Article type
Year
Abstract
Background: Reporting adverse events of pharmacological treatments is generally inadequate and incomplete in randomized clinical trials (RCTs). The safety profile of medications can be assessed more robustly in the long-term follow-up of observational studies such as cohorts. Standard methods of incorporating both study designs put much weight on the usually large observational studies, despite their, presumably, lower internal validity.
Objectives: A case study to combine safety data from RCTs and observational studies into a hierarchical Bayesian model adjusting for design-related weights.
Methods: We set up a systematic review on adverse events of erythropoiesis stimulating agents (ESAs) in off-label indications in critically ill patients. Eleven databases were searched up to April 2012. We considered RCTs and controlled observational studies in any language that compared off-label ESAs treatment with other effective interventions, placebo or no treatment in critically ill patients. We used frequentist and Bayesian models to combine studies, and performed sensitivity analyses by using different weights for the observational trials.
Results: We included 48 studies from 12,888 citations (34 RCTs; 14 observational). In order to combine data from RCTs and observational studies, we fitted a three-level hierarchical Bayesian model, which accounts for between-trial design heterogeneity. ESAs increased the risk for venous thromboembolism (VTE) in the RCTs using frequentist analyses, but had no effect on VTE in the observational studies, or when Bayesian methods were applied. We found no statistically significant difference in the mortality risk from treatment with ESAs compared to non-ESAs in a Bayesian estimate of combining data from RCTs and observational studies. Sensitivity analysis using the Bayesian approach is consistent with the main analysis.
Conclusions: A Bayesian approach can be used in a systematic review of adverse events to combine the information from all available controlled study designs by putting less weight on the observational studies.
Objectives: A case study to combine safety data from RCTs and observational studies into a hierarchical Bayesian model adjusting for design-related weights.
Methods: We set up a systematic review on adverse events of erythropoiesis stimulating agents (ESAs) in off-label indications in critically ill patients. Eleven databases were searched up to April 2012. We considered RCTs and controlled observational studies in any language that compared off-label ESAs treatment with other effective interventions, placebo or no treatment in critically ill patients. We used frequentist and Bayesian models to combine studies, and performed sensitivity analyses by using different weights for the observational trials.
Results: We included 48 studies from 12,888 citations (34 RCTs; 14 observational). In order to combine data from RCTs and observational studies, we fitted a three-level hierarchical Bayesian model, which accounts for between-trial design heterogeneity. ESAs increased the risk for venous thromboembolism (VTE) in the RCTs using frequentist analyses, but had no effect on VTE in the observational studies, or when Bayesian methods were applied. We found no statistically significant difference in the mortality risk from treatment with ESAs compared to non-ESAs in a Bayesian estimate of combining data from RCTs and observational studies. Sensitivity analysis using the Bayesian approach is consistent with the main analysis.
Conclusions: A Bayesian approach can be used in a systematic review of adverse events to combine the information from all available controlled study designs by putting less weight on the observational studies.