Skip to 0 minutes and 15 seconds Now after you’ve thoroughly assessed the NI margin and what the authors list you’re going to want to compare the ITT and per protocol results in the trial. Now it’s recommended for superiority studies that the ITT results would best reflect the study results because it’s more of what you would see out in the real world that not everybody is compliant with their patients. And just to review ITT intention-to-treat means that the that the traditional definition is that all patients who are enrolled in the trial are included in the final analysis. Now obviously how did that happen?
Skip to 0 minutes and 58 seconds Well the most common method and usually it’s a modified ITT, where it’s the last observation carried forward So you can see that if patients drop out of a trial or stop a trial sooner and those results are carried forward that’s going to make that drug whatever treatment group they’re in look worse than the other. So in superior trials they recommend that you use ITT is the basis for your information but being a literature evaluation person I think it’s great to give both. Because if there’s a difference between them then there’s probably an issue with the study design and the internal validity of a study.
Skip to 1 minute and 31 seconds Now with non-inferiority, it’s considered necessary not a recommendation necessary to list both because if you only just use ITT then you’re increasing the chance of a type 1 error. Remember that’s not a good error saying that a drug is non-inferior or therapeutically equivalent when it’s actually not. So you want to make sure they provide both results. And if they’re different again just like what you would see in a superiority trial If they’re different than there’s some issues with potentially the study design in the internal validity which you know can lead to problems in external validity.
Skip to 2 minutes and 11 seconds So let’s go back and look at an example one that I had just discussed previously when mentioning the NI margin and you can see that they only provided the ITT results Now remember this can increase the risk of a type 1 error and then combine that with the easier margin of 10% that they used they’re really setting this study up to find non-inferiority when in fact it may not be there. Alright. So let’s just summarize a little bit and I’ve got you an example here of how this data should be listed in a non-inferiority trial to make sure that you can except assess for this.
Skip to 2 minutes and 48 seconds Remember using ITT alone can increase type 1 or alpha error which again remember this is the one we’re more concerned about because you’re saying a treatment is non inferior when in fact it’s not and you can see in this excerpt that they provided both per protocol and ITT results and they listed p-values for both of them as well. So this is how the data should be presented and you can see that even though they were slightly different they both were considered to not have crossed or touch the Delta and they were found to be known inferior. So let’s discuss confidence intervals again to make sure that we’re graphing our non-inferiority information correctly.
Skip to 3 minutes and 31 seconds So just as we as we were talking about with measures there’s two different types of confidence intervals. Those the general or regular and what are known as Epidemiology or measures. So on the x-axis for general CIs that number is going to be 0 just like we talked about with measures. And then if it’s a measures which is listed as an RR OR or HR that number on the x the center of the x-axis is 1.
Skip to 3 minutes and 57 seconds Now again I don’t ever want you to forget to always check the width of your confidence intervals because I just discussed if they’re not if the sample data is not very reliable you’re not going to the authors are not going to very reliably be able to translate those results to the patient population. So there are three key pieces of information that you’re going to want to have to make that non-inferiority graph I already mentioned the type of CI because that will dictate whether we have a 0 or 1 on the center of the x axis.
Skip to 4 minutes and 28 seconds Remember our new number or threshold to not touch your cross is that Delta margin and we’re going to graph that and where we graph it will depend on how it’s presented and then we need to make sure we take the primary endpoint with the confidence interval and place that on there to see if non-inferiority was met or not. So here’s just an example of what it will look like. So or how you would graph it. Now the consort guidelines for non-inferiority trials say that the author should provide this To me, they should they say they should ultimately they should definitely provide this but they just recommend they provide it. But I can tell you that most studies don’t.
Skip to 5 minutes and 7 seconds So you’re really going to have to be proficient in graphing these to make sure that you can show whether or not inferiority was shown or not. So we’ll see that if it’s again number one thing if it to determine what the x-axis number is if it’s a general or a measures of association confidence interval you list either 0 or 1 at the bottom you place your Delta margin if it’s negative you place it on the left if it’s positive you place it on the right for general and then for good or bad outcomes for measures they’re flipped and then you’ll graph the results.
Skip to 5 minutes and 44 seconds The thing that I also like to do is you can see there’s that blue zone in the graph and this is called the zone of non-inferiority. So you can clearly see what’s in that zone and what’s not in that zone. So you can see when the Delta is on the left you can see that inferiority is shown which again you you’re not supposed to be able to say it’s inferior but it wasn’t found to be noninferior and it certainly wasn’t found to be superior and then you can see the ones that are totally in the zone but again I would look at those wide confidence intervals for some of them.
Skip to 6 minutes and 17 seconds You can see the fourth one clearly crosses that zone of not or the line of a non-inferiority and then you can see that if it was stated a priori that not only was non-inferiority shown with the last confidence interval but also superiority. Again graphs have to be set up correctly. If Delta’s on the Left favors active control drug goes on the Left. If it’s and then favors test drug would be on the right. So you can see with this one it’s the opposite. Because it’s a positive Delta. So you can see favors test drug is on the Left. Favors active control is on the right. The Blue Zone of non-inferiority is still there.
Skip to 7 minutes and 1 second So you can clearly see which ones are in there and which ones are not. Now again you can see that a lot of those confidence intervals are pretty wide so you’d have to take that into context with other clinical trials to see if that’s considered a reliable confidence interval or not. If Delta is on the left if it’s a negative value in the trial then you want to set up at the bottom of the graph that favors active control it goes to the left or new treatment worse favors test drug goes to the right and the new treatment better. If the Delta is listed on the right, then it’s the opposite.
Skip to 7 minutes and 37 seconds Favors test drug is on the left new treatment better favors active control drug is on the right new treatment worse. So graphing these from the results that you’re given is key to be able to determine non-inferiority.
Translating Confidence Intervals Graphically
Prof. Mary Ferrill clarifies how to compare ITT and Per Protocol and how to translate CIs graphically.
First, remember ITT means intention to treat, and the traditional definition is that all patients who are enrolled in the trial are included in the final analysis.
However, the most common method, modified ITT, is the last observation carried forward. If patients drop out of a trial or stop a trial sooner, those results are carried forward and make that drug look worse than the other.
For superiority studies, it is still beneficial to provide results for both ITT and PP b/c if they are different since it is probably an issue with internal validity.
As for confidence intervals, the number for general CIs on the x-axis is 0. If its measure is listed as an RR, OR, or HR, the number of the x-axis is 1.
Ultimately, when we translate CIs graphically, if delta is on the left, which means a negative value, we want to set up that favors active control goes to the left (new treatment worse). On the contrary, favors test drug goes to the right (new treatment better), and vice versa.