When AI reads a diagnosis or procedure billing code, it sees only what fits inside that administrative category — the clinical specificity is not always captured with structured data. For healthcare AI to operate safely, it needs a structured data foundation with the granularity, synonymy, and contextual relationships that clinical reasoning actually requires. MEDCIN, Medicomp’s clinical data foundation underlying the Quippe Clinical Knowledge Graph, provides more than 430,000 clinical concepts and 100 million diagnostic relevancy links that add specificity and context — the deterministic foundation that reliable, responsible AI demands.
A diagnosis code tells you how a patient gets billed. It doesn’t tell you how they’re doing – that information lives elsewhere.
For decades, the distance between codes and a patient’s full clinical picture was a manageable inconvenience. Clinicians filled the space between code and clinical reality with chart review, hallway conversations, and bedside verification. Healthcare AI has none of those workarounds. When the data it reads is a billing code, AI sees only what fits inside that category. Closing the gap is now a central reliability problem in healthcare AI, and it requires a clinical data foundation purpose-built for the way clinicians think and reason.
What a Code Leaves Out
Two ICD and CPT examples illustrate how a single classification code can cram many distinct clinical realities into one billing category – and lose the clinical context.
The International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) code C72.9, “malignant neoplasm of central nervous system, unspecified,” is a single billing category that clinically covers approximately 100 distinct conditions, each with different diagnostic workups, treatment protocols, and prognoses. Astroblastoma, a rare glial tumor of the CNS, is one of them. It has no code of its own in ICD-10. Clinical ontologies designed for precision, including MEDCIN and SNOMED CT, distinguish it as a discrete, structured concept with the specificity that diagnosis and treatment require. MEDCIN captures all 100 of the conditions within the disease hierarchy with discrete clinical terms. ICD-10 captures one.
Clinical Coding Specificity: ICD-10 & MEDCIN In brief: ICD-10-CM code C72.9 — “malignant neoplasm of central nervous system, unspecified” — is a single billing category that clinically covers approximately 100 distinct conditions with different diagnostic workups, treatment protocols, and prognoses. MEDCIN represents each of those conditions as a discrete structured concept. An AI system relying on C72.9 alone cannot distinguish between them. One that operates on MEDCIN’s clinical foundation can.
Procedural coding shows the same one-to-many relationship in a different way: anatomic location. For example, Current Procedural Terminology (CPT) code 11311 covers the shaving of epidermal or dermal lesions on six different anatomic sites: face, ears, eyelids, nose, lips, and mucous membranes. The single code corresponds to six MEDCIN terms, each identifying the site on which the procedure was performed.
Clinical Coding Specificity: CPT & MEDCIN In brief: CPT code 11311 covers shaving of epidermal or dermal lesions across six anatomic sites — face, ears, eyelids, nose, lips, and mucous membranes — without differentiating between them. MEDCIN maps six distinct clinical terms to that single code, each identifying the specific site. For AI systems making or validating clinical decisions, anatomic specificity is not optional. A code that omits it is a gap the AI cannot fill on its own.
A Growing Problem
Several developments are making this gap more consequential than at any previous point. Most immediate is the arrival of true interoperability across health systems. After decades of partial information sharing, real cross-system data exchange is just beginning to take shape, and the codes that travel between systems carry the same limitations they had inside a single chart.
Equally significant is the mounting regulatory and clinical pressure on data quality. Precision medicine, clinical trial matching, and risk adjustment all depend on a level of specificity that classification codes were never designed to provide.
The Centers for Medicare & Medicaid Services (CMS) is moving toward models in which unspecified codes lose value, and the FY 2026 ICD-10-CM update introduced nearly 500 new codes emphasizing clinical granularity. That update is an acknowledgment from within the classification system itself that its current resolution is no longer sufficient.
Designed for Clinical Thinking
Closing the gap starts with a distinction the industry has tended to blur, between three categories of medical terminology that serve fundamentally different purposes. Classification terminologies such as ICD-10 organize encounters into administrative categories. Reporting terminologies, such as CPT, support billing and program reporting. Both serve essential administrative functions, and capturing how clinicians actually think about a patient was never the job of either.
Conversely, interface terminologies are designed for clinical thinking. They mirror the granularity, synonymy, and contextual relationships of clinical reasoning, connecting symptoms, history, exam findings, tests, treatments, and diagnoses to one another at the level of detail a clinician needs. The classification and reporting codes a system uses are then derived from that richer foundation.
MEDCIN was built to be exactly that kind of interface terminology – not retrofitted to support clinical thinking but designed for it from the start more than 45 years ago. That design decision is reflected in its scale: over 430,000 clinical concepts spanning all domains of medicine, 100 million diagnostic relevancy links, and 10 million mappings to standard terminologies including SNOMED CT, ICD-10, CPT, RxNorm, and LOINC. Each concept carries the metadata that clinical thinking requires, and that AI needs to operate safely: laterality, severity, risk, complexity, and direct relevancy links across symptoms, history, tests, and treatments. MEDCIN is the deterministic foundation that reliable AI requires.
The Foundation AI Will Demand
Healthcare AI will only be as reliable as the data underneath it. For organizations and developers building intelligent clinical systems, that means treating clinical specificity as a baseline requirement and embedding validation into every workflow that creates, exchanges, or acts on clinical data.
MEDCIN provides that foundation. As the clinical data layer underlying Medicomp’s Quippe Clinical Knowledge Graph and Clinical Intelligence Engine, MEDCIN doesn’t just supply concepts — it supplies the diagnostic context along with the metadata AI needs to validate its own outputs. That structure and scope is what allows Quippe to serve as a guardrail for AI rather than simply a data source for it.
Large language models are not well suited at coding tasks. They are probabilistic by design and make things up (or “hallucinate”) convincingly. Those hallucinations within production systems in healthcare can mean unrecoverable errors.
Deterministic technologies like the Quippe Clinical Knowledge Graph have validated, fixed links between concepts and the correct codes. Deployed within EHRs, practice management systems and other health technologies, with or without LLM-based AI tools, Quippe delivers the reliable coding layer needed as part of the technical stack. This bypasses the probabilistic hallucination problem entirely, and lets LLM-based AI do what it does best – which is interacting with humans.
It is not 1980 anymore, and the data demands of healthcare AI reflect that. The clinical specificity that earlier health IT solutions could not act on is now exactly what AI tools depend on. The data foundation underneath must be precise enough to serve as a guardrail for responsible AI, protect patients, and deliver on the promise of intelligent care.
See how Quippe puts MEDCIN to work. If you’re building clinical AI applications, evaluating data infrastructure, or closing the gap between your coding environment and clinical reality, explore Quippe’s Clinical Knowledge Graph and integration options — or speak with an expert about your specific use case.