ONC HTI-1 Transparency Requirements for Health IT Developers

ONC's HTI-1 final rule created something nobody in health IT expected: a federal transparency mandate for AI in clinical software. If your certified EHR uses predictive algorithms, you now have to tell clinicians exactly what's under the hood. All 31 source attributes. For every single predictive model in your system. What's a Predictive DSI? Decision Support Intervention. ONC split DSI into two categories: evidence-based and predictive. Evidence-based DSIs are things like drug interaction alerts and clinical guidelines. The old stuff. Predictive DSIs are the new stuff: any algorithm that uses data to predict a clinical outcome. Sepsis risk scores. Readmission predictors. Deterioration alerts. Diagnostic probability models. If it takes patient data and outputs a prediction, it's a Predictive DSI. And here's the catch: the definition is broad enough to cover models that most developers don't think of as "AI." A logistic regression predicting 30-day readmission? Predictive DSI. A LACE score calculator? Probably a Predictive DSI. You don't need deep learning to trigger this rule. The 31 Source Attributes For each Predictive DSI, you have to provide 31 pieces of information.