加速失效模型及其在医学研究中的应用

【作者】 师成虎；

【导师】 何大卫； 王彤； 余红梅；

【作者基本信息】 山西医科大学 ， 流行病与卫生统计学， 2003， 硕士

【摘要】 近年来，生存分析的理论和方法已经广泛应用于社会生活的各个领域，为人类的医疗卫生保健事业和工业生产提供科学的指导依据。就医学研究来说，怎样从众多的预后因素中筛选出对疾病的发生、发展有较大影响的重要因素，已逐渐成为医学研究的热点。在这一方面，国际上当今最流行的有两大类模型：Cox比例风险回归模型和加速失效时间模型。它们各有自己的优缺点和适用范围。其中Cox模型的研究工作已经取得了很大的进展，得以为广大的医务人员所运用，而对加速失效时间模型的研究还很不成熟，在实际工作中也很少应用。但是它又是非常重要的一类模型，能用于分析许多不适合Cox模型的生存数据，所以有必要对其进行深入的研究和探讨。 本次研究结合当今国际上有关方面的最新进展，主要从模型建立方面对AFT模型进行了一些论述。在研究过程中采用了参数法和非参数法：首先结合实例，用威布尔回归模型、对数logistic回归模型、对数正态分布回归模型和广义gamma分布回归模型进行数据分析；然后采用模拟退火法和重复抽样法对数据进行了半参数分析。通过比较论证，表明无论Cox模型适用于否，加速失效时间模型都能对数据做出很好的分析。因此，作为Cox模型的备择模型(替代模型)，建议大家对其有所认识，并能在实际工作中予以应用。更多还原

【Abstract】 In recent years, the theory of Survival Analysis has been widely used in many fields of human life, and it can be used to offer scientific guidance to human health care and industrial producing process. In the scope oC medical research, interests of biological statistist is focused on how to select the significant factor that can influence the occurrence and proceeding of some disease. There are two most popular models in this area: proportional hazards model(Cox model, PHD model) and the accelerated failure time model (AFT model), they have their own advantage, disadvantage and the proper situation respectively. A great deal of valuable results have been obtained on Cox proportional hazards model. It is regretful that the proceeding on accelerated failure time model is not as successful as on Cox one, and it is seldom being used in medical research. However, it is a kind of model which has its importance and can be used to analysis survival data, particularly in situations that Cox model fails to work.In this article, we discuss the problem that how to analysis survival data with the AFT model. Four parametric models are used in our discuss, they are Weibull regression model, log logistic regression model, log normal regression model and generalized gamma regression model. We also fit the same survival data with simulated annealing method and resampling method semiparametricly.Although the classical Cox model may be not appropriate, the accelerated failure time model can fit survival data well enough. It should be regarded as an alternative to the Cox proportional hazards model in medical research.更多还原