MetaXL keeps pushing the envelope of innovation in meta-analysis. Version 1 introduced the quality effects (QE) model,
version 2 the inverse variance heterogeneity (IVhet) model, version 3 introduced the Doi plot and LFK index for the detection of publication bias, version 4 added network meta-analysis. Now version 5
adds cumulative meta-analysis to this already rich list of features
Meta-analysis is a statistical method to combine the results of epidemiological studies in order to increase power. Basically,
it produces a weighted average of the included studies results.
There are two main issues with meta-analysis: heterogeneity between studies, and publication bias. Heterogeneity is usually dealt with
by employing the random effects (RE) model. However the RE estimator, as explained in the MetaXL User Guide, underestimates the
statistical error and has a larger mean squared error (MSE) than even the fixed effects estimator. It also makes unjustifiable changes to study weights. For these reasons it is seriously flawed and
should be abandoned. MetaXL offers two alternatives to the RE model:
1) The IVhet model provides a quasi-likelihood based expansion of the confidence interval around the inverse variance weighted
pooled estimate when studies exhibit heterogeneity (without inappropriate changes to individual study weights, as the random effects model does), thus keeping the MSE lower than with the random effects
2) The QE model allows incorporating information on study quality into the analysis, thereby affording the opportunity for further reduction in estimator MSE beyond that of the IVhet model.
Much of the heterogeneity between study results is explained by differences in study quality, and it is preferable to make use of this information explicitly.
More background on these alternatives is
in our publications.
Publication bias can occur, among other reasons, because studies with ‘positive’ results are more likely
to get published than ones with ‘negative’ results. Traditionally, the funnel plot is used to detect possible publication bias, but this plot is often hard to interpret. MetaXL now offers an
alternative, the Doi plot, which is much easier to interpret.
Network meta-analysis can make multiple indirect comparisons, thus allowing to assess a range of treatment options against a common
comparator. It is a powerful technique, but it has been held back by complex methods. The MetaXL implementation is powerful, yet very easy to use.
Cumulative meta-analysis allows to analyse how
the evidence evolved over time.
Using Excel as a platform makes MetaXL-based meta-analysis highly accessible. And you still can’t beat the price!
IMPORTANT UPDATE -- 04/06/2019
A recent glitch has been discovered whereby two separate sets of studies with
the exact same pooled effect size and standard error produce Doi plots and LFK indexes that do not overlap with each other. This is due to an error in the ranking calculation within MetaXL. This glitch
only occurs when the same effect size and standard error are observed across different sets of studies.
A Stata ado file has been developed to generate a Doi plot and LFK index without the glitch.
This can be accessed by downloading LFK Stata package.zip. The downloaded file
contains: (1) a Stata ado file implementing the fix; (2) a Stata help file; and (3) a PDF which describes the problem in full and provides accompanying installation instructions.
Please note that
this fix is an alpha version as it only deals with the IOType parameters: ContSE; NumOR; and NumRR.