Modeling variance heterogeneity using the complex variation model─ applying to age-specific reference range construction

• Main Authors: Hsiao-Jung Tseng, 曾筱容
• Other Authors: Chong-Yau Fu
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/h357hy

碩士 === 國立陽明大學 === 公共衛生研究所 === 97 === In clinical medicine, some biomedical measurements are often used to construct normal ranges to offer doctors as a tool for early diagnosis. In some situations that measurements may change with age (time). When constructing such reference ranges, the age effect should be taken into account. That is so called “age-specific reference range”. The main issue of this study is to deal with the heteroscedasticity problem as well as to model variability in application to age-specific reference range construction. Parametric approaches based on regression models were used. A complex variation model was proposed to model variability and to compare with Altman’s approach (1993) of estimating variability by using absolute residuals. In this study, ultrasound data of fetal nasal bone length was used for illustration. Furthermore, simulation study was conducted to investigate the estimating effects of the complex variation modeling and two-stage estimation (modeling men and standard deviation separately) in different scenarios that data are generated with different rates of variability. Simulation results show that the estimating effect of using the complex variation model is better than the two-stage estimation as a whole. As the changing rate of variation increases, the coefficient estimates for regression obtained by the two methods are going far from the true and become more unstable. Furthermore, both precision and accuracy can be improved as the sample size becomes larger. The idea of using the complex variation model is proposed to apply to reference range construction in this study. The most advantage is that it simplifies the procedure of reference range construction and gets more accurate and reliable estimation as well.