Machine Learning Forecasts Weight Loss and App Engagement in Real-World Digital Obesity Care

Machine learning models built on real-world data from a digital weight loss application can predict clinically meaningful weight change far more accurately once early treatment information is included, according to a study published in BMC Medicine. The research, led by Christina Glasbrenner, Marco Lehmann, Christoph Höchsmann, Felix Schirmann and Karsten Koehler, analyzed data from Oviva Direkt for Obesity, a 12-week digital health application used in routine care.

The team developed models to predict three distinct outcomes: clinically meaningful weight loss, continuous relative weight change, and full retention in the app over the program period. The weight-change analyses included 17,953 adults, while the full-retention analyses covered a larger sample of 26,714 adults. The models drew on three types of information: baseline characteristics recorded before treatment, weight measured at four weeks, and weekly metrics of app engagement.

A central question was how predictive information evolves during treatment. The researchers evaluated each information source separately and in combination, quantifying how much predictive value each added and how performance changed when early-treatment data were layered onto pretreatment characteristics.

The results showed a striking hierarchy of predictive power. Baseline characteristics alone offered only limited discrimination for clinically meaningful weight loss at weeks 8 and 12, with ROC AUC values of 0.54 to 0.58, barely better than chance. Adding the four-week weight measurement transformed performance, raising the ROC AUC to between 0.86 and 0.92. App engagement metrics alone produced only modest gains, with ROC AUC values of 0.63 to 0.67, and adding engagement to models that already included four-week weight yielded little further improvement, with values of 0.87 to 0.92.

A similar pattern emerged for predicting continuous weight change. In both cases, the week 8 and week 12 predictions were driven primarily by baseline weight, four-week weight, and the self-defined target weight that participants set for themselves.

Retention told a different story. Predicting whether participants would remain fully engaged with the app improved from a ROC AUC of 0.81 using baseline characteristics alone to 0.90 when app engagement data were included, while the four-week weight measurement added no additional benefit. Engagement-related variables, such as goal setting and personalized logging recommendations, were the most influential factors for full retention.

This divergence is the study’s most consequential finding. Weight-loss response and sustained intervention use were characterized by distinct predictive information, suggesting that identifying who will lose weight and who will keep using the app are fundamentally different prediction problems. The authors conclude that these two outcomes may require different forms of early, targeted support rather than a single intervention strategy.

The researchers emphasize how their models should be interpreted. Because predictive performance improved substantially when information from the first four weeks, particularly the four-week weight, was incorporated, the week 8 and week 12 models should be understood as predictions updated during treatment rather than predictions based solely on pretreatment characteristics.

The authors position these findings as a basis for further research on the early identification of individuals at risk of suboptimal outcomes and on the development of more personalized digital obesity care. They also stress a clear limitation: external validation in independent, relevant cohorts is required before the models can be considered for clinical implementation.

Subject of Research: Machine learning prediction of weight change and app retention in a real-world digital obesity program

Article Title: Machine learning predicts weight change in a real-world digital health application

Article References: Glasbrenner, C., Lehmann, M., Höchsmann, C., Schirmann, F., & Koehler, K. (2026). Machine learning predicts weight change in a real-world digital health application. BMC Medicine. https://doi.org/10.1186/s12916-026-05320-x

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05320-x

Keywords: machine learning, digital health, weight loss, obesity, app retention, predictive modeling, Oviva Direkt, ROC AUC, engagement metrics, personalized care, BMC Medicine, Machine