Pediatrik Büyüme Eğrilerinin Model Performanslarının Gamlss Yöntemi ile İncelenmesi ve Kantil Regresyon Yöntemi ile Karşılaştırılması

Abstract

Reference pediatric growth curves constructed with the help of anthropometric measurements are one of the best tools in the examination of a child's growth and development, in the early diagnosis of disease status and in the evaluation of nutritional status. Within the scope of this thesis, growth curves for height, body weight, head circumference and body mass index measurements by gender were constructed with different statistical methods. The growth curves obtained as a result of the quantile regression method were compared with the GAMLSS (Generalized additive model for location, scale and shape) model, which was developed by using LMS, LMSP and LMST methods together. As a result of both methods, the fit of the growth curves to the data and the model adequacy were assessed with the worm plot. The difference between the LMSP and LMST methods compared to the LMS method is to take into account the skewness and kurtosis of the distribution while constructing the growth curves. While the GAMLSS model requires the assumption of normal distribution in the data, it also enables the best model selection by examining the model performances of the growth curves. On the other hand, quantile regression is an alternative method for constructing growth curves since it does not require any distribution assumption in the data. According to the results, it was observed that more smoothed flexible growth curves were obtained as a result of the GAMLSS method, but it is very important to determine the effective degrees of freedom in the model. In the quantile regression method determination of the smoothing parameter is very important and flexible growth curves were constructed with monotony restrictions, but for some anthropometric measurements it was observed that the flexibility was impaired at the end values of the curve. As a result of both methods worms are in the 95% confidence interval in the worm plot containing all age groups. It can be said that the fit of the constructed growth curves to the data is sufficient.