20 Statistical analyses Follow-up began at hospital admission with pneumonia and continued until 1 January 2013, or until death or migration, whichever occurred first. We computed the cumulative incidence (risk) of arterial thromboembolism

within 30 days of admission Z-VAD-FMK mechanism in patients with and without pre-existing AF and accounted for the competing risk of death in the analysis.21 Further comparisons were performed using Cox regression analyses to estimate crude HRs, and HRs adjusted for the thromboembolism risk factors in the CHA2DS2-VASc score. We also performed an analysis stratified by the presence or absence of previously diagnosed arterial thromboembolism to examine whether an association between pre-existing AF and development of arterial thromboembolism was related to repeated events. Among patients with and without AF, the effect of preadmission treatment with vitamin K antagonists or aspirin on risk of arterial thromboembolism was evaluated by comparing users to non-users. We repeated the analyses for patients without contraindications for anticoagulant therapy. All diagnoses coded during a given admission were assigned to the same discharge date in the

DNPR. Consequently, we were unable to assess the actual temporal relationship between pneumonia onset and arterial thromboembolism in patients treated for the two conditions during the same admission. Arterial thromboembolism might have preceded pneumonia in some patients. Thus, we reassessed the 30-day risk of arterial thromboembolism, and only considered diagnoses assigned after discharge from the index pneumonia admission. Cumulative mortality risks at 30 days and 1 year after admission with pneumonia in patients with and without AF were assessed using the Kaplan-Meier method and compared using Cox regression. The HRs were adjusted for sex and age. Further adjustments were performed for the conditions included in the Charlson Index, and for valvular heart disease, alcoholism and obesity, and GP contacts regarding preventive consultations, social-medicine-related consultations,

conversational therapy, vaccination for influenza and GSK-3 reimbursement due to chronic or terminal illness. To examine whether the association between AF and pneumonia mortality was related to coexisting cardiovascular diseases or pneumonia severity, the analyses were repeated using stratification by previous myocardial infarction, congestive heart failure, treatment with mechanical ventilation, and admission to the intensive care unit during the index admission. We also stratified the data by AF status and compared mortality at 30 days and 1 year in users and non-users of vitamin K antagonists, aspirin, β-blockers, calcium-channel blockers, digoxin, amiodarone and statins using the Kaplan-Meier method and Cox regression analyses. Crude HRs and HRs adjusted for the potential confounders described above were estimated from the Cox regression analyses.