336 638

Cited 0 times in

Hypo-exponential 분포 하에서 로그-순위 검정을 이용한 생존함수 비교에 대한 표본수와 검정력 연구

Other Titles
 Sample size and power for comparison of survival functions under hypo-exponential distribution based on log-rank test 
Authors
 고희중 
Issue Date
2017
Description
의학전산통계학협동과정 의학통계학전공/박사
Abstract
본 논문에서는 전체 생존기간의 비교를 목적으로 하는 항암제 임상시험에서 시험군과 대조군의 무진행 생존기간과 진행 후 생존기간의 분포가 각각 지수분포를 따르고 따라서 전체 생존기간의 분포는 hypo-exponential 분포를 따른다고 가정했을 때, 로그-순위 검정을 이용한 생존함수 비교에 대한 표본수와 검정력에 대하여 연구하였다. 이를 위하여 로그-순위 검정에 대한 검정 통계량을 결정하는 비 중심 평균 모수를 구함에 있어, 검정 통계량의 기댓값을 이용하는 방법(Zhang 방법)과 비 중심 평균 모수를 구성하는 각 함수의 정확한 형태를 이미 알고 있다는 사실을 이용하여 직접 적분을 적용(Schoenfeld 방법)하는 두 가지 방법을 제안하였다.
Zhang 방법에서는 시험군과 대조군의 위험비의 차이가 크면 검정력이 계획한 수준에 비해 낮게 나타났으나, 진행 후 생존기간이 9개월 이상이면 대조군의 무진행 생존기간에 관계없이 계획한 수준의 검정력을 만족하는 것으로 나타났으며, Schoenfeld 방법에서는 진행 후 생존기간과 대조군의 무진행 생존기간에 관계없이 모든 경우에 대하여 계획한 수준의 검정력을 만족하는 것으로 나타났으나, 시험군과 대조군의 위험비의 차이가 크거나 진행 후 생존기간이 3개월 또는 6개월로 짧은 경우에는 계획한 수준의 검정력을 초과하는 결과도 나타났다. 결과적으로 임상시험에 필요한 환자 수는 전체 생존기간의 분포가 hypo-exponential 분포를 따른다고 가정하면, 지수분포를 따른다고 가정하였을 때에 비해 감소하는 결과를 얻을 수 있었다.


Most clinical trials of anti-cancer agent, comparison of overall survival rate between treatment and control arms is used for primary end point and progression free survival for secondary end point. Overall survival is partitioned into two parts and expressed it as the sum of progression free survival and post progression survival. If the progression free survival and post progression survival have exponential distribution, then the overall survival has hypo-exponential distribution. In simulation results of Broglio and Berry's study(2009), the statistical power of the log-rank test for the comparison of two overall survival function decreased as the post progression survival increased when two progression free survival is different. Park SH(2013) assumed that log-rank test which is widely used to compare survival times of two groups is no longer satisfied with proportional hazards, so statistical power of log-rank test would not reached at expected level. She compared statistical power for log-rank test which is used to compare of two survival functions with Gehan, Peto-Peto, Fleming-Harrington (p=0.5,q=0.5) and Renyi type test statistics which have different weight depend on times and log-rank test show the most powerful statistical power among other tests in the simulation.
In this thesis, we assume that each progression free survival and post progression survival have exponential distribution therefore the overall survival has hypo-exponential distribution for comparing overall survival between treatment and control arms in the anti-cancer clinical trials. We proposed two methods to calculate non centrality mean parameter of log-rank test statistic using Schoenfeld(1981) and Zhang and Quan(2009). We organized log-rank test statistic, overall survival has hypo-exponential distribution and the methods to calculate non centrality mean parameter of log-rank test statistic. And we proposed a method of using expectation of test statistic and a method of using integral with an exact known form of each functions to calculate non centrality mean parameter. A simulation was designed to assess the performance of proposed methods with various conditions, such as progression free survival, post progression survival, interval unit, total study duration and power.
In Zhang method, the power satisfies the pre specified level when post progression survival is longer than 9 months regardless of the period of progression free survival of control arm, however the power is under estimated in certain situations when the difference of hazards ratio is bigger. In Schoenfeld method, all the power satisfies the expected level regardless any conditions, however the over power appears when the bigger difference of hazards ratio and the shorter post progression survival. At last, we find out small sample size need in clinical trials when the overall survival has hypo-exponential distribution against exponential distribution.
Files in This Item:
T014213.pdf Download
Appears in Collections:
1. College of Medicine (의과대학) > Others (기타) > 3. Dissertation
URI
https://ir.ymlib.yonsei.ac.kr/handle/22282913/154750
사서에게 알리기
  feedback

qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse

Links