Article type
Year
Abstract
Background: Our objective was to assess the impact study heterogeneity has on measures of publication bias in quantitative systematic reviews. We accomplished our objective by evaluating the impact that study characteristics have on Egger's measure of publication bias.
Methods: The data come from a systematic review of newer pharmacotherapies for the treatment of depression. English and non-English literature was searched using the Cochrane Collaboration Depression, Anxiety, and Neurosis Group's specialized registry. Forty-four RCTs of six weeks or greater duration that evaluated one of 32 specific newer antidepressant or herbal treatments were included. Our measure of treatment effectiveness was based on significant clinical improvement in depressive symptoms: i.e., a 50% reduction in HAMD scores. Analyses were based on a modified intent-to-treat principle. Publication bias was measured by visual inspection for funnel plot asymmetry and Egger's test. We evaluated the impact of study heterogeneity on measures of publication bias by extending Egger's meta-regression model to include measurable study characteristics that could account for the observed asymmetry in the funnel plot. Study level covariates were baseline severity, the proportion of total drop-outs in the treatment group, the proportion of total dropouts in the placebo group, and the proportion of placebo responders. Restricted maximum likelihood (REML) and hierarchical Bayesian (HB) random effect meta-regression models were used.
Results: Significant publication bias (p < .001) and study heterogeneity (p < .001) were observed among the published odds ratios measuring patients' response to newer pharmacotherapies for the treatment of depression. Addition of study-level covariates to the REML random-effects meta-regression model substantially reduced the coefficient for publication bias (p < .054). Treatment dropouts (p < .031) and the placebo response rate (p < .024) were the most important predictors of the treatment effect. Increases in both parameters were associated with an attenuation of the treatment effect. HB random-effects meta-regression produced similar results.
Conclusions: Asymmetry in a funnel plot is substantially influenced by the underlying heterogeneity among studies. Graphical or statistical tests of publication bias based on asymmetry in a funnel plot reflect the influence of study heterogeneity, as well as publicatio
Methods: The data come from a systematic review of newer pharmacotherapies for the treatment of depression. English and non-English literature was searched using the Cochrane Collaboration Depression, Anxiety, and Neurosis Group's specialized registry. Forty-four RCTs of six weeks or greater duration that evaluated one of 32 specific newer antidepressant or herbal treatments were included. Our measure of treatment effectiveness was based on significant clinical improvement in depressive symptoms: i.e., a 50% reduction in HAMD scores. Analyses were based on a modified intent-to-treat principle. Publication bias was measured by visual inspection for funnel plot asymmetry and Egger's test. We evaluated the impact of study heterogeneity on measures of publication bias by extending Egger's meta-regression model to include measurable study characteristics that could account for the observed asymmetry in the funnel plot. Study level covariates were baseline severity, the proportion of total drop-outs in the treatment group, the proportion of total dropouts in the placebo group, and the proportion of placebo responders. Restricted maximum likelihood (REML) and hierarchical Bayesian (HB) random effect meta-regression models were used.
Results: Significant publication bias (p < .001) and study heterogeneity (p < .001) were observed among the published odds ratios measuring patients' response to newer pharmacotherapies for the treatment of depression. Addition of study-level covariates to the REML random-effects meta-regression model substantially reduced the coefficient for publication bias (p < .054). Treatment dropouts (p < .031) and the placebo response rate (p < .024) were the most important predictors of the treatment effect. Increases in both parameters were associated with an attenuation of the treatment effect. HB random-effects meta-regression produced similar results.
Conclusions: Asymmetry in a funnel plot is substantially influenced by the underlying heterogeneity among studies. Graphical or statistical tests of publication bias based on asymmetry in a funnel plot reflect the influence of study heterogeneity, as well as publicatio