Multiple database approach for study of associations between frequently used drugs and community-acquired pneumonia

Background: The association between communityacquired pneumonia (CAP) and commonly used drugs such as statins, ACE-inhibitors (ACE-I), and proton pump inhibitors (PPI) has been extensively studied in different settings and populations with often conflicting results. The origin of this heterogeneity is unknown and unravelling its nature is important for clinical interpretation of these pharmacoepidemiological (PE) results. Objectives: To explore sources of heterogeneity in the association between CAP and use of ACE-I, statins, and PPIs by using the same methods in a multi-database study in multiple settings. Methods: We used data from the TI PHARMA Mondriaan project providing access to various healthcare databases from hospitals, general practices (GP), and pharmacies. Ten different case-control sets in five different populations derived from both general practitioner (GP) and hospital data have been generated (2004-2010). Patients and controls were matched on age, gender, and index year. Conditional logistic regression was used to calculate odds ratios (OR) and 95% confidence intervals (CI) for the associations between the three drug classes of interest and CAP. Crude associations were adjusted for comorbidity and drug use (semi-adjusted; common variables in all sets), and for all available confounders (fully adjusted). Results: In total, data of 38,742 cases and 118,019 controls have been studied. The mean age of the hospitalised patients was 63 years and 46-61 years for the GP patients. For statin use and pneumonia risk the semiadjusted OR varied from 0.82 to 1.38. A comparable range was observed for ACE-I and PPI use with ORs of 1.02-1.61 and 1.29-2.69, respectively. Overall, the associations were stronger for hospitalised CAP patients matched to population controls vs. GP CAP patients matched to population controls. Furthermore, prevalence of drug exposure was higher when assessed based on dispensing data vs. prescription data. Conclusions: Associations between statin, ACE-I, and PPI use and CAP risk were influenced by sampling population and data source and may explain the large heterogeneity observed between previous observational PE studies