Critical Appraisal of Healthcare Resource Use and Costs in Digital Weight Loss Intervention Participants vs. Nonparticipants

We read with interest the article by Tak et al. on health care resource utilization (HRU) and costs among participants and nonparticipants in a digital weight loss program (Noom) [1]. The study adds to a growing body of evidence on digital weight management. However, we noted several methodological and interpretative concerns. Specifically, we wish to highlight issues of inadequate propensity score balance on pre-index HRU, selection bias due to engagement-based cohort definitions, reliance on claims data with possible misclassification, and short follow-up and arbitrary time windows. Together, these limitations may overstate the reported benefits of the intervention. First, while the authors state that pre-index HRU was included in the matching, Noom participants still had significantly higher outpatient visits, ED visits, and prescriptions before enrollment. This imbalance raises concerns about residual confounding and regression-to-the-mean, especially since the apparent post-index improvements may largely reflect return from higher baseline use [2]. The absence of standardized mean differences for HRU further undermines claims of covariate balance. Second, restricting the intervention arm to “engaged” Noom users introduces strong self-selection bias. Individuals who enroll and remain active likely differ systematically in motivation, socioeconomic status, and health-seeking behaviors compared with the broad pool of nonparticipants. This risks conflating program effects with participant characteristics [3]. Third, reliance solely on claims data may misclassify key conditions. For instance, obesity was coded in only 1.1% of Noom users and 0.4% of controls, figures far below the expected prevalence, suggesting undercoding. This not only underrepresents high-risk groups but also questions generalizability. Moreover, pre-index inpatient data were “unavailable,” leaving critical gaps in baseline comparability [4]. Finally, the 6-month pre/post design is too brief to capture durable trends in utilization or costs. Seasonal variation and natural fluctuations in health care use may bias findings, and the lack of sensitivity analyses with longer follow-up further weakens causal inferences [5]. Future research should build on this work with stronger designs. Reporting covariate balance on all matched variables, conducting difference-in-differences analyses, and testing robustness over longer follow-up would strengthen validity. Importantly, intention-to-treat analyses that include all enrollees, regardless of engagement, would provide a more realistic estimate of program effectiveness. Incorporating clinical data (e.g., BMI, weight trajectories) alongside claims could mitigate misclassification and enhance external validity. While digital interventions like Noom hold promise, rigorous methodology is essential to ensure cost-saving claims are not overstated. The authors declare no conflicts of interest. All data are available in this published article.