Intention to treat analysis:

Intention to treat analysis (ITT), aka as-randomized analysis, means that the patients were analyzed in the groups they were assigned to. This is independent of whether they got the treatment or not. It is the most complete analysis of the groups that can be done.

It might be counterintuitive to count as a surgical mortality, if someone dies waiting for the surgery and the surgeon never even laid eyes on them. However, the waiting for the OR to be ready is a part of real-world surgery.

Additionally, a device might be efficacious (efficacy: it works) but might be uncomfortable to use or people might forget to put it on. If that is the case, its real-world impact might not be as good (effectiveness). i.e. a life-vest (external defibrillator) can save a patient from lethal arrhythmias, but only while it is worn. It is efficacious, but it might not be effective.

Per protocol (PP) / as treated: tend to focus more on efficacy (it works) compared to effectiveness. Therefore, it over-estimates how good treatments are. PP analysis increases our chances of a false positive study (type I error / false alarm) - seeing a difference in the study that is not true in real life. Per protocol causes cross over of patients, therefore changing the prognostic balance that we worked so hard to achieve with the randomization process.

PP or as treated might be preferred when looking at side effects - you can only have side effects from surgery if you had the surgery... (although there might be complications from waiting and not getting the surgery too)

In summary:

Intention to treat:

• Maintains the prognostic balance in the groups. (Otherwise, why randomize?!)

• It is a more conservative estimate (tends to show less difference than per protocol)

• Takes into account the whole approach - i.e. waiting for the OR to be ready.

• It is closer to the real life and takes effectiveness into account

  
  

Per protocol:

• Only counts patient outcomes if they followed all the protocol they were assigned to.

• It will not take into account other parts of the approach (see above)

• Tends to overestimate effects

• There might be a reason a patient cross-over treatment, therefore changing the prognostic balance of the groups.

  
  

As treated:

• Similar to per protocol but counts them if they got at least a dose of the treatment.

• Otherwise, similar to per-protocol with the same weaknesses.

The bottom line:

Which one is better? If only given one option, I would always go for ITT analysis. If given both ITT and PP, I usually compare them. If they show similar results, it strengthens my trust in the intervention. If the results are very different, it makes me suspicious, and I trust the ITT results more.

References:

-Livingston, E. H., & Lewis, R. J. (2020). JAMA Guide to statistics and Methods. McGraw-Hill.

-Guyatt, G, et al. (2015). JAMA Users guide to the medical literature. McGraw-Hill.