailored eHealth Review (61 interventions) [24]	87% targeted physical activity; 43% targeted nutrition; 74% used rule-based tailoring; 13% used machine learning	Limited use of data-driven methods; Insufficient reporting of methodology; Heterogeneous outcomes

The evidence indicates that while JITAIs can effectively modify intermediate behaviors like physical activity and sodium intake, translating these changes into robust clinical outcomes remains challenging [6]. Furthermore, the field is characterized by methodological heterogeneity and insufficient reporting, complicating replication and large-scale implementation [24].

Priority 1: Scaling for Diverse Populations

Current Limitations in Diversity and Scalability

Many JITAIs are developed and tested in homogenous populations, limiting generalizability. A review of dynamically tailored eHealth interventions found that over half (52.5%) were conducted in the United States, with limited representation of global populations [24]. Furthermore, a qualitative study of a JITAI for type 2 diabetes patients revealed notable differences in individual experiences with the intervention, particularly regarding perceived personalization and the relevance of tailoring variables [4] [31]. Participants expressed a need for support tailored to their specific characteristics and circumstances, indicating that current one-size-fits-all scaling approaches are insufficient.

Proposed Experimental Protocol: Scalability and Diversity

Objective: To develop and validate a culturally adapted, scalable JITAI for dietary adherence across diverse socioeconomic and ethnic groups.

Design: Multi-site, randomized controlled trial with a hybrid effectiveness-implementation design.

Participants: N=800 adults with type 2 diabetes from four distinct demographic groups:

• Group A: Low socioeconomic status, urban
• Group B: Hispanic/Latino population
• Group C: African American population
• Group D: Rural residents

Table 2: Key Components and Adaptation Strategies for Diverse Populations

JITAI Component	Standard Approach	Adaptation Strategies
Tailoring Variables	Standard EMA items (mood, location, cravings) [31]	Culturally relevant cues (e.g., traditional foods, social contexts)Language and literacy appropriatenessContextual barriers (food access, safety)
Intervention Options	Generic dietary advice	Culturally appropriate meal alternativesTailored to local food availabilityRespect for cultural dietary practices
Delivery Modality	Smartphone application	SMS-based support for limited smartphone accessMultilingual optionsLow-bandwidth compatibility

Primary Outcomes:

• Implementation: Reach, adoption, and maintenance across groups
• Effectiveness: Change in dietary adherence (24-hour dietary recalls)
• Clinical: HbA1c change at 6 months

Methodology:

• Co-Design Phase: Conduct focus groups with target populations to adapt JITAI components.
• Optimization Phase: Use microrandomized trial (MRT) designs to test adapted intervention components.
• Evaluation Phase: Conduct RCT comparing adapted JITAI versus standard care.

Priority 2: Hybrid AI-Human Intervention Models

The Case for Integration

While fully automated JITAIs offer scalability, evidence suggests that human support enhances engagement and effectiveness. A systematic review of dynamically tailored interventions found that combining algorithm-driven feedback with human guidance was associated with greater efficacy [24]. Furthermore, a meta-analysis of JITAIs and EMIs indicated that the selection of control groups significantly influences effect sizes, suggesting that human-supported interventions may serve as more appropriate comparators than waitlist controls [74].

Proposed Experimental Protocol: Hybrid AI-Human JITAI

Objective: To determine the optimal integration of automated JITAI components with human clinical support for dietary adherence.

Design: 3-arm randomized controlled trial.

Participants: N=450 adults with hypertension and poor dietary adherence.

Interventions:

• Arm 1: Fully automated JITAI
• Arm 2: Hybrid JITAI with automated messaging + monthly dietitian support
• Arm 3: Hybrid JITAI with automated messaging + algorithm-triggered dietitian support

JITAI Components:

• Decision Points: Following each EMA survey (5x/day) [3]
• Tailoring Variables: EMA items (mood, location, cravings), passive sensor data
• Intervention Options: Automated messages, self-regulation strategies, dietitian notification

The diagram below illustrates the decision workflow for the hybrid JITAI model, particularly for Arm 3.

Primary Outcomes:

• Dietary lapses per week (EMA-based)
• Systolic blood pressure change at 6 months
• Cost-effectiveness (cost per mmHg reduction)

Methodology:

• Use a machine learning algorithm to predict dietary lapse risk based on EMA and sensor data [3].
• Implement decision rules for escalating care from automated support to human intervention.
• Conduct mediation analyses to identify mechanisms of behavior change.

Priority 3: Methodological Innovations

Advanced Study Designs

Microrandomized trials (MRTs) represent a powerful methodology for optimizing JITAIs. Unlike traditional RCTs, MRTs randomize participants hundreds of times throughout the study to different intervention options at each decision point [3]. This design enables researchers to:

• Evaluate the immediate effect of each intervention component
• Examine time-varying moderation of intervention effects
• Optimize JITAI decision rules before conducting definitive RCTs

Proposed Experimental Protocol: MRT for Dietary Lapses

Objective: To optimize a JITAI for dietary adherence by identifying the most effective intervention components for preventing dietary lapses.

Design: Microrandomized trial.

Participants: N=159 adults with overweight/obesity and cardiovascular disease risk.

Procedure:

• Participants receive a 6-month behavioral obesity treatment.
• Throughout treatment, they complete EMA surveys 6 times daily assessing lapse triggers.
• When the JITAI algorithm detects elevated lapse risk, participants are randomly assigned to one of:
  • No intervention
  • Generic risk alert
  • Theory-driven intervention (4 types: education, self-efficacy, motivation, self-regulation)

Table 3: Research Reagent Solutions for JITAI Development

Reagent/Tool	Function	Example Application
Ecological Momentary Assessment (EMA)	Real-time data collection on behaviors, contexts, and states [5]	Measuring dietary lapses and triggers multiple times daily [3]
Passive Sensing Technologies	Continuous data collection without user burden [24]	Using GPS, accelerometers, and heart rate monitors to contextualize risk
Machine Learning Algorithms	Dynamic prediction of risk states and intervention optimization [3]	Classifying individual lapse risk based on multimodal data streams
Microrandomized Trial (MRT) Design	Experimental design for optimizing intervention components [3]	Testing efficacy of multiple intervention options over time

Primary Outcome: Occurrence of dietary lapse in the 2.5 hours following randomization.

Analytical Approach:

• Use weighted and centered least-squares estimation to analyze MRT data.
• Examine moderation by context (time, location, trigger type).
• Develop optimized decision rules for a future RCT.

The workflow below illustrates the sequential phases for developing and testing an optimized JITAI using an MRT design.

Advancing JITAI research for dietary adherence requires coordinated efforts across three priority areas: developing scalable interventions for diverse populations, establishing optimal hybrid AI-human models, and implementing methodologically rigorous designs. The experimental protocols outlined provide a roadmap for systematically addressing these priorities. By adopting these approaches, researchers can contribute to the development of more effective, equitable, and implementable JITAIs that ultimately improve dietary adherence and clinical outcomes in chronic disease management.

Conclusion

Just-in-Time Adaptive Interventions represent a paradigm shift in dietary adherence support, moving from static, one-size-fits-all advice to dynamic, personalized guidance. The synthesis of current research indicates that while JITAIs consistently demonstrate efficacy in promoting short-term behavioral changes—such as increased step counts and reduced sodium intake—their impact on long-term clinical endpoints like blood pressure requires further optimization. Future success hinges on overcoming key challenges: developing more sophisticated and empirically-grounded decision rules, deeply personalizing interventions to individual preferences and cultural backgrounds using frameworks like FQVT, and seamlessly integrating passive sensing to minimize user burden. For biomedical and clinical research, the next frontier involves conducting large-scale trials to validate these optimized JITAIs and exploring their integration as a core component in chronic disease management, ultimately bridging the critical gap between momentary adherence and lasting health improvement.

References

• 1. Just-in-Time Adaptive Interventions (JITAIs) in Mobile Health [https://pmc.ncbi.nlm.nih.gov/articles/PMC5364076/]
• 2. Optimizing a Just-in-Time Adaptive Intervention to Improve ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8691411/]
• 3. Optimizing a Just-in-Time Adaptive Intervention to Improve ... [https://www.researchprotocols.org/2021/12/e33568/]
• 4. Exploring the Acceptance of Just-in-Time Adaptive Lifestyle ... [https://formative.jmir.org/2025/1/e65026]
• 5. Beyond the current state of just-in-time adaptive ... [https://www.frontiersin.org/journals/digital-health/articles/10.3389/fdgth.2025.1460167/full]
• 6. Physical activity and diet just-in-time adaptive intervention ... [https://www.nature.com/articles/s41746-025-01844-3]
• 7. Protocol for a System ID Study of Just-in-Time Adaptive ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10565629/]
• 8. Just-In-Time Adaptive Interventions to Promote Behavioral ... [https://www.researchprotocols.org/2025/1/e58917]
• 9. Building health behavior models to guide the development of ... [https://pubmed.ncbi.nlm.nih.gov/26651462/]
• 10. Building health behavior models to guide the development ... [https://asu.elsevierpure.com/en/publications/building-health-behavior-models-to-guide-the-development-of-just-]
• 11. Trial Design A Just-In-Time Adaptive Mobile Application ... [https://www.sciencedirect.com/science/article/pii/S0002870325003564]
• 12. Ecological Momentary Intervention [https://ilumivu.com/markets/researchers/ecological-momentary-intervention/]
• 13. Dietary lapses are associated with meaningful elevations ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9358737/]
• 14. Longer-term effects and underlying mechanisms of dietary ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12147734/]
• 15. Overview of Dietary Assessment Methods for Measuring ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8338737/]
• 16. Identifying Dietary Triggers Among Individuals with Overweight and Obesity: An Ecological Momentary Assessment Study | MDPI [https://www.mdpi.com/2072-6643/17/3/481]
• 17. Just-in-Time Adaptive Interventions in EMA Research [https://www.expiwell.com/post/cutting-edge-applications-of-just-in-time-adaptive-interventions]
• 18. JMIR mHealth and uHealth - Using Mobile Ecological Momentary Assessment to Understand Consumption and Context Around Online Food Delivery Use: Pilot Feasibility and Acceptability Study [https://mhealth.jmir.org/2023/1/e49135/]
• 19. A comparison of personalized and fixed interval signal-contingent ecological momentary assessment to capture dietary data: a double-blinded crossover feasibility study | European Journal of Clinical Nutrition [https://www.nature.com/articles/s41430-025-01674-2]
• 20. Applying Ecological Momentary Assessment (EMA) to ... [https://www.mdpi.com/2075-4426/15/11/526]
• 21. Assessment of EMA binge-eating symptoms in Adolescents [https://www.sciencedirect.com/science/article/pii/S0195666325003654?dgcid=rss_sd_all]
• 22. Dietary Assessment for Early Childhood Dietary Intake [https://www.withpower.com/trial/phase-dietary-assessment-6-2025-1eb41]
• 23. AI-Enabled Clinical Trials: The 2025 Evidence Engineering ... [https://www.linkedin.com/pulse/ai-enabled-clinical-trials-2025-evidence-engineering-j%C3%A9r%C3%B4me-vetillard-wjhbe]
• 24. Systematic review of dynamically tailored eHealth ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12630729/]
• 25. Effects of a Novel Contextual Just-In-Time Mobile App ... [https://mhealth.jmir.org/2020/8/e16696/]
• 26. Optimizing a mobile just-in-time adaptive intervention ... [https://www.sciencedirect.com/science/article/abs/pii/S1551714425000023]
• 27. Using Personalized Intervention Criteria in a Mobile Just-in ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12129369/]
• 28. A systematic review of just-in-time adaptive interventions ... [https://ijbnpa.biomedcentral.com/articles/10.1186/s12966-019-0792-7]
• 29. Proximal and distal influences on dietary change among a ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11148607/]
• 30. Proximal and distal influences on dietary change among a ... [https://www.cambridge.org/core/journals/public-health-nutrition/article/proximal-and-distal-influences-on-dietary-change-among-a-diverse-group-with-prediabetes-participating-in-a-pragmatic-primary-care-nurseled-intervention-a-qualitative-study/87644EA3FB74607F2BC5B6AF153FBB58]
• 31. Exploring the Acceptance of Just-in-Time Adaptive Lifestyle ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11888104/]
• 32. Artificial Intelligence Applications to Personalized Dietary ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12193492/]
• 33. Sensor-based evaluation of intermittent fasting regimes [https://www.nature.com/articles/s41366-025-01889-0]
• 34. Rules [https://documentation.decisions.com/docs/rules]
• 35. Decision Tree [https://docs.decisionrules.io/doc/rules/decision-tree]
• 36. Optimizing a Just-in-Time Adaptive Intervention to Improve ... [https://pubmed.ncbi.nlm.nih.gov/34874892/]
• 37. Micro-Randomized Trials: An Experimental Design for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4732571/]
• 38. [2005.05880] The Micro-Randomized Trial for Developing ... [https://arxiv.org/abs/2005.05880]
• 39. Micro-Randomized Trials (MRTs) [http://people.seas.harvard.edu/~samurphy/JITAI_MRT/mrts4.html]
• 40. Protocol for a Series of Randomized Trials [https://www.researchprotocols.org/2025/1/e77532]
• 41. Systematic review of dynamically tailored eHealth ... [https://www.nature.com/articles/s41746-025-02054-7]
• 42. Physical activity and diet just-in-time adaptive intervention ... [https://pubmed.ncbi.nlm.nih.gov/40659778/]
• 43. Physical activity and diet just-in-time adaptive intervention to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12260109/]
• 44. Outcomes of a pilot microrandomized trial of just-in-time ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12235231/]
• 45. Enhancing Therapy Adherence: Impact on Clinical ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11766829/]
• 46. The psychosocial barriers to medication adherence of patients ... [https://bpsmedicine.biomedcentral.com/articles/10.1186/s13030-020-00202-x]
• 47. Barriers and facilitators of adherence to social distancing ... [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0239795]
• 48. Understanding barriers to and strategies for medication ... [https://bmcpulmmed.biomedcentral.com/articles/10.1186/s12890-022-01892-5]
• 49. Barriers and Facilitators to Medication Adherence among ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11395048/]
• 50. Adherence Barriers, Patient Satisfaction, and Depression ... [https://www.mdpi.com/2227-9032/13/14/1707]
• 51. Facilitators and Barriers to Interacting With Clinical ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10461946/]
• 52. Understanding “Alert Fatigue” in Primary Care: Qualitative ... [https://www.jmir.org/2025/1/e62763/]
• 53. Introduction to SMART designs for the development of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4167891/]
• 54. Personalization and localization as key expectations of ... [https://www.nature.com/articles/s41746-023-00924-6]
• 55. Personalized Nutrition: Tailoring Dietary Recommendations ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11357412/]
• 56. Artificial intelligence in personalized nutrition and food ... [https://www.frontiersin.org/journals/nutrition/articles/10.3389/fnut.2025.1636980/full]
• 57. the Fixed-Quality Variable-Type Dietary Intervention [https://pmc.ncbi.nlm.nih.gov/articles/PMC12592233/]
• 58. Technological enhancements in personalized dietary ... [https://www.sciencedirect.com/science/article/pii/S2667099225000374]
• 59. Delighting Palates with AI: Reinforcement Learning's Triumph ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10857145/]
• 60. DeepEN: Personalized Enteral Nutrition for Critically Ill ... [https://arxiv.org/html/2510.08350v1]
• 61. Reinforcement Learning-Based Food Recommendation ... [https://woodyinternational.com/index.php/jaii/article/view/210]
• 62. Integrating Food Preference Profiling, Behavior Change ... [https://www.jmir.org/2025/1/e75106]
• 63. ALOHA: developing an interactive graph-based ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6686235/]
• 64. Understanding the Relationship Between Ecological ... [https://mhealth.jmir.org/2025/1/e57018]
• 65. Current challenges and opportunities in active and passive ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12274063/]
• 66. Participant Compliance With Ecological Momentary ... [https://mhealth.jmir.org/2025/1/e52887]
• 67. Combining passive eating monitoring and ecological ... [https://www.sciencedirect.com/science/article/pii/S0195666322001817]
• 68. Unveiling overeating patterns within digital longitudinal data ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12443995/]
• 69. Promoting Behavioral Change to Improve Health Outcomes [https://pmc.ncbi.nlm.nih.gov/articles/PMC12023974/]
• 70. Rethinking endpoints in obesity clinical trials [https://www.mapi-trust.org/resources/blog/2025/11/07/rethinking-endpoints-obesity-clinical-trials]
• 71. Clinical Trial Data Visualization Market Report 2025 [https://www.researchandmarkets.com/reports/6178342/clinical-trial-data-visualization-market-report?srsltid=AfmBOooKc7od1WKDfgFqkCUSyF20wwmAJ_E6Krg6UAV-88s4vOkR2la9]
• 72. Turning data into stories: novel visualisations for clinical trials [https://www.astrazeneca.com/content/astraz/what-science-can-do/topics/clinical-innovation/data-visualisation-turning-science-into-stories.html]
• 73. Evolving Clinical Studies – Interventions to Watch in 2025 [https://communities.springernature.com/posts/evolving-clinical-studies-interventions-to-watch-in-2025]
• 74. Effectiveness of just-in-time adaptive interventions for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12481328/]
• 75. Just-In-Time Adaptive Interventions to Promote Behavioral ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11862764/]
• 76. Advancing Understanding of Just-in-Time States for ... [https://www.researchprotocols.org/2023/1/e52161/]
• 77. JMIR Formative Research - Acceptability , Relevance, and Short-Term... [https://formative.jmir.org/2025/1/e76830]
• 78. A feasibility evaluating the study and usability of the... acceptability [https://www.nature.com/articles/s41598-025-20314-0?error=cookies_not_supported&code=91d4a432-d930-41b9-9ad6-f9d93054e6dd]
• 79. A feasibility study to assess Mediterranean Diet adherence ... [https://www.nature.com/articles/s41598-022-21421-y]
• 80. Feasibility and acceptability of a dietary intervention study ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7454174/]
• 81. Supporting Sustainable Health Behavior Change [https://pmc.ncbi.nlm.nih.gov/articles/PMC11130595/]
• 82. Better understanding determinants of dietary guideline ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10348427/]