Background: Stroke prevalence is an important cross-sectional measure of disease burden for a population. However, most stroke prevalence analyses rely on self-report to identify prevalent cases. Linked administrative data may be a pragmatic means of obtaining such estimates, but the optimal methodology has not been developed.
Aims: To compare different definitions of stroke prevalence.
Methods: This study utilised linked Western Australian cardiovascular disease hospitalisation and mortality data. All strokes were identified between 01/01/1985 and 30/06/2018 from any diagnosis field (ICD-10-AM I60, I61, I62.9, I63, I64 and I69, and equivalent ICD-9-CM codes). For the standard definition, prevalent cases of stroke were identified as all people admitted to hospital for stroke in the 33-year lookback period and still alive at 30/06/2018. Alternative definitions compared shorter lookback periods and other modified parameters including a subset of ICD-10-AM codes.
Results: Using the standard prevalence definition, 26,058 prevalent cases were identified. Reducing the lookback period resulted in a decrease in prevalent cases (1.7% with a 30-year lookback, 32.9% with a 10-year lookback and 87.8% using a 1-year lookback). The reductions were similar irrespective of age and sex. Compared to the standard stroke definition, the exclusion of I69 codes (stroke sequalae) resulted in an average prevalence case loss of 7.2% for lookback periods ranging 33 to 10 years, and between 8.8% to 20.7% for lookback periods ranging 10 to 1 year. Alternative stroke definitions showed consistent results compared to the standard definition for lookback periods of at least 10 years, however, a greater/lesser loss of cases were seen for shorter lookback periods.
Conclusion: Modifications to the definition of stroke have an impact on the identification of prevalent cases. Lookback periods of less than 10 years are suboptimal for identifying stroke prevalence within a linked administrative dataset. Stroke sequelae should be included in case identification.