Author(s): Shashank Bharadwaj ; Aastha Chaudhary
Date: 2025
Origin: Street Art & Urban Creativity Scientific Journal
Subject(s): machine learning operations, MLOps, dataset shift, concept drift, statistical process control, ICD-10-CM, clinical terminology, healthcare machine learning, model monitoring, HIPAA, change control, model risk management.
Machine learning models deployed in healthcare read their inputs through administered clinical terminologies: ICD-10-CM diagnoses, CPT and HCPCS Level II procedures, LOINC laboratory observations, SNOMED CT concepts, and NDC drug identifiers. Most are revised by their maintainers on published calendars, although the cadences differ and NDC content turns over continuously with the drug market rather than on a fixed effective date. The fiscal year 2026 ICD-10-CM addenda, effective October 1, 2025, added 487 diagnosis codes, revised 38, and deleted 28. The April 1, 2026 update changed no codes at all, yet revised index entries that move documented neuroendocrine tumors from a benign category to a malignant one, altering severity capture without altering a single code. Release magnitude and release type therefore vary sharply between consecutive boundaries, and a release containing no code changes is not a null event for a deployed model.....
