有序多分类重复测量资料的广义线性混合效应模型分析

目的 探讨广义线性混合效应模型在有序多分类重复测量资料分析中的应用及SAS9.1的GLIMMIX和NLMIXED过程实现.方法 为了评价某新药治疗糖尿病神经病变的临床疗效,采用以安慰剂为对照的随机双盲临床试验.在各个随访时间记录各受试者的神经病变主觉症状总分,并根据减分率评定疗效.建立广义线性混合效应模型,并分别用线性化法和数值法积分近似法进行参数估计,利用SAS中的GLIMMIX和NLMIXED过程得以实现.结果 2种参数估计方法 结果 很接近.疗效的组间差别有统计学意义（P〈0.000 1）,试验组疗效优于安慰剂组;各个疗程间的疗效差别有统计学意义（P〈0.000 1）,且疗程越大疗效越好; 治疗前神经病变主觉症状总分对疗效有影响（P=0.061 3,接近显著性水平）,其值越高,越容易治愈,提示病情严重的患者相比病情轻微的患者治愈效果更好.另外用数值法积分近似法还给出了随机截距和随机斜率的统计显著性检验.结论 采用广义线性混合效应模型对有序多分类重复测量临床资料进行统计分析,可以更客观的进行药物疗效评价.

Generalized linear mixed models for ordered category data with repeated measurements

Objective To discuss generalized linear mixed models （GLMMS） of ordered categorical repeated measurement data, and to implement with GLIMMIX command and NLMIXED command in SAS 9.1 soft. Methods In order to assess the effectiveness of a new drug for diabetic neuropathy, the study adopted randomized, double-blind, placebo-controlled clinical method. In each follow-up time, each patient＇s neuropathy symptom score was recorded and the curative effect was evaluated by score reducing rate. We fitted parameter model applying the theory of generalized linear mixed models, and carried out the pa- rameter estimations using both linearization method and integral approximation with numerical method by GLIMMIX and NLMIXED procedure in SAS soft. Results Two kinds of parameter estimation methods were very close to the results. The differ- ence between groups was statistically significant （ P〈0.000 1 ）. The curative effects of experiment group were better than those of placebo group. The effect of time was statistically significant （ P〈0. 000 1 ） , and the longer time of the treatment required, the better the results. The neuropathy symptom score before treatment had a significant influence on curative effect （P = 0. 061 3, Close to the level of significance）, and the higher the value, the more likely to heal. It indicated that the curative effect on patients in serious condition was better than that on patients in mild condition. The random intercept and the random slope coefficients were tested for significance by integral approximation with numerical methods. Conclusion More objective results of medicine effects can be got by using generalized linear mixed models in ordered categorical repeated measurement data of clinical trials.