纵向监测连续非随机缺失数据变系数模型及其应用  

The Application of Varying-coefficient Model in Longitudinal Data with Continuity Non-randomized Missingness

作　　者：徐丽红[1] 刘志永[1] 刘桂芬[1] 罗天娥[1] 

机构地区：[1]山西医科大学公共卫生学院卫生统计教研室,030001

出　　处：《中国卫生统计》2012年第3期314-317,共4页Chinese Journal of Health Statistics

基　　金：国家自然基金项目(编号81172774);山西省自然科学基金项目(2009011005-2);国家青年科学基金项目资助(编号81001294)

摘　　要：目的阐明一种适于纵向监测时间不固定研究资料的模型分析方法。方法将贝叶斯惩罚样条函数与模式混合模型原理相结合,建立均值参数与方差成分随时间变化的变系数模型,并将其应用于社区高血压规范化管理研究。结果社区高血压规范化管理研究结果表明,对于收缩压,随观测时间延长,监测患者年龄参数和中心3呈下降趋势,性别参数呈上升趋势,而其他参数基本为一个常数;对于舒张压,性别参数为一常数,高血压病程呈上升趋势,其余参数变化情况与收缩压基本一致。随机效应基本为一个常数,收缩压的个体内误差呈下降趋势,而舒张压则略有上升。不同敏感性参数对结果的影响较小。结论变系数模型不但允许均值参数随时间变化,且允许模型中方差成分随缺失时间变化,可获得更精确的参数估计值,结果解释合理,为连续非随机缺失资料分析提供了新思路。Objective To illustrate a method used in longitu- dinal data with continuity time. Methods Combining the Bayesian pe- nalized spline function with pattern mixed model, we construced the var- ying-coefficient model with mean and variance component changing with time. This model was used in study of hypertension standardized manage- ment in community. Results The results of systolic pressure depicted that the age and center 3 parameters were in descending trend and gender was in ascending trend with the observation time, while other parameters were con- stant; The results of diastolic pressure was very close to that of systolic pressure except the gender which was a constant over time and hypertension course which showed a ascending trend. The randomized effect was a con- stant over time, while the within error of systolic pressure shows a descend- ing trend, which is reverse in diastolic pressure. There was little affection in the results of different sensitivity parameters. Contusion The varying- coefficient model allows both mean parameter and variance component changing with time, which can get more precise parameter estimates and more reasonable results than other methods. Therefore, this is a new thought in analyzing longitudinal data with continuity non-randomized missingness.

关 键 词：变系数模型 连续观测 社区高血压规范化管理 敏感性分析 

分 类 号：O212.1[理学—概率论与数理统计]