Your prof may have been too smart to figure out how to explain a concept that doesn't make any sense. As above: Simple: If you assign the group to take a pill and you know that none of them took it and you have their data from their return visit, ITT says that that data gives you information about the efficacy of the pill. That's what ITT says you must do.
👉Y was it in per protocol, just 75% instead 100%, and the experimental group get 100%, wen it was the experimental group affected and 1 fall out from experimental group happened? Pls explain
4:26 In the as-treated protocol, will not the percentage of good outcomes in the control group become 60%? Since I am moving the good outcome from the experimental group into the control group, thereby making the good outcome into a bad one?
Thank you for making a video to illustrate the concept. I still have a question. One quit from a study, according to the intention to treat principle, we keep the subject at the group that he was assigned. But when we conduct the statistical analysis, how do we calculate the time the subject spend in the study? Person-time or whole time?
Great video! But why an intention-to-treat group is better than a per-protocol group in minimizing biases of protocol-violating subjects is asserted but not explained. On its face, the opposite would seem more likely. If half of the controls used the treatment and half of the treatment group didn't use the treatment, any efficacy of the treatment should be undetectable in the intention-to-treat groups but could be uncovered in the per-protocol group. No?
Excellent explanation! Could you do a video about "concealment" of the randomisation sequence? Many people mistake the concepts of "concealing" and "blinding". Thank you!
I am still confused how if we don't know the "satisfaction" (the outcome in this video) of the dropped out subjects? In this video, you explain like the researchers know about the dropped out subjects outcome. For example, 100 subjects in control group vs 100 subjects in experimental group are followed for survival after 5 years. If there are 10 dropped out subjects in experimental group and (for the sake of simplicity) the rest of them is alive, how should we calculate the survival rate of experimental group? PP analysis: 90/90 = 100% ITT analysis: 90/100? And how about if I study their mortality rate? PP analysis: 0/90 = 0% ITT analysis: 10/100 = 10% (we count the dropped out subjects as "failure") or 0/100 (because we only include the dropped out as they were randomized, without any outcome)? Thank you
You've asked an excellent question. ITT helps with non-compliance and directs us as what to do with the data that is generated with non-compliance. My examples illustrate how, depending on their data, ITT minimizes (but does not eliminate) the impact of non-compliance. Missing data is another problem altogether, which is a disaster because it really undermines the validity of the research. For this reason we often put (arbitrary) limits on how much data can go missing before we forfeit the results of the study. I hope this helps!
LMAO I needed to clarify this concept and that was great but Im giving you a million likes because of the joke at the beginning XD XD XD now I cant think of ITT the same way ever :D
Is Intention to Treat becoming a "buzz word" in data science? A team in my analytics department at a large insurance company claimed they'd conducted an "ITT" study. After reading their paper, it was clear they had run a case-control analysis on observed data, not an ITT RCT. Maybe "Intent to Treat" sounds fancier?
Would't it make more sense to take the mean value of per protocol, as treated and ITT? That way you account for all possibilities and in the long run probably get more correct interpretation of data. Or at least my grasp says you do...
A really excellent question that I hadn't thought of before! We see a similar issue arise with meta-analysis in that we don't improve the quality of an estimate by increasing the number of inaccurate data going into the calculation. I think the old adage "Garbage in, garbage out'" probably applies here too. It points to just how terrible non-compliance in a study is on the final result... so frustrating to have to rely on ITT which (IMHO) is the worst method except for all the others (to misquote Churchill).
@@sketchyebm3043 Thanks for that response. I agree with the 'garbage in, garbage out' adage, but in this case we don't really know which is and which is not garbage. We don't know whether ITT, per protocol or as treated most accurately reflects the (majority of) drop outs. As I see it, ITT is a gamble which'll sometimes be right and other times wrong. To me it just seems more conservative and in the long term more failsafe to do a mean value. I might do both myself. One mean and one ITT. Let the reader choose. (Can I have an 'amen' for passing the buck?).
Great channel and video, I have recommended it to my classmates as it helps a lot when reading and understanding RCTs. May I please use a screenshot from 3:50 and/or 4:50 in my thesis ? References will naturally be to this video and its credits :)
@@sketchyebm3043 ahh i see! Thank you for responding ! If it's not stated should it be obvious in their analysis methods ? Or is it generally just stated
@@Mercyforthewicked Often it's just stated (or they'll say they used another technique) but sometimes not and then I go looking to see if they talk about what they did with patients who were non-compliant or didn't conform to the research methodology. Very tricky!
@@Mercyforthewicked I think it's somewhat related, but its own problem. Missing data undermines the validity of the results much like non-adherence to study protocol. There's no way to 'fix' those problems, but rather to adjust our expectations of our conclusions.
I have explained what is wrong with this in my intention-to-publish paper. I know, if I didn't write it, what am I talking about? If you can believe that you learn about the experimental even though you know that one of the people dropped out, then you can appreciate how substantial my CV is. The "best" is what he says. Never mind that it doesn't make any sense. If you knew that one of the people would not adhere for religious, political, psychological reasons, you would no include them in the trial. If you include them now, you are introducing bias, that is you are saying that what you wanted to do was more important than what happened.
Thank you for your comment Dr. Feinman. I'm sure we can agree that research participants who do not adhere to protocol present a threat to the validity of the results no matter what, mathematically, we do with their results. In the case of predictable non-adherence, I agree that having this as an exclusion criteria PRIOR to randomization is a better idea than letting them enter the study only to fail compliance. Having said that, the generalizability of the results will suffer with each additional group we exclude from the study. I have a few issues with your publication on ITT. (nutritionandmetabolism.biomedcentral.com/articles/10.1186/1743-7075-6-1) Specifically: a) A physician concluding that Diet A is equal to Diet B based on a failed superiority trial is a common error we see with clinicians, which is misinterpreting 'no significant difference found' to mean 'there is no difference'. The former being a conclusion from a failed superiority trial, the latter from a successful equivalence/non-inferiority trial. This common clinician misunderstanding is important in of itself, and would have been better off left as a separate teaching point, as opposed to muddling the application of ITT. b) The example of the CABS trial - the problem with just using as-treated post-randomization is that the reason for patients' non-compliance may be independently related to their outcome. (Perhaps the high mortality in the group who was assigned to surgery but got medicine was because these were the sickest patients in this arm of the study and someone felt they were not a good surgical candidate. Using as-treated or per-protocol would then selectively remove the sickest patients from this arm of the trial, resulting in worse bias than what we see with ITT.) c) Your comment that it is "reasonable" that scientific knowledge ignore "data that was not in the experiment" is itself based on opinion and not fact. Although it is easy mathematically to just use per-protocol or as-treated analysis, these methods likely enhance the bias from non-compliance. d) I agree that concepts or ideas are not better because they are newer. I would add that concepts or ideas are not better because they are established. As a friend says: "Be skeptical of anything you learn." Like the much quoted "Democracy is the worst form of government, except for all the others", it is my thought that ITT is the worst way to deal with non-adherence, except for all the others. Which is not to say that ITT fixes non-adherence. Glad you are drawing your own conclusions! Anthony
Can you please make more videos on clinical trials concepts? Your explanation is super simple and comprehensive.
I rarely comment on youtube, but I really felt obligated to comment on this one. Thank you for this great video.
your videos have saved me during critical appraisal. Our prof never did a great job of explaining this stuff thank-you.
Your prof may have been too smart to figure out how to explain a concept that doesn't make any sense. As above: Simple: If you assign the group to take a pill and you know that none of them took it and you have their data from their return visit, ITT says that that data gives you information about the efficacy of the pill. That's what ITT says you must do.
Thanks! A wonderful video. Even though I knew how to calculate ITT, I was struggling to understand its relevance. Much clearer now. :)
Amazing, much better than in our textbooks!
very good and straightforward
This was SIMPLY the best explanation ever. Thank you.
👉Y was it in per protocol, just 75% instead 100%, and the experimental group get 100%, wen it was the experimental group affected and 1 fall out from experimental group happened? Pls explain
Thank you so much! Your video explained the exact concepts I was having a hard time comprehending.
4:26 In the as-treated protocol, will not the percentage of good outcomes in the control group become 60%? Since I am moving the good outcome from the experimental group into the control group, thereby making the good outcome into a bad one?
Thank you for making a video to illustrate the concept. I still have a question. One quit from a study, according to the intention to treat principle, we keep the subject at the group that he was assigned. But when we conduct the statistical analysis, how do we calculate the time the subject spend in the study? Person-time or whole time?
this video is crystal clear! it would be better it explains why ITT minimizes bias
So no matter what deviations we have, randomisation purpose is preserved. Is that the summary?
That's how I understand it - it minimizes the unavoidable impact of non-compliance by preserving randomization.
Amazing explanation, many thanks :)
Great video! But why an intention-to-treat group is better than a per-protocol group in minimizing biases of protocol-violating subjects is asserted but not explained. On its face, the opposite would seem more likely.
If half of the controls used the treatment and half of the treatment group didn't use the treatment, any efficacy of the treatment should be undetectable in the intention-to-treat groups but could be uncovered in the per-protocol group. No?
Great! But what does "good outcome" mean?
Nirnaya Bhatta, it’s the desired effects of an intervention. E.g. recovery from an illness, relief of pain, ...
Excellent explanation! Could you do a video about "concealment" of the randomisation sequence? Many people mistake the concepts of "concealing" and "blinding". Thank you!
I am still confused how if we don't know the "satisfaction" (the outcome in this video) of the dropped out subjects? In this video, you explain like the researchers know about the dropped out subjects outcome. For example, 100 subjects in control group vs 100 subjects in experimental group are followed for survival after 5 years. If there are 10 dropped out subjects in experimental group and (for the sake of simplicity) the rest of them is alive, how should we calculate the survival rate of experimental group?
PP analysis: 90/90 = 100%
ITT analysis: 90/100?
And how about if I study their mortality rate?
PP analysis: 0/90 = 0%
ITT analysis: 10/100 = 10% (we count the dropped out subjects as "failure") or 0/100 (because we only include the dropped out as they were randomized, without any outcome)?
Thank you
You've asked an excellent question. ITT helps with non-compliance and directs us as what to do with the data that is generated with non-compliance. My examples illustrate how, depending on their data, ITT minimizes (but does not eliminate) the impact of non-compliance. Missing data is another problem altogether, which is a disaster because it really undermines the validity of the research. For this reason we often put (arbitrary) limits on how much data can go missing before we forfeit the results of the study. I hope this helps!
you earned a sub. I fkn suck at research and you just explained this so simple and well. Thank you so much.
REALLY helped with my paper, wonderfully explained and well illustrated! thank you!
Extremely helpful. thankyou so much!
This is the best explanation. Thank you!
LMAO I needed to clarify this concept and that was great but Im giving you a million likes because of the joke at the beginning XD XD XD now I cant think of ITT the same way ever :D
Thank you for the fantastic and concise explanation :D
thank you so much, way better than Uworld explanation. They should hire you mate!
Best ever !! Thank you
would this be the same as concept to treat?
Thank you for the video. Very intuitive and easy to understand. Awsome.
Is Intention to Treat becoming a "buzz word" in data science?
A team in my analytics department at a large insurance company claimed they'd conducted an "ITT" study. After reading their paper, it was clear they had run a case-control analysis on observed data, not an ITT RCT.
Maybe "Intent to Treat" sounds fancier?
Sorry to hear - hopefully it's not becoming a buzz word! In medical research it does imply sticking to a certain methodology.
Excellent explanation. Thanks!
You try to give the video more brightness it will be great if you do
You’re the best. God bless you
Very good and clear explanation
Would't it make more sense to take the mean value of per protocol, as treated and ITT? That way you account for all possibilities and in the long run probably get more correct interpretation of data. Or at least my grasp says you do...
A really excellent question that I hadn't thought of before! We see a similar issue arise with meta-analysis in that we don't improve the quality of an estimate by increasing the number of inaccurate data going into the calculation. I think the old adage "Garbage in, garbage out'" probably applies here too. It points to just how terrible non-compliance in a study is on the final result... so frustrating to have to rely on ITT which (IMHO) is the worst method except for all the others (to misquote Churchill).
@@sketchyebm3043 Thanks for that response. I agree with the 'garbage in, garbage out' adage, but in this case we don't really know which is and which is not garbage. We don't know whether ITT, per protocol or as treated most accurately reflects the (majority of) drop outs. As I see it, ITT is a gamble which'll sometimes be right and other times wrong. To me it just seems more conservative and in the long term more failsafe to do a mean value. I might do both myself. One mean and one ITT. Let the reader choose. (Can I have an 'amen' for passing the buck?).
@@jan-kjetiljess503 Amen!
Perfectly explained. thank you :)
This was great! Thank you!
Such a good explanation. Thank you!
Thank you
eccezionale!! forza italia!!
OMG Thank you for an explanation that makes sense!!!
excelletn video, very helpfull. Thanks!
Got it. Thank you
Amazing
Great channel and video, I have recommended it to my classmates as it helps a lot when reading and understanding RCTs. May I please use a screenshot from 3:50 and/or 4:50 in my thesis ? References will naturally be to this video and its credits :)
Thanks for the message - yes, please go ahead and use it as you see fit as long as proper credit is given. :)
thank you so much man...
Great video
Very well explained thank you
clear and helps a lot
great!! helped me a lot !
how do you do that?
I use VideoScribe for the whiteboard and iMovie for audio and editing. Takes some time to learn, but pretty user-friendly!
but how do i now they used the ITT method??
They should specify in their "methods" section that they are using intention to treat analysis. It's on my list of things I go looking for!
@@sketchyebm3043 ahh i see! Thank you for responding !
If it's not stated should it be obvious in their analysis methods ? Or is it generally just stated
@@Mercyforthewicked Often it's just stated (or they'll say they used another technique) but sometimes not and then I go looking to see if they talk about what they did with patients who were non-compliant or didn't conform to the research methodology. Very tricky!
@@sketchyebm3043 Is missing outome data a red flag for ITT analyses, or is it unrelated to the ITT
@@Mercyforthewicked I think it's somewhat related, but its own problem. Missing data undermines the validity of the results much like non-adherence to study protocol. There's no way to 'fix' those problems, but rather to adjust our expectations of our conclusions.
Excellent
Amazing!
I finally get it!!! Thank you so much!!
dident halp :(
Very well-explained video, thanks a lot :-)
fab!!!!!!!!!!!
excellent
wow sooo good finally I understood thank you soooo much
BASED BASED BASED REEEEEEE
That Canadian accent thoughhh
I have explained what is wrong with this in my intention-to-publish paper. I know, if I didn't write it, what am I talking about? If you can believe that you learn about the experimental even though you know that one of the people dropped out, then you can appreciate how substantial my CV is.
The "best" is what he says. Never mind that it doesn't make any sense. If you knew that one of the people would not adhere for religious, political, psychological reasons, you would no include them in the trial. If you include them now, you are introducing bias, that is you are saying that what you wanted to do was more important than what happened.
Thank you for your comment Dr. Feinman. I'm sure we can agree that research participants who do not adhere to protocol present a threat to the validity of the results no matter what, mathematically, we do with their results. In the case of predictable non-adherence, I agree that having this as an exclusion criteria PRIOR to randomization is a better idea than letting them enter the study only to fail compliance. Having said that, the generalizability of the results will suffer with each additional group we exclude from the study.
I have a few issues with your publication on ITT. (nutritionandmetabolism.biomedcentral.com/articles/10.1186/1743-7075-6-1)
Specifically:
a) A physician concluding that Diet A is equal to Diet B based on a failed superiority trial is a common error we see with clinicians, which is misinterpreting 'no significant difference found' to mean 'there is no difference'. The former being a conclusion from a failed superiority trial, the latter from a successful equivalence/non-inferiority trial. This common clinician misunderstanding is important in of itself, and would have been better off left as a separate teaching point, as opposed to muddling the application of ITT.
b) The example of the CABS trial - the problem with just using as-treated post-randomization is that the reason for patients' non-compliance may be independently related to their outcome. (Perhaps the high mortality in the group who was assigned to surgery but got medicine was because these were the sickest patients in this arm of the study and someone felt they were not a good surgical candidate. Using as-treated or per-protocol would then selectively remove the sickest patients from this arm of the trial, resulting in worse bias than what we see with ITT.)
c) Your comment that it is "reasonable" that scientific knowledge ignore "data that was not in the experiment" is itself based on opinion and not fact. Although it is easy mathematically to just use per-protocol or as-treated analysis, these methods likely enhance the bias from non-compliance.
d) I agree that concepts or ideas are not better because they are newer. I would add that concepts or ideas are not better because they are established. As a friend says: "Be skeptical of anything you learn."
Like the much quoted "Democracy is the worst form of government, except for all the others", it is my thought that ITT is the worst way to deal with non-adherence, except for all the others. Which is not to say that ITT fixes non-adherence.
Glad you are drawing your own conclusions!
Anthony
genius
Anyone from NBME self assessment?
Ooooout
Your explanations are good but you are too fast.
nbme 20 hahaha
You try to give the video more brightness it will be great if you do
Thank you
how do you do that?
I use VideoScribe for the whiteboard and iMovie for audio and editing. Takes some time to learn, but pretty user-friendly!
Thank You
Thank you