Bayesian Health’s FDA-cleared continuous sepsis monitoring software has secured a New Technology Add-on Payment under the Centers for Medicare and Medicaid Services’ fiscal 2027 inpatient payment rules, removing one financial barrier that can prevent hospitals from moving clinical AI beyond pilot deployments. Beginning October 1, 2026, eligible Medicare fee-for-service inpatient cases using the Bayesian Health Sepsis Flagging Device can qualify for an additional payment of up to $61.84 per discharge. Bayesian says the pathway applies across an estimated 739 Medicare Severity Diagnosis-Related Groups and can remain available for as long as three years.
The reimbursement decision follows FDA 510(k) clearance on April 30, 2026. The Class II software continuously analyzes electronic-health-record information and provides healthcare professionals with a “Sepsis Risk High” flag when its machine-learning system identifies a patient at risk of having or developing sepsis within 24 hours. FDA labeling explicitly states that the flag must be used alongside clinical assessment and laboratory data and cannot serve as the sole basis for determining whether a patient has sepsis.
What information does Bayesian Health’s sepsis AI continuously analyze?
The system draws from data already being collected within the electronic health record rather than requiring a new blood test or dedicated bedside sensor. FDA documentation says the algorithm can consider patient characteristics, comorbidities, the presenting complaint, laboratory measurements, vital signs, procedures, medication orders and specialist consultations. The software operates across emergency departments, observation units, general wards and intensive-care environments for adults aged 18 and older.
When the algorithm determines that sepsis risk has crossed its threshold, the system generates a high-risk flag within the EHR environment. Clinicians can review factors contributing to the alert before deciding whether the patient requires additional investigation or treatment. That workflow distinction matters because the FDA authorized the software as an aid to early detection and risk prediction, not as an autonomous diagnostic system that initiates antibiotics without clinician involvement.
How well did the software perform in FDA validation?
The 510(k) submission included retrospective clinical validation using 7,732 hospital encounters involving 7,298 unique patients across four hospitals. The FDA-reviewed dataset examined positive percent agreement, negative percent agreement and positive predictive value for identifying sepsis risk within the relevant clinical window. Encounter-level positive percent agreement was 79.4%, with a 95% confidence interval of 74.2% to 84.6%, while negative percent agreement reached 89.5%, with a confidence interval of 88.8% to 90.2%.
Those figures demonstrate why sepsis AI should not be portrayed as infallible. Some patients who develop sepsis will not be flagged, while some patients who receive an alert will ultimately not meet the clinical definition of sepsis. The device is intended to shift attention earlier toward patients whose evolving EHR pattern suggests risk, while clinicians remain responsible for interpreting that signal in context.
Why is continuous monitoring different from waiting until a clinician suspects sepsis?
Sepsis develops dynamically. A patient can enter the hospital with nonspecific symptoms and become progressively more abnormal as infection and organ dysfunction evolve, which means a one-time screening score may become outdated quickly. Continuous software can repeatedly reassess the patient as new laboratory values, vital signs, medications and other information enter the record.
FDA documentation notes that Bayesian’s intended-use population includes broad emergency-department and acute-care patients rather than only individuals for whom a physician has already documented a suspicion of sepsis. The agency concluded that the totality of evidence supported risk identification within the 24-hour window surrounding sepsis onset in this population. That earlier “presuspicion” position is central to Bayesian’s commercial argument because the system is designed to raise concern before the normal diagnostic process has necessarily started.
What does Medicare NTAP reimbursement actually change for hospitals?
Medicare generally pays hospitals using predetermined bundled amounts linked to diagnosis-related groups. When a new technology is introduced, those historical payment levels may not immediately reflect the additional cost of deploying it, which can discourage adoption even after FDA clearance. NTAP temporarily provides an additional reimbursement amount for qualifying technologies and cases while standard payment systems catch up.
For Bayesian Health, eligible hospitals can receive up to $61.84 per qualifying discharge beginning October 1. The amount is relatively modest compared with the total cost of a sepsis hospitalization, but the significance lies in creating a formal reimbursement mechanism tied specifically to the technology. Bayesian describes it as the first dedicated Medicare pathway for continuous presuspicion sepsis monitoring, a company claim that reflects how unusual direct reimbursement for clinical AI remains.
Will reimbursement solve the clinical AI adoption problem?
Not by itself. Hospitals frequently struggle to move AI applications from research pilots into routine care because deployment requires EHR integration, staff training, workflow redesign, alert management, governance and continued performance monitoring. A software product that produces technically accurate predictions can still fail operationally if clinicians receive too many alerts or do not know who is responsible for acting on them.
Bayesian has built its commercial proposition around embedding the system inside existing EHR workflows and monitoring performance after deployment. FDA documentation also describes a postmarket performance-management plan intended to detect degradation as patient populations and clinical environments change. Continuous monitoring of the AI itself may become increasingly important as regulators and hospitals gain more experience with predictive software operating continuously on live clinical data.
Could reimbursement accelerate a wider market for predictive hospital AI?
Potentially, because economics has been one of the missing pieces in clinical artificial intelligence. Hospitals can justify purchasing imaging equipment or laboratory analyzers through familiar reimbursement models, while AI software often enters budgets as an additional operating expense even when developers argue that it prevents expensive complications. A dedicated payment creates a clearer connection between use of the technology and hospital revenue.
Sepsis is a particularly important test case because earlier recognition can be clinically valuable and because the disease carries enormous hospital costs. If reimbursed continuous monitoring demonstrates measurable improvements in treatment timing, length of stay or outcomes, other predictive AI developers will likely use Bayesian’s pathway as a precedent. If hospitals receive reimbursement but real-world clinical benefit remains unclear, payers may become more skeptical about creating similar mechanisms elsewhere.
What should hospitals watch when NTAP begins in October?
The most informative evidence will come from adoption outside the institutions that helped develop or validate the technology. Health systems will need to measure alert burden, clinician response, timing of treatment, false-positive rates and whether benefits remain consistent across different patient populations. Financial teams will also need to determine how often cases actually qualify for the add-on payment and whether reimbursement meaningfully offsets implementation and licensing costs.
The September announcement therefore represents something broader than another AI product winning a payment code. Bayesian Health already crossed the regulatory hurdle when the FDA cleared the device in April; Medicare is now testing whether payment policy can help move predictive AI into everyday hospital infrastructure. The next question is whether widespread use confirms that continuously watching the EHR for sepsis risk improves care enough to justify making this type of algorithm a standard part of inpatient medicine.
