拟合生长标准百分位数曲线的不同方法比较

Comparison of different approaches of fitting centile curves for growth standard

目的 寻找适合于拟合各种人体测量指标的生长标准曲线的方法和途径 .方法 采用损失函数 (loss function)和残差平方和 (residual sum of square)对加权三次样条法(weighted cubic smoothing spline)和 λ-中位数 -变异系数法 (λ-median- coefficient of variation,L MS)进行对比研究 .结果 计算了两种方法拟合各组的身高 (height)、体质量 (bodymass)、体块指数 (body m ass index,BMI)的损失函数值和残差平方和 ,给出了两种方法拟合的百分位数曲线 .结论 L MS法简单易实现 ,拟合的百分位数曲线光滑 ,拟合误差小 .L MS法比加权三次样条更适合于拟合身高、体质量。

AIM To find out a method to construct growth curves for various anthropology index. METHODS Loss function and residual sum of square were used to compare thefitting results of LMS method with weighted cubic smoothing spline. RESULTS Loss function and residual sum of square for each method when applied to constructing body mass.height and BMI centile curves for children aged 0-18 years in Xi'an were calculated. Centile curves fitted by LMS method and weighted cubic smoothing spline were presented. CONCLUSION LMS is a simple method. Curves fitted by it are much smoother and the fitting errors are much fewer. Therefore LMS is more suitable for fitting centile curves for body mass, height and BMI than weighted cubic smoothing spline.