Where deterministic matching ends: cleaning and matching patient records with AI

Short summary

This article details a layered approach to patient record matching in Australian healthcare, starting with deterministic rules using Individual Healthcare Identifiers and normalised composite keys, then routing ambiguous pairs to a review queue. It cites JAMIA 2022 data showing probabilistic matching misses roughly a third of true matches while maintaining high precision. The author argues that matching design is a regulatory artefact under NSQHS Standard 6 and should be engineered, not left to vendor black boxes.

• •Deterministic matching resolves most duplicates; the grey-zone queue is where AI adds value
• •Probabilistic matching misses ~36% of true matches despite 99.95% PPV (JAMIA 2022)
• •Patient matching design is a documented compliance requirement under NSQHS Standard 6

Generated with AI, which can make mistakes.