Predicting Long-Term Weight Loss Using Self-Reported, Digitally Collected, Real-World Data After Initiation of...

We hypothesised that an exposure-response weight predictor algorithm developed from randomised controlled trial data can predict long-term individual body weight changes in both men and women in response to subcutaneous semaglutide treatment for weight management using self-reported, real-world data, collected through a digital patient support program (PSP).

An exposure-response body weight prediction model developed from clinical trials with semaglutide in people with overweight or obesity was applied to a real-world dataset from patients prescribed semaglutide (0.25-2.4 mg) by their treating healthcare provider (HCP) and enrolled into an app-based PSP. Model variables were baseline sex and body weight, and self-reported dosing and body weight during semaglutide treatment. Predictions were assessed in two scenarios. In the first scenario, body weight at 26 ± 4 weeks was predicted from baseline data with model updates at weeks 4, 8, and 16. In the second scenario, body weight at 52 ± 4 weeks was predicted from baseline data with updates at weeks 8, 16, and 28. Model bias was calculated as the difference between predicted and self-reported body weight, while precision for predicting categorical weight loss (≥ 10%, ≥ 15%, ≥ 20%) was assessed using area under the curve (AUC).

The study included 1797 WegovyCare® app users, predominantly women (81%), with a mean (SD) age of 48.0 (11.8) years. Mean (SD) self-reported body weight in the entire population was 105 (19.5) kg at baseline. In the half-year scenario, users lost an average of 15% (15.6 kg). Model bias was low (0.7-1.4 kg) and precision for predicting categorical weight loss was high (AUC 0.74-0.95). In the full-year scenario, average weight decreased by 21% (22.0 kg) with similarly low bias (-0.6 to 0.6 kg) and high prediction precision for categorical weight loss (AUC 0.75-0.92).

This study successfully applied an exposure-response weight predictor algorithm to self-reported data collected from users in the real world. Integrating weight predictors into digital PSPs may be valuable for both patients and HCPs in managing weight loss and setting or monitoring treatment targets.