Predictive value of stroke discharge diagnoses in the Danish National Patient Register

Aims: To determine the positive predictive values for stroke discharge diagnoses, including subarachnoidal haemorrhage, intracerebral haemorrhage and cerebral infarction in the Danish National Patient Register. Methods: Participants in the Danish cohort study Diet, Cancer and Health with a stroke discharge diagnosis in the National Patient Register between 1993 and 2009 were identified and their medical records were retrieved for validation of the diagnoses. Results: A total of 3326 records of possible cases of stroke were reviewed. The overall positive predictive value for stroke was 69.3% (95% confidence interval (CI) 67.8-70.9%). The predictive values differed according to hospital characteristics, with the highest predictive value of 87.8% (95% CI 85.5-90.1%) found in departments of neurology and the lowest predictive value of 43.0% (95% CI 37.6-48.5%) found in outpatient clinics. Conclusions: The overall stroke diagnosis in the Danish National Patient Register had a limited predictive value. We therefore recommend the critical use of non-validated register data for research on stroke. The possibility of optimising the predictive values based on more advanced algorithms should be considered.