This month, Bayesian Health became the first company to receive FDA clearance for an AI-powered continuous monitoring system for sepsis.
Many AI tools for sepsis detection have historically operated without an FDA review, being deployed as generalized clinical decision support software. Suchi Saria, Bayesian’s founder and CEO, framed the FDA clearance as much bigger than a regulatory milestone — she sees it as validation of her startup’s broader vision for AI-powered “real-time clinical intelligence” that helps hospitals move from reactive care to proactive care.
Saria noted that Bayesian spent years working with the FDA to establish reliable definitions for sepsis, validate its model’s performance across diverse hospital settings and patient populations, evaluate risks like missed cases and alert fatigue, and create a post-market monitoring and quality assurance program.
In her eyes, healthcare AI needs far more rigor than most tools currently receive. She said any AI tool that affects patient care should face this level of scrutiny, even if it doesn’t actually go through the regulatory process.
“Many people mistakenly see FDA clearance as the ceiling, when I think it’s only the floor. It’s the starting point,” Saria explained. “If you can’t change clinician action, then you’re not going to drive outcomes — and changing outcomes is the real goal.”
Clearance helps establish trust by showing a tool works in diverse settings, but she thinks successful clinical AI adoption ultimately depends on factors like workflow integration, usability, transparency and measurable outcome improvements.
She also pointed out that Bayesian intentionally took an evidence-first approach rather than a typical startup commercialization strategy. Instead of quickly selling an early product, the company initially focused on large real-world deployments and studies.
Bayesian formally launched in 2021, and by the following year, it had published studies in Nature Medicine involving about 750,000 patients. The research showed the startup’s AI demonstrated high clinician adoption rates and earlier sepsis detection — as well as improved outcomes like lower mortality, fewer complications and reduced length of hospital stay.
Health systems across the country use Bayesian’s sepsis tool, including Cleveland Clinic, Johns Hopkins Health System, University of Rochester Medicine and MemorialCare.
One health system leader — Dr. James Leo, chief medical officer of MemorialCare’s Physician Society — noted that implementing Bayesian’s tool wasn’t just a technology rollout.
“It was an opportunity to engage our frontline staff — both nurses and providers — to rethink early identification and treatment of sepsis across our system,” he stated.
MemorialCare ran two phases in parallel. One was the technical build inside its EHR, and the other was the clinical work of ensuring sepsis workflows optimize patient outcomes. Dr. Leo said Bayesian’s team sat down with staff members from MemorialCare’s emergency departments, inpatient units, intensive care units and quality team to map where the opportunities were — and then the startup and the health system co-designed workflows that clinicians approved before the go-live.
Dr. Leo noted that this emphasis on workflow design and clinician buy-in has been key to the tool’s success.
“Sepsis can move fast and present subtly, and a patient’s risk follows them wherever they are in the hospital. Every nurse, provider, APP and resident in scope is provisioned on the platform, which means hundreds to thousands of clinicians working from the same information and coordinating care as one team,” he remarked.
Since adopting the tool, MemorialCare is catching more sepsis patients earlier, with “more than double the sensitivity of what we were using previously,” Dr. Leo added.
He also pointed out that clinicians are working with significantly fewer electronic alerts, which he thinks is helping restore their trust in these types of AI tools. In his view, that reduction in alert fatigue has contributed to the platform’s high engagement rates among providers.
“When providers engage with the Bayesian flag, we’re seeing a 3.6% absolute mortality reduction, and time to antibiotics is cut in half when they engage within the first hour. With 90% adoption in our ED, our clinicians believe in it too — and that’s why we’re confident in taking this system-wide,” Dr. Leo declared.
The results at MemorialCare underscore Saria’s claim that clinical AI tools have to earn clinician trust — not just regulatory clearance — to actually improve patient outcomes.
Photo: Ruslanas Baranauskas/Science Photo Library
