With the prevalence of overweight and obesity at approximately 71% among US adults [1,2], there is a critical need for weight loss interventions that are both effective and scalable. Gold standard behavioral obesity treatments are effective in producing weight loss of 5%-8%. These 6-12 month treatments pair frequent counseling with evidence-based behavioral strategies to promote healthy dietary change and physical activity [3,4]. Despite their efficacy, these treatments are not easily scalable or accessible due to their high intensity, personnel costs, and time demands.

Digital health interventions can overcome these barriers by delivering weight loss programs remotely and in a standalone format (ie, without a human counseling component). They also have the potential to target broad segments of the population with overweight or obesity, including geographically, racially, and ethnically heterogeneous populations who are underrepresented in traditional obesity treatment research [5-8] despite being disproportionately burdened by obesity [9,10]. However, fully digital weight loss interventions produce only modest weight loss, ranging from 2% to 4% [11-19]. Thus, to fully capitalize on these highly scalable digital interventions, more work is needed to enhance their potency.

Self-monitoring is a well-established behavioral strategy for facilitating weight loss among adults with overweight or obesity [20-22]. It involves tracking a behavior (eg, dietary intake [23] and physical activity [24,25]) or a health-related outcome (eg, body weight [26]). Self-regulation theories, including the Social Cognitive Theory and the Control Theory, posit that behavior change occurs through a cyclical process of goal setting, paying attention to behaviors via self-monitoring, receiving feedback on how one’s current performance compares to one’s goal or past performance, creating an action plan to adjust behavior, and repeating the process to get closer to attaining one’s goal [27,28]. These behavioral strategies work in tandem to promote greater behavior change [29-32]. Self-monitoring is one of the strongest predictors of behavior change [29] and weight loss [31,33]. Numerous systematic reviews and meta-analyses have demonstrated that greater engagement in self-monitoring is consistently linked to greater weight loss [21-23,26,34,35]. Further, this relation holds across many intervention contexts, including human counseling–based and fully digital interventions [22]. However, self-monitoring engagement often declines over time [36-39]. Common barriers include time demands, perceived burden, accessibility challenges, waning novelty, lack of clarity on how to use such data to change behavior, and constraints related to literacy and numeracy [40-42].

One way to promote self-monitoring engagement is by using digital rather than paper-based methods of self-monitoring [22]. Advantages of using digital platforms for self-monitoring include immediate personalized feedback, time savings when using devices such as wireless activity trackers and smart scales, and high portability of mobile health tools, which increases the likelihood of engagement while reducing retrospective errors. In our systematic review of self-monitoring in digital weight loss interventions among adults with overweight or obesity, we found that 81% of interventions included self-monitoring of dietary intake, 82% included self-monitoring of physical activity, and 72% included self-monitoring of weight. In total, over half (54%) of the interventions included all 3 of these self-monitoring strategies [22]. In summary, self-monitoring is a theory-informed and evidence-based approach for weight loss that can be delivered via digital tools.

To the best of our knowledge, no study has systematically examined the optimal combination of self-monitoring strategies included in behavioral obesity treatment. Specifically, weight loss studies often combine multiple self-monitoring domains—tracking dietary intake, physical activity, and body weight—into 1 bundled “treatment package,” which is then compared to a suitable control in an evaluation-randomized controlled trial (RCT) [21-23,34,35]. In this classical treatment package paradigm, these trials can determine the overall efficacy of a weight loss intervention but cannot disentangle the effects of individual self-monitoring components. Thus, in trials using the classical paradigm, if an intervention demonstrates superiority, it would be unclear whether all self-monitoring strategies were needed to produce clinical impact. In addition, if an intervention shows a null effect, it would be unclear whether any of the self-monitoring strategies were beneficial, or whether their effects were dampened by other burdensome intervention components. In summary, trials using the classical treatment package approach are not able to discern whether all or none of the self-monitoring strategies are valuable.

Similarly, in comparative effectiveness trial designs, the unique effects of individual self-monitoring components can be established but not their combined effects. Past studies have used these designs to isolate the effects of 1 self-monitoring domain, including self-monitoring weight (vs not) [43,44] or self-monitoring diet (vs not) [36,45,46]. Other studies have varied the content or frequency of what is being self-monitored within 1 domain (eg, comparing simplified to detailed versions of dietary self-monitoring) [47-53]. Taken together, it is unknown whether some or all 3 of these popular self-monitoring strategies are effective.

Examining whether any interaction effects exist among the 3 most common self-monitoring strategies has the potential for enhancing the efficacy of digital interventions while limiting patient burden. Interactions can be synergistic (ie, when combined, the components result in better outcomes than expected based on the main effects alone) or antagonistic (ie, worse outcomes than expected based on the main effects alone). For instance, a synergistic interaction could occur if combining self-monitoring components provided individuals with clearer insights into changing health behaviors, leading to greater weight loss than expected. In contrast, an antagonistic interaction could occur if a burdensome self-monitoring component increased the risk of disengagement from one or more other self-monitoring components, undermining the effect of those components. This antagonistic interaction results in the combined effect being smaller than what would be expected based on the main effects of the self-monitoring components.

Intervention optimization can be used to guide the selection of self-monitoring strategies for use in weight loss interventions. To do this, the multiphase optimization strategy (MOST), an engineering-inspired framework, enables researchers to build effective multicomponent behavioral interventions through a 3-phase process of preparation, optimization of the intervention, and evaluation of the newly optimized intervention [54,55]. First, the preparation phase consists of understanding gaps in the field, selecting intervention components to test, and developing a conceptual model. Second, the optimization phase consists of conducting a fully powered optimization-RCT to rigorously test the intervention components’ unique and combined effects on the outcome of interest. Third, a subsequent evaluation phase involves testing the newly optimized intervention versus a comparator in a traditional 2-arm evaluation-RCT.

The Spark trial, focused on the optimization phase of MOST, seeks to optimize self-monitoring in a 6-month fully digital weight loss intervention for adults with overweight or obesity. Specifically, we are examining the unique and combined effects of 3 popular self-monitoring strategies (tracking dietary intake, steps, and body weight) on 6-month weight change. In doing so, we will be able to isolate their individual effects and examine whether any synergistic or antagonistic interactions exist among them. Results will inform the development of a newly optimized intervention that retains only the self-monitoring strategies that have a clinically meaningful effect in order to maximize weight loss while minimizing patient burden. We will also assess the self-monitoring strategies’ unique and combined effects on secondary outcomes, including caloric intake, diet quality, physical activity, and health-related quality of life, as well as on 1- and 3- month weight change and the relation between self-monitoring engagement and weight change. Moderators of weight loss success will be explored to understand whether certain subgroups of individuals benefit more from specific self-monitoring strategies. Using an embedded experimental design, we will test whether a self-directed web-based orientation session—completed prior to enrollment—promotes greater 6-month retention in the Spark trial. Finally, we will conduct semistructured qualitative interviews with trial participants to explore factors that impact engagement and its link to weight loss. This paper describes the study protocol, following the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) 2025 Checklist (see Multimedia Appendix 1), which are standardized guidelines for reporting clinical trial protocols [56].

The Spark trial was an optimization RCT that tests the unique and combined effects of 3 self-monitoring strategies (tracking dietary intake, steps, and body weight) in a 6-month fully digital weight loss intervention for adults with overweight or obesity. The study used a 2³ full factorial design (ie, 2 × 2 × 2), whereby each participant was randomized to 1 of 8 experimental conditions that represent all possible combinations of these self-monitoring strategies. For each strategy, half of the participants were assigned to receive it (“Yes”) while the other half did not receive it (“No”; see Table 1). All participants received a core intervention that included weekly lessons and action plans for promoting healthy eating and physical activity. We recruited 176 participants (22 per condition). The trial was preregistered in February 2022 at ClinicalTrials.gov: NCT05249465. The trial registry includes the study protocol [57]. No patient involvement occurred, though study procedures and intervention content were adapted from the principal investigator’s (PI’s) prior digital weight loss trials (ie, GoalTracker [36], Spark Pilot [47], and Ignite Pilot [58]) that gathered qualitative participant feedback on perceived satisfaction and difficulty of intervention components, feasibility of study procedures, and suggestions for improvement.

The inclusion and exclusion criteria are listed in Textbox 1.

The study procedures are depicted in Figure 1.

After the end of the intervention, a subset of participants were invited to participate in a one-to-one semistructured qualitative interview, held via Zoom videoconference. Interviews lasted 30-45 minutes. Using an interview guide, trained research team members asked participants about factors that impacted self-monitoring engagement, perceived helpfulness, and satisfaction of the intervention components (eg, goals, self-monitoring, lessons, action plans, and progress reports), contributors and barriers to weight loss, perceived changes in self-efficacy, and to what extent self-monitoring contributed to weight loss. To contextualize and tailor their questions, interviewers integrated each participant’s engagement and weight loss outcomes as well as their 6-month acceptability data. Example questions included “You tracked your food [x%] of days over the 6-month Spark program. What factors contributed to tracking that amount?” and “You rated tracking your food as [“not at all” up to “very”] helpful. What contributed to feeling this way?”

Interviews were recorded and then transcribed verbatim using a professional transcription service. Memos were used by interviewers to reflect and record emerging themes. Data saturation was achieved once no new themes emerged from the interviews, at which point we stopped recruitment for the interviews. Transcripts will be coded using an iterative team-based approach to reduce potential bias. A cloud-based software platform (Dedoose; SocioCultural Research Consultants, LLC) will be used to facilitate the coding process. A codebook was created both deductively (derived from the PI’s experience and from the research questions) and inductively (derived from an initial set of transcripts). Data will be analyzed using thematic analysis. A mixed methods approach will be used to integrate the data [118,119]. providing insight into how self-monitoring engagement and weight loss are or are not related, and if so, why (ie, mechanisms). Joint display tables will be used to depict how the qualitative data provide context to the quantitative outcomes from the trial [120]. The design and reporting of the qualitative study will adhere to the COREQ (Consolidated Criteria for Reporting Qualitative Research) checklist [121].

Unlike an RCT that compares the means of each arm to one another (1 vs 2 vs 3, and so on), in a factorial trial, the means across combinations of conditions are used to test main effects of components and their interactions [122]. In our 2³ full factorial trial (ie, a 2 × 2 × 2 with 8 conditions), we are testing whether 3 self-monitoring strategies, alone or in combination, improve weight loss at 6 months (our primary time point). To estimate the main effect of each self-monitoring component on weight change, we will examine the difference between the mean of conditions where the target component is present and the mean of conditions where the target component is absent [54]. For example, for “Tracking Diet,” the mean of experimental conditions 5, 6, 7, and 8 (ie, the component is present) will be compared to the mean of conditions 1, 2, 3, and 4 (ie, the component is absent; see Table 1). We will use effect coding (ie, –1 when a component is absent and 1 when a component is present) for analyzing the main effects and interactions of the components, as recommended for factorial designs [54]. Thus, a main effect is interpreted as the effect of a factor averaged across all other factors. Since each main effect and interaction is based on the entire sample, the factorial design is considered efficient and economical [123].

Using a power calculation, we determined that a sample size of 176 participants would be required in our factorial trial to detect a minimum clinically important difference of 2.4 kg at 6 months, which is associated with improvements in cardiometabolic risk factors [59]. We assumed an SD of 4.5 kg, based on prior 6-month weight loss outcomes in digital interventions [49,124,125]. Thus, we sought to detect a main effect or interaction effect of Cohen d=0.53. Specifically, assuming 80% power, an α of 5% in a 2-sided test, and 35% attrition at 6 months (a conservative estimate based on prior trials of fully digital weight loss interventions [22,36,50,126]), a sample size of 172 is needed. To obtain an equal number of participants in each treatment condition, we increased the sample size to 176 total participants (22 per condition). Allowing for an equal number of participants per condition permits the tests of the main effects and interactions to be uncorrelated. Even if unequal sample size across conditions occurs due to differential dropout resulting in minor imbalance, the correlations between effects should be small [122]. This power analysis was computed using the FactorialPowerPlan SAS Macro (version 1.0) [127].

Our trial uses a 2³ factorial design since it has 3 factors, each with 2 levels (Yes and No). In these types of 2^k factorial trial—those with k number of factors—all else being equal, the power for detecting a β-weight corresponding to an interaction effect is the same as that for detecting a β-weight corresponding to a main effect when using effect coding to represent the levels of the factors [54]. Thus, no additional participants are needed to detect interactions in 2^k factorial trials.

Our primary outcome is weight change, in kilograms, from baseline to 6 months. Using intent-to-treat principles, we will use linear mixed models to examine the main effects of the 3 self-monitoring components and all of their interactions on mean weight change at each follow-up time point. This model will include fixed effects for the following: time point (1, 3, or 6 months), sex assigned at birth (ie, the stratification variable), self-monitoring component, pairwise interactions of each component, and the time-by-component and time-by-pairwise interactions. We will fit an unstructured covariance matrix to take into account the correlation of weight measures within a person across time. We will use maximum likelihood to estimate the model parameters, allowing for missingness at random. Data will also be disaggregated by sex in exploratory analyses.

The PI was responsible for regular monitoring of the data, in accordance with the Data Safety and Monitoring Plan. Issues of safety were reviewed with the independent safety monitor, who was not involved in the study’s design or conduct. If they occurred, major protocol deviations were reviewed by all investigators. The PI and study statistician will have access to the final dataset.

Using the MOST framework for intervention optimization, this study seeks to balance intervention effectiveness (ie, weight loss) against efficiency (ie, minimizing participant burden). Thus, our “Optimization Objective” is to create an intervention made up of all active components without any inactive components. To decide which self-monitoring strategies should comprise our newly optimized intervention, we will use the component screening approach described by Collins [54], which iteratively sorts through the results of the factorial trial. A minimum mean weight loss threshold of 2.1 (SD 4.5) kg at 6 months will be used to decide whether to consider including any self-monitoring components alone or in combination. This magnitude equates to an α of .10. We selected a higher alpha level during this decision-making phase when constructing the newly optimized intervention because we want to lower the chance of mistakenly discarding an intervention component that is actually beneficial (ie, reduce the type 2 error rate). This rationale reflects the “decision-priority perspective” of MOST. In contrast, the “conclusion-priority perspective” is focused on drawing scientific conclusions from a trial using well-established conventions of α level ≤.05 [54,55].

Using the component screening approach, first, the main effects of each self-monitoring component on 6-month weight change will be determined. The self-monitoring strategies that meet or exceed this threshold will be tentatively included in the “screened-in” set while those not meeting this threshold will be included in the “screened-out” set. Then, lower-level interactions will be examined, first among any components in the screened-in set, followed by those in the screened-out set. Next, we will examine the interaction among all 3 self-monitoring components. For interactions meeting or exceeding the 2.1 kg threshold of weight loss, their components will be considered for inclusion in the screened-in set (even if the components, by themselves, did not demonstrate main effects). To aid in interpretation of interactions, we will plot the predicted marginal means. This visualization will provide information on whether synergistic or antagonistic interactions exist. Results of this trial will inform development of a newly optimized intervention that can be tested in a subsequent evaluation-RCT. If no self-monitoring strategies, alone or in combination, exceed our 2.1 kg weight loss threshold, then we will return to the preparation phase of the MOST framework to refine the conceptual model, brainstorm and pilot test new candidate components, adapt existing components, or strengthen the core digital health intervention.

All study procedures and human subjects research ethics were approved by the Stanford University Institutional Review Board (protocol number 64716; approval date: March 28, 2022). The participants provided written informed consent via electronic signature before enrollment. They were informed that participation is voluntary, and they could withdraw or opt out at any time. The consent form is available on the trial registry [57]. Participants were compensated a maximum of US $60 (via electronic gift cards) for their completion of assessments, as follows: US $20 at 3 months, US $30 at 6 months, and an additional US $10 for completion of all 4 dietary recalls (2 each at baseline and 6 months). Those who participated in the qualitative interviews received an additional US $25. Deidentified data will be used in study analyses and in disseminated materials.

The Spark trial will provide the first set of evidence on optimizing self-monitoring in a weight loss intervention for adults with overweight or obesity. It is common for behavioral obesity treatment to include self-monitoring of diet, physical activity, and body weight, yet it is not actually known whether all 3 of these components are necessary for weight loss or whether any are ineffective, or even detrimental. Removing any inactive components would help to minimize patient burden and effort. Using an intervention optimization framework (MOST) enables our team to build an effective and efficient fully digital weight loss intervention, which is much needed given the high prevalence of obesity, its detrimental health consequences, and the limited scalability of existing weight loss interventions. Further, to our knowledge, our trial will be the first to empirically test the impact of a self-directed web-based orientation session on trial retention. If deemed effective, this orientation session may be an affordable, brief, and scalable strategy that could be easily adapted and embedded in behavioral intervention research to enhance the validity of trial outcomes.

Several limitations exist. First, while our trial’s primary outcome is weight loss, it is possible that a component may have limited impact on weight but still improve diet quality, physical activity, or health-related quality of life. The MOST framework provides flexibility in interpreting the data in different ways depending on the Optimization Objective of interest. Thus, the intervention could be optimized for a different outcome, which would involve simply reinterpreting the trial’s existing data in light of that outcome. Second, due to funding constraints, we are unable to collect data beyond the 6-month time point, thus precluding the examination of long-term weight loss maintenance, which could be a focus of future research. Third, it is unknown whether results would generalize to digital weight loss interventions that include direct human counseling. However, by design, our trial focused on fully digital interventions (without such counseling) since they offer an opportunity to reach broad populations, including those with limited access to high-quality obesity treatment, and to deliver a weight loss program in a more scalable manner. Fourth, due to the behavioral focus on the intervention, participants are not masked to factorial condition, which could introduce bias. We aimed to establish clinical equipoise during our discussion of the various self-monitoring components at the baseline visit. Fifth, due to technical limitations, the Fitbit app is not able to reflect participants’ adaptive step goal each week. Therefore, the step goal feedback provided by the app likely differed than the feedback provided via our weekly progress reports. Sixth, our embedded experiment that is testing the effect of a self-directed orientation session may be underpowered to detect small differences in retention due to sample size constraints of the Spark trial. This is an inherent limitation of embedded experiments [67], yet given their simpler and less resource-intensive nature, they are designed to be replicated across multiple clinical trials to build up the evidence base on potential retention-promoting strategies. These data can then be evaluated together in meta-analyses.

From a research standpoint, we hope that our findings will shed light on which self-monitoring strategies should be included in behavioral obesity treatments. Considering MOST’s continual optimization principle, interventions should be continuously improved upon in an iterative fashion to enhance outcomes. Future research could seek to optimize other components of fully digital interventions, such as type of feedback, skills training materials, and gamification approaches, and could test artificial intelligence–driven strategies for tailoring intervention content based on participant preferences, needs, and treatment response. From a clinical standpoint, we hope this work will inform clinicians who provide obesity counseling as to which self-monitoring strategies they should be recommending to their patients who are seeking to lose weight in a standalone manner.

The Spark trial leverages an intervention optimization framework to understand whether self-monitoring diet, steps, or body weight maximizes weight loss, alone or in combination, which addresses a critical research gap. Ultimately, building a fully digital intervention comprised of only clinically meaningful self-monitoring strategies has potential for broad public health impact in providing scalable, potent, low burden, and far-reaching weight loss interventions. If the optimized intervention is effective, it could serve as a first-line weight loss treatment for adults with overweight or obesity.