New sepsis AI models face adoption hurdles despite improved performance

Hospitals once embraced an Epic sepsis‑prediction algorithm that promised early alerts for a condition that kills over 350,000 Americans each year. In practice the tool generated far more alerts than clinicians could handle, leading many to ignore or disable it.

Five years later a wave of new AI models is entering the market. Epic has released a re‑engineered version of its original algorithm, and several startups are piloting their own sepsis‑prediction tools in health systems.

Industry experts caution that raw performance metrics will not guarantee uptake. Successful deployment will require seamless integration with electronic health records, manageable alert volumes, and trust from clinicians who must act on the warnings.

For hospitals, the challenge now is to balance the promise of more accurate predictions with the practical realities of workflow and alert fatigue, shaping the next phase of AI adoption in critical care.