John Snow Labs 64d ago

Multimodal clinical data is one problem. Proving that AI decisions are correct — and defensible — is another. This session breaks down what it actually takes to move AI-driven HCC coding from assistive tooling to submission-grade systems — where every code is traceable, explainable, and audit-ready. Ritwik Jain and Hasham Ul Haq explain how risk adjustment really works, why up to 25% of conditions remain uncoded, and why accuracy alone is not enough without evidence linking and defensibility. From retrospective chart review to real-time prospective workflows, this talk shows how AI must operate under CMS constraints — where every HCC code must be backed by MEAT criteria and survive audit scrutiny. 📌 Applied Healthcare AI Summit 2026 — what actually works in real-world healthcare AI, from pilots to production systems. ⏱️ Key moments 00:00 What risk adjustment and HCC coding actually mean 03:40 Why HCC coding is still broken at scale 07:20 Retrospective vs prospective workflows in healthcare AI 11:00 What “defensible AI” really means (evidence, traceability, rationale) 12:10 How AI links codes to clinical evidence (MEAT, provenance, auditability) 14:30 From accuracy to audit-ready systems #HealthcareAI #HCCCoding #RiskAdjustment #AIinHealthcare #ClinicalAI #AIGovernance #HealthcareCompliance #MedicalCoding