Plastic surgery continues to advance its understanding of evidence-based medicine and execution of higher-level study designs. Meta-analyses are particularly powerful for topics for which a large definitive trial is lacking and the existing primary studies have not been summarized (e.g., systematic review). Using an example, we outline two steps in interpreting the results of a meta-analysis in plastic surgery to advance readers’ familiarity with critical appraisal: assessing the quality of primary studies (risk of bias) and elements data analysis.
Zhu et al. reviewed the evidence for anastomotic coupling devices (“couplers”) in “Mechanical versus Hand-Sewn Venous Anastomoses in Free Flap Reconstruction: A Systematic Review and Meta-Analysis.”1 The authors report an exhaustive review of the literature, clinically important outcomes, and heterogeneity and publication bias in their analyses. The meta-analysis includes nonrandomized studies. Nonrandomized studies are prone to selection bias and confounding, and are typically retrospective; they are more likely to overestimate treatment effects.2 Given that the majority of surgical literature is nonrandomized studies, it is important to include these studies in a meta-analysis. However, the results of a meta-analysis are only as valid as the primary studies the meta-analysis contains. Therefore, a formal appraisal of the limitations of these primary studies is essential; a number of tools exist to operationalize risk of bias assessment (e.g., ROBINS-I).3 In a meta-analysis, each outcome should be framed considering the uncertainty in the evidence base.3 In the article by Zhu et al.,1 readers should recognize that the quality of extant evidence is very low.
In data analysis, selecting a fixed-effects model assumes the effect of the intervention is consistent across studies (i.e., patients belong to a single population). However, nonrandomized studies will have inherent differences, including methodological (i.e., ways studies are performed) and clinical (e.g., head and neck versus breast flaps) sources of heterogeneity. Assessing heterogeneity should involve these individual study characteristics, and not rely simply on statistical analysis of their results (i.e., I2).4 In meta-analyses of nonrandomized studies, instead of selecting a fixed-effects model, a random-effects model (REM) will help account for inherent variation and report a higher fidelity estimate of effect. To illustrate this point in Zhu et al., using a random-effects model to analyze overall flap failure for couplers versus hand-sewn anastomoses, the difference is not significant: relative risk, 0.58 (95% CI, 0.33 to 1.03), p = 0.06 (Fig. 1). The evidence is not conclusive. The confidence interval is wide because the analysis remains slightly underpowered given the rarity of flap failure. A power calculation can be performed for meta-analyses, as with any other study design.5 Using a 2 percent free flap failure rate for hand-sewn anastomoses, the meta-analysis would require 2319 patients per group to detect a reduction to 1 percent when using a coupler. As is the case with this analysis, it is not always possible to attain an adequate sample size in cases of rare events.
After assessing risk of bias and elements of the statistical analysis, readers can conclude couplers likely decrease flap failure, though the quality of evidence is very low. Considering these points would improve readers’ interpretation and practical application of any meta-analysis published in Plastic and Reconstructive Surgery.
None of the authors has any relevant financial relationships or affiliations to disclose.
Christopher J. Coroneos, M.D., M.Sc.
Division of Plastic Surgery
Hamilton, Ontario, Canada
Anaeze C. Offodile II, M.D., M.P.H.
Department of Plastic Surgery
M. D. Anderson Cancer Center
Sophocles H. Voineskos, M.D., M.Sc.
Ronen Avram, M.D., M.Sc.
Division of Plastic Surgery
Hamilton, Ontario, Canada
1. Zhu Z, Wang X, Huang J, et al. Mechanical versus hand-sewn venous anastomoses in free flap reconstruction: A systematic review and meta-analysis. Plast Reconstr Surg. 2018;141:1272–1281.
2. Kunz R, Vist G, Oxman AD. Randomisation to protect against selection bias in healthcare trials. Cochrane Database Syst Rev. 2007;(2):MR000012.
3. Schünemann HJ, Cuello C, Akl EA, et al. GRADE guidelines: 18. How ROBINS-I and other tools to assess risk of bias in nonrandomized studies should be used to rate the certainty of a body of evidence. J Clin Epidemiol. E-published ahead of print February 9, 2018. PII: S0895-4356(17)31031-4. DOI: 10.1016/j.jclinepi.2018.01.012.
4. Higgins JPT, Green S. 13.6.1 What is different when including non-randomized studies? In: Cochrane Handbook for Systematic Reviews of Interventions. Version 5.1.0. 2011. London: The Cochrane Collaboration; Available at: http://handbook.cochrane.org
5. Guyatt GH, Oxman AD, Kunz R, et al. GRADE guidelines 6. Rating the quality of evidence–imprecision. J Clin Epidemiol. 2011;64:1283–1293.
Viewpoints, pertaining to issues of general interest, are welcome, even if they are not related to items previously published. Viewpoints may present unique techniques, brief technology updates, technical notes, and so on. Viewpoints will be published on a space-available basis because they are typically less timesensitive than Letters and other types of articles. Please note the following criteria:
- Text—maximum of 500 words (not including references)
- References—maximum of five
- Authors—no more than five
- Figures/Tables—no more than two figures and/or one table
Authors will be listed in the order in which they appear in the submission. Viewpoints should be submitted electronically via PRS’ enkwell, at www.editorialmanager.com/prs/. We strongly encourage authors to submit figures in color.
We reserve the right to edit Viewpoints to meet requirements of space and format. Any financial interests relevant to the content must be disclosed. Submission of a Viewpoint constitutes permission for the American Society of Plastic Surgeons and its licensees and assignees to publish it in the Journal and in any other form or medium.
The views, opinions, and conclusions expressed in the Viewpoints represent the personal opinions of the individual writers and not those of the publisher, the Editorial Board, or the sponsors of the Journal. Any stated views, opinions, and conclusions do not reflect the policy of any of the sponsoring organizations or of the institutions with which the writer is affiliated, and the publisher, the Editorial Board, and the sponsoring organizations assume no responsibility for the content of such correspondence.