Validity of stroke diagnoses in a National Register of Patients

Abstract

Background: Many registers containing routine medical information have been developed for research and surveillance purposes. In epidemiological research assessment of endpoints is often conducted via registers. In the present study we validated stroke and transient ischemic attack (TIA) diagnoses in the Danish National Register of Patients (DNRP).

Methods: Subjects from a Danish cohort study, the Copenhagen City Heart Study (n = 19,698), were crosslinked with the DNRP. The following International Classification of Disease 10th revision codes were used to identify possible strokes and TIAs: I60-I69 and G45. Two independent raters reviewed all cases. Positive predictive values of stroke, TIA and stroke subtypes were estimated by dividing the confirmed cases by the total number of cases located in the DNRP. Interrater reliability was tested using kappa statistics.

Results: Of 236 possible cerebrovascular events, 1 in 6 stroke diagnoses did not meet study criteria. The majority of events in the DNRP were registered as unspecified stroke (I64), n = 105 (44%), of which two thirds were diagnosed as ischemic stroke events by the raters. Intracerebral hemorrhage and ischemic stroke had a positive predictive value from 74 to 97%, respectively.

Conclusion: Our results show that the DNRP tends to overestimate the number of cerebrovascular events, while ischemic stroke is underestimated.