Background: Regular monitoring of stroke incidence allows for effective health services planning. Administrative data to estimate stroke incidence is cost-effective, geographically comprehensive, and continuous compared to registries, audits, but suffer from diagnostic/misclassification bias and completeness of non-hospitalised case-ascertainment compared to high-quality population-based incidence studies. We aimed to determine whether administrative data can be used to accurately determine stroke incidence.
Methods: Using data from the Integrated Data Infrastructure maintained by Statistics New Zealand, we identified first stroke events from two Auckland public hospitals’ discharge data between September 2020 and August 2021 using ICD-10-AM codes for subarachnoid haemorrhage, intracerebral haemorrhage, cerebral infarction, and unspecified stroke. Stroke incidence from the same Auckland region was compared with data from the population-based fifth Auckland Regional Community Stroke Study (ARCOS V) from September 2020 to August 2021, and included both hospitalised and non-hospitalised stroke.
Results: A total of 1,569 stroke events were identified from administrative and 1,611 from ARCOS V data. Comparisons showed 86.8% (95%CI 85.1-88.5) sensitivity and a positive predictive value of 89.3% (87.8-90.8). Sensitivity was similar for different ethnicities: 85.7% (83.6-87.8) for Europeans, 89.5% (83.3- 95.6) for Māori, and 90.1% (85.6-94.5) for Pacific Peoples. Of the 210 false-negatives, 50% had no administrative hospitalisation, 47.1% had a secondary stroke diagnosis, and very few (<1%) were referred via community sources.
Conclusion: The accuracy of administrative hospital discharge data for stroke in the Auckland region is good, and comparable across ethnic groups and approximates incidence data from ARCOS V opening opportunities for its use in future research.