Downloaded from http://journals.lww.com/jspinaldisorders by BhDMf5ePHKbH4TTImqenVHT0Rm60g2jqOa85iqlHhoQPZj1X9x8+qa01QSfJ+I/P on 02/12/2019 Guidelines on How to Perform a Meta-Analysis in Spine Surgery Strengths and Weaknesses of Design David Gendelberg, MD,* Casey A. Slattery, BA,† and Kushagra Verma, MD‡ Abstract: There are many questions in the literature that remain unanswered due to the paucity of available subjects or the large sample size needed to detect a difference. A meta-analysis con- sists of integrating together data from multiple studies into one larger data set in order to increase the subject size and power of a paper. In essence, it is a systematic review in which one uses statistical methods to summarize the results of these studies. It is important that a meta-analysis be performed in a systematic and orderly manner with the assistance of a statistician. When carried out correctly, these studies serve as powerful tools to help us better address our knowledge. Because of their complexity, they are prone to bias at multiple levels. This article will discuss the steps involved in performing a meta-analysis, select good studies, as well as explain the statistics conducted in these studies. Fur- thermore, we will discuss examples from the literature that demonstrate a good meta-analysis. Key Words: meta-analysis, methodology, spine (Clin Spine Surg 2019;32:30–31) WHAT IS A META-ANALYSIS? The complexity of spine surgery often leads to numer- ous unanswered research questions. When a large sample size is not available for a single study, a meta-analysis is a useful statistical tool. It provides an objective component of a sys- tematic review utilizing statistical methods to summarize the results of several studies. 1 It consists of integrating together unbiased quantitative data from multiple studies into one larger data set in order to increase the subject size and power of a paper. Good meta-analyses aim to include all appropriate studies, looking for heterogeneity while exploring key finding robustness. When carried out correctly, these studies serve as powerful tools to help us better address the same question. HOW IS A META-ANALYSIS PERFORMED? It is important that a meta-analysis be performed in a systematic and orderly manner with the assistance of a sta- tistician. A good meta-analysis should include a number of steps. We will describe the implementation of a meta- analysis in multiple steps, which could be performed in varying order. 1 The first step consists of framing a hypothesis with precise inclusion and exclusion criteria for the study. The next step involves performing a thorough literature re- view of all articles applicable to the question. All articles must be read to ensure they meet the inclusion criteria for the meta-analysis and have appropriate relevance. The data are then extracted from the relevant articles into a datasheet including the validity of the study (as agreed upon by the reviewers), study designs, interventions, variables of interest, sample sizes, and other desired characteristics. 1 This step must be performed by > 1 reviewer and all differ- ences must be resolved by consensus. Generally, a meta- analysis should include primarily level 1 studies when possible. Determining the quality of the information in the study list is important to determine the validity, applicability, and degree of bias. The following step consists of calculating the effect sizes and their interpretations. The effect size, is a measure of the magnitude of the difference between 2 variables. 2 Effect sizes are calculated and interpreted using “Corre- lation coefficient r” for continuous variables or “Cohen’s d statistic” for categorical variables. If the study pop- ulations, interventions, and outcomes measured between the papers are similar then the data could be pooled to- gether to perform a fixed effects model. In this case the individual studies are pooled using weights that depend on the sample size of the study. However, if there are differ- ences between the study populations (such as different demographics) then a random effects model should be performed. Since the effects of these variables cannot be reconciled between the different studies, each study is weighted equally. Because of the fact that researchers are usually interested in generalizing results across various situations and populations, the studies included in the meta-analysis are not likely to be functionally equal. Therefore, the random effects model is more commonly used. 3 If there are any potential confounders, or a third variable affecting the relationship between the 2 variables, a multiple regression analysis should be conducted. 4 Received for publication November 30, 2017; accepted July 27, 2018. From the *Harborview Medical Center; †University of Washington School of Medicine; and ‡University of Washington Medical Center, Seattle, WA. The authors declare no conflict of interest. Reprints: David Gendelberg, MD, Department of Orthopaedics, Harborview Medical Center, 325 9th Ave, Seattle, WA 98104 (e-mail: David.Gendelberg@Gmail.com). Copyright © 2018 Wolters Kluwer Health, Inc. All rights reserved. RESEARCH METHODOLOGY 30 | www.clinicalspinesurgery.com Clin Spine Surg  Volume 32, Number 1, February 2019 Copyright r 2018 Wolters Kluwer Health, Inc. All rights reserved.