Early weight change predicts outcomes in a digital obesity app

Researchers developed machine-learning models using anonymized real-world data from adults using the 12-week Oviva Direkt for Obesity app. Analyses included 17,953 adults for weight-change outcomes and 26,714 for full app retention. The models assessed baseline characteristics, weight at week 4 and app engagement to predict clinically meaningful weight loss, continuous relative weight change and retention.

For clinically meaningful weight loss at weeks 8 and 12, baseline characteristics alone had limited discrimination (area under the receiver operating characteristic curve [AUC], 0.54–0.58). Adding week-4 weight increased AUC to 0.86–0.92. Engagement alone produced more modest AUCs of 0.63–0.67, and adding engagement to models that included week-4 weight yielded little further improvement. For full retention, AUC rose from 0.81 with baseline characteristics to 0.90 with engagement; week-4 weight added no benefit.

The findings suggest that weight-loss response and app retention may require different approaches to identifying users who could benefit from support. However, these are predictive results from one app and dataset, not evidence that model-guided care improves outcomes. The authors call for external validation before clinical implementation. Two authors were employed by Oviva AG; the study received no external funding.