AI
Patient Support and 6‐Month Medication Adherence in a Digital Obesity Program: A Retrospective Analysis
Louis Talay, Connie Xu, John Alderete, Jason Hom, Marilyn Tan, Neera Ahuja ABSTRACT
Background
Real‐world glucagon‐like peptide‐1 receptor agonist (GLP‐1 RA) therapies face substantial attrition rates in commercial digital weight loss services (DWLSs). Conversational artificial intelligence (AI) has been proposed to enhance patient support at production scale, but robust evidence of its effectiveness in improving medication retention is scarce.
Objectives
To evaluate the effect of integrating an asynchronous AI patient support agent (Junebot) into a commercial DWLS on 6‐month medication adherence and to assess the independent relationship of digital engagement on retention.
Methods
This retrospective analysis evaluated 16 556 adults prescribed semaglutide within an Australian DWLS between May 2024 and October 2025. The primary endpoint was 6‐month medication adherence (≥ 6 orders fulfilled within 183 days). Analytical protocols featured a multivariate binary logistic regression on the full intention‐to‐treat cohort and three distinct 1:1 propensity score matching (PSM) sensitivity analyses to isolate era‐based and tier‐specific engagement effects.
Results
Six‐month adherence was higher post‐Junebot than in the pre‐Junebot control (53.2% vs. 47.3%; p < 0.001). However, the full cohort model ( R 2 = 0.2236) revealed that the post‐Junebot operational era was independently associated with an increase in the odds of attrition (OR: 1.178; 95% CI [1.052–1.318]; p = 0.004), a trend corroborated by the era‐matched PSM model (OR of attrition: 1.309; p = 0.038). Non‐automation factors dominated retention trajectories: high‐intensity tracking during month 1 (> 25 tracks) dramatically predicted attrition (OR: 15.753; p < 0.001), alongside program pauses (OR: 2.508; p < 0.001) and higher program cost (OR: 2.458; p < 0.001). Within‐cohort analysis demonstrated a profound, linear curve for active interaction; structured digital engagement spanning 50%–74.99% of weeks significantly optimised adherence odds across both medication‐only (OR: 32.016) and medication plus health coaching cohorts (OR: 13.311).
Conclusion
Integrating an AI digital assistant did not independently improve medication adherence; programmatic costs, program pauses and initial tracking anxiety were stronger predictors of long‐term retention. After matching post‐Junebot cohorts, moderate, active digital engagement appeared to correlate with 6‐month retention. Digital providers should prioritise proactive behavioural risk screening and financial accessibility over baseline AI integration.