多种倾向性评分匹配方法评价中医药对艾滋病患者CD4^(+)T细胞的长期影响

Impact of Chinese Medicine on the Long-Term Trend of CD4^(+)T Cell Count Among AIDS Patients by Multiple Propensity Score Matching Methods

目的 探讨在观察性临床研究中最佳的控制研究对象基线信息均衡性的倾向性评分(PS)匹配(PSM)方法,并运用最佳方法评价中医药治疗对艾滋病患者CD4^(+)T细胞的长期影响。方法 以既往艾滋病治疗数据为信息来源,收集河南省2006年参与“中医药治疗艾滋病试点项目”(中医项目)地区的艾滋病患者信息,对其CD4^(+)T细胞计数进行12年的随访。按是否参与中医项目分为中医组和非中医组。分别采用一般线性模型(Logistic回归)法、一般化加速建模(GBM)和神经网络模型(NNET)估计PS值;采用“最邻近匹配”(NNM)并设定“卡钳值”。“卡钳值”分别设定为0.01、0.2倍的两组PS值logit合并方差(σ),和0.1倍σ及“最优匹配”(OM),共12种PSM方法进行1:1无放回匹配。使用两组协变量的均值标准化差异评价PSM后两组患者基线信息的均衡性,并绘制匹配后中医组与非中医组CD4^(+)T细胞计数变化趋势。所有分析均采用R 4.1.1软件包。结果 共纳入2 783例艾滋病患者,其中中医组544例,非中医组2 239例。Log 0.01 NNM,Log 0.2σNNM模型匹配效果最好。使用Log 0.01 NNM模型数据分析中医治疗对艾滋病患者CD4^(+)T细胞影响的结果显示:在治疗初期两组艾滋病患者CD4^(+)T细胞均逐年升高,第6年CD4^(+)T细胞计数达到400个/μL左右时,增势消失。第1年中医组CD4^(+)T细胞计数均值高于非中医组(P<0.05),其余随访年两组CD4^(+)T细胞计数比较差异无统计学意义(P>0.05)。结论 使用Logistic回归计算PS值并用NNM匹配可以更好的控制观察性临床研究中的研究对象基线信息的均衡性;中医药在艾滋病患者CD4^(+)T细胞<350个/μL时,可以快速提升CD4^(+)T细胞计数。

Objective To explore the appropriate propensity score(PS) matching(PSM) method which could optimally control the baseline information balance in clinical observational studies,and use the PSM to evaluate the long-term effects of Chinese medicine(CM) on CD4^(+)T cells in acquired immune deficiency syndrome(AIDS)patients. Methods The information of AIDS patients in Henan province who participated in the "National CM–AIDS Treatment Trial Program"(CM program) in 2006 was collected in normal AIDS medical registries,and the CD4^(+)T cell count was followed up for 12 years. The patients were assigned to the CM group and non-CM group according to whether they participated in CM program or not. Logistic regress model,generalized boosted model(GBM) and neural network model(NNET) were used to estimate the PS value. Matching methods were nearest neighbor matching(NNM) with caliper value were 0.01,0.1 times logit merges variance(σ),and 0.2 times σand optimal matching(OM),a total of 12 PSM methods were used for 1:1 no-put matching. Standard difference(SD)between the 2 groups was calculated to evaluate the balance of variables after PSM,and the trend of CD4^(+)T cell count in the CM group and the non-CM group was plotted before and after PSM. R 4.1.1 software package was used for all the analyses. Results A total of 2 783 AIDS patients were enrolled,including 544 cases in the CM group and 2 239 cases in the non-CM group. The Log 0.01 NNM and Log 0.2σ NNM models have the best matching effect. Log 0.01 NNM model data were used to analyze the impact of CM treatment on CD4^(+)T cells of AIDS patients and the results showed that CD4^(+)T cells of AIDS patients in both groups increased year by year in the beginning of treatment,and the increase disappeared when the CD4^(+)T cell count reached about 400 cells/μL in the 6th year. In the first year,the mean CD4^(+)T cell count in the CM group was higher than that in the non-CM group(P<0.05),and there was no significant difference in the other follow-up years between the 2 groups(P>0.05).Conclusions Using the Logistic regress model to calculate PS value and NNM matching can better control the balance of baseline information of subjects in observational clinical studies. CM can rapidly increase the count of CD4^(+)T cells in AIDS patients when CD4^(+)T cells <350/μL.