Epic mortality model deployment highlights gaps in EHR AI validation as Omada eyes chronic care expansion

Epic's model draws on routine clinical variables including labs, vitals, and comorbidities to estimate 30- and 90-day mortality. Internal testing reportedly achieved AUROC values above 0.85, yet no external validation cohort or peer-reviewed publication has been released. This mirrors earlier Epic sepsis alerts that showed degraded performance outside the development sites.

Omada's pipeline includes AI-augmented coaching modules that incorporate mortality or hospitalization risk scores to prioritize high-need participants. The company has cited observational data linking its programs to 10-15% reductions in HbA1c, but lacks RCTs demonstrating downstream effects on mortality or acute events. Medicare Advantage plans are the primary target.

Policy analysts note that private EHR vendors and digital health firms operate under different evidentiary standards than FDA-cleared devices. Without mandated prospective testing, models risk amplifying existing disparities in care access and documentation quality across health systems.

Next steps require independent replication on multi-institutional datasets and linkage to clinical endpoints rather than surrogate markers.