Hi everyone,
Many thanks for all your replies to the message I posted on Friday. I have listed the responses I received.
Kindest regards to you all for 2018,
Kim
-----Original Message-----
From: A UK-based worldwide e-mail broadcast system mailing list [mailto:[log in to unmask]] On Behalf Of Kim Pearce
Sent: 05 January 2018 14:58
To: [log in to unmask]
Subject: Intention to Treat (ITT) : your views
Hi everyone,
Just a quick question. In an "Intention to Treat: a review" by Gupta (2011) ( http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3159210/ ) an intention to treat (ITT) analysis is said to include "all randomized patients in the groups to which they were randomly assigned, regardless of their adherence with the entry criteria, regardless of the treatment they actually received, and regardless of subsequent withdrawal from treatment or deviation from the protocol". In other words, "ITT analysis includes every subject who is randomized according to randomized treatment assignment. It ignores noncompliance, protocol deviations, withdrawal, and anything that happens after randomization." Bearing this in mind, I assume that this means that even if a patient dies during the study (that is, after randomisation), they are still included in an ITT analysis?
Thank you for your assistance on this matter.
Kind Regards,
Kim
______________________________
Kim,
What you state is essentially correct - for a pure ITT analysis, the data collected on a patient up to the time of death would be included in the analysis. HOW that data is used in terms of the primary endpoint is the question the study team has to consider. It's worth noting that most statisticians would do a modified ITT, adding the term "and received at least one dose of randomized medication."
Regards,
______________________________
Definitely - that can be the most important.
Reducto do absurdium - if I randomize lung cancer patients to receive or not receive a big hug 6 months after diagnosis the hugged group survive longer because a lot of lung cancer patients do not survive to diagnosis + 6 months.
The effect is usually a lot more subtle but can still make simple analyses highly misleading.
__________________________________
Hi Kim-
Yes indeed.
How I remember ITT - is "All randomised patients." Anything else is just ignored. Provided a patient has been randomised then he is in the ITT analysis
_____________________________________
Hi Kim,
Yes, if a patient died after randomisation they are still included in the ITT analysis. In ITT all eligible patients at randomisation are analysed according to the arms they are randomised.
There are two points that are important to mention and I came across:
-If a patient was eligible at randomisation but subsequently became ineligible during the trial, they are included in the ITT analysis. Example: patient is eligible and randomised based on the fact that the patient does not have metastases; during the trial, the patient develops metastases which would make the patient not eligible initially; the patient is still included in the ITT analysis because they were eligible initially and they just became ineligible afterwards.
-If a patient was considered at randomisation as eligible but subsequently it was found that in fact the patient was not eligible at baseline, then it is reasonable to remove the patient from the ITT analysis even though the patient was in fact randomised. Example: The patient is considered eligible according to a scan at baseline, so the patient is randomised; afterwards, a biopsy was received after randomisation and reveals that the patient was not in fact eligible at randomisation; so, in this situation, it would be reasonable to remove the patient from the ITT analysis because the patient does not belong, in the first place, to the population that is meant to be analysed.
I hope this helps
__________________________________
Yes, strict ITT means exactly what it says, so anyone randomised gets included, totally regardless of what happens to them subsequently.
Some people, including some eminent Statisticians, insist that this strict criterion should always apply. Others of us (including myself), feel that it is more sensible to undertake "modified ITT"
analyses ("mITT") which include only patients who actually get (at least one dose of, if medication) some of the randomised treatment and for whom there is at least some post-baseline data available. Academically-speaking, excluding patients who do not fulfill those mITT criteria theoretically upsets assumptions about "random allocation" of treatment, but such patients are never going to provide any actual 'information', since ALL of the post-baseline data attributed to them will necessarily be derived by some sort of imputation.
Kind Regards,
____________________________________
I would have thought that they would have to be, because could you ever be completely sure death was not due to the treatment?
_____________________________________
"Intention to treat" can mean different things to different people. As the idea is to preserve the comparability given by random allocation. I think that Gupta is correct. However, I have known pharma studies where the ITT samples are all those who receive at least one dose of the study medication. To me this is plain wrong.
It can be argued that any missing data will mean that the analysis is not ITT. I think this is correct, but as it is almost always the case that something is missing it makes ITT an ideal aspiration rather than a practicality.
In some studies, death is an adverse outcome, for example in cardiology studies, and we often use a composite outcome of death or cardiovascular event. In others, it is not, e.g. palliative care trials. Here there is much debate over whether we can treat death as an outcome and, if not, what we do about it. The participants are moribund anyway and the treatment is to alleviate suffering, not to prevent death. Can we treat the observation as missing? It is argued that there could not be an observation of the outcome, e.g. pain score, so it is not missing.
I think that if death or otherwise is not an expected treatment effect, we can treat it as missing data and use imputation, but my palliative care colleagues might disagree. If it is a plausible treatment effect, either occurrence or prevention, I would have some kind of composite outcome and count it as death, even if the unfortunate participant had been run over by a taxi.
___________________________________
in economics i have seen intended to treat analysis used to maintain a clear randomization design.
to drop observations means to make some assumptions on the process leading to this dropping.
one wishes to avoid this.
one refers to ITT when not all who were supposed to be treated did get treatment, or vice versa
when some who were not supposed to be treated did get the treatement.
importantly, the deviations from the desired treatment are not decided by the person doing the study. so when one
derives the ITT effect, then this incorporates the fact (that is not under the control of the designer)
that some will not be treated even if they are intended to.
this only works if each subject has some measured outcome. so if a patient dies then one has
to be able to assign an outcome, just as if that patient had been treated or not.
______________________________________
In the situation described, the inclusion or exclusion of deaths is not a matter of ITT but censoring or competing risk outcome. If death is not the outcome (or part of a composite outcome), you cannot include those observations in a simple analysis because you do not observe the outcome. Excluding those cases will likely create bias, unless the risk of death is the same across arms.
Death needs to be treated as a competing risk. Look at the literature on survivor average causal effect (SACE).
_________________________________________
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|