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OA北大核心CSTPCD

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 CD4T 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(o),and 0.2 times o and 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 0.2 o 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/μ L.