Autonomous Digital Lifestyle Intervention for Obesity: A Comparative Study of Software-Generated vs. Provider-Delivered Body Composition Outcomes

Abstract

Background: Many lifestyle intervention apps for weight management rely on human coaching, which increases cost and variability, or use calorie-counting approaches that lack structured, individualized guidance. Fully autonomous digital platforms that deliver personalized nutrition and exercise plans with body composition tracking remain limited. Objective: To compare an autonomous, protein-focused digital lifestyle intervention (RightBMI App) with provider-delivered lifestyle intervention for weight loss and body composition outcomes in metabolic and bariatric surgery (MBS) candidates. Materials and Methods: This prospective quasi-experimental study enrolled 160 patients at a single center (April 2024-April 2025). Patients were sequentially allocated to the RightBMI App (n=80) or provider-delivered care (n=80). Outcomes (% total body weight loss [%TBWL], % fat mass loss [%FML], % visceral fat loss [%VFL], and % muscle mass loss [%MML]) were assessed over 24 weeks using Generalized Estimating Equations (GEE) with multiple imputation and inverse probability weighting, adjusting for age, Body Mass Index (BMI), and sex. A time-to-event analysis (Kaplan-Meier and Cox proportional hazards) was performed as the primary sensitivity analysis to address informative censoring due to MBS clearance at ≥10% TBWL. Results: In GEE models, the App group showed significantly greater %TBWL at weeks 12 (+2.64%, p<0.001) and 16 (+2.47%, p=0.027), and greater %FML at weeks 12 (+2.94%, p=0.028) and 16 (+4.20%, p=0.003) compared with provider-delivered care. By week 24, outcomes were comparable (%TBWL: 20.01% vs 19.97%; %FML: 36.81% vs 35.74%). Both groups achieved substantial fat mass loss while preserving muscle mass. Time-to-event analysis showed no significant difference in time to ≥10% TBWL (log-rank p=0.23; adjusted HR 0.83, p=0.43). The App group had a higher MBS clearance rate by week 24 (90% vs 71.25%, p=0.0023) with low attrition. User satisfaction was high (mean 8.76/10). Conclusion: The autonomous RightBMI App demonstrated greater mid-study weight and fat loss compared with provider-delivered care after statistical adjustment, with comparable 24-week outcomes and effective muscle preservation. Time-to-event analysis showed no significant difference in time to achieve ≥10% TBWL. These findings support the potential of fully autonomous digital platforms as scalable alternatives to traditional provider-led interventions in MBS candidates. Limitations include the non-randomized design and use of consumer-grade bioimpedance scales.