Mount Sinai researchers have developed InfEHR, an AI system that links scattered patient data to detect rare and unusual disease patterns it can flag conditions like neonatal sepsis or post-operative kidney injury more accurately and earlier than existing clinical methods.
Glimpse:
At the Icahn School of Medicine at Mount Sinai, scientists have introduced InfEHR, a novel AI platform designed to improve detection of rare and atypical diseases by piecing together otherwise disconnected medical events from electronic health records (EHRs). InfEHR excels especially in hard cases: for example, it identifies newborns with neonatal sepsis (even when blood cultures are negative) at dramatically higher rates, and flags risk of kidney injury after surgery more effectively than standard clinical criteria. Crucially, InfEHR is built to operate without large, disease-specific datasets and seems to generalize well across hospital populations. This approach could change how we find rare risks, shift diagnoses earlier, reduce misdiagnosis, and guide more personalized interventions.
Rare diseases are famously hard to catch. Many times, symptoms are subtle, test results inconclusive and clinicians are left balancing suspicion with uncertainty. But a new tool from Mount Sinai called InfER is making strides in tipping that balance in favor of earlier, more confident detection.
The core idea behind InfER is simple but powerful: most of our medical world is documented in electronic health records. But that data is often fragmented tests done here, symptoms noted there, follow-ups in another system. InfER connects those dots. It builds a “diagnostic web” that tracks medical events over time, looking for patterns that might suggest rare or under-recognized disease, even before full criteria are met.
In a recent study, researchers tested InfER on two tough problems: neonatal sepsis (which sometimes shows up even when usual tests, like blood cultures, are negative) and post-surgical kidney injury. The results were impressive for neonatal sepsis, InfER was 12-16 times more likely to identify affected infants than standard approaches; for kidney injury, the detection rate improved significantly (4-7 times). It also has a safety-first feature: when the data just doesn’t add up, it can indicate “not sure” rather than making a false positive call.
What makes this especially exciting is that InfER doesn’t require massive, rare disease-specific datasets to work well. Instead, it leans on routine lab results (things like blood counts, kidney function, indicators like cholesterol) and histories already present in patient records. That means it has potential to scale more easily, across hospitals, across populations, with less extra cost.
Of course, there are challenges. Differences in record-keeping across hospitals, missing or inconsistent data, issues of privacy and consent, and avoiding bias in how the AI is trained or validated are all real hurdles. And in rare disease detection false positives can be harmful, causing anxiety or unnecessary testing so calibration, clinical oversight, and transparency matter a lot.
Still, InfER looks like a major leap forward. For patients whose diagnoses are often delayed for months or years, tools like this can help shorten that wait. For clinicians, it can offer insight, backup, and a safety net in tricky diagnostic situations. And for health systems, earlier detection often means more effective treatment, fewer complications, and better outcomes.
“By linking the pieces of scattered lab results, symptoms, and medical history, we can now see disease risk in a whole new light even when traditional tests come back negative,”
By
HB Team
