BACKGROUND: The evaluation of methods and risk of bias in observational studies included in systematic reviews is essential to interpretation of results. In the context of the SAFEGUARD project, we conducted a systematic review of observational studies on the use of glucose-lowering drugs and the risk of cardiovascular events.
OBJECTIVES: To describe the methodological issues of included studies, using the RTI item bank (RTI IB).
METHODS: The RTI IB consists of 29 items organized in 11 domains. We added another 4 items (a 12th domain) to evaluate aspects specific to this research question. Methodological issues covered by the 12 domains were summarized in study population, exposure, outcome, analysis, immortal-time bias, confounding by indication and residual confounding, and formulary restrictions. Two investigators independently evaluated methods and risk of bias in 44 included studies.
RESULTS: Although some studies did not describe the study population in detail, only four studies were at high risk of bias in this domain. Exposure definition was very heterogeneous across studies and the information on how exposures were assessed was incomplete. The most frequent outcomes were acute myocardial infarction and heart failure. Very few studies conducted validation by comparing source medical records. Statistical methods were evaluated as appropriate in 57% of studies. Overall, 43% and 30% of the studies were at high risk of confounding by indication or of residual confounding, respectively; 25% of the studies were at risk of immortal-time bias. Few studies provided information on formulary restrictions. Dose information was very limited, and there was no information on duration.
CONCLUSIONS: There was great heterogeneity across the included studies regarding outcomes, exposures of interest, and the comparison groups used. Confounding could affect a large proportion of the studies. The large number of reference groups in the included studies could make quantitative synthesis of risk estimates challenging.
