I have a Weibull proportional hazard model with fitted parameters $\lambda$ and $k$ and a time dependent covariate $x(t)$ with fitted coefficient $\beta$, such that the hazard function is given by,

$h(t) = \frac{k}{\lambda} (\frac{t}{\lambda})^{k-1} e^{\beta x(t)}$

Now, suppose I have an observation that has survived until time $s$. I am interested in predicting the probability of survival until some future time $t$ given knowledge of $x(t)$ for all future $t\geq s$ (but not past $t<s$).

Normally this would be calculated as

$Pr(T > t | t > s) = \frac{S(t)}{S(s)}$

However, I cannot know $S(t)$ and $S(s)$ because these depend on values of the hazard function before $s$, which I don't know because that depends on past values of $x(t)$.

But, it seems like I should still be able to get $Pr(T > t | t > s)$ without ever having to calculate the survival function, since I know all future values of the hazard function.

Edit: I believe I have a solution:

$\frac{S(t)}{S(s)} = \frac{e^{-\int_0^t h(\tau)d\tau}}{e^{-\int_0^s h(\tau)d\tau}} = e^{\int_0^s h(\tau)d\tau - \int_0^t h(\tau)d\tau} = e^{-\int_s^t h(\tau)d\tau}$

Which only depends on values of the hazard function after $s$. Does this work?


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.