## metaHelper versus the Cochrane handbook – assessing the performance of a new web-based tool to calculate statistical measures needed for meta-analyses

**Session Type**

**Category**

**Speakers**

**Authors**

^{1}, Schwarzer G

^{2}, Tölch U

^{3}, Gartlehner G

^{4}

^{1}University Krems, Austria

^{2}Universitätsklinikum Freiburg, Germany

^{3}Charité, Germany

^{4}Universität Krems, Austria

**Description**

**Background:** Research publications often do not report outcomes in sufficient detail required for meta-analysis. For example, articles might report means with confidence intervals but not standard deviations as a measure of dispersion. In such cases, systematic reviewers need to calculate the desired measure from the reported information in the publication. The Cochrane handbook provides guidance and mathematically simple formulas for such calculations. Nevertheless, research suggests that errors are frequent and a reason for the lack of replicability of meta-analyses. To address this issue, we have developed metaHelper, an easy-to-use web application (www.metahelper.eu) that calculates or transforms statistical measures which are commonly used for evidence synthesis (figure 1). metaHelper is based on an R package (available at https://github.com/RobertEmprechtinger/metaHelper) used in the web application which provides input fields depending on the required arguments of the underlying R functions. The goal of metaHelper is to make calculations or transformations easier and more reliable for systematic reviewers. It also offers a log that saves calculations.
**Objectives:** To test the accuracy and utility of metaHelper compared with the conventional approach of using statistical formulas provided in the Cochrane handbook.
**Methods:** To test the performance of the web application of metaHelper, we will randomize medical students who have completed a course in clinical epidemiology to the metaHelper or Cochrane handbook group. The latter will receive the relevant sections of the handbook to enable them to calculate the specific statistical measures. A covariate-adaptive randomization will take sex, completed academic years, and age into account. The primary outcomes are the proportion of correct calculations and the time needed to conduct calculations. Statistical analyses for group differences will be conducted within the R environment using Bayesian methods. We will estimate the posterior probabilities of the differences in outcome measures between groups.
Results and **Conclusions:** metaHelper has the potential to make statistical calculations for systematic reviewers easier and more reliable. Results of the evaluation study will be available at the time of the Cochrane Colloquium.