2
$\begingroup$

I'm reading an article where the authors use an extended Cox-Weibull hazard model with time-varying covariates to model survival time of an individual. This individual is allowed to start at week $m$, and covariates are available at weekly granularity.

Assuming the hazard is represented as $\lambda\left(t ; Z(t)\right)=\lambda_0(t) e^{\beta^\top Z(t)}$, where the baseline is Weibull, they provide an expression for the probability an individual survives at least $m^\prime-m$ weeks, given they started at week $m$, through $$\tag{1} S(m^\prime-m|m)=\exp(-\lambda B(m,m^\prime))$$ where $$ B(m,m^\prime)=\sum_{i=m+1}^{m^\prime}[(i-m)^c-(i-m-1)^c]\exp(\beta^\top Z(i)). $$ I'm having trouble justifying expression (1), however, and feel it should follow from the relation between survival and hazard functions, namely $$ S(m^\prime -m|m)=\exp\left(-\int_0^{m^\prime -m} \exp (\beta^\top Z(u)) \lambda_0(u)~\mathrm d u\right). $$ Using the fact the covariates $Z(t)$ are a step function, as they change only at the start of each week, I feel as if I can calculate what I'm after through something like $$ \exp\left(-\left[\exp (\beta^\top Z(0)) \int_0^{1} \lambda_0(u) ~\mathrm d u+\ldots+\exp(\beta^\top Z(m^\prime-m-1)) \int_{m^\prime-m-1}^{m^\prime-m} \lambda_0(u) ~\mathrm d u\right]\right), $$ but it doesn't seem like this will correspond to the expression in the paper.

$\endgroup$
2
  • $\begingroup$ Can you be explicit with how the Weibull is parameterized please? $\endgroup$ Sep 29, 2022 at 2:06
  • $\begingroup$ I should have specified that, but for this Weibull parameterization, the hazard is $h(x|k,b)=bkx^{k-1}$, with a CDF of $1-e^{-bx^k}$ $\endgroup$
    – statian
    Sep 29, 2022 at 14:18

1 Answer 1

3
$\begingroup$

The authors may be expressing that if an individual starts at week $m$ then they have survived up until week $m$, that is, they have the survival function $S(t | m) = P(T \ge t | T \ge m)$. This conditional survival function can be simplified to look like: $S(t | m) = \frac{S(t)} {S(m)}$. Their expression looks like this ratio.

$\endgroup$
3
  • $\begingroup$ I thought about this as well. I might be naive, but this felt wrong to me because it's not that the individual survived until week $m$, but rather they were born at that time. Are they equivalent? $\endgroup$
    – statian
    Sep 29, 2022 at 15:31
  • 1
    $\begingroup$ Typically the time variable in survival analysis is not an absolute time, but a measure of duration. So it's strange to say they were born at time $m$, normally you would say they were born at time 0, and record from there. However, it sounds like subjects could enter study at week $m$, which would imply survival between time 0 and time m. This is called late entry in the literature. $\endgroup$ Sep 29, 2022 at 18:10
  • $\begingroup$ I see, this is very helpful. In this case, I worded it as born because these individuals weren't yet added to the population until time $m$, so it feels weird to condition upon survival time from 0 to $m$ in this case $\endgroup$
    – statian
    Sep 29, 2022 at 19:09

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.