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

Area of Science:

• Pharmacology and Endocrinology
• Digital Health and Patient Support
• Data Science and Predictive Modeling

Background:

• Predicting individual body weight changes is crucial for effective weight management.
• Real-world data from digital patient support programs (PSPs) offer valuable insights.
• Subcutaneous semaglutide is a treatment for weight management.

Purpose of the Study:

• To test if an exposure-response weight predictor algorithm, developed from clinical trials, can predict long-term body weight changes in real-world semaglutide users.
• To validate the algorithm using self-reported data from a digital PSP.
• To assess the algorithm's performance in predicting weight loss over 6 and 12 months.

Main Methods:

• Applied a semaglutide exposure-response weight prediction model to real-world data from an app-based PSP.
• Used baseline sex and body weight, plus self-reported dosing and weight during treatment, as model variables.
• Assessed prediction accuracy by comparing model predictions to self-reported weights and calculating area under the curve (AUC) for categorical weight loss.

Main Results:

• The study included 1797 users, with an average baseline weight of 105 kg.
• In the 6-month scenario, the model showed low bias (0.7-1.4 kg) and high precision (AUC 0.74-0.95) for predicting weight loss.
• In the 12-month scenario, users lost an average of 21% (22.0 kg), with similarly low bias (-0.6 to 0.6 kg) and high prediction precision (AUC 0.75-0.92).

Conclusions:

• The exposure-response weight predictor algorithm was successfully applied to real-world, self-reported data.
• Integrating such predictors into digital PSPs can aid patients and healthcare providers in managing weight loss.
• The findings support the use of predictive algorithms for setting and monitoring weight management treatment targets.