Latest / This Is Why with Dr. Busti / Independent vs Paired Groups Explained (Choose the Right Test)

Transcript

Independent vs Paired Groups Explained (Choose the Right Test) · This Is Why with Dr. Busti

• 0:00
  SPEAKER_00: All right, biostatistics.
• 0:01
  SPEAKER_00: You either love it or you hate it, but we're gonna make it easier clinically relevant.
• 0:06
  SPEAKER_00: I'm Dr.
• 0:06
  SPEAKER_00: Busai and welcome to This Is Why show dedicated to helping healthcare professionals understand why they do what they do because it makes the how easier.
• 0:16
  SPEAKER_00: How?
• 0:16
  SPEAKER_00: How do you read a paper, properly interpret that, and apply the results to your patient?
• 0:22
  SPEAKER_00: If you're a resident or clinician and you're doing clinical research, how do you develop a study protocol and define your protocol variables appropriately so that you arrive at the correct statistical test?
• 0:38
  SPEAKER_00: All right, now just to show you where we are in the larger overall plan in this series of biostatistics, right?
• 0:47
  SPEAKER_00: There's part one, so that's where we're at before the study starts.
• 0:52
  SPEAKER_00: What did the researcher, what were they supposed to do in defining their protocol?
• 0:57
  SPEAKER_00: And that's part of that statistical analysis plan or the SAP.
• 1:01
  SPEAKER_00: Part two is the act the completion, the summary statistics that include descriptive and inferential statistical analysis, traditional things like p-values and such, right?
• 1:11
  SPEAKER_00: And then if it's a diagnostic test related study, then what are those measures?
• 1:17
  SPEAKER_00: Okay, so here we are in part one.
• 1:19
  SPEAKER_00: We're in the SAP, okay, the statistical analysis plan, and we're specifically going to be focusing our attention very briefly on the types of study groups.
• 1:28
  SPEAKER_00: Now listen, guys, each of these topics can stand alone like this one.
• 1:33
  SPEAKER_00: But I and that's why I keep orienting you, okay?
• 1:36
  SPEAKER_00: But they can also be used in a series if you need that, and that's available on our YouTube channel, podcast channel, and at thisiswhy.health for you if that's uh useful.
• 1:49
  SPEAKER_00: So, you know, we've already covered things like intro to statistics, getting oriented to the big picture.
• 1:55
  SPEAKER_00: Where do you start, where are you ending, or trying to get to?
• 1:58
  SPEAKER_00: The type of data that you are gonna have to you know intervene evaluate at the end of the study is what's that study endpoint, okay?
• 2:07
  SPEAKER_00: And then the types of study groups, are they independent or are they related to each other?
• 2:12
  SPEAKER_00: And then we have other topics like understanding parametric versus non-parametric statistical analysis, and then moving on eventually to those inferential statistics of how do you interpret the results?
• 2:24
  SPEAKER_00: Okay, how do you interpret those results?
• 2:27
  SPEAKER_00: And those are things like the 95% confidence interval, the effect size, and then if it's a diagnostic-related study, things like precision, accuracy, sensitivity, specificity, positive, negative, predictive values.
• 2:38
  SPEAKER_00: That's not where we're going to be spending our time here.
• 2:41
  SPEAKER_00: We are going to be trying to focus our attention on figuring out which of the four columns are we in on this table.
• 2:52
  SPEAKER_00: Remember, I have this for those of you who can't see it or that joining us from our podcast channels.
• 2:58
  SPEAKER_00: This is a table on the standard statistical tests that are typically done.
• 3:03
  SPEAKER_00: Okay, for most studies that we typically read.
• 3:07
  SPEAKER_00: It's a decision tree, if you will, but it's in the form of a table.
• 3:11
  SPEAKER_00: Okay.
• 3:12
  SPEAKER_00: And I organize the type of data by nominal, ordinal, continuous.
• 3:16
  SPEAKER_00: Over on the far left, each row represents one of the types of scales of measure.
• 3:20
  SPEAKER_00: And then each of the columns represents how many groups and if they're independent or related.
• 3:25
  SPEAKER_00: And so we have to be able to discern that.
• 3:36
  SPEAKER_00: Okay, when they do more than two groups, it can get a little wonky in the way they organize it, like the ISIS 2 trial with aspirin.
• 3:44
  SPEAKER_00: There's a 4x4.
• 3:46
  SPEAKER_00: Okay, so you just have to get oriented to that part.
• 3:49
  SPEAKER_00: And then you have to ask yourself: are they related or are they independent of each other?
• 3:53
  SPEAKER_00: And do you know how to discern that?
• 3:56
  SPEAKER_00: Because that drives which statistical test is going to be used.
• 4:03
  SPEAKER_00: Okay?
• 4:03
  SPEAKER_00: And so this table is available for download for free at thisiswhy.health.
• 4:09
  SPEAKER_00: And the in the description of the episode, there is a link that can point you to that if you want that for studying purposes, or you're developing a protocol and you want to use it to guide you.
• 4:21
  SPEAKER_00: Okay, so what are the types of groups?
• 4:24
  SPEAKER_00: We have those, the number of groups is straightforward in most cases, but you sometimes do have to read if they combine groups.
• 4:32
  SPEAKER_00: Okay.
• 4:34
  SPEAKER_00: And they should tell you that.
• 4:36
  SPEAKER_00: But it's in the obviously, it's not necessarily a visual, it's written out.
• 4:41
  SPEAKER_00: So you have to kind of read it, the method section, pretty clearly on the study group, to create a little bit of that structure.
• 4:50
  SPEAKER_00: And then sometimes it's also codified a little bit better in the baseline characteristics.
• 4:56
  SPEAKER_00: But independent groups or samples are where the patients are not the same in the two or more groups.
• 5:06
  SPEAKER_00: They're just different people.
• 5:07
  SPEAKER_00: Okay?
• 5:08
  SPEAKER_00: And this is classic for an RCT.
• 5:12
  SPEAKER_00: Randomized controlled trial.
• 5:15
  SPEAKER_00: You are randomly assigned to one of or more groups, right?
• 5:22
  SPEAKER_00: So wait, one or you're you're assigned to a group, but it's good, there may be two or three groups that you are more that you get assigned to, but you only get assigned to one and you don't cross over.
• 5:33
  SPEAKER_00: So when you're doing a comparison from group one to group two, and maybe group two to three, and one and three, if you have three groups or four, then those are those patients serve as separate individuals.
• 5:47
  SPEAKER_00: Now, the randomization process is put into place, if done correctly, to try to closely match them naturally.
• 5:58
  SPEAKER_00: So you're not the researcher isn't handpicking and placing the patients with a desired probably endpoint that they can foresee happening based on behaviors or other confounders.
• 6:11
  SPEAKER_00: We don't want that.
• 6:12
  SPEAKER_00: That's where there's double blinding and double dummy study designs become very relevant in controlling for some of those outside influences.
• 6:22
  SPEAKER_00: But the randomization should allow for different people to be closely matched up.
• 6:28
  SPEAKER_00: And how do we know that happens?
• 6:30
  SPEAKER_00: Well, the baseline characteristics of the study, if you did a well-designed randomized randomization process that was appropriate, your patient should look fairly similar.
• 6:42
  SPEAKER_00: And they'll do statistical analysis to see if there is a difference and if there's an area where it is.
• 6:49
  SPEAKER_00: And that might be a confounder or something that influences the internal validity of the results that you as a reader need to discern.
• 7:01
  SPEAKER_00: Now, what about paired or related groups?
• 7:04
  SPEAKER_00: Okay, now these are where the patients actually are not different.
• 7:09
  SPEAKER_00: Let's say I put a medication in my ro my right eye versus my left eye.
• 7:15
  SPEAKER_00: Okay?
• 7:16
  SPEAKER_00: Well, I'm the same biological person, right?
• 7:20
  SPEAKER_00: So my anatomy, my physiology, my immune system, my my uh genetics all are going to respond in the same way.
• 7:30
  SPEAKER_00: So if I were to test something in one eye over the other, right?
• 7:34
  SPEAKER_00: Or what if they're so highly matched that the patients begin to look and act like the other one?
• 7:41
  SPEAKER_00: You can do that.
• 7:42
  SPEAKER_00: Even with randomization.
• 7:43
  SPEAKER_00: You may say, well, wait a minute, Dr.
• 7:45
  SPEAKER_00: Basil, you just told me that if they're biologically two different people, that they're that's independent.
• 7:51
  SPEAKER_00: Yes, I agree.
• 7:52
  SPEAKER_00: But depending on who you talk to and how deep in the minutiae of biostatistical analysis you want to go, there are scenarios where you can match people so tightly that they act and basically look like they're paired or the same person, even though their genetics are technically not.
• 8:13
  SPEAKER_00: I mean, what's the main thing that makes us different?
• 8:16
  SPEAKER_00: SNPs, single nucleotide genetic polymorphisms are the one thing.
• 8:20
  SPEAKER_00: Well, those the majority of our DNA are very similar.
• 8:24
  SPEAKER_00: We vary by these small SNPs, okay?
• 8:27
  SPEAKER_00: And they are profound, but they're relevant.
• 8:30
  SPEAKER_00: Okay, but you can match people so tightly that you begin to look the same.
• 8:36
  SPEAKER_00: That's why I highlight that one there as well.
• 8:38
  SPEAKER_00: But the classic one is a randomized crossover trial where the patients get randomly selected to go one uh treatment and non-treatment group, they get the intervention, the independent variable.
• 8:52
  SPEAKER_00: We finish the study, we assess the dependent variable or the outcome or the end point at that point, and then we do a washout period.
• 9:01
  SPEAKER_00: Let that intervention, that exposure, that independent variable go away.
• 9:06
  SPEAKER_00: And then when it's sufficient washout period has occurred where that lingering effect is still not present, so we don't drag it over, we then cross those patients over.
• 9:15
  SPEAKER_00: So what ends up happening in a crossover study is the patient serves as their own control.
• 9:23
  SPEAKER_00: And why is that relevant?
• 9:26
  SPEAKER_00: Relevant because you can enroll a fewer number of people in a study and maintain good and adequate power to protect yourself from making a type 2 error, okay?
• 9:39
  SPEAKER_00: So it matters, but it takes longer to do the study because you're basically doing the study twice.
• 9:46
  SPEAKER_00: You just have to randomly, or you have to cross those patients over, and now they go into the other intervention.
• 9:51
  SPEAKER_00: And then you follow them again forward in time and you measure that endpoint, right?
• 9:57
  SPEAKER_00: Your dependent variable, you measure it again, and then you're going to do your summary statistics, descriptive inferential analysis, based on the fact that these patients are serving as their own controls.
• 10:10
  SPEAKER_00: That means typically their confounders are the same, right?
• 10:15
  SPEAKER_00: Patient, unless something's changed during the time frame, and that could be relevant, but in most cases the patient serving as their own controls is a good thing because it reduces confounder influence.
• 10:28
  SPEAKER_00: Okay?
• 10:29
  SPEAKER_00: Identical twins would be the other one rare scenario that you might encounter.
• 10:34
  SPEAKER_00: So it's important that you define for that.
• 10:37
  SPEAKER_00: But if they say this is a randomized controlled trial, that right out of the gate tells you you have, and there's only two groups, you have two independent groups.
• 10:50
  SPEAKER_00: You're right here in a typical randomized controlled trial of two groups, that's the column you're in.
• 10:56
  SPEAKER_00: You just now need to discern what type of data variable is the endpoint, the dependent variable.
• 11:05
  SPEAKER_00: And I'm gonna say it over and over again for those of you watching this series, because I'm gonna drill it in, chisel it into that brain.
• 11:13
  SPEAKER_00: Okay?
• 11:14
  SPEAKER_00: But but that's that first step.
• 11:17
  SPEAKER_00: Which column are you in?
• 11:19
  SPEAKER_00: Or if you're already knowing what type of data endpoint you have, what row are you in so that you then discern, okay?
• 11:26
  SPEAKER_00: I've either got two or more groups, right?
• 11:30
  SPEAKER_00: So if I got three or more, I know where I'm at.
• 11:32
  SPEAKER_00: I'm gonna be in one of these two.
• 11:33
  SPEAKER_00: Now I just need to decide are they related or independent.
• 11:36
  SPEAKER_00: That is what's great about this decision table, is it streamlines it down for you, makes it a little bit easier.
• 11:44
  SPEAKER_00: Does this encompass all of the statistical analysis that biostatisticians do?
• 11:49
  SPEAKER_00: No, of course not.
• 11:50
  SPEAKER_00: I mean, I'm not gonna insult that, those people, those experts.
• 11:54
  SPEAKER_00: This is a basic level, but it will carry you pretty far and allows you the framework to assess a paper or to write your protocol and then get appropriate consultation if it's needed, especially if you have some nuances to the study design that you're going to use.
• 12:15
  SPEAKER_00: So this is a core concept that you correctly recognize the type of study groups because that directly influences whether you're gonna use what statistical tests you're going to use.
• 12:29
  SPEAKER_00: Okay.
• 12:29
  SPEAKER_00: Now that may not matter whether it's parametric, non-parametric, but the fact of the matter is you've got to be in the right row, the column, in order to choose from that then list of statistical tests to do your analysis at the end of the study.
• 12:44
  SPEAKER_00: And if you do everything right, it should work out.
• 12:46
  SPEAKER_00: Okay, and that's what makes this great.
• 12:49
  SPEAKER_00: So I hope that you found that orientation and that additional detail on the types of study groups to be helpful in the big picture and framework, okay, of where you are in if you're developing the protocol or if you're reading the paper.
• 13:06
  SPEAKER_00: Was it done right?
• 13:07
  SPEAKER_00: So give us a like if you uh are on our channel at YouTube.
• 13:10
  SPEAKER_00: We really appreciate the feedback.
• 13:12
  SPEAKER_00: If there's anything that we can do for you, any questions, leave them in the comments.
• 13:15
  SPEAKER_00: We I do answer those.
• 13:17
  SPEAKER_00: Okay.
• 13:18
  SPEAKER_00: Uh we will get back to you because we all learn from each other.
• 13:22
  SPEAKER_00: Okay.
• 13:23
  SPEAKER_00: And again, these tools are available over at thisiswhy.health for you for free.
• 13:28
  SPEAKER_00: If you need to practice, you know, questions for you're preparing for a board exam or something.
• 13:33
  SPEAKER_00: We have Q Bank questions over at high yield med reviews.
• 13:36
  SPEAKER_00: Okay.
• 13:37
  SPEAKER_00: Uh, and there's discounts that are available at thisiswhy.health if you need that kind of stuff to prepare for a board exam.
• 13:43
  SPEAKER_00: So thanks again for joining us.
• 13:44
  SPEAKER_00: We'll see you next time.