Transforming complex survey data into actionable insights: a data-driven approach for evaluating digital health programmes

This study evaluates engagement and impact across five digital health programmes using a hybrid data analytics approach. Logistic regression revealed that engagement with the digital programmes varied significantly by age, gender, health status, disability, and physical activity levels, with distinct predictors identified for each programme type. BERTopic revealed key user motivations such as information seeking, community support and barriers such as lack of awareness, platform aversion to engagement across the programmes. Sentiment analysis indicated generally positive perceptions but proved limited for in-depth understanding of programme usefulness. Zero-shot classification, leveraging the ‘facebook/bart-large-mnli’ model and the COM-B framework, identified ‘Capability’ and ‘Motivation’ as primary impact categories reported by users. The study underscored the importance of manual refinement to improve interpretability in automated analyses and the context-specific limitations of topic modelling with brief textual responses. Additionally, addressing missing data and imbalanced categorical variables was critical for maintaining analytical integrity. Future research should explore automated refinement techniques and employ mixed methods designs to enhance understanding of programmes effectiveness and user engagement.