Deterioration Index Version 2 (K260008) — FDA 510(k) | Innolitics

Open original

Epic Systems got FDA clearance for Deterioration Index Version 2, an AI tool that analyzes vital signs to predict cardiovascular events. It's not meant to work alone—doctors use it alongside other clinical data.

  • Epic had to validate the algorithm on independent datasets and prove its predictions actually correlate with real patient outcomes
  • The device requires detailed labeling about what sensors work with it, how far in advance it predicts events, and that it's only adjunctive to standard care
  • FDA imposed eight specific control requirements covering software validation, human factors testing, clinical data assessment, and performance across different patient populations and environments

Epic Systems got FDA clearance on September 18, 2026, for a software tool called Deterioration Index Version 2, classified as a Class II medical device. It's an AI-powered predictor designed to analyze cardiovascular vital signs and flag when a patient's condition might worsen or a cardiac event could occur. The catch—and this matters—is that it's explicitly adjunctive. It can't stand alone and make treatment calls. It needs to work alongside other vital sign data and patient information that clinicians are already using. This is a deliberate regulatory boundary: the device assists judgment rather than replacing it. The FDA imposed strict special controls that essentially demand Epic prove the thing actually works and won't mislead doctors. The company had to document every algorithm detail, explain how sensor hardware affects accuracy, and describe what happens when something breaks or a user misreads the output. They needed to validate their prediction model against independent data—not just the training set—and show that the algorithm's calculations are sound. They also had to run a human factors study to catch ways clinicians might misinterpret what the device is telling them. This wasn't a rubber-stamp approval; the regulatory bar was high. On the labeling side, Epic has to be crystal clear about what the device actually measures, which sensors it works with, and the time window before predicted events (accounting for patient differences). They need to spell out every assumption baked into the calculations and flag which sensor factors could throw off results. The label must include specifics on how it performed across different patient populations in their validation studies—age, race, gender, condition—so clinicians know whether it was tested on people like their own patients. The clinical data assessment itself had to be representative of real-world use, meaning they couldn't cherry-pick clean data sets; they had to show it handles messy, variable input from actual hospital environments.

Ask about this article