Direct assessment of completeness of ascertainment in a stroke incidence study.
Coull AJ., Silver LE., Bull LM., Giles MF., Rothwell PM., Oxford Vascular (OXVASC) Study None.
BACKGROUND AND PURPOSE: Validity of comparisons of stroke incidence between studies or time periods depends on the completeness of ascertainment. Ascertainment cannot be reliably assessed indirectly by statistical methods, such as capture-recapture. We report the first use of direct methods to determine the completeness of different ascertainment strategies in a population-based stroke incidence study (Oxford Vascular Study). METHODS: We assessed completeness of 2 different ascertainment strategies: the core methods common to most previous incidence studies and core plus supplementary methods used in some studies (including access to carotid and brain imaging referrals and assessment of patients referred as "transient ischemic attack" or "recurrent stroke"). We assessed completeness of ascertainment in 2 ways. First, we searched anonymized primary care electronic patient records of the whole study population (n=90,542). Second, we interviewed and followed-up a high-risk subset of our study population: all patients who had an acute coronary or peripheral vascular event or a related elective investigation or intervention. RESULTS: 126 strokes were ascertained by the core plus supplementary methods, of which only 108 were identified by the core methods alone. Only 2 additional incident strokes were identified by access to primary care electronic patient records of the whole study population. Assessment and follow-up of 1103 high-risk individuals (5.5% of our total study population aged older than 60 years) identified 16 incident strokes. However, all 16 had already been ascertained by the core plus supplementary methods. CONCLUSIONS: The core methods of ascertainment used in some stroke incidence studies lead to significant underascertainment. However, direct assessment of ascertainment suggests that the supplementary methods used in recent studies can lead to near-complete ascertainment.