ABC | Volume 115, Nº2, August 2020

Original Article Jesus et al. Meta-analysis: body adiposity and apolipoproteins Arq Bras Cardiol. 2020; 115(2):163-171 Search Strategies and Eligibility Criteria Two independent researchers searched the articles in the PubMed, Embase, Lilacs, Web of Science, Ovid and Science Direct databases from December 16, 2016 to July 20, 2017 using descriptors as proposed in the Medical Subject Headings (MESH): exposure (body adiposity and related terms: “obesity” OR “overweight” OR “Abdominal obesity” OR “Central obesity” OR “Waist circumference”), and the outcome (ApoB and ApoA1 levels and related terms: Apolipoprotein OR ApoB OR “Apo B” OR “Apoprotein B”; ApoA OR “ApoA” OR “Apoprotein A”). The descriptors were combined with Boolean operators “or” and “and” in all databases. The search strategy considered the orienting question of investigation, structured by the acronym Population, Exposure, Comparison and Outcome (PECO). Only the terms for the Exposure (E) and Outcome (O) components were defined, in order to avoid unwanted specificity, restricting the selection of studies. The eligibility criteria for inclusion of an article in the study were: original studies conducted with humans, prospective observational design involving children and adolescents aged between 5 and 19 years old and analysis of the relationship between adiposity and ApoB and ApoA1 levels. The studies were supposed to provide information on exposure and outcome, with the adoption of mean as a measure of occurrence and respective standard deviation. There were no restrictions as to year, place and language of publication. The selection of articles was based on the information displayed in the heading and abstract, adopting the eligibility criteria available in a standardized form. Duplicates were removed manually. In the following step, the articles that were maintained were read in full. In case of disagreements, a third reviewer was invited for a consensus meeting. We performed manual search in the reference lists of the selected articles to identify possible studies not included in the electronic search. Exclusion Criteria Studies conducted with pregnant women, nursing mothers, individuals with CVD, diabetes mellitus, arterial hypertension and who had undergone bariatric surgery were excluded. The gray literature—defined mainly as abstracts of congresses and conferences, and academic, governmental and industry reports 14 —was included only in the search on Ovid. The authors of the articles that did not report the mean ApoB and ApoA1 serum levels and the standard deviation according to adiposity were contacted via email and, in case of no response, they were kept in the systematic review and excluded from the meta-analysis. Data Extraction Two independent reviewers read all the eligible articles in full and listed in a standardized spreadsheet the ones meeting the criteria, taking note of the first author’s surname, year of publication, sample size, mean age of participants, sex, and body fat measurement, data on presence or absence of excess adiposity in the end of follow-up, mean and standard deviation of serum ApoB and ApoA1 of participants with and without excess adiposity in the end of follow-up. Bias risk Assessment Two independent reviewers assessed the risk of bias according to the Research Triangle Institute Item Bank (RTI – Item bank). 15 This tool is organized into 29 items, aimed at assessing bias in observational studies, of which six of them were applied in this study: Q1 – study design; Q2 – explicit inclusion and exclusion criteria; Q3 – inclusion and exclusion criteria with valid and reliable measures; Q5 – equal recruitment strategy; Q6 – sample size; Q7 – level of detail in the description of the exhibition; and Q14 – exposures assessed using valid and reliable measures. For all questions, answers considered were 1) yes, 2) no or 3) not applicable. A high risk of bias was identified when the study had two or more negative or not applicable answers; and low risk of bias when it presented less than two negative points or not applicable. Methodological Quality Assessment Methodological quality was assessed using the criteria proposed in the Newcastle Ottawa 16 scale, which consists of three domains: 1. Selection: in this domain, sample representativeness, determination of exposure and absence of selection bias are identified (article can be scored with up to four stars); 2. Comparability between groups: articles can be scored with up to two stars; 3. Outcome: analysis of outcomes, evidence of exposure, assessment of losses and adequacy of follow-up time (article can be scored with up to three stars), totaling nine stars. For this work, a minimum of six stars was adopted to classify an article with good methodological quality. 17 Statistical Analysis Four studies were selected to perform the meta-analysis. According to Higgins & Green, 14 the meta-analysis can be performed by combining two or more different studies. Thus, the descriptive data of the outcome variables according to body adiposity at the end of the follow-up were collected by two independent authors and input in an Excel ® spreadsheet. The simple measure used in the meta-analysis was the difference between the weighted mean (Weighted Mean Difference-WMD) of ApoB and ApoA1 between individuals with and without excess body adiposity, and respective confidence intervals (CI), afterwards put in a forest plot graph. This measure can be used as summary statistics in meta-analyses when the outcome measures in all studies are on the same scale. 14 To calculate the global WMD, random effects models were used, appropriate for studies with high heterogeneity. This model assumes that studies were conducted differently, forming a random sample of hypothetical population, so there is not one value only that estimates the measure of association, but rather a distribution of values. 18,19 It is known that heterogeneity is common in observational studies and this influences the measure of association. Thus, the assumption of homogeneity was tested by the Q-Cochran test and the magnitude of heterogeneity was interpreted by the percentage of variation between the studies considered, measured with I 2 statistic (Higgins inconsistency test). An I 2 less than 50% was considered indicative of moderate 164