Adjusted survival curve estimation using covariates

Covariate adjustment techniques, such as Cox's proportional hazard model, are used frequently in the analysis of clinical trials to compare survival of patients in two or more treatment groups. A visual display of survival curves for each treatment group is often plotted using the Kaplan-Meier method even though statistical adjustment techniques are used in making treatment comparisons. As a result, a statistically significant difference may be achieved between treatment groups although the survival curves appear similar. One solution to this problem was suggested by Hankey and Myers who derived survival curves accounting for patient characteristics based on Mantel's statistic. In this paper a method for estimating survival curves with covariates is described for use with Cox's model. An example is provided to illustrate its usefulness in providing to the medical audience a graphical display which is consistent with the achieved level of statistical significance.