目的 :识别 Cox回归模型中的强影响点。方法 :从似然距离的观点出发 ,应用局部影响的分析方法 ,介绍 Δk诊断量和最大影响曲率度量诊断量。结果 :对参数估计有强影响的点 ,两种诊断量的绝对值远远大于其它点。结论 :此法可有效地识别影响参数估计的强影响点 ,从而避免参数估计时得出错误的结论。如参数较多时 ,最大曲率度量方法更为方便 Purpose : To identify strong influence points in the Cox regression model. Methods: From the point of view of likelihood distance, the method of local influence analysis was used to introduce the Δk diagnosis quantity and the maximum influence curvature metric diagnosis quantity. Results: The point where there is a strong influence on the parameter estimation, the absolute value of the two diagnostic quantities is much larger than other points. Conclusion: This method can effectively identify the strong influence points that affect the parameter estimation, so as to avoid erroneous conclusions when estimating the parameters. If there are more parameters, the maximum curvature measurement method is more convenient