Quantifying stroke incidence and mortality is crucial for disease surveillance and health system planning. Administrative data offer a cost-effective alternative to “gold standard” population-based studies. However, the optimal methodology for establishing stroke deaths from administrative data remains unclear. We aimed to determine the optimal method for identifying stroke-related deaths in administrative datasets as the fatal component of stroke incidence, comparing counts derived using underlying and all causes of death (CoD).

Using whole-population multijurisdictional person-level linked data from hospital and death datasets from South Australia, the Northern Territory, and Western Australia, we identified first-ever stroke events between 2012 and 2015, using underlying CoD and all CoD to identify fatal stroke counts. We determined the 28-day case fatality for both counts and compared results with gold standard Australian population-based stroke incidence studies.

The total number of incident stroke events was 16,150 using underlying CoD and 18,074 using all CoD. Case fatality was 24.7% and 32.7% using underlying and all CoD, respectively. Case fatality using underlying CoD was similar to that observed in four Australian "gold standard" population-based studies (20%–24%).

Underlying CoD generates fatal incident stroke estimates more consistent with population-based studies than estimates based on stroke deaths identified from all-cause fields in death registers.