Skip to 0 minutes and 14 seconds So here with this example it’s looking at a drug for weight loss and type 2 diabetes. They had three primary endpoints which think about the fact that we should really have three power calculations. you can see that rarely happens, but they had three different primary outcomes and they haven’t several secondary outcomes. ANCOVA was their primary statistical method that they used. They also used regression analysis which don’t worry too much about that cuz we haven’t talked about that yet.
Skip to 0 minutes and 43 seconds And then the results are listed there so you’re gonna go through remember and you’re going to look at each of the tables you’re gonna find the data types in the baselines The outcomes specifically paying attention to the primary outcome and to the adverse reactions and then you’re going to go look for this statistical test in the written portion and in the legends to make sure you found all the different things and then you’re going to look to see if they use the correct statistical tests for the different areas.
Skip to 1 minute and 14 seconds So again here’s the baseline demographics you can see that they had only the two data types and the baselines and in this case you see that they didn’t have any p-values for these and this is kind of controversial one way or the other some advocate that you need to make sure there’s no differences between the groups to make sure that’s not a confounder.
Skip to 1 minute and 34 seconds Some say you shouldn’t I’m the part of the of the site of why not give us the information if you did it in the first place, it wouldn’t take up that much more space to give us a p-value to tell us that those differences there weren’t significant differences between the groups because what happens if you have a group of patients who are diabetic and if you’re looking at A1C at baseline and one significantly higher than the other wouldn’t that ultimately adversely affect the results and I would say yes a lot of those wait that’s the whole reason why they’re listed there. So I like to see p-values here.
Skip to 2 minutes and 7 seconds Alright so the outcomes again you you’re gonna focus on the primary outcomes. But you’re obviously going to look at all of them and so you can see now with these we have all three types we’ve been a ratio we have ordinal and nominal. So again we talked about you could just use two tests to cover all these you could just use an ordinal test that was appropriate for the data type or for the statistical design in the number of groups and you could just use one phenomenal but I find that most of time they don’t do that.
Skip to 2 minutes and 34 seconds Alright so here’s the results are the adverse reactions which they didn’t have in the other study we talked about for the most part ADRs should be nominal that the patient did experience it or they didn’t and so if you’re just gonna want to make sure that you list ADRS in your nominal section of your chart. So again you’re gonna fill out this chart each time. So hope I’m hoping that you did this on your own before it ice glows the answers to make sure that you’ve really gotten the concepts down. So so again we’re gonna look at the end is greater than thirty in each group and again that’s to make sure whether we can use parametric tests or not.
Skip to 3 minutes and 13 seconds Look to see if it was independent or dependent and you need to really make sure you look carefully to make sure they didn’t look at some of the before and after in each group as well as between group differences. Then you look at the number of groups and this was three and the comparisons are again were more than six so teach us would not be appropriate to use in this case. But they know would be if you had parametric data.
Skip to 3 minutes and 34 seconds So then again from those those charts that we went through and picked them we’re going to put each one of the data types in the appropriate box and then again you can either list what they used first or list what they could have used whichever way you’d prefer to do that and so you’re gonna list it what they said they used And so then you’re gonna look to see what could have been used to see if they match up. Well you can see that they said they used ANCOVA for the parametric statistics which you can pretty much see that those were all just the baseline characteristics that wasn’t any of the primary endpoints.
Skip to 4 minutes and 12 seconds So yes that one was correct and that’s the one that you’re least concerned about is the baseline characteristics yes you’d like it to be correct because that helps you to determine the how well they they were able to analyze everything else. you can see that here where we get our little sticky wicket that these were our primary at least that the first one was a primary outcome and they didn’t list any statistical tests that they used for it.
Skip to 4 minutes and 36 seconds So you’d have to say NO that they weren’t that p-value associated with that would not be able to be interpreted correctly and then you can see for the other primary endpoint which was the five and ten percent change in weight loss that they used logistic regression which again we don’t know if that’s okay but some of them may not have been appropriate for those.
Skip to 4 minutes and 59 seconds So you can see again going back to this chart that for someone of some of them they were less than 0.05 and thereby the primary outcome of weight loss they use logistic regression which we’ll talk about later that’s measures of association which is yes/no data which would have been correct and it was significant so that data was okay but we know that they used the wrong statistical test for the other primary outcome as well as for this quite a few of the secondary outcomes. So you can’t make any conclusions about those.
Skip to 5 minutes and 32 seconds So in conclusion just to make sure that I know I spent a lot of time going over or very short time going over a lot of information that you may need to go back and look at some basic articles on statistical tests but a thank you follow these easy steps you should be able to discern whether the statistical tests either helped or didn’t help a study but just remember the the design of a study the internal validity is much more important than the statistical tests that are chosen. So the best statistics won’t save a bad study and the worst statistics can’t totally invalidate a good study but they can hinder potentially full interpretation of the results.
Skip to 6 minutes and 16 seconds And then remember and we’ll go through this more as we go through this series that you need to make sure your critique all articles of study and not just the statistics. I found from teaching this over the years that the statistics believe it or not or more black and white then a lot of the other critiquing part of an article. So that’s what people will hone in on. where there’s other components that are equally if not more important if you use the wrong dose or if you set up your your study incorrectly again the best statistics aren’t going to save those.
Skip to 6 minutes and 49 seconds So make sure that you do a really good critique of the full study especially being pharmacist making sure it’s right studies, right duration, right dose, all of these different types of things. And then you’re gonna put it in context of looking at everything else. How does this drug if you’re looking at any drug compared to the safety profile of other drugs is the efficacy good or better. What about the cost? This one’s huge. Because something may be more effective than something else but it fits ten times the cost. It’s not necessarily going to be something you’re going to use plus look and see we’ll talk about this later about clinical practice guidelines. How does this fit in there?
Skip to 7 minutes and 28 seconds Does it add or do nothing does it take away from what the guidelines currently say? And then also you want to look at patient convenience and say a study is really efficacious but it’s a patient’s don’t want to take it or it has to be given by IV that that would be much more difficult for patients to use. So I hope that I’ve helped you at least a little bit to be able to understand those statistical words that you see in a study and how you might be able to evaluate them and I hope that you will use this tool the next time you’re reviewing an article
Summary: Results & Conclusion
In this video, Prof. Mary Ferrill points out the conclusion of this section.
First, we are given an example of a drug for weight loss and type 2 diabetes. Again, we have to distinguish the data types of the results of outcomes and ADRs.
After that, we can start to fill the table. Repeat the same steps as the previous video.
Do you get to the conclusion? If not, please go back to the previous video to make sure that you are able to fill out this table.
To sum up, the design of a study internal validity is much more important than the chosen statistical tests. The best statistics won’t save a bad study, and the worst statistics can’t totally invalidate a good study. However, they can hinder potentially full interpretation of the results. Make sure that you do a good critique of the full study.
This is the end of week 1. If you have any questions, please share them below.
We will discuss how to apply clinical practical guidelines considering patients’ convenience in the next week.