Deployed worldwide. Validated nowhere.

The study

Epic's sepsis prediction algorithm was deployed at scale across hospital systems internationally, designed to identify patients at risk of sepsis in real time.

The failure

The algorithm missed two in every three sepsis cases — and 86% of the alerts it did raise were false alarms, eroding clinical trust and patient safety.

Full case study: Wong et al., 2021 — Journal of the American Medical Informatics Association

The critical gap

Pre-implementation validation methods for good AI governance were not applied. A prospective silent trial — assessing performance without affecting patient care — would have identified these failures before the algorithm reached patients at scale.

How we can help

Our Pre-Implementation Validation and Silent Trial Design service helps research teams plan and execute robust validation approaches before deployment — so performance failures are caught early, not after go-live.