BACKGROUND: Following the publication of a meta-analysis reporting a risk of acute myocardial infarction (AMI) with rosiglitazone that led to severe restrictions being placed on its use, several observational studies of the association were reported. The lifting of restrictions in the United States in 2013 makes a review of these studies pertinent. OBJECTIVE: To evaluate the quality of population-based observational studies of the rosiglitazone-AMI association. METHODS: PubMed and Embase literature databases were searched for observational studies evaluating the association that were published between 2006 and 2010. Publications satisfying the inclusion criteria were reviewed using the Checklist for Retrospective Database Studies. RESULTS: Nineteen studies satisfied the inclusion criteria. Reasons for the research design and data source were absent or unclear in 18 (95%) and 16 (84%), respectively. Administrative data were used exclusively in 14 (74%). Baseline periods for prior diagnoses and medications varied widely. Reimbursement constraints on rosiglitazone use were reported in only seven studies (37%), although all were likely to have been impacted by them. What was being tested in half of the rosiglitazone treatment comparisons lacked specificity and clarity. All relied on risk ratios and, for 90% of the comparisons, the ratios were between 0.5 and two - a level at which residual confounding can lead to spurious significance. CONCLUSION: Important deficiencies existed in the rosiglitazone studies suggesting that standards for methods and reporting of observational safety analyses need improvement. In particular, detailed clinical data should be included when the risk of confounding by indication is likely to be high.
BACKGROUND: Following the publication of a meta-analysis reporting a risk of acute myocardial infarction (AMI) with rosiglitazone that led to severe restrictions being placed on its use, several observational studies of the association were reported. The lifting of restrictions in the United States in 2013 makes a review of these studies pertinent. OBJECTIVE: To evaluate the quality of population-based observational studies of the rosiglitazone-AMI association. METHODS: PubMed and Embase literature databases were searched for observational studies evaluating the association that were published between 2006 and 2010. Publications satisfying the inclusion criteria were reviewed using the Checklist for Retrospective Database Studies. RESULTS: Nineteen studies satisfied the inclusion criteria. Reasons for the research design and data source were absent or unclear in 18 (95%) and 16 (84%), respectively. Administrative data were used exclusively in 14 (74%). Baseline periods for prior diagnoses and medications varied widely. Reimbursement constraints on rosiglitazone use were reported in only seven studies (37%), although all were likely to have been impacted by them. What was being tested in half of the rosiglitazone treatment comparisons lacked specificity and clarity. All relied on risk ratios and, for 90% of the comparisons, the ratios were between 0.5 and two - a level at which residual confounding can lead to spurious significance. CONCLUSION: Important deficiencies existed in the rosiglitazone studies suggesting that standards for methods and reporting of observational safety analyses need improvement. In particular, detailed clinical data should be included when the risk of confounding by indication is likely to be high.