Patient Record

Patient history lives in the memories of whoever treated the patient last. When a new physician asks 'what medications is this patient on?' the answer is 'check with Dr. Patel — she saw him last month.' Paper charts may exist in storage somewhere, but nobody can find them quickly enough to matter during a clinical encounter.

None — AI cannot perform any clinical analysis because no patient record exists in any accessible system.

Patient records exist but vary wildly by provider. One physician dictates notes into a recorder, another scribbles on paper forms, a third uses a personal Word document. The front desk has a folder with the patient's insurance card and a faxed referral from two years ago. Finding the patient's complete medication list means calling the pharmacy.

AI could potentially digitize paper records via OCR, but cannot reliably assemble a complete patient history because records are scattered across formats and locations with no consistent structure.

All providers use the same EHR and patient records have standard sections — demographics, problem list, medication list, allergies, encounter notes. The record is findable and consistently structured. But the depth varies: some providers write detailed notes while others enter the minimum required. Historical records from before the EHR migration are scanned PDFs that nobody reads.

AI can generate basic patient summaries and flag medication allergies from the structured fields, but cannot reason across the full clinical picture because scanned legacy records and inconsistent note depth limit what is machine-readable.

Patient records are comprehensive and current in the EHR with structured problem lists, coded diagnoses (ICD-10), medication histories with start/stop dates, and encounter notes that follow documentation templates. A care coordinator can pull up any patient and see their complete clinical picture without calling anyone or opening another system.

AI can perform clinical decision support — flagging drug interactions, suggesting diagnoses based on symptom patterns, and generating care gap alerts. Cannot yet predict clinical trajectories because historical progression patterns are not systematically encoded.

Patient records are schema-driven with formal entity relationships — every diagnosis links to the encounter where it was identified, every medication links to the prescribing provider and clinical indication, and every lab result links to the ordering context. An AI agent can ask 'show me all HbA1c trends for this diabetic patient correlated with medication changes' and get a structured answer.

AI can perform predictive clinical analytics — forecasting disease progression, recommending treatment adjustments based on population comparisons, and generating personalized risk scores. Fully autonomous clinical decisions are possible for protocol-driven scenarios like insulin dose titration.

The patient record is a living, continuously updating entity. Every clinical observation, device reading, lab result, and provider interaction flows into the record in real-time. The record self-documents — when a nurse takes vitals, the record updates before she finishes charting. AI agents consume patient record events as a continuous stream and reason over the complete longitudinal context as it evolves.

Can autonomously manage routine clinical monitoring, trigger real-time alerts for deterioration, generate discharge summaries, and maintain care plans as a living document. The patient record is a real-time knowledge base, not a static chart.

Ceiling of the CMC framework for this dimension.