I generated a mixture cure model with Weibull distributed survival times (Weibull mixture cure model) . This is my baseline survival function conditional on the Bernoulli distributed covariate $X_i$ : $ S_0(t_i \vert x_i) = exp (-(e^{\gamma_0 + \gamma_1x_i}) t^{\alpha}$) where $e^{\gamma_0 + \gamma_1x_i}$ is the scale parameter and $\alpha $ is the shape parameter.

I generated the data with about 30% censoring, P(x=0) = 0.2 (probability of being cured when x = 0) and P(x=1)=0.3 (probability of being cured when x = 1), $\alpha = 1.5$, $\gamma_0 = 0.5 $ and $\gamma_1 = 0.0 $.


After generating the data, my aim is fit the true model (the Weibull mixture cure model which I did using nlm function in R to minimize the negative log likelihood), the Cox PH model and the Weibull PH model to the data. I recorded p-values under $\gamma_1 = 0.0 $ and did this 1000 times to find the empirical Type I error (to validate the Wald test statistic).

Null hypothesis : $\gamma_1 = 0 $ Alternative hypothesis : $ \gamma_1 = 1$

It worked fine for the true model and the estimates, the empirical Type I error is 0.040 which is within 95% confidence interval for 0.05 significance level (0.0365, 0.0635).

It did not work fine for the Cox PH model and I am assuming that the Cox PH assumptions were violated.

I was however expecting the Weibull PH model to somehow have empirical Type I error within the 95% significance level for me to compare the estimates with the true model. Under $\gamma_1 = 0.0$, I received 0.915 of empirical type I error for the Weibull PH model (way too high) . Under the $\gamma_1 = 1.0$, I received empirical power which is around 0.032. Is this expected? At the end of the day, I want to see if Weibull PH would still be reasonable model when it ignores cure proportions.

I tried to find literature that specifically compares mixture cure model with semiparametric/parametric PH models through simulations but could find none.

Has anyone done such simulations before to compare two survival models please enlighten me?

Thank you so much.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.