Construção e validação de equações para predição da massa corporal e estatura em idosos

Abstract

The height (H) and body mass (BM) are measures widely used in nutritional assessment and verification of health status. However, it's often impossible to measure elderly subjects by conventional methods, if necessary, seek alternative able to predict these measures. In this sense, the use of regression equations are viable alternatives for obtaining these anthropometric variables, since they are inexpensive, practical and noninvasive. The purpose of this study was to test in a sample of elderly of both sexes, living in the city of Maceió / AL, cross-validation of equations to predict BM Chumlea and H. As cross-validation of these equations was not possible, we tried to develop and validate new predictive models, specific to the sample used. For this study used a sample of 280 elderly of both sexes (62 men and 218 women) selected in an objective probabilistic and not subject to the following anthropometric measurements: H, BM, circumferences, skinfolds, bone lengths and diameters. The data were verified by statistical analysis with the aid of Pearson's linear correlation, multiple linear regression and paired t test. For women, the results of equations proposed by Chumlea showed differences between means (p <0.05), as well as values of constant error (CE), total error (TE) and standard error of estimate (SEE) in addition to limits allowed for BM (4.49, 9.62 and 4.57 respectively) and H (5.18, 15.70, 9.99, respectively). For men though there is no difference between means (p> 0.05), the values of CE, TE and SEE also extrapolated the limits considered for validation to BM (-0.35, 4.32 and 8.38) and H (- 3.45, 4.85 and 4.12). For women, the models developed for the population from which the sample was taken, showed correlation coefficients (r), coefficient of determination (r²) and standard error of estimate beyond the limits of validation for BM (-0.35; 4.32 and 8.38) and H (-3.45, 4.85 and 4.12), thus not being subjected to internal validation process. The models for men were subjected to validation as the correlation and SEE values were statistically significant, however, did not pass the test of cross-validation, due to the fact that they show differences between means (p <0.05). Additionally, the models did not differ, had values of CE, TE and SEE over the limits of validation. The results allow us to corroborate the scientific literature to suggest the use of predictive methods with caution. In the sample used, the equations of Chumlea not evaluate accurately the BM and H of the subjects and therefore should not be used in our population. In addition, the proposed predictive models did not show statistically significant results, demonstrating the need for further studies in an attempt to develop regression models more accurate for our population from the use of anthropometric measures not used in this study.