- 期刊
Research on the Test Method of Proportional hazard Hypothesis in Cox Model
摘要
With the growth of world economy, the development of health care, the change of disease spectrum and the improvement of life expectancy, the follow‐up studies on clinical trials and epidemiology of tumor, chronic diseases and senile diseases are becoming more and more important. The data of these clinical trials and follow‐up studies can be sorted into survival data. Survival analysis is a subject that makes statistical inference on one or more non negative random variables, studies survival phenomenon and response time data and their statistical laws. At present, Cox proportional hazard regression model is still the most commonly used method for multivariate analysis of survival data. Due to the wide range of application of Cox model, analysts often ignore its application condition proportional hazard assumption, which directly affects the stability of the model. This paper systematically discusses the common methods of Cox model to test the proportional hazard hypothesis from two aspects: graphical method and hypothesis testing method. This paper studies six graphical methods, namely direct observation method, Cox & K‐M comparison method, log‐log image method, various graphical methods based on cumulative hazard function, Schoenfeld residual diagram and score residual diagram. In addition, five hypothesis testing methods are studied, which are linear correlation test, time covariate test, weighted residual score test, omnibus test and a new cubic spline function method. This paper explores the principles of these methods and compares their advantages and disadvantages.