J Bourbeau1, S D Aaron2, N C Barnes3, K J Davis4, Y Lacasse5, G Nadeau6. 1. Respiratory Epidemiology and Clinical Research Unit, Research Institute of McGill University Health Centre and McGill University, Montreal, Quebec, Canada. Electronic address: jean.bourbeau@mcgill.ca. 2. Ottawa University, Ottawa, Ontario, Canada. 3. GlaxoSmithKline, Brantford, Middlesex, UK. 4. GlaxoSmithKline, Collegeville, PA, USA. 5. Institut universitaire de cardiologie et de pneumologie de Québec, Université Laval, Québec, Québec, Canada. 6. GlaxoSmithKline, Mississauga, Canada.
Abstract
An increased risk of non-fatal pneumonia has been documented in COPD patients treated with inhaled corticosteroids (ICS) in randomized clinical trials. Retrospective database analyses have been conducted to evaluate this signal in larger populations treated in the community. To understand how methodological choices may influence results in observational studies, we compared two recent Canadian studies which used health administrative databases from Quebec and Ontario and came to opposite conclusions on the risk of pneumonia in ICS treated COPD patients. Explanations for why the results of these studies diverged are explored. The Suissa analysis used RAMQ data from Quebec and showed an increased relative risk of serious pneumonia for current users of ICS compared to non users, RR = 1.69 (95% confidence interval, 1.63-1.75). The Gershon analysis used ODB data and showed no difference for pneumonia hospitalisation, RR = 1.01 (0.93-1.08). Reasons for differences in study findings include lack of validated definitions of COPD, poor selection of relevant exposure groups, channeling and confounding biases, and failure to perform on-treatment analyses for safety. CONCLUSION: Our study identifies methodological features that need consideration to increase robustness and minimize threats to internal validity of retrospective health administrative database studies.
An increased risk of non-fatal pneumonia has been documented in COPDpatients treated with inhaled corticosteroids (ICS) in randomized clinical trials. Retrospective database analyses have been conducted to evaluate this signal in larger populations treated in the community. To understand how methodological choices may influence results in observational studies, we compared two recent Canadian studies which used health administrative databases from Quebec and Ontario and came to opposite conclusions on the risk of pneumonia in ICS treated COPDpatients. Explanations for why the results of these studies diverged are explored. The Suissa analysis used RAMQ data from Quebec and showed an increased relative risk of serious pneumonia for current users of ICS compared to non users, RR = 1.69 (95% confidence interval, 1.63-1.75). The Gershon analysis used ODB data and showed no difference for pneumonia hospitalisation, RR = 1.01 (0.93-1.08). Reasons for differences in study findings include lack of validated definitions of COPD, poor selection of relevant exposure groups, channeling and confounding biases, and failure to perform on-treatment analyses for safety. CONCLUSION: Our study identifies methodological features that need consideration to increase robustness and minimize threats to internal validity of retrospective health administrative database studies.