1
$\begingroup$

Can we include post index event in logistic regression if teh event is before the end point?

I have to do a regression to understand which patients stay on medication. I have taken the index date as the date they start with Medicine x. All characteristics of patients before starting the medication x ( baseline char) are used to predict weather they stay on tretment till 1 year. I am interested in adding a variable to see the effect of medicine at 6 months to this regression. Can I do that or is it not statistically sound because essentially this variable is post index ?

Stay on treatment for 1 year ( 1 or0) ~ patient characteristics before starting treatment + effect of of treatment at 6 months

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

If you know the status of each individual at 6 months, you could treat this as a discrete-time survival model, where the "event" is stopping use of the medication. You would effectively have two binomial outcome models, one for the 0- to 6-month interval and the second for the 6- to 12-month interval. The second model could incorporate information about the status at 6 months, as it would only be applied to those who are still "at risk" of stopping the medication during those last 6 months. You can think about that as setting up a second "index" time at 6 months for that second model, with most patient characteristics the same as for the first model but adding the extra information available at 6 months for the second model.

Improve this answer
$\endgroup$
3
  • $\begingroup$ I am sorry, I am not sure if I understand (not a stats person). Are you suggesting two separate models? or can both these binomial models be incorporated into one? $\endgroup$
    – Riya Arora
    Commented Sep 22 at 11:53
  • $\begingroup$ @RiyaArora this could be done either way if there are only two observation times for "stay on medication." In general, a discrete-time survival model is a binomial regression that includes data for each time interval along with some model of changes over time, possibly allowing for changes over time in how predictors are associated with outcome. With only two time intervals in your case (0 to 6 months; 6 to 12 months), that's equivalent to two models, one for the 0 to 6 month interval and the second model (restricted to those still on medication at 6 months) for the 6 to 12 month interval. $\endgroup$
    – EdM
    Commented Sep 22 at 16:19
  • $\begingroup$ See pubmed.ncbi.nlm.nih.gov/35883032 $\endgroup$
    – kjetil b halvorsen ♦
    Commented Oct 6 at 20:31

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.