Article type
Year
Abstract
Background: We recently published a systematic review of 142 randomized controlled trials (RCTs) of quality improvement (QI) interventions for diabetes care. We performed meta-analysis and meta-regression of RCTs comparing QI interventions that included a specific component of interest (and 0, 1, or more other QI components) versus usual care. While easier to specify, this analytical approach has a number of limitations. We found that QI interventions improved diabetes care in general, but could not clearly distinguish the effectiveness of various components within QI interventions.
Objectives: To determine whether hierarchical meta-regression analyses lead to different interpretations than meta-analysis or simple meta-regression.
Methods: We performed hierarchical metaregression analyses to estimate additive intervention effects, test for non-additive (e.g., synergistic or antagonistic) intervention effects and test for effect modifiers. At the first level, we modeled the outcome of interest (e.g., mean HBA1c value) within each arm of each RCT. At the second level we modeled between study variability (heterogeneity). The mean effect in each study arm is regressed against the 12 QI component interventions of interest plus usual care. We then extended the model to include interaction terms. We tested each of the 66 possible pairwise interactions.
Results: In the meta-analysis, QI interventions were found to reduce HBA1c by 0.37% and a subgroup analysis found that baseline HBA1c was a significant effect modifier. In the simple meta-regression, we found HBA1c was reduced by 0.33% and no differences were found between component QI interventions. In the hierarchical meta-regression, the mean reduction in HBA1c was 0.36%. Examining pairwise interactions revealed that interventions with case-management, team change, patient education, and promotion of self-management had greater effects in patient groups with higher baseline A1c.
Conclusions: In reviews of complex interventions, hierarchical meta-regressions may provide additional insights and generate new hypothesis for further study than traditional meta-analytical techniques.
Objectives: To determine whether hierarchical meta-regression analyses lead to different interpretations than meta-analysis or simple meta-regression.
Methods: We performed hierarchical metaregression analyses to estimate additive intervention effects, test for non-additive (e.g., synergistic or antagonistic) intervention effects and test for effect modifiers. At the first level, we modeled the outcome of interest (e.g., mean HBA1c value) within each arm of each RCT. At the second level we modeled between study variability (heterogeneity). The mean effect in each study arm is regressed against the 12 QI component interventions of interest plus usual care. We then extended the model to include interaction terms. We tested each of the 66 possible pairwise interactions.
Results: In the meta-analysis, QI interventions were found to reduce HBA1c by 0.37% and a subgroup analysis found that baseline HBA1c was a significant effect modifier. In the simple meta-regression, we found HBA1c was reduced by 0.33% and no differences were found between component QI interventions. In the hierarchical meta-regression, the mean reduction in HBA1c was 0.36%. Examining pairwise interactions revealed that interventions with case-management, team change, patient education, and promotion of self-management had greater effects in patient groups with higher baseline A1c.
Conclusions: In reviews of complex interventions, hierarchical meta-regressions may provide additional insights and generate new hypothesis for further study than traditional meta-analytical techniques.