Skip to main content
Journal of Southern Medical University logoLink to Journal of Southern Medical University
. 2021 Mar 20;41(3):391–398. [Article in Chinese] doi: 10.12122/j.issn.1673-4254.2021.03.11

基于Adaptive Elastic Net与加速失效时间模型的亚组识别方法的应用拓展

Subgroup identification based on accelerated failure time model combined with adaptive elastic net

Hongxia WEI 1, Pei KANG 1, Yingxin LIU 1, Fuqiang HUANG 1, Zheng CHEN 1,*, Shengli AN 1,*
PMCID: PMC8075779  PMID: 33849830

Abstract

Objective

To solve the problem of identifying subgroup in a randomized clinical trial with respect to survival time, we present a strategy based on accelerated failure time model to identify the subgroup with an enhanced treatment effect.

Methods

We fitted and compared univariate accelerated failure time (AFT) models and penalized AFT models regularized by adaptive elastic net to identify the candidate covariates. Based on these covariates, we utilized change-point algorithm to classify the patient subgroups. A two-stage adaptive design was adopted to verify the treatment effect in certain subgroups. Simulations were conducted across different scenarios to evaluate the performance of the models.

Results

As the correlation between covariates differed, the power of the models with an adaptive design was stable. In the two-stage adaptive design, the power of the models was the highest when the two significance levels (α1 and α2) were allocated to be 0.035 and 0.015, respectively. A better effect of the responder subgroup was associated with a higher power of the models. For a fixed sample size, the power decreased as the covariate number to sample size ratio increased, but the power showed a stable trend when the ratio was above 1. The univariate models showed different distribution patterns of the parameters for different survival time, while their distribution was relatively stable in the penalized AFT models.

Conclusion

The correlation between the covariates does not affect the performance of univariate AFT models and penalized AFT models. (0.035, 0.015) can be used as a reference for the significance level of an adaptive design. For small differences in the treatment effect between the responder and the non-responder, the penalized AFT model including the main effect of covariate (Penalized, Eq_in) outperforms the univariate AFT model excluding the main effect of covariate (Univariate, Eq_ex). Univariate, Eq_ex performs better when the covariate number to sample size ratio is less than 1, but is outperformed by Penalized, Eq_in when the ratio is above 1. The parameter distribution of survival time has a greater impact on the univariate models but a smaller impact on the penalized models.

Keywords: accelerated failure time model, adaptive elastic net, adaptive design, survival data, subgroup identification


近年来,世界各地的监管机构和卫生技术评估机构越来越重视从试验(治疗)中识别获益的患者亚群,即亚组识别[1];当传统临床随机对照试验中组间疗效无差异时,通过亚组识别,若能识别出实际上在试验组中存在明显获益的亚组人群,则可以有效减轻疾病对该亚组人群的危害[2]。目前,国内外亚组识别的研究已较多[3-5],但涉及生存数据亚组识别的研究仍较少。现有生存数据亚组识别的研究,多是基于Cox模型[6-8],而使用此模型需要满足等比例风险假定,但此假定有时不成立;对于高维的小样本数据,Cox模型也不够稳定。加速失效时间(AFT)模型已被证实适用于高删失和高维数据[9],且无需满足等比例风险假定,因此可考虑将AFT模型作为Cox的替代模型。此外考虑到协变量之间的相关性问题,Engler和Li(2009)[10]提出了基于Elastic Net的惩罚AFT模型。但Elastic Net和其他惩罚模型(如lasso、岭回归)有一个共同的缺点:不能确保变量选择的一致性和系数估计的渐近正态性,即不具有Oracle性质[11]。针对此问题,Zou和Zhang(2009)[12]提出了Adaptive Elastic Net。Khan & Shaw(2016)[13]也基于结合Adaptive Elastic Net的AFT模型,对高维数据进行变量选择。目前这些惩罚模型的应用局限于筛选协变量[9, 10, 13],Kang(2019)[14]改进常规应用中的适应性正则化模型,将基于Adaptive Elastic Net的AFT模型应用于生存数据进行亚组识别,并探讨了其在不同的样本量、删失率、亚组比例下的性能。本文在Kang(2019)所研究的基础上进行拓展,进一步探讨协变量间相关性、获益亚组与非获益亚组间的疗效差异、样本量与待选协变量个数比例、生存数据的参数分布以及适应设计两个阶段的检验水准分配等对亚组识别的影响。

1. 方法

生存数据亚组识别分析策略包括3个部分:亚组相关协变量筛选,亚组选择和临床效用性评价。

1.1. 亚组相关协变量筛选

1.1.1. 单变量AFT模型

假设临床试验中,包含一个试验组和一个对照组,且患者中存在两个亚组:获益亚组(g+)和非获益亚组(g-)。令p表示待选协变量个数;n表示样本量;TR表示组别变量(TR =0,对照组;TR =1,治疗组);xij为第i个患者的第j个协变量;zij = xij∙TRi为第i个患者的第j个协变量与组别变量的交互项;时间Ti = min (ti, ci),其中ti和ci分别是出现终点事件和删失的时间;δi是第i个患者的结局指示变量δi = I(ti ≤ ci),当第i个患者出现终点事件则δi =1,当第i个患者删失则δi =0。

为评估亚组之间治疗效果的异质性,对治疗和亚组之间交互作用进行检验是一种常用的统计方法。本研究中,拟合了单变量AFT模型;每次只纳入一个协变量xij、治疗变量TRi,以及交互效应zij:

1.1.1. 1

其中,Yi表示观察结局,Yi = log (Ti),其中Ti是第i个患者生存时间;σ是尺度参数; εij是随机误差;β1j是协变量的回归系数;β2j是治疗变量的回归系数;β3j是协变量与治疗变量交互项的回归系数。若交互项有统计学意义,则xij为亚组相关协变量。

有研究表明同时纳入协变量xij与组别变量TR主效应时,评估交互作用的效能往往较差[15-16]。因此,学者Freidlin和Simon[17]提出了剔除协变量xij主效应的方法。本研究所对应的模型为:

1.1.1. 2

此外,若指定随机误差项εij的分布,则称为参数加速失效时间模型。若误差项服从标准正态分布、极值分布(尺度参数σ =1)、极值分布(尺度参数σ ≠ 1)、logistic分布,则分别对应着对数正态AFT模型、指数AFT模型、Weibull AFT模型、logistic AFT模型。Cox和Oakes(1984)[18]认为Weibull分布同时具有比例风险和加速失效时间的性质;Vanderhoef(1982)[19]和Scheme(2003)[20]、Sayehmir(2008)[21]等学者的研究表明Weibull回归模型的拟合效果较好。因此本文的单变量AFT模型考虑了Weibull回归模型,即f (ε) = exp (ε - exp (ε))。

此处的单变量模型,通过Benjamini-Hochberg算法控制错误发现率(false discovery rate,FDR)[22],以实现多重校正。

记U ={x1, x2, …, xk} 为预先指定的错误发现率q = 0.05时,单变量AFT模型所筛选的亚组相关协变量的子集。

1.1.2. 惩罚AFT模型

上述的单变量模型,模型每次只考虑一个协变量,没有考虑协变量之间的相关性问题;若建立多变量模型,则得到:

1.1.2. 3

若在上式中剔除协变量xj主效应,则得到:

1.1.2. 4

但是上述的两个模型通常只适合于协变量较少而样本量较多的数据;对于协变量数量超过样本量的数据,特别是高维的小样本(HDLSS)数据,常规模型不再适用。针对这个问题,Kang(2019)在上述模型参数求解时引入了Adaptive Elastic Net[14]。最终Adaptive Elastic Net结合AFT模型的目标函数为:

1.1.2. 5

其中,1/2是归一化约束;Inline graphic是适应性权重;Yi是第i个患者的对数生存时间;α是常数项;X是协变量矩阵;β是未知的p ×1的回归系数矩阵;λ1和λ2为Adaptive Elastic Net的惩罚参数。

1.2. 亚组选择

亚组选择是指在患者人群中识别出对特定治疗手段有较好疗效的亚组人群g+。对于生存数据,其分为两步:首先基于上述模型所筛选的亚组相关协变量计算每个患者的预测得分;其次是寻找预测得分的截断点,把患者分成g+和g-两个亚组。采用Hinkley等[23]学者提出的change-point算法寻找截断点[24]。

1.3. 临床效用性评价

亚组分析中,二阶段适应性设计常被用以评估治疗对亚组人群的有效性[25-26]。第一阶段,按检验水准α1,比较全部患者试验组和对照组的治疗效应;如果P ≤ α1,则结论是治疗对全部患者有效;否则继续第二阶段,按α2,比较获益亚组人群g+中试验组和对照组的治疗效应;如果P ≤ α2,则认为治疗对获益亚组人群g+ 有效(图 1)。检验水准α = α1 + α2(α取值为0.05)。

1.

1

二阶段适应性设计[14]

Two-stage adaptive design[14].

2. 模拟与实例

2.1. 模拟

2.1.1. 模拟设计

假定有p个协变量,将其分成两部分:前k个协变量作为亚组相关协变量,在获益亚组g+和非获益亚组g-中,分别来自于N(μ1, σ12) 和N(μ2, σ22) 分布;其余的p - k个协变量来自于N(μ2, σ22)分布。其中,p,k分别设为100,10;μ1, μ2, σ1, 和σ2的取值分别设为6,8,0.5和0.25。n为样本量,生存时间由以下AFT模型模拟产生:

2.1.1. 6

其中,截距β0,取值为2;σ是度量参数,其取值为1;εi是随机误差,并假定其服从广义F-分布,例如Weibull分布、指数分布、对数正态分布、对数logistics分布、gamma分布等。根据前期研究[14],得知回归系数β2取为3.2,β3j取-0.04。

记是否包含协变量主效应的公式分别为Eq_in、Eq_ex,考虑4种亚组相关协变量筛选模型:是否包含协变量主效应的单变量模型和惩罚模型,记为(“Univariate,Eq_in/Eq_ex”和“Penalized,Eq_in/ Eq_ex”)。亚组选择分类器基于change-point算法。

存在亚组的条件下,模拟研究适应性设计的检验效能。每个数据集中,二阶段适应性设计中的第一阶段检验有统计学意义(P < α1),或第二阶段检验有统计学意义(P < α2),则适应性设计的检验效能记为1,否则记为0。通过模拟1000次,计算平均的检验效能。

各种情形设置如下:(1)设置低维、高维数据;分别产生样本量不超过协变量个数(n ≤ p)、样本量远大于协变量个数(n > p)数据;其中,样本量不超过协变量个数(n ≤ p)的数据中待选协变量个数p =600,样本量远大于协变量个数(n > p)的数据中p =100,亚组相关协变量个数k均为10;(2)设置不同相关程度的协变量;亚组相关协变量间的相关性和非亚组相关协变量间的相关性设置为r =0、0.3、0.5;(3)调整适应性设计中两个阶段的检验水准α1和α2;由于所确定亚组治疗效果可能比整个研究人群大得多,因此在严格的显著性水准下对第二阶段的亚组进行分析仍可能具有较大的检验效能。所以考察了以下几种不同的组合(α1,α2)=(0.015,0.035),(0.02,0.03),(0.025,0.025),(0.03,0.02),(0.035,0.015),(0.04,0.01);(4)在治疗仅对获益亚组g+有效(u00 = u10 = u01 < u11)的情况下,选取不同的回归系数β2和β3j的取值;其中,β2的取值越大,代表获益亚组与非获益亚组间的治疗效果差异越大;此处考察如下几种组合(β2,β3j)=(0.8,-0.01),(1.6,-0.02),(2.4,-0.03),(3.2,-0.04),(4.0,-0.05),(4.8,-0.06);这几种组合下均是u00 = u10 = u01 =2;当β1 =0.8、1.6、2.4、3.2、4.0、4.8时,u11 =2.2、2.4、2.6、2.8、3.0、3.2;(5)此前Kang的研究和以上研究中待选协变量个数为100和600,选取设置可能不够严谨、客观。本研究考虑待选协变量个数与样本量(n =400)的不同比例(0.25∶1, 0.5∶1, 1∶1, 1.5∶1, 2∶1, 2.5∶1)下,模型的检验效能表现;(6)研究模拟产生生存时间的AFT模型的随机误差εi的不同分布(指数分布、极值分布、对数正态分布)对模型的影响。除了情形(3)以外,其他情形中α1 =0.03、α2 =0.02;(7)所有模拟设定样本量为400,删失率为30%,亚组比例为20%。

2.1.2. 模拟结果

不同情形下模型在适应性设计中的检验效能如图 2~4所示。

2.

2

不同情形下惩罚AFT模型在适应性设计的检验效能

Power of AFT models in subgroup identification in different scenarios. In panel A, the correlation among covariates are set to be 0, 0.3, 0.5 respectively. In panel B, Alpha1 is the significance level α1 of the 1st stage of adaptive design. In panel C, Beta is the coefficient β2.

4.

4

生存时间参数分布不同时惩罚AFT模型在适应性设计的检验效能

Power of penalized AFT model in subgroup identification when the parameter distribution of survival time differs. "Exponential" corresponds to exponential AFT model, and "Weibull" corresponds to WeibullAFT model, and "Lognormal" corresponds to lognormalAFT model.

由图 2可知,图 2A中随着协变量间相关性的提高,各模型的检验效能均呈平稳的状态。图 2B中,适应性设计两个阶段的检验水准不同的组合(α1,α2)=(0.015,0.035),(0.02,0.03),(0.025,0.025),(0.03,0.02),(0.035,0.015),(0.04,0.01)下,不含协变量主效应的惩罚性AFT模型(Penalized,Eq_ex)的检验效能呈现平稳的状态,含主效应的单变量模型(Univariate,Eq_in)的检验效能呈现缓慢上升的状态;其他两个模型在α1取值为0.015~0.03时检验效能也是平稳的状态,但在α1取值为0.035时达到最高点,而在0.04时模型的检验效能下降8%左右。图 2C中,表现最好的也是“Penalized,Eq_in”与“Univariate,Eq_ex”。各个模型的检验效能随着回归系数β2的取值增大而增大;在n > p的数据中,当β2的取值为0.8~2.4时,“Penalized,Eq_in”与“Univariate,Eq_ex”的检验效能较接近;而当β2的取值大于2.4时“,Univariate,Eq_ex”的检验效能高于“Penalized,Eq_in”。在n ≤ p的数据中,当β2的取值为0.8和3.2时,“Penalized,Eq_in”与“Univariate,Eq_ex”的检验效能较接近;而当β2的取值为1.6~2.4时,“Penalized,Eq_in”的检验效能高于“Univariate,Eq_ex”;而当β2的取值大于3.2时“,Univariate,Eq_ex”的检验效能高于“Penalized,Eq_in”。即,当β2的取值(获益亚组与非获益亚组间的治疗效果差异)较小时,“Penalized,Eq_in”较有优势,其他情况下则是“Univariate,Eq_ex”的效果更好。

由图 3可知,当待选协变量个数与样本量(n =400)的比例为0.25∶1,0.5∶1,1∶1,1.5∶1,2∶1,2.5:1时,“Penalized,Eq_ex”的检验效能较低且呈现平稳的趋势。其他三个模型的检验效能,在比例为0.25∶1,0.5∶1,1∶1时缓慢下降,下降了5%~10%;当比例为1.5∶1,2∶1,2.5∶1时,平缓下降。即样本量与待选协变量个数的比例不影响“Univariate,Eq_in”的检验效能,而其他三个模型的检验效能会随着比例的增加而缓慢下降;当比例小于1时“,Univariate,Eq_ex”的效能更高;其他情况下则是“Penalized,Eq_in”的效能更高。

3.

3

不同的协变量个数下惩罚AFT模型在适应性设计的检验效能

Power of penalized AFT model in subgroup identification in different number of sensitive covariates. Ratio is the ratio of the number of covariates to the sample size (0.25∶1, 0.5∶1, 1∶1, 1.5∶1, 2∶1, 2.5∶1). The dotted line with inverted triangle (reference line) represents the powers of conventional design. The number of sensitive covariates is 10.

由图 4可知,相比于指数AFT模型产生生存数据时模型的检验效能,Weibull AFT模型产生数据时,各个模型的检验效能大幅下降至10%~20%左右。而对数正态AFT模型产生生存数据时,各个模型的检验效能都提高了,差距随着删失率的增大而增大;而且“Univariate, Eq_in”的检验效能超越了“Penalized, Eq_in”。其中,产生生存时间的模型类型对单变量模型的影响最大;对数正态AFT模型产生生存数据时“,Univariate, Eq_ex”的检验效能提高了0%~20%,而“Univariate, Eq_in”的检验效能提高了10%~25%,惩罚模型的效能提高了5%左右。产生生存数据的模型对单变量模型的影响较大,可能是因为单变量模型是参数模型,比较容易受到生存数据的参数分布的影响。

2.2. 实例

数据来自艾滋病临床试验组175(ATCG 175)研究[27]。在这项研究中,2139名艾滋病毒感染者被随机分到4个不同的治疗组:齐多夫定(ZDV)单一疗法、ZDV+ 地达诺新(ddI)、ZDV+扎西他宾(zal)和ddI单一疗法。本研究选取联合用药组(ZDV+ddI)和单独用药组(ddI)的病人作为研究对象,两组病人的基线协变量均衡可比[31]。其中,有522名感染者接受了ZDV+ddI治疗,559人接受了ddI单一疗法。两组中的删失率分别是80.27%和77.10%。出现以下终点事件的时间作为观察结局:(i)CD4 T细胞计数至少下降50%;(ii)指示向艾滋病进展的事件;(iii)死亡。协变量包括患者的年龄和体质量、CD4和CD8计数、Karnofsky得分、先前接受抗逆转录病毒疗法的天数、是否患血友病、同性性行为、静脉注射吸毒史、研究前接受非齐多夫定抗逆转录病毒治疗、在治疗开始前30天内使用齐多夫定、在治疗开始前使用齐多夫定、种族、性别、抗逆转录病毒史、有无症状指标。前6个协变量为连续性变量,其他为二分类变量。

经检验满足比例风险假定(P =0.479),两组生存曲线比较选用log-rank方法,以及建立Cox模型计算风险比及其95%可信区间[28]。两组间生存曲线没有统计学差异(HR=0.84,P =0.181)(图 5)[28]。

5.

5

两组间的生存曲线

Survival curves of the ZDV+ddI group and ddI group.

虽然尚不能认为ZDV+ddI组的治疗效果好于ddI组。但在ZDV+ddI组的病人中,可能存在获益亚组,因此考虑用不含协变量主效应的单变量模型(Univariate, Eq_ex)以及含协变量主效应的惩罚模型(Penalized, Eq_in),再结合十折交叉验证(10-fold cross validation)对该数据进行亚组分析,其中,单变量模型通过Benjamini-Hochberg算法控制错误发现率(FDR)。

亚组识别结果如下所示。表 1中,第3~4行展现了对照组与试验组的中位生存时间和组内样本量。适应性设计的第一阶段(α1 =0.03)的结果(第1行)显示,对照组和试验组间生存时间没有统计学差异(P =0.181);因此,继续适应设计的第二阶段分析(α2 =0.02),结果(第2行)显示:“Univariate, Eq_ex”和“Penalized, Eq_in”所识别的获益亚组中的两组间差异比较的P值分别为0.369和0.549。图 6、7给出了两个模型所识别的4个亚组的生存曲线图,结果显示试验组中获益人群(T+)和对照组中获益人群(C+)之间的治疗效果没有差异(图 6),即两种疗法的获益人群相同,但“Univariate, Eq_ex”所识别的获益人群(T+和C+)的治疗效果显著优于非获益人群(T- 和C-)(P < 0.001,图 7)。其中,“Univariate, Eq_ex”和“Penalized, Eq_in”所识别的获益亚组分别为270+293=563 (52.08%)和141+183=324(29.97%)。

1.

数据ACTG175的亚组识别结果

Subgroup analysis of dataACTG175

Object Univariate, Eq_ex Penalized, Eq_in
Univariate, Eq_ex: The univariate model excluding main effect; Penalized, Eq_in: The penalized model including main effect; Overall test: The 1st stage of biomarker adaptive design(α1 =0.03); Subgroup test: The 2st stage of biomarker adaptive design(α2 =0.02). *P value of Logrank test, §Median survival time: Days (group size or subgroup size).
Overall test (P-value) * 0.181 0.181
Subgroup test (P-value) * 0.369 0.549
ZDV+ddI§ 1014 (522) 1014 (522)
ddI§ 1000 (559) 1000 (559)
ZDV+ddI g+§ 1048.5 (270) 1015(141)
ZDV+ddI g-§ 992.5 (252) 1013(381)
ddI g+§ 1020 (293) 999 (183)
ddI g-§ 991(266) 1001 (376)

6.

6

四个亚组中不含协变量主效应的单变量AFT模型(A)、含协变量主效应的惩罚AFT模型(B)所识别的亚组生存曲线图

Survival curves for subgroups identified by Univariate, Eq_ex (A), and Penalized, Eq_in(B). The four subgroups correspond to responder in ZDV+ddI group (T+), non-responder in ZDV+ddI group (T-), responder in ddI group (C+) and non-responder in ddI group (C-).

7.

7

两个亚组中不含协变量主效应的单变量AFT模型(A)、含协变量主效应的惩罚AFT模型(B)所识别的亚组生存曲线图

Survival curves for 4 subgroups identified by Univariate, Eq_ex (A), and Penalized, Eq_in (B). The two subgroups correspond to responder in both arms (T+ and C+), non-responder in in both arms (T- and C-).

3. 讨论

目前Adaptive Elastic Net与AFT模型的结合主要应用于协变量筛选,Kang[14]的创新之处在于对其进行改进并应用于生存数据的亚组识别。本文在此基础上,进一步探讨了协变量间的相关性、获益亚组与非获益亚组间的治疗效果差异、样本量与待选协变量个数比例、生存数据的参数分布以及适应设计两个阶段的检验水准分配的影响。

对于样本量大于协变量个数(n > p)的数据,在n较小、删失率较高或者亚组比例较小等不利情况下,含主效应的惩罚模型在亚组选择上相对有优势;而其他情况下,则是不含主效应的单变量模型更优;对于n ≤ p,含主效应的惩罚模型在亚组选择上相对有优势。此外,(0.035,0.015)可以作为二阶段适应形设计显着性水准的设置参考;协变量间的相关性不影响模型在适应性设计中的检验效能;生存数据的参数分布对单变量模型的影响较大,对惩罚模型的影响较小。获益亚组与非获益亚组间的治疗效果(对数生存时间log(T))差异小于0.6(n > p的数据中)或者小于0.8(n ≤ p的数据中)时,基于Eq_in的惩罚模型较有优势。

实例数据分析中,模型“Univariate, Eq_ex”的亚组识别结果显示,两种疗法的获益人群差异无统计学意义。另外,不管是接受联合用药还是单独用药,所识别的获益人群的治疗效果都优于非获益人群。这可能是由于该数据中样本量n大于协变量个数p,模型“Univariate, Eq_ex”的检验效能更高。

前期研究结果显示,在预先设定的检验水准(α1 = 0.03,α2 =0.02)下,亚组选择中表现最好的模型在适应性设计中的检验效能并未显示出优势,提示这可能与两个阶段检验水准的设定有关。本研究发现,固定样本量(400),删失率(30%)和亚组比例(20%)时,(α1,α2)=(0.035,0.015)时亚组识别能力较好的两个模型(Univariate, Eq_ex、Penalized, Eq_in)的检验效能最高,但样本量、删失率和亚组比例等因素如何影响适应性设计检验水准的设置有待进一步研究。

当治疗有效且存在获益亚组时,我们仅模拟了治疗对获益亚组人群g+有效的情况(u00 = u10 = u01 < u11);后续可拓展到治疗对获益亚组g+和非获益亚组g−都有效,但获益亚组g+ 的效果更好(u00 = u10 < u01 < u11)的情形。另外,本研究提出的亚组识别方法仅限于识别两个亚组,即获益亚组和非获益亚组;后续可延伸到识别三个及以上亚组的情况。

Cox比例风险模型和AFT模型是分析生存数据最为常用的两种模型[28-29],这两种模型潜在的假设是:在随访时间足够长的情况下,每个病人最终都将经历终点事件。然而,近年来,随着治疗效果的提高,临床试验中所收集到的生存数据常常存在被“治愈”的患者,即这部分患者(获益亚组)不可能发生相应的终点事件[28, 30-32]。换言之,生存数据中患者的获益会存在两种情况,第一种是患者仅是生存时间延长了,最终仍会发生结局事件;第二种是患者治愈了,最终不会发生结局事件[33-36]。本研究仅考虑了第一种情况,对于第二种情况下疾病的治疗,我们更关注的不是生存时间的延长而是治愈率的提高,若仍然使用传统的Cox比例风险模型或者AFT模型进行分析,可能会得出不恰当的结论,因此后续拟将本研究所提出的分析策略拓展到潜在的亚组人群为治愈亚组的情形。

Biography

韦红霞,在读硕士研究生,E-mail: 1936806707@qq.com

Funding Statement

国家自然科学基金(71673126), 南方医科大学科研启蒙项目(B219339020)

Supported by National Natural Science Foundation of China (71673126)

Contributor Information

韦 红霞 (Hongxia WEI), Email: 1936806707@qq.com.

陈 征 (Zheng CHEN), Email: zchensmu@163.com.

安 胜利 (Shengli AN), Email: ASL0418@126.com.

References

  • 1.Pocock SJ, Assmann SE, Enos LE, et al. Subgroup analysis, covariate adjustment and baseline comparisons in clinical trial reporting: current practice and problems. Stat Med. 2002;21(19):2917–30. doi: 10.1002/sim.1296. [Pocock SJ, Assmann SE, Enos LE, et al. Subgroup analysis, covariate adjustment and baseline comparisons in clinical trial reporting: current practice and problems[J]. Stat Med, 2002, 21(19): 2917-30.] [DOI] [PubMed] [Google Scholar]
  • 2.Kehl V, Ulm K. Responder identification in clinical trials with censored data. Comput Stat DataAnal. 2006;50(5):1338–55. doi: 10.1016/j.csda.2004.11.015. [Kehl V, Ulm K. Responder identification in clinical trials with censored data[J]. Comput Stat DataAnal, 2006, 50(5): 1338-55.] [DOI] [Google Scholar]
  • 3.Brookes ST, Whitley E, Peters TJ, et al. Subgroup analyses in randomised controlled trials: quantifying the risks of false-positives and false-negatives. http://europepmc.org/abstract/med/11701102. Health TechnolAssess. 2001;5(33):1–56. doi: 10.3310/hta5330. [Brookes ST, Whitley E, Peters TJ, et al. Subgroup analyses in randomised controlled trials: quantifying the risks of false-positives and false-negatives[J]. Health TechnolAssess, 2001, 5(33): 1-56.] [DOI] [PubMed] [Google Scholar]
  • 4.Lipkovich I, Dmitrienko A, B R. Tutorial in biostatistics: data-driven subgroup identification and analysis in clinical trials. Stat Med. 2017;36(1):136–96. doi: 10.1002/sim.7064. [Lipkovich I, Dmitrienko A, B R. Tutorial in biostatistics: data-driven subgroup identification and analysis in clinical trials[J]. Stat Med, 2017, 36(1): 136-96.] [DOI] [PubMed] [Google Scholar]
  • 5.Schnell PM, Müller P, Tang Q, et al. Multiplicity-adjusted semiparametric benefiting subgroup identification in clinical trials. Clin Trials Lond Engl. 2018;15(1):75–86. doi: 10.1177/1740774517729167. [Schnell PM, Müller P, Tang Q, et al. Multiplicity-adjusted semiparametric benefiting subgroup identification in clinical trials [J]. Clin Trials Lond Engl, 2018, 15(1): 75-86.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Noll DR, Degenhardt BF, Johnson JC. Multicenter osteopathic pneumonia study in the elderly: subgroup analysis on hospital length of stay, ventilator-dependent respiratory failure rate, and in-hospital mortality rate. http://www.ncbi.nlm.nih.gov/pubmed/27571294. JAm OsteopathAssoc. 2016;116(9):574–87. doi: 10.7556/jaoa.2016.117. [Noll DR, Degenhardt BF, Johnson JC. Multicenter osteopathic pneumonia study in the elderly: subgroup analysis on hospital length of stay, ventilator-dependent respiratory failure rate, and in-hospital mortality rate[J]. JAm OsteopathAssoc, 2016, 116(9): 574-87.] [DOI] [PubMed] [Google Scholar]
  • 7.Chen YC, Lee UJ, Tsai CA, et al. Development of predictive signatures for treatment selection in precision medicine with survival outcomes. Pharm Stat. 2018;17(2):105–16. doi: 10.1002/pst.1842. [Chen YC, Lee UJ, Tsai CA, et al. Development of predictive signatures for treatment selection in precision medicine with survival outcomes[J]. Pharm Stat, 2018, 17(2): 105-16.] [DOI] [PubMed] [Google Scholar]
  • 8.Motzer RJ, Ravaud A, Patard JJ, et al. Adjuvant sunitinib for highrisk renal cell carcinoma after nephrectomy: Subgroup analyses and updated overall survival results. Eur Urol. 2018;73(1):62–8. doi: 10.1016/j.eururo.2017.09.008. [Motzer RJ, Ravaud A, Patard JJ, et al. Adjuvant sunitinib for highrisk renal cell carcinoma after nephrectomy: Subgroup analyses and updated overall survival results[J]. Eur Urol, 2018, 73(1): 62-8.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Khan MHR, Bhadra A, Howlader T. Stability selection for lasso, ridge and elastic net implemented with AFT models. http://www.ncbi.nlm.nih.gov/pubmed/31586968. Stat Appl Genet Mol Biol. 2019;18(5):1515. doi: 10.1515/sagmb-2017-0001. [Khan MHR, Bhadra A, Howlader T. Stability selection for lasso, ridge and elastic net implemented with AFT models[J]. Stat Appl Genet Mol Biol, 2019, 18(5): 1515.] [DOI] [PubMed] [Google Scholar]
  • 10.Engler D, Li Y. Survival analysis with high-dimensional covariates: an application in microarray studies. http://search.ebscohost.com/login.aspx?direct=true&db=aph&AN=36585608&site=ehost-live. Stat Appl Genet Mol Biol. 2009;8:Article14. doi: 10.2202/1544-6115.1423. [Engler D, Li Y. Survival analysis with high-dimensional covariates: an application in microarray studies[J]. Stat Appl Genet Mol Biol, 2009, 8: Article14.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.黄登香. Elastic Net方法在几类模型变量选择中的应用[D]. 广西大学, 2014.
  • 12.Zou H, Zhang HH. On the adaptive elastic-net with a diverging number of parameters. http://www.ncbi.nlm.nih.gov/pubmed/20445770. Ann Stat. 2009;37(4):1733–51. doi: 10.1214/08-AOS625. [Zou H, Zhang HH. On the adaptive elastic-net with a diverging number of parameters[J]. Ann Stat, 2009, 37(4): 1733-51.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Khan MHR, Shaw JEH. Variable selection for survival data with a class of adaptive elastic net techniques. Stat Comput. 2016;26(3):725–41. doi: 10.1007/s11222-015-9555-8. [Khan MHR, Shaw JEH. Variable selection for survival data with a class of adaptive elastic net techniques[J]. Stat Comput, 2016, 26(3): 725-41.] [DOI] [Google Scholar]
  • 14.康 佩, 许 军, 黄 福强, et al. Adaptive Elastic Net结合加速失效时间模型在亚组识别中的应用. http://www.j-smu.com/oa/DArticle.aspx?type=view&id=2019101200. 南方医科大学学报. 2019;39(10):1200–6. doi: 10.12122/j.issn.1673-4254.2019.10.11. [康佩, 许军, 黄福强, 等. Adaptive Elastic Net结合加速失效时间模型在亚组识别中的应用[J]. 南方医科大学学报, 2019, 39(10): 1200-6.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Brookes ST, Whitely E, Egger M, et al. Subgroup analyses in randomized trials: risks of subgroup-specific analyses; power and sample size for the interaction test. J Clin Epidemiol. 2004;57(3):229–36. doi: 10.1016/j.jclinepi.2003.08.009. [Brookes ST, Whitely E, Egger M, et al. Subgroup analyses in randomized trials: risks of subgroup-specific analyses; power and sample size for the interaction test[J]. J Clin Epidemiol, 2004, 57(3): 229-36.] [DOI] [PubMed] [Google Scholar]
  • 16.Tian L, Alizadeh AA, Gentles AJ, et al. A simple method for estimating interactions between a treatment and a large number of covariates. http://europepmc.org/abstract/med/25729117. JAm StatAssoc. 2014;109(508):1517–32. doi: 10.1080/01621459.2014.951443. [Tian L, Alizadeh AA, Gentles AJ, et al. A simple method for estimating interactions between a treatment and a large number of covariates[J]. JAm StatAssoc, 2014, 109(508): 1517-32.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Freidlin B, Simon R. Adaptive signature design: an adaptive clinical trial design for generating and prospectively testing a gene expression signature for sensitive patients. Clin Cancer Res. 2005;11(21):7872–8. doi: 10.1158/1078-0432.CCR-05-0605. [Freidlin B, Simon R. Adaptive signature design: an adaptive clinical trial design for generating and prospectively testing a gene expression signature for sensitive patients[J]. Clin Cancer Res, 2005, 11(21): 7872-8.] [DOI] [PubMed] [Google Scholar]
  • 18.Breslow N, Cox DR, Oakes D. Analysis of survival data. Biometrics. 1985;41(2):593. doi: 10.2307/2530888. [Breslow N, Cox DR, Oakes D. Analysis of survival data[J]. Biometrics, 1985, 41(2): 593.] [DOI] [Google Scholar]
  • 19.Vanderhoeft C. Accelerated failure time models: an application to current status breast-feeding data from Pakistan. http://www.ncbi.nlm.nih.gov/pubmed/12339001. Genus. 1982;28(1/2):135–57. [Vanderhoeft C. Accelerated failure time models: an application to current status breast-feeding data from Pakistan[J]. Genus, 1982, 28 (1/2): 135-57.] [PubMed] [Google Scholar]
  • 20.Nardi A, Schemper M. Comparing Cox and parametric models in clinical studies. Stat Med. 2003;22(23):3597–610. doi: 10.1002/sim.1592. [Nardi A, Schemper M. Comparing Cox and parametric models in clinical studies[J]. Stat Med, 2003, 22(23): 3597-610.] [DOI] [PubMed] [Google Scholar]
  • 21.Sayehmiri K, Eshraghian MR, Mohammad K, et al. Prognostic factors of survival time after hematopoietic stem cell transplant in acute lymphoblastic leukemia patients: Cox proportional hazard versus accelerated failure time models. J Exp Clin Cancer Res. 2008;27:74. doi: 10.1186/1756-9966-27-74. [Sayehmiri K, Eshraghian MR, Mohammad K, et al. Prognostic factors of survival time after hematopoietic stem cell transplant in acute lymphoblastic leukemia patients: Cox proportional hazard versus accelerated failure time models[J]. J Exp Clin Cancer Res, 2008, 27: 74.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. http://www.ams.org/mathscinet-getitem?mr=1325392. J Royal Stat Soc: Ser B Methodol. 1995;57(1):289–300. [Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing[J]. J Royal Stat Soc: Ser B Methodol, 1995, 57(1): 289-300.] [Google Scholar]
  • 23.Hinkley DV, Hinkley EA. Inference about the change-point in a sequence of binomial variables. Biometrika. 1970;57(3):477–88. doi: 10.1093/biomet/57.3.477. [Hinkley DV, Hinkley EA. Inference about the change-point in a sequence of binomial variables[J]. Biometrika, 1970, 57(3): 477-88.] [DOI] [Google Scholar]
  • 24.Killick R, Eckley IA. Changepoint: AnRPackage for changepoint analysis. http://www.nrcresearchpress.com/servlet/linkout?suffix=refg37/ref37&amp;dbid=16&amp;doi=10.1139%2Fcjfas-2015-0175&amp;key=10.18637%2Fjss.v058.i03. J Stat Soft. 2014;58(3):18637. [Killick R, Eckley IA. Changepoint: AnRPackage for changepoint analysis[J]. J Stat Soft, 2014, 58(3): 18637.] [Google Scholar]
  • 25.Wang SJ, Hung HM, O'Neill RT. Adaptive patient enrichment designs in therapeutic trials. Biom J. 2009;51(2):358–74. doi: 10.1002/bimj.200900003. [Wang SJ, Hung HM, O'Neill RT. Adaptive patient enrichment designs in therapeutic trials[J]. Biom J, 2009, 51(2): 358-74.] [DOI] [PubMed] [Google Scholar]
  • 26.Jiang W, Freidlin B, Simon R. Biomarker-adaptive threshold design: a procedure for evaluating treatment with possible biomarkerdefined subset effect. J Natl Cancer Inst. 2007;99(13):1036–43. doi: 10.1093/jnci/djm022. [Jiang W, Freidlin B, Simon R. Biomarker-adaptive threshold design: a procedure for evaluating treatment with possible biomarkerdefined subset effect[J]. J Natl Cancer Inst, 2007, 99(13): 1036-43.] [DOI] [PubMed] [Google Scholar]
  • 27.Hammer SM, Katzenstein DA, Hughes MD, et al. A trial comparing nucleoside monotherapy with combination therapy in HIV-infected adults with CD4 cell counts from 200 to 500 per cubic millimeter. AIDS Clinical Trials Group Study 175 Study Team. N Engl J Med. 1996;335(15):1084–90. doi: 10.1056/NEJM199610103351501. [Hammer SM, Katzenstein DA, Hughes MD, et al. A trial comparing nucleoside monotherapy with combination therapy in HIV-infected adults with CD4 cell counts from 200 to 500 per cubic millimeter. AIDS Clinical Trials Group Study 175 Study Team[J]. N Engl J Med, 1996, 335(15): 1081-90.] [DOI] [PubMed] [Google Scholar]
  • 28.黄福强. 含治愈个体生存资料的分析策略和亚组识别研究[D]. 南方医科大学, 2019.
  • 29.Gill RD. Multistate life-tables and regression models. Math Popul Stud. 1992;3(4):259–76. doi: 10.1080/08898489209525345. [Gill RD. Multistate life-tables and regression models[J]. Math Popul Stud, 1992, 3(4): 259-76.] [DOI] [PubMed] [Google Scholar]
  • 30.Zhang J, Peng Y. A new estimation method for the semiparametric accelerated failure time mixture cure model. Stat Med. 2007;26(16):3157–71. doi: 10.1002/sim.2748. [Zhang J, Peng Y. A new estimation method for the semiparametric accelerated failure time mixture cure model[J]. Stat Med, 2007, 26 (16): 3157-71.] [DOI] [PubMed] [Google Scholar]
  • 31.Corbière F, Commenges D, Taylor JM, et al. A penalized likelihood approach for mixture cure models. Stat Med. 2009;28(3):510–24. doi: 10.1002/sim.3481. [Corbière F, Commenges D, Taylor JM, et al. A penalized likelihood approach for mixture cure models[J]. Stat Med, 2009, 28(3): 510-24.] [DOI] [PubMed] [Google Scholar]
  • 32.Taylor JM. Semi-parametric estimation in failure time mixture models. Biometrics. 1995;51(3):899–907. doi: 10.2307/2532991. [Taylor JM. Semi-parametric estimation in failure time mixture models[J]. Biometrics, 1995, 51(3): 899-907.] [DOI] [PubMed] [Google Scholar]
  • 33.Boag JW. Maximum likelihood estimates of the proportion of patients cured by cancer therapy. http://onlinelibrary.wiley.com/doi/10.1111/j.2517-6161.1949.tb00020.x. J Royal Stat Soc: Ser B Methodol. 1949;11(1):15–44. [Boag JW. Maximum likelihood estimates of the proportion of patients cured by cancer therapy[J]. J Royal Stat Soc: Ser B Methodol, 1949, 11(1): 15-44.] [Google Scholar]
  • 34.Maetani S, Gamel JW. Parametric cure model versus proportional hazards model in survival analysis of breast cancer and other malignancies. Adv Breast Cancer Res. 2013;2(4):119–25. doi: 10.4236/abcr.2013.24020. [Maetani S, Gamel JW. Parametric cure model versus proportional hazards model in survival analysis of breast cancer and other malignancies[J]. Adv Breast Cancer Res, 2013, 2(4): 119-25.] [DOI] [Google Scholar]
  • 35.Othus M, Barlogie B, Leblanc ML, et al. Cure models as a useful statistical tool for analyzing survival. Clin Cancer Res. 2012;18(14):3731–6. doi: 10.1158/1078-0432.CCR-11-2859. [Othus M, Barlogie B, Leblanc ML, et al. Cure models as a useful statistical tool for analyzing survival[J]. Clin Cancer Res, 2012, 18 (14): 3731-6.] [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Zhang Y, Shao Y. Concordance measure and discriminatory accuracy in transformation cure models. Biostatistics. 2018;19(1):14–26. doi: 10.1093/biostatistics/kxx016. [Zhang Y, Shao Y. Concordance measure and discriminatory accuracy in transformation cure models[J]. Biostatistics, 2018, 19(1): 14-26.] [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Journal of Southern Medical University are provided here courtesy of Editorial Department of Journal of Southern Medical University

RESOURCES