From transactional logs to clinical bundles: Applying Apriori to neurology outpatient orders

Private hospitals face rising operating costs while being expected to provide timely, high-quality diagnostic services. In outpatient neurology, diagnostic workups frequently combine laboratory panels and neuroimaging, and incomplete completion of ordered tests can delay diagnosis and reduce operational efficiency. This retrospective single-center case study applied association rule mining to 14,000 diagnostic test-order line items spanning 145 distinct tests, extracted over five months in 2025 from a private hospital in Konya, Türkiye and aggregated into de-identified encounter-level baskets. Apriori was applied with a minimum support of 0.01 and a minimum confidence of 0.30. Post-processing retained one-to-one rules and removed symmetric duplicates, reducing 124,039 raw rules to 116 candidates for clinical review. A neurology specialist grouped clinically coherent items into a 20-test HIS order set. At baseline, 1,445 of 2,408 diagnostic orders were completed (60.01%; Wilson 95% CI, 58.04%–61.95%); during the seven-month follow-up, 1,996 of 2,683 orders were completed (74.39%; 95% CI, 72.71%–76.01%). This was an absolute increase of 14.39 percentage points (Newcombe-Wilson 95% CI, 11.82–16.93) and a 23.97% relative increase (RR, 1.240; 95% CI, 1.192–1.290). The baseline and follow-up daily series comprised 114 and 144 active clinic days. Diagnostic-order volume decreased from 21.12 ± 9.15 to 18.63 ± 9.14 orders/day (mean difference, -2.49; 95% CI, -4.75 to -0.23; p=.031), while completed tests/day calculated from the aggregate counts increased from 12.68 to 13.86. Fewer tests were ordered, but a larger proportion was completed; this pattern is compatible with more selective ordering, although the uncontrolled design is descriptive and non-causal. Association rule mining combined with clinical governance may support locally tailored order-set design and monitoring.