I'm trying to investigate several independents on a dataset with time-to-event data for a treatment. I'm measuring time-to-treatment effect (where treatment effect is a binary parameter).

Treatment effect always occurs within the first two month. But for some subjects, the treatment is ineffective altogether. I do have a long follow-up for these subjects. These subjects make up less than 10% of the data.

I want to model both time-to-treatment effect, and amount of subjects never reaching treatment effect.

I can reasonably state that the hazards are proportional, and if a dependent will have a lower hazard of reaching time-to-event, it will have a higher fraction of subjects never reaching the event.

Can I use a Cox proportional hazards model to model this data?

If I use a Cox proportional hazards model, which time do I use for subjects never reaching the event?

Can I just censor patients never reaching treatment effect at 60 days, since they're no longer at risk for experiencing treatment effect, or do I include them for the full length of the follow-up, even though they're no longer at risk?

  • $\begingroup$ Do you have multiple visits for (at least some of) these patients such that, at each visit, you determine whether or not the treatment is effective? If yes, do you know the actual dates of these visits? $\endgroup$ Commented Feb 7, 2019 at 14:59
  • 1
    $\begingroup$ I have multiple visits, but for most (if not all) of these patients, treatment effect occurs while they're still admitted to the hospital, and it's evaluated on a short interval. I have the actual dates. For the patients never reaching treatment effect, I have data on the last visit for a separate analysis on patient survival. $\endgroup$
    – Erik A
    Commented Feb 7, 2019 at 15:06
  • $\begingroup$ If treatment effect occurs while patients are still admitted to the hospital, then why not use the hospital discharge date as the cut-off for your assessment of treatment effectiveness? Will that mean there is a single date of assessing treatment effectiveness for hospitalized patients or will there still be multiple assessment dates even while the patient is in the hospital? $\endgroup$ Commented Feb 7, 2019 at 15:12
  • $\begingroup$ I guess this comes back to how you define "treatment effectiveness" - best to have a clear definition of what it means for the treatment to be effective (which would include the time scale for assessing effectiveness). $\endgroup$ Commented Feb 7, 2019 at 15:14
  • $\begingroup$ I don't think I can censor them at discharge, since I know they won't get treatment effectiveness after being discharged, and if I censor them, I will model them as having equal risk of treatment effectiveness as others at that timepoint (while in fact I know they will never reach it, and others are still at risk at that timepoint since admission duration is not fixed). For my analyses, I assume treatment data is continuously assessed, and the timepoint of treatment effectiveness is an accurate value. $\endgroup$
    – Erik A
    Commented Feb 7, 2019 at 15:33


Your Answer

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

Browse other questions tagged or ask your own question.