Five years after Epic Systems rolled out a sepsis-prediction algorithm that hundreds of hospitals adopted and then quietly muted, a new wave of challengers is trying to succeed where the electronic health record giant stumbled. The most notable is Bayesian Health, a startup with roots at Johns Hopkins, which this month announced FDA clearance for its sepsis-flagging device. The clearance marks one of the first times a challenger to Epic’s dominant EHR-embedded model has cleared a formal regulatory bar the incumbent’s tool never had to meet in the same way.
A Tool That Cried Wolf
Sepsis kills more than 350,000 people in the United States every year, making early detection one of the highest-stakes problems in hospital medicine. Epic’s original sepsis model, introduced roughly five years ago, promised to flag at-risk patients before their condition worsened. In practice, clinicians say the tool generated so many alerts that most were tuned out, and some hospitals disabled it outright. Independent researchers, including a team at the University of Michigan, later found that much of the model’s apparent predictive power came from indirectly detecting clinicians’ own suspicion of sepsis already reflected in the chart, rather than catching cases doctors had missed.
Epic Retools, Rivals Circle
Epic has since released a retooled version of its algorithm, but the stumble opened the door for competitors who argue they can do better with newer techniques and independent validation. Bayesian Health’s device is built to flag deteriorating patients using vital-sign and lab trends, and its Johns Hopkins origins give it a research pedigree the company is leaning on as it courts hospital buyers. Separately, a University of California San Diego-affiliated team is testing whether large language models can mine free-text clinical notes for early sepsis signals that structured data alone might miss, a different technical bet than either Epic or Bayesian Health is making.
Ohio Hospitals as a Testing Ground
At least one Ohio health system is piloting an AI sepsis-prediction tool outside Epic’s native offering, part of a broader pattern of hospitals quietly testing multiple vendors against each other before committing. That kind of parallel evaluation was rare five years ago, when Epic’s EHR dominance made its bundled algorithm the default choice for many systems simply because it required no extra procurement effort. Hospital informatics leaders say the sepsis debacle changed that calculus, pushing more institutions to demand outside validation data before trusting any single vendor’s alert.
Why Performance Alone May Not Decide the Winner
Health-tech analysts covering the space have framed the competition with a pointed observation: in the battle of sepsis algorithms, performance alone doesn’t predict victory. Epic’s grip on hospital IT infrastructure, and the fact that its sepsis tool ships inside the EHR system most physicians already use, gives it a structural advantage that a better-performing standalone product from a startup has to overcome. FDA clearance helps newcomers like Bayesian Health signal credibility to skeptical hospital committees, and emerging Medicare payment structures tied to AI-assisted care could further level the playing field by reimbursing outcomes rather than favoring whichever tool is easiest to install.
Proponents See Overdue Competition, Critics Urge Caution
Supporters of the new entrants argue that competition is exactly what hospital AI needs after years of Epic’s alert fatigue problem going largely unaddressed; independent FDA review and external validation, they say, are the accountability the first generation of sepsis tools lacked. Skeptics counter that sepsis prediction has repeatedly overpromised: the same University of Michigan research that undercut Epic’s model warns that any algorithm trained on data contaminated by clinicians’ existing suspicions risks looking accurate in retrospective studies while adding little real predictive value at the bedside. Some clinician groups also worry that a crowded field of competing sepsis tools could recreate the alert-fatigue problem at scale, just distributed across more vendors instead of concentrated in one.
For now, hospitals are left to run their own bake-offs, weighing Epic’s convenience against newer entrants’ regulatory clearances and independent data. How Medicare’s evolving payment rules for AI-assisted diagnostics shake out over the next year could determine whether Bayesian Health and the LLM-based challengers actually dent Epic’s hold, or whether hospitals simply wait for Epic’s next retooled version and stick with what’s already installed. Either way, sepsis detection has become an early test case for how much scrutiny hospital-embedded AI tools will face once they move from research pilots into the everyday clinical workflows that decide how quickly a patient gets treated.
