Why AI and probabilistic matching are needed to fix duplicate patient records

Healthcare organisations routinely accumulate duplicate patient records when multiple systems capture the same person with inconsistent or incomplete details, such as typos, old addresses, or missing identifiers. Australia's NSQHS Standard 6 requires at least three approved patient identifiers at every point of care, and organisations must document their patient-matching processes in a way that can withstand external assessment. The Individual Healthcare Identifier, a lifelong 16-digit number tied to Medicare or DVA enrolment, offers a more reliable matching anchor than a Medicare number, though known defects in the national system mean returned identifiers can still be stale. A layered matching approach — starting with deterministic rules, then probabilistic scoring, then human review for ambiguous cases — reduces erroneous merges but still misses a significant share of true matches. Research published in JAMIA found that even well-tuned probabilistic matching missed roughly one in three duplicate pairs, meaning unresolved duplicates remain a patient safety risk when clinicians rely on incomplete records.