This study used a prospective single-arm convergent parallel mixed methods design [16] in which quantitative and qualitative data were collected simultaneously, and with equal priority, analyzed independently, and subsequently integrated at the interpretation stage to obtain a more complete understanding of PNA usability and treatment satisfaction than either method could provide alone. This design was selected because evaluating the usability and treatment satisfaction of a novel prototype requires both the breadth of standardized quantitative measurement and the depth of qualitative exploration to capture the contextual and experiential dimensions of clinical implementation that validated scales alone cannot adequately reflect [17,18].

This trial enrolled overweight adults (BMI ≥28 kg/m² before GLP-1 RA initialization) who had achieved ≥5% weight loss with GLP-1 RA to support ongoing weight management over 12 consecutive weeks of PNA-assisted remote nutritional counseling using the Oviva lifestyle-tracking app [19]. Remote nutritional counseling was delivered by 6 certified dietitians representing the full complement of practitioners within the program.

Participants were consecutively referred via the dietitians from the outpatient weight specialist clinic at the University Hospital Bern between August 2023 and May 2024. Inclusion criteria were being aged ≥18 years; having a minimum of 16 weeks of GLP-1 RA therapy in combination with remote nutritional counseling and use of the lifestyle tracking app, ensuring participants had passed the initial rapid, predominantly drug-induced weight loss phase; and achieved a clinically significant weight loss of ≥5% (for baseline BMI 28‐35 kg/m²) or ≥7% (for baseline BMI ≥35 kg/m²) at the time of enrollment. The 16-week minimum also ensured that sufficient historical data from the lifestyle-tracking app were available to inform the PNA tool’s energy balance model from day 1 of the intervention. Exclusion criteria were the inability to provide written informed consent; any physical or psychological condition likely to interfere with study conduct or interpretation of results; self-reported pregnancy, planned pregnancy within 3 months, or breastfeeding; concomitant participation in another interfering trial; and insufficient proficiency in German (as this study’s related questionnaires were only available in German).

Of 105 individuals screened for eligibility, 81 met the inclusion criteria and were enrolled. The sample size was calculated for the primary patient outcome (treatment satisfaction), assuming a proportion of P=.20, a margin of error of 0.1, and a 95% confidence level (z=1.96), yielding a required sample size of 62 completed questionnaires. Accounting for an anticipated dropout rate of 20%, a total of 80 participants were planned for recruitment.

This study was approved by the Ethics Committee of Bern (2023‐00864) and registered at ClinicalTrials.gov (NCT05997771). All participants provided written informed consent before any study-related procedures. All data were deidentified, and no participant compensation was provided.

The PNA is a prototype clinical decision-support tool designed to visually present patients’ current dietary patterns, energy balance, and body weight trends to enhance nutritional counseling and deliver actionable, individualized recommendations aligned with each patient’s weight goals and nutritional needs. The PNA integrates 3 primary data sources: frequent body weight measurements, logged food intake, and physical activity data from a lifestyle-tracking app. These inputs feed into a mathematical model that forecasts weight trajectories by estimating energy intake and expenditure [7].

To facilitate interactive use, the PNA prototype was implemented as an interactive web-based dashboard via Google Looker Studio, visualizing model outputs through six integrated panels (Figure 1):

To support realistic and individualized weight goal setting, the PNA enables adjustment of the output estimation timeframe via the dashboard’s upper control bars. As a PDF printout of the PNA output could be distributed to patients via email by their dietitian, a clear disclaimer indicating that the information is to be reviewed by a health care professional only and must not replace professional medical or dietary advice was included.

The dietary data outputs and visualization components of the PNA were developed through an iterative expert review process involving senior registered dietitians with dedicated prototype development roles and without involvement in the prospective usability evaluation. Feedback was systematically incorporated across multiple development cycles to ensure the clinical plausibility and interpretability of the tool’s dietary pattern assessments and recommendations. Formal criterion validation against an established dietary assessment reference method was not conducted.

One month before this trial’s period, all dietitians received comprehensive training on this trial’s purpose and objectives, including a 1-hour hands-on demonstration of PNA system navigation and functionality delivered by the research team. Following the training, all dietitians had access to the PNA dashboard to get used to it before the intervention started. To support ongoing reference and skill development, dietitians were provided with multiple supplementary training resources, including a 20-minute video tutorial of how to use the dashboard, a comprehensive procedure manual, and a condensed 1-page reference guide. No formal competency assessment was conducted following training, but dietitians were encouraged to consult the research team for technical queries or other PNA-related questions arising before the intervention started but not during this trial to avoid outcome interference by the research team.

The PNA prototype was evaluated within the established remote nutritional counseling framework provided by Oviva, which consists of a professional platform enabling certified dietitians to access insights from their patients’ app-based lifestyle-tracking (eg, weight, nutrition, and activity logs) and provide feedback through various communication channels (video calls, phone calls, and secure messaging).

The lifestyle-tracking app for patients enables dietary intake tracking using photographs or text entry, physical activity monitoring (entered manually or synchronized via connected devices such as Apple Health, Google Fit, or Fitbit [Google LLC]), and body weight measurements (entered manually or via connected TRANSTEK Teleweight [Guangdong Transtek Medical Electronics Co, Ltd] scales) [19,21]. Food items in photographs were automatically recognized and assigned 1 of 117 predefined food tags by a proprietary deep-learning image recognition algorithm that has been quality validated previously [22]. Automatically assigned food tags could be discarded, confirmed, or modified by users. Only user-confirmed tags were incorporated into the PNA model. Participants were encouraged to track body weight at least once weekly using connected scales or manual entry.

To access the PNA dashboard, dietitians were required to use an access-restricted link to Google Looker Studio. They were encouraged to access the dashboard during both the preparation and delivery of each nutritional counseling session. Dietitians had the option to share the PNA report with patients in PDF format via email when they determined that patients would benefit from having access to their personalized nutritional insights. Dietitians were asked to track their time of PNA usage (including time used for both preparation and consultation) for each patient within a study-specific document. To preserve real-world conditions, dietitians worked at their typical pace without additional time accommodations to discuss or share the PNA outputs.

Patient sociodemographic and socioeconomic characteristics were collected at study initiation via an online questionnaire. Medical history, comorbidities, and anthropometric measurements, including body weight at the commencement of GLP-1 RA therapy, along with treatment-specific information, were extracted from electronic medical records. All available body weight measurements, dietary intake logs, and physical activity data (both pre and during the PNA trial period) were automatically extracted from the lifestyle-tracking app and used for the weight prediction model and descriptive analysis of app-engagement. Descriptive information regarding remote counseling sessions during this study’s period was retrieved from the digital platform used by the dietitians.

Using the validated Healthcare Systems Usability Scale (HSUS), dietitians assessed the PNA’s usability upon study completion (online). Items 2 and 22 were excluded as they were not applicable to this study’s scope. Each item was evaluated using a 5-point Likert scale (1-5), where higher scores indicate greater agreement. We calculated the mean score, SD, and minimum-maximum ranges for each of the 4 subscales and the overall scale score, which represents the mean of the 4 subscale scores. Additionally, we computed the usability percentage score, calculated as the relative value of the overall score in relation to the theoretical maximum score of the 19 items multiplied by 5 [23].

Patient treatment satisfaction was partially assessed by the German language validated adult version of the Nutritional and Dietetic Patient Outcomes Questionnaire (NDPOQ-A) [24] completed by the patients online before and after 6 weeks and 12 weeks of PNA-assisted remote nutritional counseling. The 15-item NDPOQ-A uses a 5-point Likert scale ranging from 0 to 4 points, where higher points indicate greater agreement with statements concerning treatment satisfaction. The total questionnaire score ranges from 0 to 60. The overall mean score and SDs were calculated for all assessment time points.

Additionally, patients’ perceived impact of the PNA on health behavior was assessed after 12 weeks of PNA-assisted remote nutritional counseling using the “perceived impact” subscale of the user version of the Mobile Application Rating Scale (uMARS) [25]. This subscale consists of 6 items rated on a 5-point Likert scale, with item 5 concerning help-seeking excluded as it did not apply to this study’s scope, resulting in a total score range of 5 to 25. Mean scores, SDs, and minimum-maximum ranges were calculated for individual items and the total subscale score.

Patient-reported general self-efficacy was assessed using the validated General Self-Efficacy Scale (GSE) from Imperial College London [26] before and after 6 weeks and 12 weeks of PNA-assisted remote nutritional counseling. The GSE comprises 10 items rated on a 4-point Likert scale ranging from 1 to 4, yielding a total score range of 10 to 40, where higher scores indicate greater self-efficacy. Mean scores and SDs were calculated for all assessment time points.

Qualitative feedback regarding the PNA dashboard was collected using a self-designed open-ended feedback questionnaire administered to both patients and dietitians online at study completion. The questionnaire assessed content quality, perceived usefulness, and visualization effectiveness of the 6 PNA dashboard panels.

Two focus groups explored PNA prototype experiences: all 6 dietitians discussed tool usability and utility in semistructured interviews, while 6 patients (selected from those who provided additional written informed consent for focus group participation and were available at the scheduled interview date at the end of this trial) shared experiences with PNA-assisted remote nutritional counseling. The semistructured interview guides are presented within Multimedia Appendix 1.

All interviews (duration of 60 min each) were conducted remotely and recorded via Zoom (version 6.2.11, Zoom Video Communications, 2024) by a member of the research team (MS) and were subsequently transcribed verbatim for qualitative analysis by an independent member of the research team.

Verbatim transcripts from focus group interviews and responses to open-ended feedback questionnaires were analyzed using a thematic analysis approach [27] applying a hybrid deductive-inductive methodology [28]. This qualitative strand was conducted within a pragmatic paradigm, in which the approach and methods were selected based on their utility for contextualizing and explaining quantitative outcomes rather than adherence to a single philosophical tradition consistent with the integration goals of the convergent mixed methods design [16].

The first author independently coded all transcripts using the MAXQDA software (version 24, VERBI GmbH) across 4 iterative rounds of coding. Consistent with a deductive orientation at the thematic level, an initial set of themes was defined a priori based on this study’s conceptual framework and the domains covered by the interview guides and feedback questionnaires: (1) “prior expectations and concerns,” (2) “perceived utility and satisfaction,” (3) “usability challenges,” and (4) “suggestions for clinical integration.” Within this thematic structure, codes were developed inductively from the transcript data, allowing the specific content and nuance of participant responses to shape the coding and subthemes without being constrained by predetermined categories. Subsequent coding rounds involved reviewing, refining, and consolidating codes to ensure consistency and analytical depth. Analytic decisions, including code definitions and revisions, were documented as memos within MAXQDA throughout the process.

Peer discussions to enhance the trustworthiness of the analysis were conducted in two complementary settings: a series of qualitative analysis workshop sessions involving 5 external faculty members, and multiple internal research team meetings, during which coauthors provided critical appraisal of emerging codes and subthemes and their interpretive decisions. A dedicated mapping session was subsequently conducted to organize finalized codes into subthemes within the 4 a priori themes, with the resulting code and subtheme structure reviewed and agreed upon by the coauthors.

It is acknowledged that the primary analyst (FO) is a member of the research team involved in PNA development, which may have introduced familiarity bias in the interpretation of dietitian and patient feedback. This was mitigated through the iterative coding and regular peer discussions, including members not directly involved in the tool development described above.

Quantitative and qualitative data were analyzed independently by the first author before being brought together in a convergent synthesis. Integration occurred at the interpretation stage through a narrative comparison of quantitative outcomes and qualitative theme content, with discordances and convergences between strands used to inform a more comprehensive understanding of PNA usability and implementation.

All primary analyses were conducted using complete cases with available data for each outcome.

Descriptive statistics for patient characteristics and lifestyle behavior include means, SD, frequencies, and percentages. For summarizing PNA-assisted consultations, means, SDs, medians, and minimum-maximum ranges are reported.

Repeated-measures ANOVA was used to assess differences in total questionnaire scores between time points (at intervention start, after 6 weeks, and after 12 weeks of PNA-assisted nutritional counseling). Paired t tests evaluated pre-post changes in percent weight change and BMI following PNA-assisted remote nutritional counseling.

For the descriptive report of engagement with the lifestyle-tracking app, all participants with available data were included, regardless of questionnaire completion. To ensure data quality of body weight measurements collected via the lifestyle-tracking app, recorded weights outside the range of 40‐400 kg were excluded. To address artifacts inherent to self-reported and smart-scale data (eg, synchronization errors from third-party use), subsequent body weights were also excluded if the relative change from the preceding measurement exceeded a predefined plausibility threshold. Specifically, a measurement was excluded if the proportional deviation from the previous weight exceeded 4% plus 1% per week elapsed between the two measurements. These parameters were validated by visual inspection to remove obvious outliers while accommodating natural fluctuations and rapid weight loss.

To assess the robustness of the complete-case analyses, sensitivity analyses using multiple imputation by chained equations were conducted for outcomes with participants lost to follow-up (LTFU). To explore the plausibility of the missing completely at random (MCAR) assumption, baseline characteristics between LTFU participants and completers were compared, using Wilcoxon rank-sum tests for continuous variables and Fisher exact test for binary categorical variables. For categorical variables with more than two levels, P values were estimated by Monte Carlo simulation (10,000 replicates). For the NDPOQ-A treatment satisfaction score and GSE self-efficacy score, missing time point values were imputed using predictive mean matching with observed scores at other time points serving as mutual predictors (mean 50 imputations, 10 iterations). As repeated-measures ANOVA cannot be directly applied to multiply-imputed data, the imputed questionnaire analyses used linear mixed models with time point as a fixed effect and a random intercept per participant, with pooled estimates derived across imputed datasets following Rubin Rules. For body weight, missing measurements ±15 days from the 12-week study end date (day 84 relative to the intervention start date) were imputed using predictive mean matching with weight values reported within the time frame of ±15 days of intervention day 0, mean mid-study weight (days 20‐65), number of weight recordings, age, BMI at treatment start, sex, and GLP-1 RA status as predictors. All imputation analyses assumed data were missing at random. Complete-case and imputed results are presented in parallel; concordance between the two approaches was used to assess the potential impact of missing data on study conclusions.

All statistical analyses were performed using R (version 4.3.0; R Foundation for Statistical Computing).

Quantitative and qualitative data were collected concurrently and analyzed separately before being integrated at the interpretation stage.

A total of 81 adults were enrolled to receive PNA-assisted remote nutritional counseling over 12 weeks. Three participants were excluded from the analysis as they did not complete any study-related questionnaires at the intervention start, resulting in 78 participants (58/78, 74% women) included in the analytic sample. In total, 73 participants completed the primary treatment satisfaction outcome questionnaire, exceeding the prespecified minimum sample size. Loss to follow-up for individual outcomes is detailed in Figure S1 (Multimedia Appendix 1).

At enrollment, participants had a mean BMI of 32.7 (SD 4.4) kg/m² and had achieved a mean weight loss of −13.3% (SD 6.7%) after 10 (SD 5) months of GLP-1 RA therapy. All participants were receiving liraglutide (Saxenda) as their GLP-1 RA regimen, with daily doses ranging from 0.6 to 3.0 mg. Key patient characteristics at the time of study enrollment are summarized in Table 1.

During the 12-week study period, participants received between 3 and 5 remote consultations. Consultation modalities included secure chat-based sessions (228/360, 63%), video calls (71/360, 20%), and phone calls (61/360, 17%). The intervention was delivered by 6 dietitians, each managing between 3 and 30 patients (median 7, IQR 11, patients per dietitian). PNA usage time per consultation ranged from 2 to 12 minutes. Dietitians shared PNA-generated outputs with participants in 50% (180/360) of consultations.

During this trial’s period, participants recorded a mean of 12.7 (SD 8.5) meal and/or beverage logs per week, corresponding to an average of 30.8 (31.8) food tags weekly. The mean number of activity logs per week was 2.7 (SD 3.3), with 90% (2.4/2.7) of these entered manually. Weekly self-recorded body weight measurements average 1.2  (SD 1.0). Compared to the weeks before receiving PNA-assisted nutritional counseling, meal frequency logging decreased while the mean number of food tags per week remained stable during the intervention (Table S4 in Multimedia Appendix 1).

Body weight and BMI remained stable over the 3-month study period, with no significant changes observed (−0.2%, SD  2.5%, P=.56, for body weight; –0.2%, SD  3.4%, P=.64 for BMI). Multiple imputations of LTFU did not have an effect on the results (Table S2 in Multimedia Appendix 1). MCAR tests comparing baseline characteristics of participants with or without weight follow-up data confirmed the MCAR assumption (Table S3 in Multimedia Appendix 1). Due to supply chain disruptions affecting Saxenda availability, 21% (16/78) of participants discontinued treatment during this study’s period. Percent weight loss over the intervention period stratified for GLP-1 discontinuation or continuation is displayed in Table S3 (Multimedia Appendix 1).

The HSUS questionnaire indicated moderate overall usability for the PNA (62%, range: 45%‐67%), with higher scores for ease of use and user control, and lower scores for patient safety as well as decision and work effectiveness (Table 2). When asked whether they would recommend the PNA to a colleague, 50% (3/6) responded “yes,” and 50% (3/6) responded “no.”

Patients reported a mean perceived impact score of 19.1 of 25 (SD 3.5) on the uMARS scale at the end of the intervention (Table 3). Patient-reported treatment satisfaction assessed by the total raw score of the NDPOQ-A (Table S6 in Multimedia Appendix 1), did not change before and after PNA-assisted nutritional counseling (41.8, SD 7.3, vs 41.2, SD 8.4, P=.78). Similarly, no significant change in patient-reported self-efficacy was observed on the GSE questionnaire (21.8, SD 4.5 vs 22.0, SD  4.0; P=.57, see Table S6 in Multimedia Appendix 1). Sensitivity analysis, including MCAR comparisons of baseline characteristics of the participants with or without LTFU of the treatment satisfaction outcome, followed by multiple imputations, showed no difference from the complete case analysis (Tables S5 and S6 in Multimedia Appendix 1). When patients were asked if they would recommend PNA-assisted remote nutritional counseling to a friend with a similar condition, 78.2% (n=61) responded “yes,” 9% (n=7) “no,” 3.8% (n=3) “I do not know,” and 7.7% (n=6) did not respond.

Integrating the quantitative and qualitative strands detected both convergence and divergence. Convergence was observed in dietitians’ usability assessments: moderate overall HSUS scores (62%), with lower ratings for patient safety, decision and work effectiveness, aligned with qualitative accounts of limited trust in lifestyle advice panels due to data completeness concerns, insufficient algorithmic transparency, and inadequate workflow integration. Both strands independently identified incomplete understanding of tool outputs as a key adoption barrier.

Divergence was identified for treatment satisfaction and self-efficacy, where quantitative measures showed no significant postintervention improvement, contrasting with qualitative reports of meaningful subjective progress, increased tracking motivation, and enhanced dietitian communication. This divergence is likely partly attributable to the limited dashboard exposure, as the PNA was shared with only 50% (39/78) of patients.

Partial convergence was observed for perceived patient impact: a mean uMARS score of 19.1/25 (SD 3.5) and 78.2% (61/78) willingness to recommend PNA-assisted counseling were broadly consistent with qualitative reports of patients valuing weight trajectory visualizations and personalized dietary feedback, though qualitative data added that motivational impact was feature-dependent and that direct dashboard access represented a universally expressed unmet need not captured quantitatively. The qualitative had reported increased motivation for app-tracking during the PNA intervention were not reflected quantitatively (Table S4 in Multimedia Appendix 1).

In this study, we evaluated the usability, treatment satisfaction, and weight effects of the PNA, a prototype decision-support tool designed to facilitate dietitian-led remote nutritional counseling for personalized weight management in adults achieving significant weight reduction with GLP-1 RA therapy.

Given the growing recognition that personalized nutrition is critical for patients on GLP-1 RA therapy to maximize health outcomes, prevent nutritional deficiencies, manage side effects, and support sustainable weight management [9], tools such as the PNA could help address important needs in this area, such as providing quantitative and qualitative advice based on app-based lifestyle tracking.

Overall, PNA-assisted counseling was feasibly delivered within routine care, and patients maintained clinically significant weight loss over the 12-week intervention. The mixed methods design of this study allowed for a more nuanced interpretation of these findings than quantitative measures alone would permit. Notably, with respect to the patient satisfaction and self-efficacy findings, a discordance was identified between quantitative null findings and positive qualitative signals, which is a well-recognized phenomenon in mixed methods usability studies of novel health technologies [29,30]. This underscores the limitation of applying conventional efficacy metrics to prototype evaluations and illustrates the added value of a convergent parallel mixed methods approach in capturing the full complexity of technology implementation. The limited PNA dashboard exposure to 50% (39/78) of patients likely attenuated measurable effects, and the universal patient desire for direct dashboard access represents a clear priority for future development despite being designed as a clinical decision support tool for dietitians as primary users.

Moreover, our qualitative findings recognized pronounced feature-specific variations in dietitians’ assessment of the PNA’s utility, highlighting the need for targeted improvements across different tool components. The weight prediction panel was identified as the most valued and frequently used feature, with dietitians successfully integrating it into both consultation preparation and real-time patient interactions to support both weight loss and maintenance phases. This feature’s success appears driven by its primary dependence on continuous weight tracking—data that patients found comparatively low-burden to record consistently (1/wk)—combined with its ability to provide concrete, motivational visualizations that patients could readily understand and engage with. While the prediction model also incorporates food and activity data to refine trajectory estimates, its robustness to incomplete dietary logging meant that clinically meaningful predictions remained accessible even when food tracking adherence was suboptimal. In contrast, panels whose outputs were more directly dependent on complete food and activity tracking data encountered greater skepticism with dietitians citing incomplete data inputs, limited food tag options, insufficient algorithmic transparency, and the model’s inability to account for portion sizes or body composition changes were identified as key concerns undermining their trust.

Technical implementation barriers, including the prototype’s separate hosting platform, manual data entry requirements (of target weights and weight surveillance timelines), and the incomplete understanding of the PNA outputs, particularly regarding the underlying algorithms and appropriate interpretation of predictions, further constrained adoption and limited the perceived clinical utility. The latter knowledge gap may have prevented effective communication of results to patients and reduced confidence in tool recommendations, creating a barrier to adoption even when data quality was adequate.

These findings align with established frameworks for digital health adoption barriers, which identify technical, organizational, and user-level barriers as key determinants of implementation success [29-31]. The PNA’s integration into clinical practice will require iterative codevelopment with dietitians and patients, seamless embedding within existing digital platforms, and structured onboarding that builds algorithmic literacy and clinical confidence in tool outputs. Previous research has shown that apps and platforms developed in collaboration with dietitians are more trusted and accepted by both professionals and patients [32,33]. Our observations also highlight the need for sufficient time for training and adaptation when adopting new technology, including the provision of continuous training materials [30,34].

App engagement, including regular tracking of lifestyle behaviors and body weight, is essential not only for accurate weight prediction [5,35] but also for optimizing outcomes during GLP-1 RA–based weight management therapy, as recent evidence demonstrates that greater engagement with digital platforms significantly enhances weight loss [36]. The NICE (National Institute for Health and Care Excellence) in the United Kingdom also recognizes the important role of digital tools in weight management and provided guidance on their integration into clinical practice [37]. According to this guidance, digital aids should promote weight goal achievement through ongoing engagement, personalized feedback, and tailored advice. While the PNA was designed to meet these requirements, bringing it into daily use will require further refinement and robust efficacy evaluation in its final form.

Several technical enhancements could substantially increase the PNA tools’ utility. Technical improvements, such as the use of computer vision and depth-sensing technologies for quantitative nutrient analysis, enabling 3D reconstruction of food portions (rather than relying on standard serving size assumptions) [38], as well as mobile voice-added apps for food intake reporting [39], are promising. Additionally, large language models offer new opportunities to refine nutritional recommendations based on individual profiles (eg, recipe suggestions tailored to specific requirements) [40,41]; however, optimization of nutritional composition estimations is still required [42]. Further development should include options for direct patient interaction with the tool and prompts emphasizing that consistent lifestyle tracking improves the quality and accuracy of support provided. In the future, our energy balance model could incorporate individual body composition data from connected smart scales, moving beyond generic assumptions and further enhancing the tool’s utility for optimizing both body weight and composition—key factors for long-term health and quality of life. Additionally, integrating self-reported GLP-1 RA dosages tracked within the lifestyle app into the PNA dashboard could support personalized dose adjustments based on predicted trends. Manufacturer-linked companions such as WeGoTogether (Novo Nordisk Inc) and Lilly Health (Eli Lilly and Company) offer comparable functionality, but remain unavailable outside of the United States, leaving a clear gap that the PNA could cover.

In the context of the nutritional priorities associated with GLP-1 RA therapy, several areas for further development of tools such as the PNA can be identified. To address the risk of micronutrient deficiencies under energy restriction, future iterations should incorporate monitoring of micronutrients such as B vitamins, calcium, vitamin D, iron, folate, and trace minerals such as zinc [9,12-15]. Specific dietary adjustments can further meaningfully reduce GLP-1 RA–related gastrointestinal side effects, especially during initiation and dose escalation [9]. Psychological factors such as disordered eating [43], depression [44], and quality of life [45] represent further dimensions that the PNA could eventually incorporate. Finally, just-in-time interventions using behavioral change techniques, triggered by detection of dietary lapses or deviations from target weight trajectories [46,47], also represent a promising avenue for future development.

Several limitations in this study’s design should be acknowledged. The single-arm trial design limits the ability to assess clinical efficacy; while our primary focus was on usability and treatment satisfaction, future health technology efficacy assessments of the finalized product should be conducted in randomized controlled trials. The dietitians involved (predominantly middle-aged women) may not represent the broader professional community, particularly younger practitioners who may be more receptive to digital tools. The requirement for a minimum of 16 weeks of prior GLP-1 RA therapy in combination with lifestyle-app tracking before enrollment may limit the generalizability of findings to individuals earlier in their treatment trajectory or those with limited prior engagement with lifestyle tracking, as participants had by definition already passed the initial weight loss phase and accumulated sufficient data to inform the PNA tool. The patient focus group sample may introduce a positive selection bias, as participation required additional written informed consent and evening availability, potentially overrepresenting more motivated patients. However, as open-ended feedback questionnaires, which formed an equally important qualitative data source, were completed by 76 of 78 patients, the overall qualitative findings are unlikely to be substantially affected by this sampling limitation. Finally, supply chain issues with GLP-1 RA led to abrupt drug discontinuation in at least 21% (16/78) of the participants, complicating the interpretation of intervention effects on weight. Although abrupt pharmacotherapy withdrawal is known to predispose to weight regain [10,11], this was not observed in our cohort. However, the absence of expected weight regain following pharmacotherapy discontinuation in our cohort cannot be definitely attributed to PNA-assisted nutritional counseling due to the lack of a control group.

Clinical decision support systems that integrate digital lifestyle tracking and advanced analytics may help address nutritional priorities during GLP-1 RA weight management. This could support scalable and adaptive dietitian-led nutrition care, given their limited time and capacity to analyze large, multidimensional data streams. Further development, rigorous user-centered design, and robust clinical efficacy testing are necessary to fully realize the potential of such systems to improve health outcomes.