![]() The methodology for defining the experiments was based on the guidelines for reporting mathematical models in clinical medicine. In this paper, we explain the derivation of the proposed model and provide a comparison of the two current z-score determination models, the model proposed by WHO (using first order statistical moments) and the approximation proposed by CDC (inspired on Cole’s LMS method ) with the model based on GPR. Due to the Gaussian nature of the anthropometric measurements, we propose a new tool for calculating z-scores for HFA, WFA and BMIFA by using a Gaussian Process Regressions (GPR) model, without smoothing the curves to empirical data. The eruption of big data and machine learning in health has contributed to the development and implementation of new mathematical models to support clinical decisions. ![]() Still, the CDC-LMS method is widely used by software tools and clinicians rely on it to follow-up patients’ growth and nutritional status. The CDC-LMS method does not guarantee a good fit to the empirical data and moreover the tails of the distribution (values below 3% and over 97%) are not used. Fundamentals of this method were employed by the CDC on an inverse approach for determining LMS parameters and percentiles-smoothing. LMS deals with skewed distributions by adjusting parameters, but it can lead to poor fitting on small population samples. LMS parameters are coefficients estimated from growth data, smoothed and then computed to map the values to percentiles (and z-scores). In 1990, Cole proposed the LMS equation, a method for z-score and percentile determination. Historically, a variety of parametric and nonparametric methods were employed to determine z-score values, but such models did not allow the calculation of percentiles or equivalent z-scores for other than the selected smoothed percentiles. As an alternative, but less widely used, linear regression models have been proposed to estimate z-scores and to identify implausible z-score values. For values in the obesity range, BMI z-scores have been found unsatisfactory because the statistical method used to construct the growth charts compresses the z-score scale. These charts are based on the National Health and Nutrition Examination Survey data from the 1960s through the 1980s to determine the distribution of height, weight and BMI in children, which varies by age and sex. In 2002, the CDC published growth charts for several anthropometric measurements. Z-score values below -3 indicate severe wasting and stunting. Similarly, moderate wasting (low weight-for-height (WFH)), stunting (low height-for-age (HFA)) are defined as z-score between -3 and -2 SD. ![]() Moderate malnutrition is defined as a weight-for-age (WFA) between -3 and -2 SD below the mean of the WHO child growth standards. WHO proposes the calculation of z-scores for the analysis and interpretation of anthropometric values either for population-based and individual assessment, and suggests z-scores as a sex-independent variable that can be grouped by combining sex and age groups. Z-score charts (also known as centile growth charts) are used in paediatric growth follow-up and to compare anthropometrical variables to detect the presence of malnutrition or disease. Z-score equal to 0 means an average value, while a z-score of +1 means the value is one SD above the mean value of the population. For a normal distribution, a z-score represents the distance in SDs of a given value to the mean value of the distribution. Although curve-fitting may be imprecise, normal distributions are the most popular because they are scalable to the mean and standard deviation (SD). ![]() Īnthropometric measurements may have different distributions for different populations. Using the most precise methods to calculate z-score is important because of the risk of misclassification and its additional consequences. The use of z-scores in medicine and paediatrics is widespread to accurately assess growth through anthropometric measurements such as height, weight and Body Mass Index (BMI). ![]()
0 Comments
Leave a Reply. |