Could you explain how using a grace period to assign exposure group as described in Hernan's target trial article does not introduce immortal time? (e.g. how in the statin treatment article, patients were defined as statin users if they started using a statin within 6 months)
I understand why the grace period is needed, given that it may be hard to find adequate numbers of incident users of the exposure at time 0, but am confused as to why it doesn't introduce immortal time.
Thanks!
Hi Jean,
We will answer your question after class tmrw. It's a good question, however, to answer it may involve giving away at least part of an answer to another question in the problem set.
Thank you for your patience!
JMC
Hi Jean:
Thank you for posting this question and for participating in class yesterday. Since we did not have time to discuss your question in class, we want to answer your question here and remind you that the bonus question has some examples illustrating the analysis using clones or copies of the study participants. The two examples that we added to the answer key shows the importance of the grace period.
The strategy of a grace period that Hernan proposes goes together with generating clones of each eligible participant. In the study of statin use, there were two treatment strategies: initiation of statins or no initiation of statins. Each participant had two copies in the dataset, the authors assigned a treatment strategy at baseline to each copy or replicate. For a person who started statins within 6 months, the copy that was assigned to the statin arm will be no censored and her follow-up time is counted in the person-time of the statin group. The other copy will be censored when she initiates statins and this shorter time will be assigned to the no statin use. By creating these clones all eligible participants are contributing part of their time to each treatment strategy, then preventing immortal time bias. You need a grace period to know when to do the censoring in the clone that does not follow the assigned treatment strategy.
