On May 12, 2026, Bayesian Health announced that its continuous sepsis-monitoring platform had received FDA 510(k) clearance — the first clearance ever granted for an AI system that watches hospitalized patients around the clock for signs of a condition that kills an estimated 11 million people worldwide every year. The device, called TREWS (Targeted Real-Time Early Warning System), was built by researchers at Johns Hopkins University and commercialized by the Baltimore-based startup Bayesian spun out of that lab in 2023.
A disease that hides until it is almost too late
Sepsis is notoriously difficult to catch early because its warning signs — a slightly elevated heart rate, a subtle drop in blood pressure, a fever that comes and goes — look like dozens of other things a hospitalized patient might be experiencing. By the time a clinician is confident enough to say the word “sepsis” out loud, the infection has often already progressed into organ dysfunction, and every hour of delayed treatment measurably raises the odds of death. TREWS attacks that problem differently than a human ever could: it continuously ingests vital signs, lab values, and nursing notes from the electronic health record and recalculates a patient’s sepsis risk in real time, flagging deterioration long before it would register as a clinical suspicion.
What the data actually showed
The clearance rests on a prospective study published in Nature Medicine that tracked TREWS across multiple hospital systems, including Johns Hopkins facilities in the Baltimore area and Cleveland Clinic, where the tool was piloted starting in 2023. In that research, TREWS identified sepsis cases with 82% sensitivity, and — the number that matters most to hospital administrators — patients were 18% less likely to die when clinicians responded to a TREWS alert within an hour of it firing. That is a meaningful reduction in mortality for a disease where speed of treatment is often the single biggest variable separating survival from death.
Why the FDA route matters here
Bayesian pursued the clearance through the FDA’s 510(k) pathway, the same lower-friction route used for the majority of AI-enabled medical devices reaching the market in 2026. That distinction is not just regulatory trivia: earlier sepsis-detection tools operated as decision-support software that hospitals could adopt without formal FDA sign-off, leaving their claims about accuracy largely unverified by an outside regulator. TREWS is different — it is the first continuous AI sepsis monitor to have its performance claims independently reviewed and cleared by the agency, which industry analysts say gives hospital risk-management committees a much easier case to make when deciding whether to fund a system-wide rollout.
The alert-fatigue problem nobody has fully solved
Continuous monitoring tools carry a well-known hazard: if they fire too often, nurses and physicians tune them out, and the alerts become background noise rather than an early warning. Clinicians at Bayesian’s pilot sites say the system has been engineered specifically to reduce that risk by suppressing lower-confidence alerts and prioritizing the ones most likely to represent a genuine deterioration, a design choice credited with keeping the tool in active use rather than becoming another ignored pop-up on a crowded hospital dashboard. Even so, critics of alert-based AI tools broadly caution that a device is only as good as the workflow around it — a hospital that does not staff up its rapid-response teams to actually act on early warnings will not see the mortality benefit no matter how accurate the algorithm is.
What comes next
Bayesian is now pursuing a New Technology Add-on Payment designation from the Centers for Medicare and Medicaid Services, with a decision expected in early August 2026 and a payment pathway that could open as soon as October 2026 if approved — a reimbursement mechanism that would make it financially easier for cash-strapped hospitals to adopt the tool at scale. For an industry that has spent years testing sepsis algorithms that promised much and delivered inconsistent results in the real world, TREWS’s FDA clearance and its published mortality numbers give hospital systems something rarer than another AI pilot: independently verified evidence that watching every patient, all the time, for a disease that often announces itself too late, can actually save lives. Hospital administrators evaluating the tool say the reimbursement decision expected this month will likely determine how quickly that evidence translates into wider adoption beyond the handful of academic medical centers where TREWS has already been running.
