中文摘要：

目的 比较四种方法估计含有Ⅱ型区间删失数据的回归模型参数最大似然估计值,探讨在不同样本量情况下四种方法估计结果的准确性和稳定性。方法 对含有Ⅱ型区间删失的生存数据分别拟合Weibull参数回归模型和Cox-PH半参数回归模型,并结合EMICM算法得到模型参数的最大似然估计值;应用组均值插补法将区间删失数据填补为右删失数据,进一步用传统的Cox回归以及建立伪观察值的方法估计生存函数,模拟样本量分别为50、200、500例。结果Weibull回归模型的参数最大似然估计值分别为β_1=0.496,β_2=-0.366;β_1=0.680,β_2=-0.586;β_1=0.620,β_2=-0.504。Cox-PH半参数回归模型的参数最大似然估计值为β_1=0.652,β_2=-0.469;β_1=0.683,β_2=-0.538;β_1=0.629,β_2=-0.511。填补为右删失数据后传统Cox回归方法得到的参数最大似然估计值分别为β_1=0.203,β_2=-0.227;β_1=0.641,β_2=-0.514;β_1=0.545,β_2=-0.446。用Pseudo-observations得到的参数最大似然估计值分别为β_1=0.217,β_2=-0.275;β_1=0.796,β_2=-0.601;β_1=0.561,β_2=-0.468。结论 在不同样本量情况下,拟合Weibull参数回归模型,Cox-PH半参数回归模型结合EMICM算法估计的参数最大似然估计更准确更稳定。

英文摘要：

Objective Our study compares four methods to estimate the maximum likelihood estimators of regression model, which contains the type Ⅱ interval censored failure data, and discusses the accuracy and stability of the results under the conditions of different sample size. Methods The sample size was 50,200 and 500. We use the weibull regression model and Cox-PH model to fit the type Ⅱ interval censored failure data. EMICM arithmetic is used to estimate the maximum likelihood estimations of covariates. The imputations of fight censored is estimated by traditional Cox model and pseudo-observations. Results In the first method, the estimators of parametrics based on weibull model were β1 = 0. 496,β2 = -0. 366 ;β1 = 0. 680,β2 = - 0. 586 ;β1 = 0. 620 ,β2 = - 0. 504. While the estimators of parametrics based on Cox-PH model were β1 = 0. 652 ,β2 = - 0. 469 ; β1 = 0. 683 ,β2 = -0. 538 ;β1 = 0. 629 ,β2 = -0. 511, respectively. After filling for the failure time data, the estimators based on the cox model wereβ1 = 0. 203 ,β2 = 0. 227 ;β1 = 0. 641 ,β2 = - 0. 514 ;β1 = 0. 545 ,β2 = - 0. 446. The estimators of paramet- fics based on pseudo-observations were β1 =0. 217 ,β2 = -0. 275 ;β1 = 0. 796 ,β2 = - 0. 601 ;β1 = 0. 561 ,β2 = - 0. 468. Conclu- sion Under the conditions of different sample size, fitting of weibull parameters regression model, Cox-PH semi-parametric re- gression model combining EMICM algorithm to estimate maximum likelihood parameters were more accurate and stable.