Background: Large linked datasets are increasingly being used to determine stroke incidence in Australia. We aimed to determine how lookback periods affect the number of first-ever strokes estimated.
Methods: We obtained emergency department and hospital admissions for stroke 2001-2019 in Victoria and New South Wales (NSW), supplemented National Death Index data. Diagnosis codes in hospital administrative datasets and cause of death codes were used to identify episodes of stroke (ICD-codes I60-I61, I62.9, I63-I64). Hospitalisations and death registrations with overlapping dates were considered a single episode, with episodes containing any records with a principal diagnosis of stroke categorised as stroke episodes. For each year 2006-2012, up to a 5-year lookback period was used to determine whether an episode was a first-ever stroke (incident stroke).
Results: Of >2.5 million records, approximately 300,000 had a principal diagnosis of stroke (Victoria=126,437; NSW=167,894). After merging records based on overlapping dates, 106,113 episodes of stroke in Victoria (approximately 7,000/year) and 122,003 episodes in NSW (approximately 11,000/year) were identified. The drop in first-ever stroke counts with a 1-year lookback period was 1.09%, 2.46% with a 2-year lookback, 3.51% with a 3-year lookback and 4.28% with a 4-year lookback. Approximately 4.3% in Victoria and 5.5% in NSW had a prior stroke episode in the previous 5-years.
Conclusions: Unlinked data provide a substantial overestimation of the number of first-ever strokes occurring. Wider access to linked datasets with longer data series will enable more regular and accurate calculation of incidence of stroke