Figure 3

Marginal Survival function. Estimate of the marginal survival function for the prostate cancer data using a transformation model estimated with pseudo-values. 32 pseudo-times were considered. Each subject is therefore replicated 32 times in the dataset. The typical estimation of the baseline risk function is through indicator variables. In this case 32 coefficients should be included in the model. A B-Spline was instead used with one knot at the median of the unique failure time distribution resulting in 4 bases plus the intercept for modeling the baseline risk. The figure shows the estimated marginal survival with superimposed the Kaplan-Meier estimate.