AI-Powered Clinical Decision Support Integrated with Epic EHR

Clinical providers at a large regional health system were making encounter decisions with incomplete information — not because the data did not exist, but because it was not accessible at the right moment. A patient's full history, prior diagnoses, medication interactions, and relevant clinical notes lived across multiple systems, and a physician in the middle of an encounter did not have the time to navigate all of them. The result was decisions made with a partial picture, which contributed to unnecessary follow-up visits and, in some cases, avoidable readmissions.

The health system had a significant investment in Epic as their primary EHR, as well as Dragon Medical for voice-assisted clinical documentation. The goal was not to replace these systems but to build an intelligence layer on top of them — one that could aggregate the relevant patient data from every available source and present it to the provider contextually, within the existing clinical workflow. Any solution had to be fast enough for real-time use, accurate enough to trust, and unobtrusive enough to actually be used during an active encounter.

Beyond the encounter context, the health system also wanted to reduce readmissions — a metric tied directly to both patient outcomes and Medicare reimbursement penalties. Understanding which discharge patients were at elevated risk for readmission required predictive capability that did not exist in their current tooling.

Nextekk built a clinical decision support platform on Azure AI Foundry, connecting into Epic via its FHIR API and integrating Dragon Medical voice analytics as an additional data signal. The platform aggregates patient history, diagnostic patterns, medication history, prior encounter notes, and voice-documented clinical observations into a unified patient context model that is available in real time during the encounter.

At the moment a provider begins an encounter, the system surfaces a structured summary of the most clinically relevant information — flagging conditions, interaction risks, and historical patterns that the provider should be aware of. The interface was designed to work within the existing Epic workflow rather than alongside it, minimizing the context-switching burden that had caused previous clinical decision tools to see low adoption.

The readmission risk model was built using Azure Machine Learning, trained on historical patient data including diagnosis codes, social determinants, prior readmission history, and discharge disposition. High-risk patients are flagged at discharge with specific contributing factors surfaced for the care coordination team, enabling targeted follow-up rather than generic post-discharge protocols. Dragon Medical analytics were integrated to identify documentation patterns correlated with readmission risk, adding a signal that had not previously been available.

In the first 90 days following deployment, the health system recorded a 12% reduction in hospital readmissions — a result that exceeded the initial project target and carried meaningful financial implications given Medicare reimbursement structures tied to readmission rates. The readmission risk model achieved high precision on the validation cohort, and care coordination teams reported that the flagged risk factors gave them actionable information rather than a generic risk score.

Provider adoption of the encounter decision support exceeded expectations. The contextual patient summary was integrated into the Epic workflow without requiring additional logins or screen switching, which drove consistent use from day one. Providers reported spending less time navigating between systems and more time in direct conversation with patients. The Dragon Medical integration surfaced voice-documented clinical signals that had previously been unstructured and therefore unavailable for analysis, adding a new dimension to the patient context model that will continue to be refined over time.