0
$\begingroup$

I would be very grateful for a gentle sanity check on adjusting survival times in my comparison groups. The scenario: I am using Cox regression to calculate HR risks over some covariates of interest.

My population is observed for six months, starting from the index event when a patient makes a medical complaint. There are a few competing events: death, leaving the study, lost-to-following-up, and remission. The population is split into two groups:

  • exposed (cases) when a patient has a record of drug exposure within 60 days from the Index date.
  • not exposed (controls) those patients without a record of drug exposure during the observation period.

All the patients in my exposed (cases) group have an immortal time bias because we know they can't have any competing risk (death, remission, etc.) until they've first had exposure within the initial 60 days after Index.

As far as I understand, there are several approaches to adjusting the survival times of the exposed group. However, my question is, do I need to do anything to the survival-times of the not-exposed (control) group? Thanks.

Update 1.

To answer my question, I will use a "Landmark" approach to treat the immortal time bias. This involves adjusting survival times in both groups by the same about of time.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

The problem is that your "exposed" cases were in the "non-exposed" group until the time of exposure. This suggests that you would be better off with a multi-state model that includes the transition from the "non-exposed" to the "exposed" state. That would also help evaluate systematic differences between those who enter those two states, which would tend to confound any simple comparisons between those who happened to end up in those two groups by 60 days. Then you can evaluate the transitions from each of the "non-exposed" and "exposed" states to the other states of interest.

The R vignette on competing risks shows how to set up such models and use them to estimate the probability of state occupancy as a function of time and covariates.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thank you for that information and the link. I haven't really gone into too much detail about competing risks, which I have factored into my analysis as per Terry Therneau's box on extending Cox regression. However, this vignette is 2022, I haven't seen it before. Thanks. $\endgroup$
    – Anthony Nash
    Dec 15, 2022 at 20:39

Your Answer

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

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