Medical coding feels modern — it lives inside EHRs, runs on payer rules engines, and increasingly leans on large language models. But the underlying idea, that we should classify diseases consistently across institutions and across countries, dates back more than a century. Knowing the lineage makes it much easier to understand why the code sets look the way they do, and where they are going next.
1893 — The Bertillon Classification
In 1893, the French statistician Jacques Bertillon presented the Bertillon Classification of Causes of Death to the International Statistical Institute. It was the first widely adopted system for classifying mortality across countries, and it is the direct ancestor of every modern ICD revision. The motivation was simple: comparing public-health statistics across nations was impossible until everyone agreed on what to call each disease.
1948 — The WHO takes over and ICD-6 expands beyond mortality
After World War II, the newly formed World Health Organization took ownership of the classification. ICD-6, published in 1948, was the first revision to include both morbidity (illness) and mortality, broadening the scope from death certificates into hospital and clinic statistics. From here on the ICD became a living, internationally maintained standard.
1979 — ICD-9 and the U.S. clinical modification
The U.S. adopted ICD-9 in the late 1970s and developed its own clinical modification (ICD-9-CM) for inpatient hospital reporting. Volume 3, the procedure-coding section, was a U.S. invention — the WHO's ICD did not originally contain procedure codes, only diagnoses.
Around the same time, the American Medical Association published the first edition of CPT (Current Procedural Terminology) in 1966, and CPT became the dominant outpatient and physician-procedure code set. HCPCS was layered on top in the 1980s for items CPT did not cover (durable medical equipment, supplies, ambulance, certain drugs).
2015 — The U.S. transitions to ICD-10
After more than a decade of delays, the U.S. transitioned from ICD-9 to ICD-10 on October 1, 2015. The diagnosis code set jumped from about 14,000 codes to about 70,000, and a brand-new procedure code set, ICD-10-PCS, replaced ICD-9-CM Volume 3 with a 7-character procedural grammar.
The transition was controversial at the time but is now broadly viewed as a success — the additional specificity has improved analytics, audit precision, and risk-adjustment accuracy.
Today — ICD-11, AI, and the next chapter
ICD-11 was endorsed by the WHO in 2019 and is being adopted by countries on different timelines. The U.S. has not yet committed to a transition date, but the code set is more digital-native than ICD-10 and supports a much richer post-coordination model (combining base codes with extension codes for laterality, severity, and stage).
Layered on top is the AI revolution. Computer-assisted coding (CAC) tools have existed for decades, but they were rule-based NLP systems that suggested codes and left assembly to the human. Modern large-language-model approaches go further — they can read a chart, propose the full coded output for an encounter, and explain the reasoning. That is the chapter we are writing right now.
Curious how AI fits into a 130-year-old profession? Our pipeline grounds every code in the official corpus and shows the chart span that justified it — bringing modern automation without breaking the audit trail the field was built on.