Intention to Treat in Propensity Matched Cohort Studies?

Hi all,

I am doing an internal study to compare two surgical treatments (let's call them A & B). These are transplants, so there is a definite failure possibility (ie, the artery or vein to the transplant may clot, and the transplant thus "fails"). This is different from the outcome being measured and compared. Obviously, if the transplant fails, we expect the measured variable to perform poorly.

I used propensity score matching to compare outcomes from both groups. However, in group A there is 1 transplant failure. N is designed based on a priori power analysis. There are also no more patients in group A that I could try to find another patient via score matching for the comparison.

My question is, is it appropriate to keep the failure patient in the study? I generally think of ITT as it applies in RCTs, but would this be an example of ITT in a propensity matched study?

I do have an MS Statistician involved in this project, but wanted to reach out for more opinions because I have not encountered this situation before.

Thank you!
Last edited:


Less is more. Stay pure. Stay poor.
Your context isnt quite clear to me. Two treatments in a sample of transplant patients, correct? A and B should suffice for treatments, but just to make sure i dont confuse things can you provide a short description of them (A is an immunosuppressor and B is another type of suppressor). What is the outcome? So clot is a related competing event? Also what model do you plan to us at the end (survival analysis)?
Group A: Received one type of transplant lymphoid tissue
Group B: Received a different type of transplanted lymphoid tissue

Transplants were done to the same site, for the same indications, with the measured outcomes being volumetric differences in the extremity and standard patient reported outcomes.

The transplanted tissue requires an arterial and venous anastomosis at the recipient site. These may clot, and it is generally assumed that the transplanted tissue becomes nonfunctional at this point.

Thank you


Less is more. Stay pure. Stay poor.
Volumetric differences before after, between limbs? How does clot play into this? Since ITT deals with treatments and clot seems post treatment.
Correct re non-randomization. For more clarification:

For the treatment of extremity lymphedema, lymph nodes from several different sites of the body may be transplanted into the affected extremity in order to help alleviate it. The lymph nodes are harvested from one site of the body with the artery and vein that feed them. They are then transplanted to the recipient (affected) extremity. This involves connecting the lymph node artery and vein to an artery and vein at the recipient site. With any transplant, a thrombus may form in the re-connected artery and vein. Excess lymphatic fluid enters the lymph node and drains to the vein; this is how it helps alleviate the edema.

If the artery or vein clot, the lymph node may not survive; and definitely if the vein clots, the primary drainage pathway is blocked. In this way, we expect that the transplanted lymph nodes which suffer from a clot in the artery or vein (or both) will function less well than those that do not... because they are essentially unable to perform their function.

I'm measuring volumetric differences of the same limb, before and after treatment, so that I can determine which site of harvested lymph nodes (A or B) have worked best in this population. All of these are previously completed cases, and this is a retrospective, propensity matched cohort study.

So, lymph node transfer (A or B) is the treatment. Clot is a post-treatment effect that we assume influence outcomes. And measured outcomes are volumetric differences pre- and post- treatment of the affected extremity.

My question is, then: Is it appropriate to say this is an ITT analysis, since one case suffered a complication, which would affect the outcome? Or, should this case not be included, since it is not the same as the others, which had a successful transfer (analogous to a per-protocol analysis)?
Last edited:


Less is more. Stay pure. Stay poor.
I don't think your logic with ITT is relevant. ITT would be you assigned a patient to treatment A, but they got treatment B - which is impossible here.

What is your sample size? I apologize for not rereading your prior post - but if there is only one patient with a clot, I would just post hoc exclude them and report this in the results. The reader can use their own perspective on how the exclusion may impact the results. Why are you matching, which can throw away cases. Given you have an able sample size you can control for confounders in the model or (in my opinion usually better) use inverse propensity treatment weights.