Atrial fibrillation in patients with ischaemic stroke in the Swedish national patient registers: how much do we miss?

AIMS Data from national discharge registers are commonly used to estimate prevalence and incidence of atrial fibrillation (AF) in epidemiology studies. However, sensitivity and specificity of register-based AF diagnosis have not been evaluated. We sought to assess the validity of AF diagnosis in the Swedish Patient Register against electrocardiography (ECG) documentation of AF. METHODS AND RESULTS The study sample comprised of 336 patients [median age 76 (interquartile range (IQR) 67-82 years, 136 female] with first-ever ischaemic stroke, enroled in the Lund Stroke Register from March 2001 to February 2002 and 1 : 1 age- and gender-matched control subjects without stroke from the population register. Data was exported from the patient register in October 2011 (the end of follow-up). Atrial fibrillation documentation by ECG was assessed using an electronic archive containing all ECGs taken in the hospital catchment area starting in 1988. A total of 7247 ECGs were reviewed, with the median number of ECGs per person being 7.5 (IQR 3-15). Atrial fibrillation was detected by ECG in 190 patients; and in 188 patients by linkage with patient register. In most patients, AF was documented first by ECG data, with median time to register diagnosis being 16 days (IQR 3-859). Specificity of AF diagnosis in the Swedish Patient Register was 93%, sensitivity was 80%. CONCLUSION Despite the high specificity, AF diagnosis in the Swedish Patient Register assessed in the population of ischaemic stroke patients and age- and gender-matched control subjects has modest sensitivity, which may result in underestimating prevalent and incident AF cases if only register data are used for identification of subjects with AF in epidemiology studies.