Analysis of heterogeneity in a systematic review using meta-regression technique. Issue 2 (June 2019)

• Record Type: Journal Article
• Authors: Lo, Kenneth; Stephenson, Matthew; Lockwood, Craig
• Is Part Of: International journal of evidence-based healthcare. Volume 17:Issue 2(2019)
• Publication Date: 2019-06
• Subjects: Meta-analysis -- meta-regression -- robotic rehabilitation -- stroke -- systematic review; Nursing -- Periodicals; Midwifery -- Periodicals; Evidence-based medicine -- Periodicals; 616.005
• DOI: 10.1097/XEB.0000000000000163
• Languages: English
• ISSNs: 1744-1595

Abstract

ABSTRACT: Aim: Heterogeneity is an important consideration in systematic reviews, as high heterogeneity may imply that it is not suitable to perform meta-analysis. The degree of variation could be caused by clinical or methodological differences among the studies, or it could be due to the randomness of chance. Methods of assessing heterogeneity are calculating a statistical test for heterogeneity (the I 2 value), visual evaluations of forest plots, conducting subgroup analysis or meta-regression. We conducted meta-regression on data of our previous systematic review on the effectiveness of robotic rehabilitation, and in this article, we present the findings and discuss its implications. Method: In our meta-regression plots, plotted on the x -axis was the trial covariate (duration of intervention group therapy), and plotted on the y -axis was the effect size measure (standardized mean differences), with positive effect sizes favouring robotic intervention. Analysis using random effects was applied, and each study symbol was sized in proportion to its precision (inverse-variance weighting). Results: Differences were observed in the meta-regression plots between the subgroups of therapy ratio = 0 and therapy ratio more than 0 for upper limb movement, lower limb walking and activities of daily living. For upper limb movement, positive linear relationships were found for both subgroups. However, in terms of the strength of the relationship, a stronger relationship was found for therapy ratio = 0. For lower limb walking, opposing linear relationships were found in both subgroups: therapy ratio = 0 had a negative linear relationship, whereas therapy ratio more than 0 had a positive linear relationship. For activities of daily living, positive linear relationships were found for both subgroups, but a stronger linear relationship was found for therapy ratio = 0. Conclusion: From the meta-regression analysis, we found that differing levels of linear relationships and the varying spread of effect sizes across positive and negative ranges were the likely sources of heterogeneity. This was especially so in the meta-regression of lower limb walking, which showed opposing directions of linear relationships. The wider spread of effect sizes for therapy ratio = 0 could indicate that some robotic devices were more effective than others. In addition, for therapy ratio more than 0, the effect sizes were mainly found in the positive region, which implied that adding conventional training to robotic training was generally positive for robotic devices.