Stanford Medicine updates My Heart Counts app

Per a Stanford Medicine news release, the research platform My Heart Counts has received a redesigned app with an upgraded interface and an embedded AI coaching study, and Stanford reports the platform has engaged over 100,000 participants since its 2015 launch with Apple partnership (Stanford Medicine). The My Heart Counts research site highlights recent and forthcoming publications, including a 2025 NPJ article describing MHC-Coach, an AI health coach reported to outperform human-crafted messages in promoting personalized physical activity (My Heart Counts research page). A medRxiv preprint and a sciencedirect summary provide the study's design and rationale, describing an embedded randomized trial that evaluates text-based coaching prompts generated by a large language model and a modular, cross-platform smartphone application (medRxiv; ScienceDirect).

Editorial analysis - technical context: Public sources describe the intervention as LLM-driven, text-based coaching integrated into a smartphone study framework. The available summaries indicate the trial architecture uses modular software and cross-platform accessibility to deliver automated coaching prompts; the authors present this as an autonomous intervention layer on top of standard activity and survey data streams (medRxiv; ScienceDirect). The research site lists MHC-Coach specifically and references peer-reviewed and preprint evaluations that compare model-generated messages to human-crafted messages (My Heart Counts research page; NPJ Cardiovascular Health 2025 listing).

Academic groups have increasingly embedded algorithmic interventions in digital trials to test causal effects on behavior. The My Heart Counts program is notable because it moves a large, long-running digital cohort (reported 100,000+ participants) toward randomized evaluation of an AI coaching layer, rather than retrospective analysis alone (Stanford Medicine; My Heart Counts research). For practitioners, this shift means more experimental evidence on automated behavior-change messaging will be available, which could inform how teams design, validate, and monitor production coaching systems in health settings.

For practitioners: observers will want to review the full trial manuscripts and preprints for details on model training/fine-tuning, prompt engineering, safety and hallucination mitigation strategies, consent and opt-in flows, outcome definitions, and statistical effect sizes (medRxiv; ScienceDirect; My Heart Counts research). Key indicators to follow in the manuscripts include reported effect sizes for physical activity, subgroup analyses (age, baseline activity, comorbidities), retention and engagement metrics, and any reported adverse events or message safety incidents. Also watch for data-sharing statements and released code or model artifacts that would enable reproducibility and independent benchmarking (My Heart Counts data release history).

The scraped Stanford and My Heart Counts pages and the linked preprint describe wearable-data integration and LLM-driven coaching, but the available sources do not document hospital EHR connectivity or production deployment into clinical workflows; public reporting focuses on smartphone and wearable data streams and on the embedded trial design (Stanford Medicine; My Heart Counts research; medRxiv). If claims of EHR integration appear elsewhere, they are not present in the scraped sources used for this summary.

Editorial analysis: The My Heart Counts updates reflect a broader research trend testing automated, conversational or text-based interventions in randomized settings. Practitioners and researchers should treat the forthcoming full manuscripts and any released datasets or code as the primary artifacts for assessing clinical validity, reproducibility, and operational risks before considering translation to deployed clinical or consumer systems.