AI-Driven Clinical Decision Support for a Top-10 U.S. Health System
What was at stake
The health system's clinicians were spending an average of 22 minutes per patient on diagnostic review, relying on fragmented data across siloed EHR modules. Missed and delayed diagnoses were contributing to preventable adverse events, and the organization needed a way to surface evidence-based recommendations in real time without disrupting existing clinical workflows.
How we delivered
Clinical Workflow Discovery
Embedded with physicians, nurses, and clinical informaticists across three flagship hospitals to map diagnostic decision points, data dependencies, and EHR interaction patterns — identifying 47 high-impact intervention opportunities.
ML Model Development & Validation
Trained a multi-modal ensemble model on 2.1M de-identified patient records, incorporating lab results, imaging metadata, and clinical notes. Validated against board-certified physician panels achieving 94.7% accuracy with a false-negative rate under 1.2%.
Epic EHR Integration
Built a FHIR-compliant middleware layer that connects to Epic via CDS Hooks, surfacing real-time recommendations directly within the physician's existing workflow — no tab-switching, no separate application.
Phased Rollout & Monitoring
Deployed across 120+ care sites in three phases over 16 weeks, with real-time model performance dashboards, clinician feedback loops, and automated drift detection to maintain accuracy post-launch.
Measurable impact, verified by the client
Technologies we used
“We evaluated four vendors over six months. TPWITS was the only team that understood both the clinical workflow and the ML engineering required. Diagnostic accuracy improved to 94.7%, and our clinicians adopted the tool within weeks — not months.”
Power your next digital move.
Whether you need AI expertise, cloud infrastructure, or a full digital transformation, our team is ready to help you build what's next.