Determinação e validação de limiares de acelerômetros para a estimativa da intensidade da atividade física em adolescentes

Abstract

O presente estudo foi desenvolvido com o objetivo de derivar limiares para valores de counts de tres modelos de acelerometros (ActiGraph GT3X, Actical e RT3) e testar a validade de cinco conjuntos de limiares da versao uniaxial do acelerometro ActiGraph para distinguir diferentes intensidades da atividade fisica em adolescentes. Setenta e nove adolescentes de 10 a 15 anos de idade participaram do estudo. Dados de counts dos acelerometros e de consumo de oxigenio (VO2) foram coletados em repouso e em onze atividades fisicas de diferentes intensidades. Os tres acelerometros foram posicionados na cintura e o VO2 foi mensurado pelo sistema metabolico portatil Cosmed K4b2. Escores de MET¡¦s foram computados individualmente e adotados como criterios para: a) derivar limiares relacionados as atividades sedentarias (SED), leves (LEV), moderadas a vigorosas (MOD) e vigorosas (VIG) para os acelerometros ActiGraph (GT3X), Actical e RT3 e; b) validar limiares da versao uniaxial do ActiGraph propostos por Evenson et al., Mattocks et al., Puyau et al., Treuth et al. e Vanhelst et al. Receiver Operating Characteristic Curves (ROC) foram utilizadas para determinar limiares para os tres modelos de acelerometros no ponto onde a sensibilidade e a especificidade foram maximizadas. A validade dos cinco conjuntos de limiares foi testada por curvas ROC e coeficientes ponderados de Kappa (k). O alcance dos limiares para SED, LEV, MOD e VIG foi de 0-180, 181-756, 757-1111 e >1111 counts.15 s-1 para o vetor magnitude do ActiGraph, 0-17, 18-440, 441-872 e >872 counts.15 s-1 para o Actical e 0-5.6, 5.7-20.3, 20.4-32,1 e >32.1 counts.s-1 para o vetor magnitude do RT3, respectivamente. Para os tres acelerometros foi observada uma discriminacao quase perfeita das SED e das MOD (ROC > 0,97) e excelente discriminacao das VIG (ROC > 0,90). Areas sob a curva ROC indicaram melhor discriminacao das MOD pelo ActiGraph e Actical quando comparados ao RT3 (P<0,05). A predicao das intensidades nao foi melhorada quando as analises foram estratificadas por idade e sexo. Quanto a validacao dos conjuntos de limiares da versao uniaxial do ActiGraph, limiares propostos por Evenson et al. e Vanhelst et al. apresentaram melhor concordancia (k=0,84 e 0,82) que aqueles propostos por Mattocks et al. e Puyau et al. (k=0,78 e 0,70). Em geral, acuracia de classificacao das SED e MOD foi excelente (ROC>0,90). Limiares de Evenson et al., Treuth et al. e Mattocks et al. exibiram melhor predicao das SED que limiares de Vanhelst et al. e Puyau et al. (P<0,0001). Para MOD, limiar de Mattocks et al. apresentou menor acuracia que os demais limiares (P<0,01). A predicao das VIG pelos limiares de Evenson et al. e Vanhelst et al. foi boa (ROC=0,88 e 0,89) e significantemente melhor que Treuth et al., Mattocks et al. e Puyau et al. (ROC=0,57 a 0,78). Coletivamente, limiares de Evenson et al. estiveram associados as melhores estimativas das SED, MOD e VIG. Em conclusao, os limiares derivados no presente estudo para os acelerometros ActiGraph GT3X, Actical e RT3 apresentaram elevada capacidade de predicao da intensidade da atividade fisica e podem ser utilizados para monitorar a atividade fisica habitual de adolescentes. Dentre os limiares existentes para a versao uniaxial do ActiGraph, limiares de Evenson et al. demonstraram melhores indicadores de validade e sao recomendados para a estimativa da intensidade da atividade fisica em adolescentes.

The present study was developed with the aim of determining thresholds for values of counts of three accelerometer models (ActiGraph GT3X, Actical and RT3) and test the validity of five sets of thresholds for the uniaxial version of the ActiGraph accelerometer to distinguish different physical activity intensities among adolescents. Seventy-nine adolescents aged 10 to 15 years participated in the study. Data from accelerometer counts and oxygen consumption (VO2) were collected at rest and at eleven different physical activity intensities. The three accelerometers were positioned at the adolescents¡¦ waist and VO2 was measured using the Cosmed K4b2 portable metabolic system. MET's scores were individually computed and adopted as criteria to: a) determine thresholds related to sedentary (SED), light (LEV), moderate to vigorous (MOD) and vigorous activities (VIG) for ActiGraph (GT3X) Actical and RT3 accelerometers and b) validate thresholds for the uniaxial version of the ActiGraph accelerometer proposed by Evenson et al., Mattocks et al., Puyau et al., Treuth et al. and Vanhelst et al. Receiver Operating Characteristic Curves (ROC) were used to determine thresholds for the three accelerometer models at the point where sensitivity and specificity were maximized. The validity of the five sets of thresholds was tested by ROC curves and weighted kappa coefficients (k). The range of thresholds for SED, LEV, MOD and VIG were 0-180, 181-756, 757-1111 and> 1111 counts.15 s-1 for the magnitude vector of ActiGraph accelerometer, 0-17, 18-440, 441 -872 and > 872 counts.15 s-1 for Actical and 0-5.6, 5.7-20.3, 20.4-32,1 and > 32.1 counts.s-1 for the magnitude vector of RT3, respectively. For the three accelerometers, an almost perfect discrimination of SED and MOD (ROC> 0.97) and excellent discrimination of VIG (ROC> 0.90) was observed. Areas under the ROC curves indicated better discrimination of MOD by ActiGraph and Actical when compared to RT3 (P <0.05). The prediction of intensities was not improved when the analyses were stratified by age and sex. In relation to the validation of the sets of thresholds for the uniaxial version of ActiGraph, thresholds proposed by Evenson et al. and Vanhelst et al. showed better agreement (k = 0.84 and 0.82) than those proposed by Mattocks et al. and Puyau et al. (k = 0.78 and 0.70). In general, the classification accuracy of SED and MOD was excellent (ROC> 0.90). Thresholds proposed by Evenson et al., Treuth et al. and Mattocks et al. showed better prediction of SED than those proposed by Vanhelst et al. and Puyau et al. (P <0.0001). For MOD, Mattocks et al. threshold showed lower accuracy than the other thresholds (P <0.01). The prediction of VIG by Evenson et al. and Vanhelst et al. thresholds was good (ROC = 0.88 and 0.89) and significantly better than by Treuth et al., Mattocks et al. and Puyau et al. (ROC = 0.57 to 0.78). Altogether, Evenson et al. thresholds were associated with better estimates of SED, MOD and VIG. In conclusion, the thresholds derived in this study for ActiGraph GT3X, Actical and RT3 accelerometers showed high capacity for predicting physical activity intensity and can be used to monitor the usual physical activity intensity in adolescents. Among the existing thresholds for the uniaxial version of the ActiGraph accelerometer, Evenson et al. thresholds showed better validity indicators and are recommended for estimating the physical activity intensity in adolescents.