Say I am interested in looking at the 5 year occurrence of an serious adverse drug reaction including mortality occurring in a group of patients being treated for a disease.

I would imagine that because we are looking at the odds of whether someone is alive or not/experienced hospitalisation for the ADR or not at the time point of 5 years, then that would require logistic regression.

I imagine Cox may be more suited to a question asking what the survival is post-cancer, as it looks at the time to event.

However, when I look at papers modelling adverse drug reactions, I see a Cox proportional hazards model is frequently used to estimate hazard ratios.

Any guidance on what I am misunderstanding would be much appreciated.

  • 1
    $\begingroup$ Some units may fail before the end of the study (say 5 years) and with the logistic regression, there is no way you can account for this. $\endgroup$
    – utobi
    Nov 14, 2022 at 10:38

1 Answer 1


Survival analysis allows for individuals who are lost to follow up or haven't yet finished 5 years of observation to provide data. It also provides information about time-to-event that a simple logistic regression can't provide.

That said, such data are sometimes evaluated with a set of logistic regressions over a set of defined time intervals (e.g., each year out to 5 years). That's discrete-time survival analysis, which does provide information about time-to-event (at a coarse time scale) and incorporates some information from those lost to follow up or not yet ruching 5 years.

  • $\begingroup$ Thank you, this makes sense. I guess it will depend on the outcome then? If the outcome is looking at a dichotomous risk of either developing the event or not, LR could be used despite the limitation of not accounting for censoring. Otherwise, the outcome of the study using a Cox model would be more along the lines of event-free survival within 5 years or each year out to 5 years if using a discrete-time survival analysis. Let me know if I understood this correctly. $\endgroup$
    – Savannah
    Nov 14, 2022 at 17:18
  • $\begingroup$ @Savannah there are many dangers in not accounting for censoring. This review by Leung et al., Annu. Rev. Public Health 18:83–104 (1997) is still a good reference. Other than that, you seem to understand correctly. $\endgroup$
    – EdM
    Nov 14, 2022 at 17:38

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.