Article type
Year
Abstract
Background: The DerSimonian and Laird approach (DL) is widely used for random effects meta-analysis, but this often results in inappropriate type I error rates. The method described by Sidik and Jonkman (SJ) is known to perform better when trials of similar size are combined. However evidence in realistic situations, where one trial might be much larger than the other trials, is incomplete.
Objectives: We aimed to evaluate the relative performance of the DL and SJ methods when studies of different sizes are combined, and to develop a simple process to convert results from DL to SJ.
Methods: We evaluated the performance of the SJ versus DL approach in meta-analyses of 2–20 trials with varying sample sizes and between-study heterogeneity, and allowing trials to have various sizes, e.g. 25%, 50% or 75% of the trials being 10-times larger than the smaller trials.
Results: The SJ method consistently resulted in more adequate error rates than the DL method. When the statistical significance level was 5%, the SJ error rates remained below 12%. For DL they could be over 30%. DL, and, far less so, SJ had more inflated error rates when the combined studies had unequal size and there was between-study heterogeneity (Figs 1 and 2). We also show how DL results can be easily converted into SJ.
Conclusions: The SJ method performed consistently well, and can easily be applied routinely in meta-analyses. Extra caution is needed when there are = ¡ 5 studies of very unequal sizes. Figure 1. Error rates for risk ratios, for the DerSimonian-Laird (DL) and Sidik-Jonkman (SJ) meta-analysis methods, different values of I2 and one large trial (10 times larger than other trials). Vertical bars refer to the minimum and maximum error rates over the group sizes. The lines connect the means of these error rates.
Objectives: We aimed to evaluate the relative performance of the DL and SJ methods when studies of different sizes are combined, and to develop a simple process to convert results from DL to SJ.
Methods: We evaluated the performance of the SJ versus DL approach in meta-analyses of 2–20 trials with varying sample sizes and between-study heterogeneity, and allowing trials to have various sizes, e.g. 25%, 50% or 75% of the trials being 10-times larger than the smaller trials.
Results: The SJ method consistently resulted in more adequate error rates than the DL method. When the statistical significance level was 5%, the SJ error rates remained below 12%. For DL they could be over 30%. DL, and, far less so, SJ had more inflated error rates when the combined studies had unequal size and there was between-study heterogeneity (Figs 1 and 2). We also show how DL results can be easily converted into SJ.
Conclusions: The SJ method performed consistently well, and can easily be applied routinely in meta-analyses. Extra caution is needed when there are = ¡ 5 studies of very unequal sizes. Figure 1. Error rates for risk ratios, for the DerSimonian-Laird (DL) and Sidik-Jonkman (SJ) meta-analysis methods, different values of I2 and one large trial (10 times larger than other trials). Vertical bars refer to the minimum and maximum error rates over the group sizes. The lines connect the means of these error rates.
