Article type
Year
Abstract
Background: In many Randomised Controlled Trials (RCTs) a continuous variable is collected at baseline and again at follow-up. In this circumstance there exist many methods to estimate the intervention effect. Three common methods involve comparing, between intervention groups, follow-up scores, change scores, and adjusted change scores (calculated from a regression model). When carrying out a Meta-Analysis (MA), review authors will commonly encounter estimates of intervention effect calculated from a variety of methods. While we know theoretically pooling these different estimators using the Mean Difference (MD) is reasonable, in practice it is not known if this is true, or by how much these choices impact on results.
Objectives:
1. To summarise methods that are used to analyse RCTs.
2. To empirically examine if MA of intervention effects calculated using a mix of estimators (e.g. from change scores and follow-up scores) provides a pooled estimate of intervention effect, similar to that calculated using the same estimators.
Methods: One hundred Systematic Reviews (SRs) which met the inclusion criteria were randomly selected from SRs appearing in the 2004, Issue 4 edition of The Cochrane Library. From each SR, one MA was randomly selected. Each included MA had at least four trials, and had employed the MD. The original publications of the included trials were sourced, and data was extracted, where possible, which allowed estimation of the intervention effect using a different estimator. Pooled estimates of intervention effect were then calculated using a) intervention estimates calculated from a mix of methods, and b) intervention estimates calculated using the same method. These pooled estimates of intervention effect were then compared. The ratio of the 95% confidence intervals surrounding the pooled intervention estimates was calculated as a measure of relative precision.
Research status: This work is of practical importance to many review authors, and it is intended that results of this research will be included in future editions of the Cochrane Handbook of Reviews of Interventions. While this research is not yet complete, the authors wish to inform both review authors and methodologists of this ongoing research, and will present available results.
Objectives:
1. To summarise methods that are used to analyse RCTs.
2. To empirically examine if MA of intervention effects calculated using a mix of estimators (e.g. from change scores and follow-up scores) provides a pooled estimate of intervention effect, similar to that calculated using the same estimators.
Methods: One hundred Systematic Reviews (SRs) which met the inclusion criteria were randomly selected from SRs appearing in the 2004, Issue 4 edition of The Cochrane Library. From each SR, one MA was randomly selected. Each included MA had at least four trials, and had employed the MD. The original publications of the included trials were sourced, and data was extracted, where possible, which allowed estimation of the intervention effect using a different estimator. Pooled estimates of intervention effect were then calculated using a) intervention estimates calculated from a mix of methods, and b) intervention estimates calculated using the same method. These pooled estimates of intervention effect were then compared. The ratio of the 95% confidence intervals surrounding the pooled intervention estimates was calculated as a measure of relative precision.
Research status: This work is of practical importance to many review authors, and it is intended that results of this research will be included in future editions of the Cochrane Handbook of Reviews of Interventions. While this research is not yet complete, the authors wish to inform both review authors and methodologists of this ongoing research, and will present available results.