Preface to First Edition - lib

ANALYSIS USING R 207R> data("GBSG2", package = "ipred")R> plot(survfit(Surv(time, cens) ~ horTh, data = GBSG2),+ lty = 1:2, mark.time = FALSE, ylab = "Probability",+ xlab = "Survival Time in Days")R> legend(250, 0.2, legend = c("yes", "no"), lty = c(2, 1),+ title = "Hormonal Therapy", bty = "n")Probability0.0 0.2 0.4 0.6 0.8 1.0Hormonal Therapyyesno0 500 1000 1500 2000 2500Survival Time in DaysFigure 11.3Kaplan-Meier estimates for breast cancer patients who either receiveda hormonal therapy or not.R> ci exp(cbind(coef(GBSG2_coxph), ci))["horThyes",]2.5 % 97.5 %0.7073155 0.5492178 0.9109233This result implies that patients treated with a hormonal therapy had a lowerrisk and thus survived longer compared to women who were not treated thisway.Model checking and model selection for proportional hazards models arecomplicated by the fact that easy-to-use residuals, such as those discussed inChapter 6 for linear regression models, are not available, but several possibilitiesdo exist. A check of the proportional hazards assumption can be done bylooking at the parameter estimates β 1 , ...,β q over time. We can safely assume© 2010 by Taylor and Francis Group, LLC