Retrospective intention-to-treat analysis? what does it mean?

#1
Hi everyone:),
Intention-to-treat analysis is often used in randomized controlled trials. I was wondering how its principle can be applied in some retrospective studies? For instance:
Sapisochin G, et al. Single HCC in Cirrhotic Patients: Liver Resection or Liver Transplantation? Long-term Outcome According to an Intention-to-treat Basis. Ann Surg Oncol (2013) 20:1194–1202

Does it always have to be used in studies comparing different treatments or can it also be applied in retrospective case-control studies?

Thank you for the attention and reply!
 

hlsmith

Omega Contributor
#2
You can use it whenever. The principle is that you just keep the subject in the treatment group they were assigned to when conducting analyses, even if they don't adhere. The idea is that it would mirror expected outcomes seen in the real world.


Applying it to retrospective data, would mean that you are certain which treatment they were assigned and certain which treatment they took. These two things may not always be readily available in retrospective data or there could be questions about the validity of documentation and the truth.
 
#3
Excuse my ignorance, i am not sure how you can do intention to treat analysis in retrospective study design!? can someone link an article discussing this or elaborate more on it? thanks in advance.
 

ondansetron

TS Contributor
#4
Excuse my ignorance, i am not sure how you can do intention to treat analysis in retrospective study design!? can someone link an article discussing this or elaborate more on it? thanks in advance.
You can look at the first clinic note for when a treatment was supposed to be initiated, and you can see when that was and what the treatment was. You can follow clinic notes to see if the patient had been compliant through the period of study. (For completeness of analysis, although ITT would not need the compliance information, technically).

This assumes that patient assignment to a treatment is accurately documented at the initial prescription.

Overall, it mirrors a retrospective cohort study in someways (at least, I think it does-- I believe you need to sample on the treatment assignment).
 
#5
I am trying to understand how intention-to-treat can be relevant to retrospective studies; Randomization is essential to render meaning to an ITT analysis as it balances deviations in both arms. I am not sure how this would apply to the retrospective design.
I find that ITT in retrospective design can be very misleading as the circumstances of deviation/dropout/change of treatment cannot be balanced in retrospective design; am I wrong in my understanding?
 

ondansetron

TS Contributor
#6
I am trying to understand how intention-to-treat can be relevant to retrospective studies; Randomization is essential to render meaning to an ITT analysis as it balances deviations in both arms. I am not sure how this would apply to the retrospective design.
I find that ITT in retrospective design can be very misleading as the circumstances of deviation/dropout/change of treatment cannot be balanced in retrospective design; am I wrong in my understanding?
I think these are valid points. Of course, randomization is the better approach, but attempting to account for these "imbalances" can be done but will still leave the study open to criticism.

I'm just stating how it could be done. I think it is similar to a retrospective cohort study, in a way.
 
#7
I think these are valid points. Of course, randomization is the better approach, but attempting to account for these "imbalances" can be done but will still leave the study open to criticism.

I'm just stating how it could be done. I think it is similar to a retrospective cohort study, in a way.
Thank you. I have seen this use of ITT in quite few papers, and no one seems to question it.
 
#8
Thank you. I have seen this use of ITT in quite few papers, and no one seems to question it.
Is it ITT in a retrospective study? ITT in a prospective study or RCT is a lot better because it mimics the real world situation where a physician may prescribe, but patients may end up as noncompliant in one way or another. I don't think ITT is bad.

The BS to watch for is ""modified" ITT or something like that. There is ITT or there is per protocol. ITT is analysis in the group you were assigned. Per protocol is analysis if you followed the protocol for treatment completely. People have tried to make other "modified" versions, which may be reasonable, but they're not ITT and they're not PP. It should be named something else if anything, but often it's done to make the results appear a different way when you see people define Per protocol as you are assigned to A and you took at least 2 pills of A.

In medicine, many people don't know a whole lot about statistics and view it as a peripheral thing like grammar. They blindly follow publications assuming they are right because they are published.
 

hlsmith

Omega Contributor
#9
Yes, you can use ITT, per protocol, as treated, or local average treatment effects. You just apply them, though in retrospective data it is important to think about possible limitations. Is it perfect, probably not. Does it provide some information, yes. Thus sensitivity analyses may be warranted.