Accuracy of death certification and hospital record linkage for identification of incident stroke.
BMC Medical Research Methodology 2008 ; 8: 74.
Sinha S, Myint PK, Luben RN, Khaw KT
DOI : 10.1186/1471-2288-8-74
PubMed ID : 19000303
PMCID : PMC2605452
URL : https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/1471-2288-8-74
There is little information on the validity of using record linkage with routinely collected data for case ascertainment of stroke in large population-based studies in the UK. We examined the accuracy of these routine record linkage approaches for identifying incident stroke cases in a large UK population-based study, the European Prospective Investigation into Cancer (EPIC)-Norfolk cohort.
We examined a sample of hospital records of incident stroke cases identified by linkage with two routine data sources, death certificates and a national hospital record linkage system (ENCORE), using predefined study criteria. Two senior Specialist Registrars with clinical experience in stroke medicine examined the hospital records and searched for the evidence of stroke recorded in these records between 1993/97-2003.
Of 520 incident strokes identified between 1993/1997-2003 using record linkage systems in the EPIC-Norfolk, a sample of 250 medical case notes were examined between March and July 2004. Using the predefined study criteria, there were 191 definite strokes (76%), 20 probable strokes (8%), 11 possible strokes and 11 cases of transient ischaemic attacks (4% each) i.e. 233/250 (93%) with possible or definite stroke or transient ischaemic attacks. Stroke could not be verified using hospital records in 13 cases (5%) and 4 cases (2%) had other diagnoses: 3 cases of vascular dementia and 1 case of benign intracranial hypertension. The diagnosis of stroke in 185 out of 250 cases identified in the EPIC-Norfolk (74.0%) was supported by radiological evidence using WHO criteria.
Death certificates and hospital record linkage in this British prospective study have a high accuracy or positive predictive value in correctly identifying incident stroke cases.