I am not sure how to interpret the results given from the Cox PH model and the Kaplan-Meier curves. The analyses are conducted in R.

In an observational study, 172 patients are categorized based on tumor profileration status (Ki67): 0-2%, 3-4%, 5-10% and >10% (an index of tumor activity).

I am investigating the correlation between Ki67-status and the event of tumor recurrence.

library(survival) library(survminer)

coxph(formula = Surv(time.to.recurrence, event.recurrence) ~ 1 + 
    factor(ki67), data=p)

  n= 172, number of events= 30 
   (4 observations deleted due to missingness)

                  coef exp(coef) se(coef)     z Pr(>|z|)   
factor(p$ki67)1 0.1109    1.1172   0.5781 0.192  0.84793   
factor(p$ki67)2 0.7136    2.0412   0.6065 1.176  0.23942   
factor(p$ki67)3 1.8830    6.5731   0.6774 2.780  0.00544 **

I want to test if the proportional hazard assumption is violated:

                   rho chisq     p
factor(p$ki67)1  0.064 0.123 0.726
factor(p$ki67)2 -0.151 0.702 0.402
factor(p$ki67)3 -0.222 1.450 0.229
GLOBAL              NA 4.686 0.196

By using ggcoxzph(), I have the following print:

Shoenfeld residuals


1) Each covariate is non-significant correlated to time - thus, time independent supporting that the proportional hazard assumption is valid and the model is fit for use

2) The graphic visualizes a non-random pattern further supporting the PH assumption

The Kaplan-Meier analysis was subsequently conducted

Call: survfit(formula = Surv(time.to.recurrence, event.recurrence) ~ 1 + 
    factor(ki67), data = p, conf.type = "log")

   4 observations deleted due to missingness 
                  n events median 0.95LCL 0.95UCL
factor(p$ki67)=0 30      4     NA      NA      NA
factor(p$ki67)=1 86     12     NA     124      NA
factor(p$ki67)=2 44      9     NA     118      NA
factor(p$ki67)=3 12      5     56       7      NA

The Kaplan-Meier curves were then computed:

Kaplan-Meier Curves

1) How can this graph be interpreted? Intuitively, that curves should not be crossed if the PH assumption is valid?


1 Answer 1


The only crossings of survival curves in your data are between the first 2 groups (0-2% and 3-4% Ki67+), between which there is no significant difference in survival (HR 1.1, p-value 0.85). This is not of concern with respect to the proportional hazard assumption. If you took the 116 cases initially at risk in those 2 groups and randomly broke them into groups of the sizes of those two groups (30 and 86) I suspect that you would find similar "crossings" in Kaplan-Meier curves, as one or another group randomly has higher survival fractions at different times. That is, what you see is normally expected variability between essentially identical survival curves.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.