Background: Population-based administrative registers could be used for identifying heart failure (HF) cases. However, the validity of the classification obtained from administrative registers is not known.
Design: The validity of HF diagnoses obtained by record linkage of administrative databases in Finland was assessed against classification by three independent physicians.
Methods: Data from the nationwide registers in Finland - the Hospital Discharge Register, Causes of Death Register, Drug Reimbursement Register, and pharmacy prescription data - were linked with the FINRISK 1997 survey data. Cases with hospitalizations before the survey date with HF as one of the discharge diagnoses, cases with special reimbursement for HF drugs before the survey date and cases with the use of furosemide before the survey date were classified as HF in the registers. All these cases, cases with baseline brain natriuretic peptide > 100 pg/ml, and cases with use of digoxin were independently assessed by two physicians as HF/no HF. Discrepant cases were solved by a third physician. This classification was considered as the gold standard, against which the registers were assessed.
Results: The specificity of the registers was 99.7% (95% CI 99.5-99.8%), positive predictive value 85.9% (95% CI 79.7-90.5%), negative predictive value 97.9% (95% CI 97.6-98.2%), and sensitivity 48.5% (95% CI 42.9-54.2%).
Conclusions: Classification obtained from administrative registers has high specificity and can be used in follow-up studies with HF as an end point. Sensitivity is modest and administrative data should be used with caution for surveillance.