Interpreting pooled estimates in systematic reviews involving continuous variables

Article type
Authors
Thorlund K1, Johnston B1, Furukawa T2, Patrick D3, Schunemann H4, Walter S1
1McMaster University, Canada
2Kyoto University, Japan
3University of Washington, United States
4McMaster University, Hamitlon
Abstract
Background: The Cochrane Handbook suggests that, when individual studies use different instruments to measure the same construct, authors present results in standard deviation units (standardized mean difference - SMD). The SMD is, however, limited by vulnerability to differential variability in populations enrolled, and challenges in interpretation.

Objectives: To summarize and apply approaches to enhance interpretability of pooled estimates of continuous variables in meta-analyses.

Methods: We reviewed available methods and applied these methods to three example systematic reviews. Here, we provide examples from one, a review of respiratory rehabilitation in which meta-analysis resulted in an SMD of 0.72.

Results: We identified the following methods, grouped them into three categories and evaluated their strengths and limitations: A) Summary estimates derived from the pooled SMD. SMD can be converted into the units of the most frequently used instrument (1.02, where 0.5 represents a minimally important difference), into a risk difference (0.28 to 0.40, depending on the formulae applied) or into an odds ratio (3.2 to 3.7). These approaches remain vulnerable to differential variability in populations. B) Summary estimates derived from the individual trial summary statistics. Conversion into units of the most frequently used instrument (0.72). A ratio of means approach, one alternative in this category, is not applicable when, as here, measurement is of change. C) Summary estimates derived from the individual trial summary statistics and knowledge of MIDs. Presentation of results in MID units (1.75 MID units). Estimates of risk difference (0.31) or odds ratio (3.4). Risk differences are ideal for trading off desirable and undesirable consequences of alternative management strategies.

Conclusions: Systematic review authors dealing with continuous variables should, in addition to or as an alternative to the SMD, present results using one or more of the analytic approaches that enhance interpretability.