Objective The proportional hazards assumption and the influential cases in Cox regression for survival data were assessed with software packages SAS 8.2 and SPSS 11.5 and comparison of the differences between the two results were made.Methods The assessing of proportional hazards assumption used Kaplan-Meier survival curve,the curve of log[-logS(t)]to t and time-covariate test.Schoenfeld residual,weighted score residual,martingale residual,deviance residual,likelihood displacement and maximum influence curvature were used to identify influential cases.Software packages SAS 8.2 and SPSS 11.5 were performed.Results The graphical methods and formal tests can assess the proportional hazards assumption to some extent,other methods were very quick and convenient except the method of log[-logS(t)]to t which needed extra SAS program.SAS can output all diagnostics to identify influential cases including Schoenfeld residual,weighted score residual,martingale residual,deviance residual,likelihood displacement and maximum influence curvature,while SPSS can only output some of these diagnostics.The partial residual of SPSS is the Schoenfeld residual of SAS.Conclusion It was necessary to test the proportional hazards assumption and to detect the influential cases in survival data.The efficacy of diagnostics was basically the same for SAS and SPSS.SPSS was better than SAS in testing of proportional hazards assumption,but in detecting the influential points SAS was better. |