复发事件资料的生存分析在临床试验中的应用及SAS实现

Application of Survival Analysis to Recurrent Event Data in Clinical Trials with SAS

目的探讨复发事件资料的生存分析在临床试验中的应用。方法为了评价某新药治疗成人慢性乙型肝炎的有效性,采用多中心、随机、双盲双模拟、阳性药物平行对照的Ⅱ期临床试验,对238例受试者的HBVDNA进行重复观测。以HBVDNA阴转作为终点事件,记录各次阴转时间,分别用总时间模型(total time model)和边际模型(marginal model)进行复发事件资料的生存分析,用稳健的夹心方差估计方法对模型的回归系数进行估计,并进行组间比较,用SAS中的PHREG过程实现。结果服用试验药比服用对照药,HBV DNA更容易阴转,且时间更短,尤其是服药后的前期(24周内),随着随访时间的增加,试验药和对照药对于HBV DNA阴转的影响逐渐接近。治疗前的HBVDNA值越大,治疗后越不容易阴转,且时间越长,随着阴转次数的增加,治疗前HBV DNA值的大小对治疗后阴转的影响越来越小。结论对于复发事件资料的生存数据,应根据实际情况采用合适的模型,才能更客观地进行评价。

Objective To discuss the application of survival analysis to recurrent event data in clinical trials. Methods In order to e- valuate the effectiveness of the new drug for chronic hepatitis B, a multi- center, randomized, double-blind, double-dummy, positive parallel con- trolled,Phase Ⅱ clinical trial was adopted in 238 patients with the repeated observations of HBV DNA. The total time model and marginal model were applied to analysis recurrent event data with the HBV DNA negative con- version as the endpoint event and meanwhile,the time of each HBV DNA negative conversion was recorded. The robust estimator of the variance of the regression coefficients was used to adjusting for the correlation among outcomes on the same subject. Corresponding analysis methods were pro- grammed with PHREG procedure of SAS software. Results HBV DNA turn to negative after taking the new drug more easily and more quickly than taking the positive control drug especially in 24 weeks, and with the increase of the follow-up time, the influence between the new drug and con- trol drug to HBV DNA negative conversion changed more and more close- ly. the greater the value of HBV DNA before treatment, the more hardly and more slowly HBV DNA after treatment to turn to negative, and with the increase of the numbers of HBV DNA negative conversion, the influence of HBV DNA value before treatment to HBV DNA negative conversion changed more and more little. Conclusion It can get more objective effect assessment when we choose appropriate survival analysis model ac- cording to the actual situation for recurrent event data.