Epic has announced Comet, an AI system built on Cosmos data, designed to predict what might happen next in a patient’s journey. Think readmission risk, length of stay, ASCVD events, even early cancer signals — all modeled from 100 billion+ de-identified patient events.
The promise? Clinicians won’t just react to today’s labs or diagnoses; they’ll plan for tomorrow’s likely scenarios. Starting February 2026, Comet will be available in a virtual lab for Cosmos researchers, with the potential to reshape care planning from reactive to anticipatory.
As someone who has lived through more “game-changing” technologies than I can count, here’s my take:
- Integration is the mountain. Predictive insights are valuable only if they surface seamlessly inside workflows—not in a separate dashboard clinicians won’t open during a 12-minute encounter.
- Trust is earned, not claimed. Accuracy benchmarks are fine, but clinicians want to know why the model predicts a readmission or complication. Black-box forecasts don’t inspire adoption.
- Equity matters. Cosmos is vast, but not every community looks like the aggregate dataset. Will Comet be as effective in smaller, under-resourced hospitals as it is in large, academic systems?
- Cost and control. Will this be bundled into Epic’s roadmap, or yet another add-on module with a hefty price tag? And how do we evaluate it against best-of-breed AI vendors already in play?
- Upgrade dependency. Every new Epic capability comes with a minimum code baseline. For many health systems already behind on upgrades, that means major lift: regression testing, retraining, and potential downstream impacts on ancillary systems. If Comet requires staying within one or two versions of Epic’s quarterly cycle, that alone could limit adoption. Falling behind on code is no longer just technical debt—it becomes a barrier to innovation.
My bottom line: Comet could be a step toward a more anticipatory, personalized healthcare system—but only if it moves from hype to practical, explainable, and equitable use at the bedside. Until then, it’s a promising constellation, not yet a guiding star.
What do you think—would your clinicians trust an AI-simulated “likely future” enough to change their decisions today?
