15) (table 4). Discussion In this population-based study of 88 315 patients hospitalised with pneumonia, we found that compared with patients without AF, patients with pre-existing AF had a 60% higher risk of arterial http://www.selleckchem.com/products/INCB18424.html thromboembolism within 30 days of admission and 50% higher

mortality after 30 days and 1 year. However, after controlling for other prognostic factors associated with AF, the excess risk of arterial thromboembolism and mortality was found to be related to major differences in age and coexisting diseases, rather than to AF directly. There was a robust association between preadmission use of vitamin K antagonists and reduced risk of arterial thromboembolism following pneumonia in patients with AF. We also observed markedly reduced risks of death in patients with AF who were treated with statins, β-blockers or vitamin K antagonists. However, compared with non-users, there was an increased risk of death in users of amiodarone and digoxin. Strengths and limitations The risk of selection bias was minimal in this study because we used population-based data with virtually complete follow-up. A positive predictive value of 90% has been estimated for the pneumonia diagnoses recorded in the DNPR; 87% of the cases represent community-acquired infections.22 Atrial fibrillation is also accurately coded in the DNPR. Positive predictive value estimates range from 93% to 99%.23–25 Since AF and

atrial flutter share the same code in the DNPR, we were unable to distinguish between these arrhythmias. However, of patients coded with atrial fibrillation/flutter, only 5–6% have atrial flutter.23–25 The coding system also did not allow for differentiation between paroxysmal, persistent and permanent AF. Finally, the DNPR’s

positive predictive value for ischaemic stroke is 88–100%. Of the patients coded with unspecified stroke, 57–70% have ischaemic stroke.26 27 Since we used filled prescriptions as a measure of actual drug intake, non-adherence or treatment discontinuation before admission could have biased the effect of any given drug towards the null value. However, we concluded that this source of bias was a low concern for the results of this study. We did not have data on in-hospital medications; thus, we were unable to assess whether pre-admission treatments were continued, discontinued or altered (eg, change Cilengitide of vitamin K antagonists to low-molecular heparins) during admission. Any bias introduced by the discontinuation of a drug during admission would be directed towards the null. Use of the prophylactic drugs (eg, statins and β-blockers) could be markers for unmeasured, but greater, health awareness among the patient population. However, the welfare structure of the Danish society reduces the risk of confounding by differences in socioeconomic status; in Denmark, statin users have unhealthier lifestyles than statin non-users.