Latest / This Is Why with Dr. Busti / Types of Data Explained: Nominal, Ordinal, or Continuous?

Transcript

Types of Data Explained: Nominal, Ordinal, or Continuous? · This Is Why with Dr. Busti

• 0:00
  SPEAKER_00: All right, we're in our biostatistics series, and this is going to be specifically about the type of data that you are having to read or put together in a research protocol.
• 0:12
  SPEAKER_00: I'm Dr.
• 0:12
  SPEAKER_00: Bustaye, and welcome to This Is Why, a show that is dedicated to helping all of us understand why we do what we do, because that makes the how much easier.
• 0:22
  SPEAKER_00: And the how can be if you're developing a research protocol or you're doing your analysis, how to do that.
• 0:29
  SPEAKER_00: If you're a clinician and you don't do research, but you need to read it and know how to interpret it, then how do you properly read a paper and then how do you apply that to the patient that you are privileged to serve?
• 0:45
  SPEAKER_00: That's what this is all about.
• 0:46
  SPEAKER_00: Getting oriented to the series, we have part one, which is what we do before the study starts.
• 0:53
  SPEAKER_00: Again, keeping us oriented.
• 0:54
  SPEAKER_00: Part two is at the completion of the study, that's our summary statistics.
• 0:59
  SPEAKER_00: And then if it's a diagnostic study and that involves a diagnostic test, then what part of that are we going to be focused on?
• 1:08
  SPEAKER_00: So here we are in part one, again, where the we are focusing our attention right now on this particular topic, which is part of the statistical analysis plan.
• 1:17
  SPEAKER_00: So there's the first one which is getting oriented.
• 1:20
  SPEAKER_00: There's now what we're going to be doing, the type of data.
• 1:23
  SPEAKER_00: Are you dealing with the primary endpoint of a study that is nominal, ordinal, or continuous?
• 1:29
  SPEAKER_00: May it be influenced by the way the hypothesis was worded?
• 1:33
  SPEAKER_00: What is the purpose or primary objective of the study?
• 1:36
  SPEAKER_00: That's why it matters because it literally helps you pick the right statistical test.
• 1:43
  SPEAKER_00: And then there's other components of the series on study groups, considering parametric versus non-parametric statistical analysis, and then the additional components that really are set up to make sure that we're not making mistakes or errors.
• 1:57
  SPEAKER_00: That's the alpha, the risk of making a type 1, beta, a power, which is the risk of making a type 2 error.
• 2:03
  SPEAKER_00: Okay.
• 2:04
  SPEAKER_00: And those get part of at the beginning.
• 2:06
  SPEAKER_00: And then there's that statistical analysis where we do a summary statistics after we've completed the study and we've collected the data.
• 2:14
  SPEAKER_00: That's descriptive statistics, inferential statistics, and then of course the diagnostic test if it was involving a diagnostic study where we cover things like sensitivity specificity, positive, negative, predictive values.
• 2:27
  SPEAKER_00: Alright, so our first task is to tackle that type of data, and you'll see the term analogy scale of measure.
• 2:35
  SPEAKER_00: And this is probably one of the biggest areas of confusion.
• 2:40
  SPEAKER_00: What type of data is my variable in the study?
• 2:46
  SPEAKER_00: Okay.
• 2:46
  SPEAKER_00: Now remember, we're talking about variables.
• 2:49
  SPEAKER_00: We have the independent variable and we have the dependent variable.
• 2:52
  SPEAKER_00: And the independent variable is typically the intervention, the exposure.
• 2:56
  SPEAKER_00: It's usually categorical.
• 2:57
  SPEAKER_00: You either gave them the drug or you didn't.
• 3:00
  SPEAKER_00: They were either exposed to something or they were not.
• 3:03
  SPEAKER_00: Okay, so in most studies, we're dealing with categorical, independent variables.
• 3:09
  SPEAKER_00: But what we're talking about right here, right now, in more depth, is what is the endpoint?
• 3:17
  SPEAKER_00: That's typically what is the primary objective.
• 3:20
  SPEAKER_00: And many times that's the last sentence in the introduction of a paper.
• 3:26
  SPEAKER_00: Okay.
• 3:27
  SPEAKER_00: Or if you're in a protocol, they'll say the primary objective was, the secondary objectives are.
• 3:34
  SPEAKER_00: So you've got to be able to find that information very clearly because the way they ask the question or the orientation around that variable influences whether it's categorically being treated, okay, or quantitatively being treated.
• 3:52
  SPEAKER_00: Now let's dive a little bit deeper into categorical.
• 3:55
  SPEAKER_00: That's typically a non-numerical measurement.
• 3:59
  SPEAKER_00: Okay, so that's where we're getting that nominal, ordinal kind of data.
• 4:03
  SPEAKER_00: We're going to be using non-parametric statistical analysis.
• 4:07
  SPEAKER_00: And then quantitative is a numerical.
• 4:10
  SPEAKER_00: It's usually something that we can measure.
• 4:11
  SPEAKER_00: And most commonly that's a continuous data point.
• 4:15
  SPEAKER_00: But here's the kicker, guys.
• 4:17
  SPEAKER_00: If you don't understand what those are, which is what we're covering here right now, I might ask a question differently.
• 4:27
  SPEAKER_00: I may take a continuous data variable that is traditional, Lee, a continuous data endpoint, and turn it into nominal.
• 4:36
  SPEAKER_00: That's right.
• 4:38
  SPEAKER_00: By the way I handle the endpoint of that variable.
• 4:43
  SPEAKER_00: Okay.
• 4:44
  SPEAKER_00: And at the end of this lecture, that is going to make more sense to you.
• 4:48
  SPEAKER_00: I promise.
• 4:49
  SPEAKER_00: Stick with me.
• 4:50
  SPEAKER_00: Okay.
• 4:50
  SPEAKER_00: Now, what I'd like to do next is to walk you through a table.
• 4:54
  SPEAKER_00: I like tables because it wraps my mind around how are they similar?
• 4:58
  SPEAKER_00: How are they different?
• 4:59
  SPEAKER_00: When do I use this?
• 5:00
  SPEAKER_00: When do I use that?
• 5:02
  SPEAKER_00: It's also great for studying.
• 5:05
  SPEAKER_00: Okay, so those of you joining us on our podcast channel, this is a types of data variables table.
• 5:12
  SPEAKER_00: Okay, and it's broken down into two big sections, those that make up non-parametric statistical data endpoints for analysis, and those that are parametrically treated.
• 5:24
  SPEAKER_00: Okay, continuous.health.
• 5:34
  SPEAKER_00: The link is available in the description of the episode for you.
• 5:39
  SPEAKER_00: Okay, so feel free to use it, share with your friends, colleagues, give it to students, residents.
• 5:45
  SPEAKER_00: If you're working on a protocol, you know, share with your team.
• 5:48
  SPEAKER_00: That's what it's there for.
• 5:50
  SPEAKER_00: If you're studying for a board exam, it's going to be very helpful for you for that.
• 5:54
  SPEAKER_00: Okay.
• 5:54
  SPEAKER_00: Board exams love to ask questions of content in this table.
• 5:59
  SPEAKER_00: I know that because I've taken, being a nurse, a pharmacist, and a physician, I've taken almost every board exam there is.
• 6:06
  SPEAKER_00: And I can tell you, I learned some of that process by taking those tests, seeing that like that's what's in here.
• 6:14
  SPEAKER_00: Okay.
• 6:14
  SPEAKER_00: But it also summarizes what you need to do.
• 6:17
  SPEAKER_00: Okay, so let me orient you to the rows.
• 6:20
  SPEAKER_00: So the first section up here is the non-parametric data.
• 6:24
  SPEAKER_00: This is where the population does not follow a normal distribution.
• 6:27
  SPEAKER_00: The data from that sample does not follow a normal distribution or bell curve.
• 6:32
  SPEAKER_00: Okay, and I'm gonna show you that, what that looks like here.
• 6:34
  SPEAKER_00: We're gonna talk about SKU data.
• 6:36
  SPEAKER_00: And then the second part down here is not parametric, where the data from the population or sample follows a normal distribution.
• 6:44
  SPEAKER_00: That's your orientation to this table, the biggest level.
• 6:49
  SPEAKER_00: Then you're gonna see nominal ordinals predominantly non-parametric, but you see this continuous is listed twice.
• 6:56
  SPEAKER_00: That could be confusing.
• 6:59
  SPEAKER_00: I get it.
• 7:00
  SPEAKER_00: I don't like it.
• 7:02
  SPEAKER_00: But I left it in there on purpose with that little asterisk there for a reason.
• 7:08
  SPEAKER_00: When your continuous data, which we'll get to in a minute, does not meet the assumptions or the criteria for being continuous or normally distributed, so it fails to meet those criteria, then you take that continuous variable and you treat it in the same way as you would other non-parametric data endpoints.
• 7:32
  SPEAKER_00: It's a it just is a little bit more conservative way of protecting your integrity of the data in the sense because something's messing with the results at some re for some reason, some confounder.
• 7:44
  SPEAKER_00: Maybe you know what it is, maybe you don't.
• 7:46
  SPEAKER_00: You should be looking for it.
• 7:48
  SPEAKER_00: Okay, but that's why that particular row of continuous is listed inside of the non-parametric division of this table.
• 7:58
  SPEAKER_00: It's not meant to be there to be confused.
• 8:00
  SPEAKER_00: I didn't make a mistake, because it very clearly says right here, what is the descriptor?
• 8:07
  SPEAKER_00: So pay attention to that stuff.
• 8:10
  SPEAKER_00: It's there for a reason, because those are the nuances, and again, board exams, oh man, they really like that kind of stuff.
• 8:17
  SPEAKER_00: You know, throwing the curveball.
• 8:20
  SPEAKER_00: Now, why does it matter to learn the type of data of the variable or endpoint that you are reading about or that you want to study or are studying?
• 8:30
  SPEAKER_00: Because that drives you to picking the right statistical test.
• 8:34
  SPEAKER_00: What row do you fit in?
• 8:37
  SPEAKER_00: Okay, so each row up on this table, this is the typical standard statistical test decision kind of table or tree.
• 8:45
  SPEAKER_00: All right, guides you.
• 8:47
  SPEAKER_00: What type of data do you have that puts you in the right row?
• 8:50
  SPEAKER_00: And then how many groups, how many groups do you have?
• 8:52
  SPEAKER_00: And whether they're related or independent, you draw a line over, down, and boom.
• 8:56
  SPEAKER_00: You're going to be using one of those tests.
• 8:59
  SPEAKER_00: If there's only one, then that's the one you use.
• 9:01
  SPEAKER_00: Now, there's reasons that you pick one over the other.
• 9:05
  SPEAKER_00: Some produce different types of results or give you a different type of perspective.
• 9:11
  SPEAKER_00: If you have small sample sizes or small uh data variables in certain groups, you might use something like a Fissures exact over a chi-square.
• 9:21
  SPEAKER_00: That, you know, so there's those nuances that we'll dive into as well, but at some point in this series.
• 9:26
  SPEAKER_00: But I want you to recognize that you want to get to the type of data variable that you're just trying to, you're studying or you're reading about because it puts you in the right row.
• 9:37
  SPEAKER_00: That's why this is so important.
• 9:40
  SPEAKER_00: And if you miss that, you make a continuous chain of mistakes.
• 9:46
  SPEAKER_00: So you must be oriented to these critical decisions of the statistical analysis plan or the SAP.
• 9:53
  SPEAKER_00: Okay?
• 9:54
  SPEAKER_00: So when we talk about parametric data from a sample, then we assume that it follows a normal distribution where there's that bell curve, okay?
• 10:03
  SPEAKER_00: That's what you see here, and where the mean equals the median equals the mode.
• 10:10
  SPEAKER_00: They're roughly equal.
• 10:12
  SPEAKER_00: Now they don't have to be exact to the decimal point, but they're pretty darn close.
• 10:20
  SPEAKER_00: So in you then and what are these descriptors?
• 10:23
  SPEAKER_00: Well, that's part of descriptive analysis.
• 10:27
  SPEAKER_00: The descriptive statistics.
• 10:29
  SPEAKER_00: Remember a summary statistics?
• 10:31
  SPEAKER_00: You have the first group is descriptive and then the inferential, right?
• 10:35
  SPEAKER_00: So that descriptive analysis is talking about what is the variability of the data points, what is the dispersion, how do they relate?
• 10:43
  SPEAKER_00: How do the data samples relate to each other from the sample?
• 10:46
  SPEAKER_00: Because remember, you took a sample out of the larger population to study because you can't study the whole world.
• 10:53
  SPEAKER_00: So does your patient sample reflect back on the larger group?
• 11:00
  SPEAKER_00: And if not, many times the data then becomes skewed.
• 11:04
  SPEAKER_00: It's not following a normal distribution.
• 11:07
  SPEAKER_00: Okay?
• 11:07
  SPEAKER_00: And that's where it's non, you start thinking about non-parametric analysis that needs to be done.
• 11:14
  SPEAKER_00: And that's why we calculate and describe the mean, median, and mode, or we do histograms and other things like that.
• 11:19
  SPEAKER_00: We're looking at it and seeing, does it kind of fit?
• 11:22
  SPEAKER_00: Or are there outliers skewing the data?
• 11:25
  SPEAKER_00: So in this particular diagram here, you can see the lighter gray line representing the bell curve, and then the darker line representing the skewed data.
• 11:34
  SPEAKER_00: This is positively skewed.
• 11:37
  SPEAKER_00: That means something is over on the right side, over here, drawing that mean or average over.
• 11:47
  SPEAKER_00: There's an outlier.
• 11:49
  SPEAKER_00: Okay, and typically what that means is the mean is now going to be greater than the median.
• 11:54
  SPEAKER_00: Remember, the median is the middle most, which is typically described by the interquartile range.
• 12:01
  SPEAKER_00: So where's the middle most in that skewed data?
• 12:04
  SPEAKER_00: Well, it's going to be over here.
• 12:06
  SPEAKER_00: So you see here that the mean, again, because if you go down this way on the x-axis, the number is getting larger.
• 12:12
  SPEAKER_00: So therefore the mean is going to be greater than the median, because here's the median right here.
• 12:18
  SPEAKER_00: So the mean is greater than the median, is greater than the mode.
• 12:23
  SPEAKER_00: Okay?
• 12:24
  SPEAKER_00: And this is where the tail is over on the right, so it's skewed to the right, and therefore the mean is greater than the median.
• 12:32
  SPEAKER_00: Those are great questions for board exams.
• 12:36
  SPEAKER_00: And they may just give you data and ask you to interpret that basic fundamental thing.
• 12:43
  SPEAKER_00: But you should do the same thing.
• 12:45
  SPEAKER_00: Because if I have data that was supposed to be a bell curve that is not, well then that meets, remember, that fails to meet the assumptions.
• 12:53
  SPEAKER_00: Let's say it's continuous data.
• 12:55
  SPEAKER_00: Then it fails to meet that, and I got to go up to that.
• 12:58
  SPEAKER_00: That's what this is looking like.
• 12:59
  SPEAKER_00: So if you then skew it to the other side where it's skewed to the left, okay, now the mean is now less than the median because the middle most is over here.
• 13:11
  SPEAKER_00: The inner quartile range is over on the further to the right, it's a larger number.
• 13:18
  SPEAKER_00: Okay.
• 13:18
  SPEAKER_00: And so skewed to the right, skewed to the left is really about orientation around those measures of central tendency.
• 13:28
  SPEAKER_00: Okay?
• 13:28
  SPEAKER_00: And so in your table, I have put those together.
• 13:32
  SPEAKER_00: So let's walk through the table in more depth.
• 13:34
  SPEAKER_00: And again, for those of you listening to us on our podcast channels, I'll try my best to walk you through it.
• 13:40
  SPEAKER_00: But you just kind of vision, we've got two main rows, and we're in the non-parametric kind of bucket, not following a normal distribution.
• 13:51
  SPEAKER_00: So on the first row is nominal data, the second row is ordinal data.
• 13:56
  SPEAKER_00: And I'm gonna walk you through some examples here of real, like practical endpoints that help you to understand that and to see that.
• 14:05
  SPEAKER_00: So follow me for a moment, okay?
• 14:08
  SPEAKER_00: Nominal is categorical, it's typically binary or dichotomous.
• 14:13
  SPEAKER_00: You are or you're not.
• 14:14
  SPEAKER_00: It's a yes or a no in most cases.
• 14:18
  SPEAKER_00: Okay?
• 14:19
  SPEAKER_00: It does not have a numerical variable, it's not measurable.
• 14:23
  SPEAKER_00: You don't, there's no sense of ranking.
• 14:25
  SPEAKER_00: You either are or you're not.
• 14:26
  SPEAKER_00: So think about mortality.
• 14:28
  SPEAKER_00: At the end of the study, we want to know if this study improves uh survivability, improves mortality.
• 14:38
  SPEAKER_00: Well, what's the question at the end of the study?
• 14:41
  SPEAKER_00: Is subject 200 dead or alive?
• 14:45
  SPEAKER_00: They're dead or alive.
• 14:46
  SPEAKER_00: Yes, no.
• 14:48
  SPEAKER_00: It's dichotomous.
• 14:50
  SPEAKER_00: They're in one or the other.
• 14:51
  SPEAKER_00: You cannot be in between.
• 14:54
  SPEAKER_00: So guess what?
• 14:54
  SPEAKER_00: There's not an average.
• 14:56
  SPEAKER_00: So when you get over to central tendency over here, there's no mean.
• 15:02
  SPEAKER_00: It's gone.
• 15:03
  SPEAKER_00: You can't be between the two.
• 15:06
  SPEAKER_00: Variability, there's none.
• 15:08
  SPEAKER_00: It's qualitative.
• 15:09
  SPEAKER_00: We're describing it.
• 15:11
  SPEAKER_00: Okay?
• 15:11
  SPEAKER_00: So they typically follow, represent straightforward yes, no.
• 15:15
  SPEAKER_00: Mortality is probably the most common.
• 15:18
  SPEAKER_00: Did they achieve their LDL goal on this new lipid-lowering drug?
• 15:23
  SPEAKER_00: Yes or no.
• 15:24
  SPEAKER_00: They either achieved the goal or they didn't.
• 15:28
  SPEAKER_00: Did they achieve their blood pressure goal or they didn't?
• 15:32
  SPEAKER_00: Did they lose a certain amount of weight or they didn't?
• 15:35
  SPEAKER_00: Alright, so the end point of that variable is a yes or no dichotomous, not measurable.
• 15:43
  SPEAKER_00: You put it in one bucket over another, there's no sense of ranking.
• 15:46
  SPEAKER_00: Whereas when you go to ordinal data, there is a sense of order to the information, the data.
• 15:55
  SPEAKER_00: There is a category, it's still not numerical in most cases, right?
• 16:00
  SPEAKER_00: You're falling into a category.
• 16:02
  SPEAKER_00: Think about the pain: mild, moderate, severe.
• 16:08
  SPEAKER_00: So you can't measure it objectively like a lab.
• 16:12
  SPEAKER_00: Okay, so there's no numerical value, there's no real measurable number.
• 16:19
  SPEAKER_00: Okay?
• 16:19
  SPEAKER_00: And the magnitude of the difference between those various orders, there's a sense of rank, but the magnitude of difference is not the same.
• 16:30
  SPEAKER_00: That is huge, that is key, that is essential for you to understand.
• 16:37
  SPEAKER_00: So think about the name ordinal data.
• 16:39
  SPEAKER_00: There's a sense of order and a sense of ranking, but the magnitude of difference from one ranking or order to the next is not the same.
• 16:50
  SPEAKER_00: Huge.
• 16:51
  SPEAKER_00: That's key.
• 16:52
  SPEAKER_00: So again, you can't have a mean.
• 16:55
  SPEAKER_00: You can't be between these things.
• 16:58
  SPEAKER_00: Like pain is between mild to moderate.
• 17:03
  SPEAKER_00: Okay.
• 17:04
  SPEAKER_00: Well, what is that?
• 17:05
  SPEAKER_00: I don't know.
• 17:06
  SPEAKER_00: New York Heart functional classification 2.5.
• 17:11
  SPEAKER_00: What is that?
• 17:12
  SPEAKER_00: I don't know.
• 17:14
  SPEAKER_00: Right?
• 17:14
  SPEAKER_00: So there's no, when you look at the central tendency, there's no mean.
• 17:19
  SPEAKER_00: You just have mode and the median.
• 17:21
  SPEAKER_00: So you can get an interquartile range.
• 17:24
  SPEAKER_00: That can be helpful.
• 17:27
  SPEAKER_00: Right?
• 17:27
  SPEAKER_00: Typically, visuals that we use are a box and whisker plots.
• 17:32
  SPEAKER_00: Okay.
• 17:32
  SPEAKER_00: One of the classic examples of this is the New York heart of functional classification for heart failure.
• 17:39
  SPEAKER_00: You can use stage cancer staging, uh, you know, BMI classifications.
• 17:44
  SPEAKER_00: Those are sort of classic ones.
• 17:45
  SPEAKER_00: So let's look at that.
• 17:46
  SPEAKER_00: So again, we're we have skewed data.
• 17:48
  SPEAKER_00: Our mean and median and mode are not lined up, depending on if it's skewed to the left or to the right.
• 17:55
  SPEAKER_00: Okay.
• 17:56
  SPEAKER_00: That's how we know.
• 17:58
  SPEAKER_00: But when we look at things like New York Heart Association's functional classification for heart failure, and you go, you look at go from class one to two to three to four.
• 18:10
  SPEAKER_00: Well, we know four is worse than three, and three is worse than two, and two is worse than one.
• 18:16
  SPEAKER_00: So there's a sense of order, there's a sense of ranking, but the magnitude of difference between them when you line them up are not the same.
• 18:28
  SPEAKER_00: The magnitude of difference is not the same.
• 18:31
  SPEAKER_00: You're gonna see here in a minute in the example I give for continuous, there's a sense of order, there's a sense of rank, and the magnitude is the same between each step.
• 18:44
  SPEAKER_00: So we come back here, we then, okay, we're at a non-parametric bucket.
• 18:51
  SPEAKER_00: So when we come back to our statistical analysis decision table, not saying you have to use this one, but it's one that represents the majority of the things that you will probably see and need to know.
• 19:05
  SPEAKER_00: It will get you by for sure.
• 19:07
  SPEAKER_00: It served me well from decades.
• 19:10
  SPEAKER_00: Okay.
• 19:11
  SPEAKER_00: But you're following in one of these buckets here.
• 19:14
  SPEAKER_00: It's where we've been spending our time.
• 19:16
  SPEAKER_00: So those rows then represent the statistical tests that you would use.
• 19:22
  SPEAKER_00: All you have to now discern is how many groups do you have and whether they're independent or related.
• 19:26
  SPEAKER_00: And you just draw a line over, draw a line down, there you are.
• 19:29
  SPEAKER_00: Do it.
• 19:30
  SPEAKER_00: Take a paper, do it.
• 19:32
  SPEAKER_00: It's so straightforward and it's it's actually kind of nice that it makes it that easy, especially when you understand the or and you're oriented to it.
• 19:41
  SPEAKER_00: Now, what about the continuous variable?
• 19:46
  SPEAKER_00: So remember, continuous variable here, the assumption is that it follows a normal distribution, which is that bell curve.
• 19:55
  SPEAKER_00: Now, continuous data is measurable and it's a numerical endpoint or value.
• 20:01
  SPEAKER_00: So that first point is huge.
• 20:03
  SPEAKER_00: There's a sense of ranking, and it can go on to infinity.
• 20:08
  SPEAKER_00: Okay?
• 20:08
  SPEAKER_00: And the magnitude of difference between each level is the same forever.
• 20:14
  SPEAKER_00: So therefore, in the central tendency, I can get a mean, median, and a mode.
• 20:20
  SPEAKER_00: I can also describe it as an interquoral tile range.
• 20:23
  SPEAKER_00: I can generate a standard deviation in this level here.
• 20:27
  SPEAKER_00: Okay, if I have a mean, as long as the mean and the median are not too far off.
• 20:34
  SPEAKER_00: Now remember, they don't have to be pinpoint exact.
• 20:38
  SPEAKER_00: But going back to that skewed data, the reason we do these descriptive statistics is to see if there is a difference and the data is skewed.
• 20:47
  SPEAKER_00: And so that's where frequency distribution graphs can give us that guidance.
• 20:51
  SPEAKER_00: We can look at the numbers.
• 20:53
  SPEAKER_00: A good publication will give you that data.
• 20:57
  SPEAKER_00: They aren't hiding anything.
• 20:59
  SPEAKER_00: But if they're assessing continuous data endpoints as a median and they're reporting the median, they're telling you there was something skewing the data.
• 21:10
  SPEAKER_00: That's what they're telling you, without saying it.
• 21:13
  SPEAKER_00: Because some people don't want to expose that their data is a little bit dirty, right?
• 21:19
  SPEAKER_00: Something's contaminated it.
• 21:22
  SPEAKER_00: So me, median, and mode, normal distribution, follows a bell curve, relatively equal.
• 21:29
  SPEAKER_00: Blood pressure is a great example of that.
• 21:31
  SPEAKER_00: So there's a sense of order, a sense of ranking.
• 21:34
  SPEAKER_00: Certainly, uh the higher the number, the potentially the worse the blood pressure could get at some point.
• 21:40
  SPEAKER_00: There's obviously a window of normal that's acceptable.
• 21:45
  SPEAKER_00: But every point of difference is one millimeter of mercury pressure.
• 21:50
  SPEAKER_00: It doesn't change.
• 21:51
  SPEAKER_00: So the magnitude of difference is the same, which is different than ordinal data.
• 21:57
  SPEAKER_00: That's key.
• 21:58
  SPEAKER_00: And so things that you can measure.
• 22:00
  SPEAKER_00: Measure that are numerically measurable, like labs, vital signs, are classic.
• 22:07
  SPEAKER_00: Think about lipid profiles.
• 22:10
  SPEAKER_00: Your your what was your total cholesterol and the average reduction at the end of the study?
• 22:17
  SPEAKER_00: Right?
• 22:20
  SPEAKER_00: Hot changes in blood pressure.
• 22:22
  SPEAKER_00: But here's the thing here that's so important.
• 22:24
  SPEAKER_00: Listen here, guys.
• 22:25
  SPEAKER_00: Pay attention to this comment.
• 22:28
  SPEAKER_00: Blood pressure right now in this format.
• 22:31
  SPEAKER_00: If I ask the question, what is the average reduction from the use of this independent variable or drug, right, is intervention, at the end of the study?
• 22:44
  SPEAKER_00: Well, that's taking the mean.
• 22:46
  SPEAKER_00: That means I'm treating that endpoint as continuous.
• 22:53
  SPEAKER_00: But if I say, did this patient who received the independent variable, i.e.
• 22:59
  SPEAKER_00: the intervention of this new drug, did they achieve their blood pressure systolic goal of less than 130 or not?
• 23:07
  SPEAKER_00: So at the end of the study, I'm sitting there assessing each individual patient, did you achieve your goal blood pressure of less than a systolic less than 130?
• 23:16
  SPEAKER_00: Yes or no.
• 23:18
  SPEAKER_00: So you see how I turned a continuous traditionally treated variable into a nominal endpoint.
• 23:30
  SPEAKER_00: That's huge, guys.
• 23:32
  SPEAKER_00: If you did not get that, rewind it, watch this.
• 23:35
  SPEAKER_00: That must stick.
• 23:37
  SPEAKER_00: Because again, it puts you in the different row, doesn't it?
• 23:41
  SPEAKER_00: It changes you from being in a parametric statistical test to a non-parametric statistical analysis.
• 23:47
  SPEAKER_00: Totally two different study uh study results.
• 23:52
  SPEAKER_00: You do not want to make that mistake, nor do you want to present that to a publication, you'll get rejected.
• 24:00
  SPEAKER_00: I mean, a good peer reviewer will flat out reject that paper right out of the gate for you making that catastrophic error.
• 24:08
  SPEAKER_00: So, this is why, in the a priori or pre-study initiation stage, you'd codify that.
• 24:17
  SPEAKER_00: You don't get it wrong.
• 24:18
  SPEAKER_00: And that's why it's so important to understand what is the hypothesis of this study, what is the null and alternative, and how am I asking the question, what is the orientation around that question?
• 24:29
  SPEAKER_00: Because it puts me into one of these rows.
• 24:32
  SPEAKER_00: So when we're in our parametric continuous, assuming that it meets all the assumptions, which is typically sample sizes, usually over 30, the data is following a normal distribution, the mean, median, and mode look relatively equal, there's no outliers.
• 24:48
  SPEAKER_00: Okay, we're sticking.
• 24:50
  SPEAKER_00: Now, that classically or historically, people would traditionally, you know, especially with two independent groups, you'd choose a student's T-test.
• 24:58
  SPEAKER_00: The Welsh's T-test is more real-world data becoming more popular.
• 25:04
  SPEAKER_00: Okay.
• 25:06
  SPEAKER_00: And so it assumes that there's some degree of inequality.
• 25:11
  SPEAKER_00: And so it's more real world, isn't it?
• 25:13
  SPEAKER_00: Because our patients and our clinics and our practices are not clinical trial environments in most cases, where people get treated differently.
• 25:24
  SPEAKER_00: Okay?
• 25:24
  SPEAKER_00: So you can see that the row that you're on is so important.
• 25:28
  SPEAKER_00: So that is why choosing the right designation for the data that's being evaluated in that primary and secondary endpoints is so critical for picking the right statistical tests.
• 25:43
  SPEAKER_00: Because without it, you are going to land in a whole different bucket.
• 25:46
  SPEAKER_00: And I'm telling you right now, board exams, all over that.
• 25:51
  SPEAKER_00: Because that makes sense.
• 25:52
  SPEAKER_00: If they it makes you're that's more critical analysis, that you understand the other components that get you to a point.
• 26:01
  SPEAKER_00: But if you can memorize that decision table, which is I know kind of busy, there's some stuff on that, but once you get used to it, it's really not bad.
• 26:10
  SPEAKER_00: You memorize that, you're gonna be good for journal clubs, you're gonna be good for your protocols, you're gonna be in good shape for your board exams.
• 26:19
  SPEAKER_00: I know because I used it.
• 26:22
  SPEAKER_00: And I've successfully published and I've successfully passed on my board exams on the first try.
• 26:27
  SPEAKER_00: Okay, so prepare yourself properly up front and you will achieve that success.
• 26:33
  SPEAKER_00: Thank you guys so much for joining.
• 26:35
  SPEAKER_00: I hope this is helpful.
• 26:36
  SPEAKER_00: Okay, please feel free to share this with your friends, your colleagues.
• 26:40
  SPEAKER_00: Give us a like.
• 26:41
  SPEAKER_00: We really appreciate the support.
• 26:42
  SPEAKER_00: Throw out a comment, I will reply back to you.
• 26:46
  SPEAKER_00: Okay, share your experience, maybe share your perspective.
• 26:50
  SPEAKER_00: Again, I'm not a biostatistician, don't pretend to be one.
• 26:54
  SPEAKER_00: But you know, I did my training at Oxford and studied evidence-based healthcare and have been involved in this for two decades.
• 27:01
  SPEAKER_00: So there's components of this that I've I feel comfortable with.
• 27:04
  SPEAKER_00: But yeah, I mean, we all learn something different from each other and different perspectives.
• 27:08
  SPEAKER_00: That's the value of sharing content like this and helping each other out.
• 27:13
  SPEAKER_00: So thanks so much.
• 27:13
  SPEAKER_00: We'll see you next time.