AI Patient Record Validation Agent

How Cassidy automates this using AI

Step 1: Ingest and normalize patient data

The Workflow triggers on incoming ADT feeds, CCD(A) documents, or FHIR API payloads. Cassidy normalizes demographics and clinical data against internal standards and US Core expectations—applying data quality rules for formatting, required fields, and address standardization.

Step 2: Execute identity matching and generate worklists

Cassidy runs deterministic and probabilistic matching algorithms against your EMPI, scoring candidate pairs and applying configurable thresholds. Potential duplicates route to HIM stewardship worklists, while survivorship rules determine attribute-level sources of truth for the golden record.

Step 3: Validate FHIR conformance

The Workflow validates each resource instance against US Core profiles, checking Must Support elements, terminology bindings, and profile-specific constraints. Cassidy flags IG version mismatches, missing required content, and ValueSet errors—generating conformance reports with actionable remediation guidance.

Step 4: Extract clinical evidence and link to diagnoses

Cassidy pulls source documents tied to encounter dates and rendering providers, then uses NLP to locate supporting passages for each diagnosis. The system applies MEAT criteria and HCC mapping rules, de-duplicating conflicting evidence and classifying findings as supported, unsupported, or needs review.

Step 5: Generate audit-ready deliverables

The Workflow produces evidence dossiers with exact page citations, timestamps, and Provenance trails. Cassidy routes insufficient documentation to provider query workflows and pushes demographic corrections back to source systems—creating a complete remediation log for compliance.