随访研究中两个问题的探讨

发布时间：2018-03-26 12:18

本文选题：临床随访研究　切入点：删失率　出处：《南京医科大学》2017年硕士论文

【摘要】：背景与目的:临床随访研究中以生存时间作为主要疗效指标较为常见,而数据删失是生存资料的重要特征之一。有研究表明,过高的删失率将导致疗效估计偏差,检验效能降低。实际研究中,往往组间的删失率水平不同,有时还相差较大,这种情况下对疗效评价产生何种影响尚无定论。此外,很多临床试验中观察的事件(例如肿瘤研究中发生进展)只有在访视时才能发现,因此,并不知道准确的事件发生时间,即区间删失。本文主要探讨临床试验中这两个问题:(1)组间不同删失率情况下常规疗效评价方法的统计学性质,(2)区间删失时,生存时间的不同定义方法对参数估计结果的影响。方法:采用Monte-Carlo模拟方法,基于指数分布和Weibull分布分别产生生存时间,在不同情况下,分别采用log-rank检验及Cox比例风险回归模型进行相应的参数估计和组间疗效的比较。论文分为两个部分,第一部分:在组间删失率相同和不同时,从假设检验的一类错误和检验效能、参数估计的准确度估计偏差等三个方面来评价对疗效评价的影响;第二部分:区间删失时,事件发生时间不确定,比较3种不同的生存时间的定义方法,对中位生存时间和风险比估计的影响。结果:1)在生存分析中,删失率的变化,会导致第一类错误率变化,且随删失率的增大,检验效能逐渐下降;2)相比于两组删失率相等的情况,两组删失率不等时,一类错误增加,HR的偏倚增大;且两组删失率相差越大,第一类错误率越大,检验的结果越不可信。3)区间删失时,生存时间的不同定义对中位生存时间有影响,在随访间隔不是很大时,对中位生存时间之差以及HR的估计影响不大。4)采用Cox比例风险回归模型进行疗效评价时,所得结果与log-rank检验相似。结论:1)临床随访研究中,总删失率应控制在30%以内,组间删失率之差应控制在10%以内,否则将导致一类错误膨胀。2)区间删失时,事件发生时间不确定,应尽可能缩短每次访视的时间,无论采用何种生存时间定义,不宜估计各组的中位生存时间,但估计中位生存时间之差或HR是可行的。
[Abstract]:Background & objective: survival time is common in clinical follow-up studies, and data deletion is one of the important characteristics of survival data. Test effectiveness is reduced. In actual studies, the level of deletion rate is often different among groups, and sometimes there is a big difference. In this case, the impact on the evaluation of curative effect has not been decided. The events observed in many clinical trials (such as progress in cancer studies) can only be detected during the visit, so the exact time of the event is not known. This paper mainly discusses the statistical properties of the conventional evaluation methods of curative effect under different deletion rates in clinical trials. Methods: Monte-Carlo simulation method is used to generate survival time based on exponential distribution and Weibull distribution respectively. Log-rank test and Cox proportional risk regression model were used to estimate the parameters and compare the efficacy of the two groups. The effect of hypothesis test on the evaluation of efficacy is evaluated from three aspects: the error of hypothesis test and the efficiency of test, the error of estimating accuracy of parameter estimation, etc. The second part: when the interval is censored, the time of event occurrence is uncertain. The effects of three different definition methods of survival time on median survival time and risk ratio estimation were compared. Results: in survival analysis, the change of deletion rate will lead to the change of error rate of the first kind, and increase with the increase of deletion rate. Compared with the two groups with the same deletion rate, the bias of HR increased when the deletion rate of two groups was not equal, and the greater the difference between the two groups, the greater the error rate of the first group. The less reliable. 3) interval deleted, the different definitions of survival time had an effect on median survival time, and when the interval of follow-up was not very large, The difference of median survival time and the estimated impact of HR. 4) when the Cox proportional risk regression model was used to evaluate the efficacy, the results were similar to that of the log-rank test. Conclusion: in the clinical follow-up study, the total deletion rate should be controlled within 30%. The difference of deletion rate between groups should be controlled within 10%, otherwise it will cause a class of errors to expand. 2) when the time of occurrence is uncertain, the time of each visit should be shortened as far as possible, no matter what definition of survival time is used. It is not suitable to estimate the median survival time of each group, but it is feasible to estimate the difference of median survival time or HR.
【学位授予单位】：南京医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R195.1

【参考文献】