PREDICTIVE EQUATIONS OF MAXIMUM OXYGEN CONSUMPTION BY SHUTTLE RUN TEST IN CHILDREN AND ADOLESCENTS: A SYSTEMATIC REVIEW

ABSTRACT Objective: To systematically review the literature as for the level of evidence of predictive equations of VO2peak through the 20-meter shuttle run test (20m-SRT) in children and adolescents. Data sources: Searches were conducted independently by two researchers, according to the procedures adopted by PRISMA, in the electronic databases MEDLINE via PubMed, ScienceDirect, Web of Science, LILACS and SciELO, for articles published until September 2017 in English and Portuguese. The inclusion criteria were: original studies, abstract available, using predictive equations of VO2peak through 20m-SRT, conducted with adolescents and/or children, non-athletes, and mentioning correlation analysis between predicted and measured VO2peak. The level of evidence of equations was based on the risk of bias of the studies using the following criteria: sample number, sample characteristics, and statistical analysis. Data synthesis: Eighteen studies were selected, in which fifteen equations were found and analyzed. The studies had been conducted with samples composed of subjects of both sexes, aged 8 to 19 years. Equations of Léger and Matsuzaka had their level of evidence classified as high, and estimation ranged between r=0.54-0.90 and r=0.65-0.90. Equations by Ruiz, Barnett and Matsuzaka had their level of evidence classified as moderate, and estimation ranged between r=0.75-0.96, r=0.66-0.84 and r=0.66-0.89, respectively. Conclusions: Matsuzaka’s equation presented satisfactory parameters for estimates of VO2peak in children and adolescents. Although not explored in equations, body adiposity and pubertal stage are significantly associated with cardiorespiratory fitness in children and adolescents.


INTRODUCTION
Cardiorespiratory fitness (CRF) is an important health marker in children and adolescents 1 , as it reflects cardiopulmonary efficiency for oxygen and musculoskeletal distribution during exercise or physical activity. 2,3 Studies have shown that children with low CRF tend to maintain this condition over the years, which adversely affects their functional capacity to perform daily activities and quality of life. 4,5 In addition, low CRF is associated with an increase in risk factors for cardiovascular diseases and metabolic changes related to pediatric morbidity and mortality in adults. 6,7 Thus, CRF analysis is a measure of health status of the child and adolescent population. 6 It provides relevant information to the diagnosis and prognosis of cardiometabolic risk factors. 6 Moreover, it serves as an instrument in individualized therapeutics and exercise prescription. 8 Oxygen consumption (VO 2 ) is considered the main index to determine CRF. 9 In children and adolescents, the peak of oxygen consumption (VO 2peak ) is generally used, defined as the peak of VO 2 reached at the end of maximum effort period. 9 VO 2peak can be measured by direct methods by ergospirometric analysis in maximum tests conducted in laboratory with different ergometers or in field, by sport activity simulation. 10,8 From direct testings, authors have proposed equations that assess VO 2peak by indirect methods, which can be performed in maximum or submaximal tests, thus increasing practicality and reducing the costs of evaluations. 11 In epidemiological studies, indirect field tests are mostly indicated because they usually require low cost, short time of execution and ease of simultaneous application in a larger number of individuals. 12,13 The 20-meter shuttle run test (20m-SRT), conceived and described by Léger et al. 14 for the adult population, is one of the field protocols most used in children and adolescents. 6 20m-SRT is considered a simple method, as it requires few equipment, can be performed in space-limited environments, and allows to assess several individuals at the same time, which can increase participants' motivation. 2,6 A systematic review including about 319,000 children and adolescents from 32 countries reported the performance achieved at the 20m-SRT as directly related to health indicators in children and adolescents. 5 In the last decades, the 20m-SRT was included in several batches of physical fitness tests such as EUROFIT and FITNESSGRAM, 6 resulting in the need to improve VO 2peak predictive equations through this test for the child and adolescent population. 15,16 Equations were therefore developed using mathematical regression models or artificial neural networks, and including biological characteristics such as age, sex, body mass and performance in the test. 17,18 On the other hand, prediction of VO 2peak by equations may vary in measurements depending on the characteristics of the sample, especially age group, stage of sexual maturation, gender, and body composition. 16,18 So, in order for an equation to be considered appropriate, it must have adequate validity, that is, produce little variation range between estimate values. 19 Batista et al. 19 pointed out the relevance of analyzing the level of evidence of equations developed to estimate VO 2peak in children and adolescents, and contributed to this review in a more careful and orderly manner.
Therefore, it is not clear which equation establishes better accuracy for estimates based on the different characteristics of the child and adolescent population, or which variables are important to predict VO 2peak , because so far, the findings of different studies have not been systematically analyzed. Thus, the objective of this study was to systematically review the literature to assess the level of evidence of equations intended to predict VO 2peak through the 20m-SRT in children and adolescents.

METHOD
This work was conducted in compliance with recommendations by the Preferred Reporting Items for Systematic Review and Meta-analyzes: the PRISMA statement, 20 from August to September 2017.
Five online databases were selected according to the field of knowledge and scientific relevance worldwide: Medical Literature Analysis and Retrieval System Online (MEDLINE) via PubMed, ScienceDirect, Web of Science, Latin American and Caribbean Literature in Health Sciences (LILACS), and Scientific Electronic Library Online (SciELO). We also searched the reference lists of articles selected that were related to the topic.
The search strategies were defined after identification and selection of search descriptors, based on DECS (BIREME health sciences descriptors) and MESH (Medical Subject Headingscontrolled vocabulary used for indexing articles for PubMed). In this way, the following keywords were chosen in English and Portuguese: Cardiopulmonary Exercise Test, Oxygen consumption, Children and Adolescents. The keywords were combined using "AND" and/or "Boolean" terms and the period of study publication was set until September 2017.
After using the selected descriptors, the studies in duplicity were discarded and the inclusion (1, 2 and 3) and non-inclusion (4, 5 and 6) criteria were applied to screened studies, upon reading of the headings and abstracts: • Original studies with transversal or longitudinal design, or clinical trial. After this step, articles classified as eligible were read and analyzed in full; studies were excluded for several reasons: sample presenting a diagnosed pathology, sample made up of adults and adolescents who were analyzed together, adapted shuttle run protocol, no direct measurement of VO 2peak , no correlation analysis, and/or no VO 2peak prediction.
The criteria for bias risk assessment were adapted by Batista et al. 19 and three parameters were observed: number of participants, sample description and statistical analysis. In order to measure, each parameter was assigned a score of 0 to 2 points.
As for the number of participants, the studies were classified as "0", when the sample had less than 10 participants; "1", between 11 and 50 participants; or "2" more than 51 participants. Age, sex, health status, physical fitness level, pubertal status, body composition, physical activity level and ethnicity were considered when analyzing sample characteristics. Based on these aspects, the studies were scored as "0" when less than four characteristics were described; "1" for four characteristics; or "2" for more than four features. The studies were classified as "0" when presenting no regression analysis or error measures; "1", when presenting regression analysis and/or error measures; and "2" when more than three statistical analyzes were present, or Bland-Altman plot and/or analysis of variance (ANOVA) of repeated measurements. The studies were all categorized according to the scores received: high risk of bias (0-2 points), moderate risk of bias (3 and 4 points) and low risk of bias (5 and 6 points).
Subsequently, the validity of the identified equations was assessed based on the evidence-level criteria expressed by Castro-Piñero et al.: 21 • Strong evidence: equations validated by three or more studies with low risk of bias. • Moderate evidence: equations validated by two studies with low risk of bias. • Limited evidence: equations validated by studies with high risk of bias, inconsistent results among several studies, regardless of risk of bias, or the results of a single study.
Some characteristics of the samples were highlighted in the studies, such as age, gender and number of subjects. The values of correlation coefficient (r) and standard estimation error (SEE) in mL/kg.min were extracted when available. Estimate range variation (ΔER) of each equation was determined by the description from the lowest to the highest correlation coefficient obtained by the equation between the studies. To facilitate identification, we chose to name the equations with the name of the first author of the study in which it was validated. When one author had identified two or more equations in a single study, each equation was accompanied by (a), (b) or (c).
The steps of the process of research, selection, analysis, application of bias risk parameters, and data extraction were independently performed by two researchers (FJMJ and ICJ), and, in case of disagreement, a third researcher (NL) was asked to decide on divergent points.

RESULTS
In total, 2,125 studies were found using the combination of selected descriptors, but 194 were discarded for being duplicates. Afterwards, the inclusion and non-inclusion criteria were applied and 64 studies were considered eligible in full, ending the selection with 14 articles for qualitative synthesis. In addition, four studies relating to the theme identified in other reference lists of articles selected were included, so 18 studies were selected. The process of studies selection is outlined in Figure 1.
As for moderate evidence, the equation by Ruiz 16 showed a low ΔER, while Barnett's (a) 31 and Matsuzaka's (b) 23 resulted in high association values, but low ΔER, respectively. In addition, Barnett's equation (a) 31 had higher correlation values for girls.

DISCUSSION
This systematic review gathered 18 studies in which fifteen equations were identified. Among these, different variables were employed, including sample characteristics and performance in 20m-SRT. Two equations had a strong level of evidence, 23,29 three were classified as moderate evidence 17,23,31 , and nine as limited evidence. 18,25,27,31,33,35 Our findings show that Matsuzaka's (a) 23 equation tends to have higher predictive reliability and a high level of evidence for both genders and may be a potential equation to estimate the VO 2peak in eutrophic boys.
As previously presented, children and adolescents with high VO 2peak levels tend to have risk factors related to cardiovascular diseases, obesity and the metabolic syndrome reduced. 5,7 Thus, the accuracy of equations to estimate VO 2peak is relevant, since it provides valuable information for the diagnosis and prognosis of cardiometabolic risk factors. 6,8 Access to a practical and inexpensive method is important; the 20m-SRT has fulfilled this requirement with strong level of evidence. 6,19 This test requires cheap resources and infrastructure that is easily accessible in schools, clubs and academies. In addition, it can be considered practical and efficient, as it allows the evaluation of several people at the same time. 6 According to our findings, Léger's equation 29 was primary to estimate VO 2peak in children and adolescents in the literature. This equation, which uses age and performance in 20m-SRT as variables, was more popular in studies and presented strong evidence. However, it presents a considerable ΔER between correlation values, being frequently inferior to r = 0.60. 15,18,22,[32][33][34] This variation can be explained by differences in gender between subjects in the sample. Although Léger et al. 29 found no significant predictive value for gender, other studies demonstrate a strong association between this component and cardiorespiratory fitness in children and adolescents. 18,32 On the other hand, the Matsuzaka's equation (a) 23 , with strong evidence, obtained values of estimate validity with lower  Barnett (a) 31 and Matsuzaka (b) 23 were classified with moderate level of evidence, they showed relevant estimates of validity. In particular, Ruiz's 17 equation presented the lowest ΔER among the estimation results. In addition, it matched Matsuzaka's (a) 23 , taking the greater number of characteristics of the sample included into account (sex, age, body mass, height and 20m-SRT performance). This equation was evaluated by a few studies, but seems to be a promising tool that should be better studied.
When considering only studies with low risk of bias, findings become more evident. Léger's equation 29 continues to present higher ΔER compared to Matsuzaka's (a) 23 , among equations with strong evidence; Ruiz 17 obtained higher correlation values and lower ΔER, in comparison to the other equations of moderate evidence. When analyzing data by gender in samples, only the Léger's 29 equation reached strong evidence, despite having low correlation values and high ΔER, showing underestimation of VO 2peak prediction for females and males. Therefore, it was not possible to define the validity of the specific equations by gender, since few studies have provided isolated correlational information and analysis with this variable. Despite this, Barnett's (a) 31 may be a potential equation to estimate VO 2peak in girls and Matsuzaka's (a) 23 in boys, since they were shown to have higher correlational values for the respective groups.
Among equations with strong and moderate level of evidence, the Matsuzaka's (a), 23 Matsuzaka's (b) 23 and Ruiz's 17 equations were the ones that used the largest number of variables from the sample and obtained a lower ΔER with high correlation values. Léger 29 and Barnett (a) 31 inserted fewer variables and found higher ΔER values. The use of more than one characteristic of the sample, such as gender, body mass, stature or BMI, in equations tends to result in higher values of association between predicted and measured VO 2peak . This trend was also noted in other studies. 37,38 From this point of view, moderate associations between VO 2peak and BMI, body mass and gender were identified. 25,32,33 According to Saint-Maurice et al., 38 BMI tends to have a larger influence on CRF in children and adolescents, which can explain 30 to 34% of the variance between VO 2peak estimates found with predictive equations. In this perspective, equations that do not take BMI into account tend to overestimate the CRF of individuals in high nutritional status. 38 Although not yet explored in prediction equations, the body fat percentage shows a significant association with CRF in both children and adolescents. 39 Correlational values of r=-0.60 for both genders, r=-0.48 to -0,53 for boys and r=-0.24 to -0.40 for girls evidence this variable as a strong predictor for males and moderate predictor for females. 27,40,41 Although chronological age has often been used to characterize physical fitness profile, the different stages of sexual maturation tend to relate to different physical fitness characteristics in children and adolescents. 42,43 Girls, specifically, demonstrate significant differences in CRF in different stages of sexual maturation, often presenting decreased VO 2peak as their stages of sexual maturation progress. 44 However, this variable has not yet been tested in prediction equations.
In addition, children and adolescents of different economic classes, sedentary behavior profiles and habitual physical activity levels may present differences as to health-related parameters. 40,44,45 However, information about the use of these variables ∆ER: estimate range variation; strong evidence: more than three studies with low risk of bias; moderate evidence: two studies with low risk of bias; limited evidence: several studies with high risk of bias, wide range of variation or only one study. to predict VO 2peak in children and adolescents is still limited, and new studies on the topic should be developed to better understand the influence of these variables on VO 2peak prediction. This study has some limitations for analysis that should be listed, such as lack of information on sample characteristics, adiposity level, level of physical activity and sedentary behavior; especially related to correlation analyses for samples adjusted by gender. These limitations have turned the identification of the best predictive equation for different groups of children and adolescents into a challenge.
Therefore, future research should be able to provide more information on the sample, such as ethnicity, length of time with sedentary behavior, physical activity level, aspects of body composition and stage of sexual maturation, as well as to promote correlations with CRF. Thus, doubts about the association between these variables and VO 2peak prediction can be better understood, allowing more accurate equations to be elaborated. It is also important that further research be conducted to verify the reproducibility of the equations proposed by Ruiz, 17 Barnett (a), 31 Matsuzaka (a) 23 and Matsuzaka (b), 23 identified in this review as promising but poorly explored in studies. These should be tested for both children and adolescents with different nutritional status, as well as gender-specific variations.
In the present study, we were able to point out the equations with better validity of VO 2peak prediction for children and adolescents, as well as to identify aspects that hindered a more satisfactory analysis to elect a definitive equation. The suggestions presented in this review contribute to a more accurate elaboration and description of future studies, which contributes to the expansion of the scientific and practical knowledge about predictive equations for VO 2peak in children and adolescents.
In conclusion, our findings suggest that using more than one sample feature in equations tends to exert higher association values between predicted and measured VO 2peak . Matsuzaka's equation (a), in this sense, tends to have the strongest level of evidence with greater precision of estimation in children and adolescents. Although not explored in prediction equations, body fat percentage and sexual maturation stage are shown to have relevant associations with CRF in children and adolescents and further analyses of these variables in other equations are encouraged. However, new research should be conducted to evaluate the reproducibility of equations considered by this review as promising, as well as to improve the understanding about the relationship between anthropometric variables, body composition components and sexual maturation stages with VO 2peak prediction in children and adolescents.