Name:
Statistics for Postgraduate Orthopaedic Exams ( Part 1)
Description:
Statistics for Postgraduate Orthopaedic Exams ( Part 1)
Thumbnail URL:
https://cadmoremediastorage.blob.core.windows.net/6856162c-079c-4e5a-a785-001f7b82bcd9/videoscrubberimages/Scrubber_1.jpg
Duration:
T00H31M34S
Embed URL:
https://stream.cadmore.media/player/6856162c-079c-4e5a-a785-001f7b82bcd9
Content URL:
https://cadmoreoriginalmedia.blob.core.windows.net/6856162c-079c-4e5a-a785-001f7b82bcd9/Statistics for Postgraduate Orthopaedic Exams ( Part 1).mp4?sv=2019-02-02&sr=c&sig=iy5xmyzCLQ3w6hbiUD8aIRi%2FoLOnGzQoYmG1yP9xnTI%3D&st=2024-12-08T19%3A15%3A33Z&se=2024-12-08T21%3A20%3A33Z&sp=r
Upload Date:
2024-05-31T00:00:00.0000000
Transcript:
Language: EN.
Segment:0 .
Thank you, everybody, for joining us. This is now the 1st of January 2020. Happy new year to everybody. And once again, I have to say thank you to all our mentors who work really hard throughout the last year and will continue to work hard now. Today, we were lucky to have Hani ghazi, who from Cork University hospital, who's going to talk to us about the research methodology and the statistics.
And one of our other mentors is Amjad Medina. He's joining us as well today to help us with the five sessions in the later part of this presentation, which will not be recorded as per standard. We'll also have a Mustafa. No, sorry. I apologies. We'll also have Fouad joining us soon as well.
Honey, if you would like to start. OK, thank you very much. Mr Vogel is also one of our mentors who's just joined us. Thank you. Also, medical staff in Cork in South of Ireland. So our topic today is the statistics and the research methodologies.
So all the views here are my view and no interest, no interest to be declared. So I will start with the data, which is the second that is a kind of terminology, so you have to know the language of sophistication and to do any research. So the first thing is the data, what the data is, observation of variance. So we'll go to your data, maybe numbers, maybe what, maybe like figures, so if it's no numerical like blood pressure, like visit numbers or maybe a categorical mild moderate, severe like disease stage like we had stage, all these things are categorical.
So the next step you start to you, you put your data in a graph and you start to interpret this graph. And if your data is normally distributed, so you will find this curve, which is like a bell shaped curve and been described by a car, and Frederick goes, which is mathematicians, German mathematicians and who describe if this normally distribute our data, it would be like a shape of the bill, and it's named after him.
But sometimes this data may be skewed to the right or the left. And you have a tail. So for example, if you are doing an ACL reconstruction, we commonly is done among young age group. So we will find the tails with a less number of data to the right because with the extreme of age, no much easier reconstruction.
And that's called positively skewed distribution. And the reverse in the total hip replacement or joint replacement. We do that mostly in old age groups. You will find the tail to the left because in younger age groups you will do a less number of hip replacement or joint replacement.
So next to how to measure the tendency, the central tendency of your data after you redraw the graph, so you have three definitions, three terms to know. The first is the mode, which is the most frequent data. For an example, the most frequently day is ofsted's for your hip fracture patients is about one week. But maybe the discharges start from two days, maybe to one month.
But in one week, most of your patients being discharged from your hospital, so one week is the mode. And then the second, the next was the median, which is a 50% you're not concerned about the most frequently, but in 50% of the patients say example, if you have a 100 patients to the 50 percent, you will be discharged with B in two weeks. So that is the median and the mean is the average show. So we'll calculate you will do the sum of all discharge for all your patients and divided that by the numbers, and that is a mean.
OK, if your data are normally distributed, the mean and the mode and the median should be the same. And this is a graph here. So the first one in the upper one, it's the median and mean it's one line, so it's normally distributed the bell shaped curve in the lower 1 to the left side. There is a difference in the lower one on the right one.
Is there a different distribution of your data? You will find the mean and the median mode is not the same. You have a three lines there. OK, so after drawing your, Uh, your bell or your diagram for data, so you have a variance which if you divide your graph two squares. The corrected some of this square in the middle of the mean is a variance.
And if you got the square root of this variance, that is the standard deviation. So one standard deviation is about 68% of your data. And two is 95% of your data. And three standard deviation, as you can see here in the graph, it's about 99.7% So in 95% of your data is only two standard deviation, and that is very important because the next step will show the confidence interval.
The confidence interval you include only 95 percent, which is 2 standard deviation for each side of the mean. And that is a confidence interval. So it is the interval that include value of probability 95% or two or two standard deviation. And important in the confidence intervals, if you have your sample size with high number of populations, so we will find your confidence intervals that 95% will go back here.
You will find the 95% is a narrow, you resist. You will find this line. It will be narrow because you have a large sample of size if you have a small sample of size of populations. You will find this 95% white and that is the difference between narrow and wide confidence intervals. So after collecting your data, how to interpret your data, so the next step, what is the null hypothesis?
So its assumption is made. That is no. Is the different seen occur by chance? There is no different with any difference. You detect it's only by chance. So how much any difference, so any something you created in your data by chance that's called P value is a probability occurred by chance. Usually we accept less than 5% That is AP value.
So the lesser the p value is a stronger evidence. Because if there is like any value by chance is less, that means that you have strong evidence and vice versa. So in to subject, we accept up to 5% of beef of p value and you will find this number in all of our searches like v equal 0.5% 0.001. But you will never find it more than 1% So next, there is error, so there is a type I and type Ii and type III.
So type one, you will incorrectly reject the null hypothesis. So null hypothesis. We say that this assumption that anything in your data by any difference, by chance. So if you incorrectly rejecting this hypothesis that is a type I. If you incorrectly accept the null hypothesis is the type Ii and there is a rare type called type III when the researcher correctly reject the null hypothesis but incorrectly attribute the causes of that and this type III.
Just one second. So, nick, is the power is a probability of finding significant association if it's truly exist. So like, for example, if you are, you have a drug or medications treatment for arthritis and you compare this medication against placebo or nothing.
So you measure how much significant association or any like a change happens by your new drug or new medications. That is the power. So the stronger the effect, the more power. It's a power. If it's truly exist, and estimation of the probability of the state will be able to detect through effect of the intervention. And what is power is power is 1 minus type Ii error.
So a. And another thing is part analysis, and our analysis is the method of determining the number of subjects needed in the study. So this is if you are going to do a research about something. You have to know what is the exact number needed? To detect valuable, to be for your research, and that's called part analysis, how you can get that. There is a lot of programs, or you can seek the assistance of a statistician to help you to get that, but it's a lot of programs you can use.
You can interview your prevalence of your disease, your incidence, your confidence intervals and the program. You will give you the number required for your research and factors. Affecting analysis is a science of difference between the means the spread of data, the acceptable level of significance with AP value sample size variability in observations and experimental designs like using subjects versus between subject and type of data.
It is parametric or non-parametric, its number or four figures, or according to that. So next, what does the study design? You have an observational or experimental so like observational, like the investigator, observe like polyethylene after a pulmonary embolism, after total hip replacement, so we will observe the complication of something experimental. No, you're obliged maneuver, you observe the outcome.
So we'll give the patient heparin versus something that will see the DVT prophylaxis. So will apply. You will. You will give something. You will give a new drug, new total hip new prosthesis. New something that's experimental. And there is a study timelines, if your data collection of your data from the past, it is a retrospective.
This is from we are going to start to collect the data in the next five years will be prospective study it or at the mean time. It's a cross-sectional study, so examine a patient at a 1 point of time without follow up. So, for example, you see the vision in the fracture clinic today, that is a cross-sectional study.
So what is the significance of testing to test for statistical significance? Think about the following. What type of data have been used in the study? What is the sample size? Are the group distributed normally or not? Do the data need to be transformed so that we can make normality assumption? Are the group interdependent or not?
And it will go again to the parametric or non-parametric test, the test, so parametric tests assume that it were sample from normal population. And the observation must be independent and the population must have the same variance. So there is specific characters for each test you are going to use. And this is mostly it's very common in the MCU. So this.
Few different type of test. The first one, which is a t test, so you will do like you will see the systolic blood pressure or blood pressure before and after, so period of observation on a single subject, so blood pressure before and after. It's only that is the B birth test. If there is a multiple observation, you will see something ANOVA or a one way variance test.
And here it is, a test of the before and after. The nexus unpaired t-test the umpire can use by 2 random sample like, for this example, will the mean of two independent sample they persuade in high standard and the low in high and the low socioeconomic status. So you have two independent sample, so we have to use no test, not the party test.
Chi-square test, this is if you are testing for qualitative data, the test is unreliable if any expected value is less than 5. So in this scenario, you have to use a feature exact test, but. Uh, in chi-square test, so we can examine like boys and girls blue, green, pink color of eyes, so you have more than one variant.
And that's chi-square test. Next to in the research, what is the level of evidence? So the first will start with a systematic review of all randomized controlled studies. This is the highest evidence. Next to it, it is a randomized controlled trial, then will design a controlled trial without randomization, then well-designed case study or cohort study.
Then a qualitative study or expert opinion at the base of the pyramid. So we'll start with the systematic review and meta analysis. The meta analysis is a quantitative in one word is quantitative estimation of the study. So it is a statistical analysis that combines the results of multiple scientific studies that address the same question.
For example. A arthroscopic versus many approach for rotator cuff repair. So you will go through auront all randomized controlled studies in the past and you will collect the data together and you will do your meta analysis. So this is praise my charge. This is the first step in the meta analysis after going to the websites above made and always both sides to see how many studies like have the same topics you will do.
Prisma Prisma charts the first deal. You will get the whole studies here. And they would start to exclude a study, so at the end by their. You you will exclude all studies, according to inclusion and exclusion criteria for your meta analysis. And then we will do a forest plot, so. You would put your data from all your studies, and you will find that and sometimes may this force of media obedience exam and you ask to interpret.
So we start from left to the right. Here is in the left is the name of the studies like Jonathan, italo, Al and so on. So n is the number of patients. So this is the study compared between the words shoulder and the hip arthroplasty. This is a meta analysis about that. We'll go there and you will find like squares and line the line. Here is a confidence interval.
So the narrowest line is a narrow confidence intervals and the widest is wide confidence interval and the square here is the weight of the study. So the bigger the square, the more weighted study. And you have to hear you have a 0 at this here, and you have to the right and the left. So all is study in favor of reverse shoulder and all study in the left here are in favor of hemiarthroplasty.
So if you ask you to interpret, so you're more you study more the studies. Are in favor of reverse shoulder also blasted and all missed, most of the study are the same way with a no confidence interval. And it is another false plot you see here. No, there is one study in the left, so it is study in favor of the control.
And this one. And you have CCC here by Adam is the big square. That means more weighted study. And this is study here by arash, 36. You will find how wide is the confidence interval. And how small is the square? So this list weighted study? So that is a false plot and how to interpret that. And we'll go there, you will find here that the word outrage you.
So the next question would be, what is the order issue? So it's operation is a probability of having a risk factor in those who developed conditions, so if you have 100 patients who are smokers, 30 of them died or got or got lung cancer. So what is the other issue? The risk ratio is 30 out of 100. 30% The other issue? No so you will.
You will take the 30 from the 100 will be 70, and they will divide the 30 by the 70. That is the duration. OK, so remember this example about smoking and lung cancer. And 30 and hundreds and how to remember the outrage. OK, so here is another force to plot, and sometimes you may ask you about this number here you find here, is it a small number?
This I2 equals zero. What is that? It is a heterogeneity. So what is the heterogeneity and what is important? It is important. It's very it's very common problem among meta analysis. And when you are trying to combine studies that are not matching with each other. So remember, you can combine apples, whatever the colors like green apples or red apples, but you couldn't combine orange and apples like that.
So we accept in the heterogeneity to be from 0 to 25 percent, and that is the low heterogeneity. It's a medium from 25 to 75 and is very high after 75. So if you find this, I will go back here. If if it's just from the forest plot, if you find the eye to here, it's like 96% You will know that this meta analysis is not good.
And this author combined like a study not matching with each other. And this is the review manager, it's a very common program if you want to do, like any meta analysis, that's free and it is a Cochrane review website. And what is systematic review, if we say about the meta analysis, it is a quantitative, less systematic review is a quality.
So we'll go deep through each study and to identify the methods, the populations and everything in this study. And you write your review. And that is why most of the studies meta analysis plus systematic review, you know, most combine both systematic and meta analysis. So next will go to the randomized controlled trial, which is the top after the meta analysis, randomized controlled trial in the top of the evidence.
So definition, so what is the randomized controlled trial is the group of patients are randomized to either receive or not receive an intervention. Like, you have an ulcer arthritis, you will divide the patient, some group of patients will receive your medication, your new drug and other patients will receive placebo and they will combine the results between both. And here the world is randomized, so what is the randomization?
Randomization is you have to ensure that all prognostic variable, both known and unknown, will probably be distributed equally among the treatment groups. And this avoids bias in the treatment. So To be sure that the patient involved within each group should be randomized, because sometimes if you are going to do a study for a new procedure or new hip replacement and you choose your patients, young patients fit skinny patients.
You avoid all obese patients, all that are bias. But if you have to do a real controlled trial, your patient, your sample should be randomized. And there is a different way is of randomization. Uh, the first one is assemble, you ask the computer by the date that we egawa like patient admitted in the first second for 6 and so on would be in the group And the patients coming in the third, fifth, 7.
And so on in the month will be group B This is assembled and the next will be stratified like you will ensure equal distribution between treatment among all age, sex in everything that's stratified and block. You will do two blocks with a number and in each block you are sure that you have the same equal number of patients equal number of age or number of sex. So that is the best one.
OK, so next, so because it's the heavy meal, so I divide it into two lectures, so the next lecture, we'll talk about the bias other type of study like cohort study, case control, study case. Here's a critical appraisal. The phases of clinical trials, how to conduct trial. How to conduct research will speak about screening tests and survival analysis, like kaplan-meier survival analysis and funnel plot as well.
OK, thank you very much. Thank you, honey. Anyone, anyone have any questions? So one of the questions is from Atif Mahmud, he's asking. He thought level 1 evidence was meta-analysis. Yes meta analysis of well done randomized of RCC is the best evidence, because if I have five well done randomized controlled trials.
I combined in one study in meta analysis, that is the best, the best evidence. Mason, thank you, and the. The question I would like to ask you if you don't mind, is could you explain the connection between the heterogeneity and the relation to I2, as you showed on the I2 is the most common way which is in the reef man, which is the most common program used in the meta analysis.
A reef man. So when you put your data in this program, so the eye to it is a measure of heterogeneity. So if I two, it's zero, $0.25 that is the measurement of the heterogeneity. So the lower the I two, no from the five 0 to 20 five, that's low. 25 to 75 media, more than 70 five, it's a high.
OK, so the higher the heterogeneity, if you say high heterogeneity, you're saying greater than 75% Yes, that means that number is traject. Yeah Yeah. OK that means that the studies like oranges and apples, not apples. OK, that's good. OK, thank you. The your next quote, sorry.
The next question is what added information does confidence interval give you compared to a standard deviation? OK if you did a study, if OK, if you did a study about six patients and you will give me, the complication rate is 0% OK could I rely on this, your sample sizes only 10 patients. I couldn't rely on that.
How I can know from like interpreting your graphs that your sample size is low, your comfort confidence interval, which is determined the certainty of your result. Is why it? So that is why we use a confidence interval. Another example, if you started. Your study with 100 patients and you got, for example, for a total hip, for the new total hip replacement, and you did the revision in the first two years 4 to 25 or 30 patients and you excluded these 30 patients that you did the revision from the end result.
So for an example, you are measuring the Oxford score pre-operative and in two years or five years for your new hip replacement. And you start to do the study with 100 patients, and after two years or five years, you give me like only follow up in 50% Only 50 patients and you excluded 30 patients. You did the revision and the 20 patients, lots of follow up.
That is very wide confidence intervals, and I would know that. OK, that's good. Another way of describing the confidence interval is, for example, if I told you that there's a 10% risk of heart attack associated with breathing air in an inner city compared to air from a village.
My confidence interval is 9 to 11% That's a very narrow confidence interval, so you can be reliant that number is correct. But if I said to you, my confidence interval is between minus 10% and 15% That means it actually might be even healthy, the possibility is the number you might be even healthier than air in a city is better for you because you have a minus 10% risk of a heart attack.
However, that might indicate that you have a small number of patients and therefore your confidence interval is really big. Or there is it's a spurious connection in my comment, for thing, is to know the definition once you start with the definition of any topic. If you are asked about AP value, you need to know what you value if the definition for emphasis. You are almost half half away through your path and then you can.
You can negotiate about the discussion later on. I think even the examiners are not they're not kind of very they don't get hold of. They don't have a very strong grip of statistics. So no, your definition or anything like if you are asked about the standard deviation, once you know what standard deviation. You can say that definition and then you can build your discussion afterwards.
Thank you for the presentation. Thank you. Thank you. And one point for me, like, I think it's a difficult topic and you covered it very well. Again, I'm just saying, if you get the opportunity to draw something, just draw it at the same time. So like, if you are lucky, they ask you what sensitivity and specificity.
So as you are speaking about them, just make their be simple for Dubai to chart and expand accordingly. That will save time and also give good impression to the examiners that the subject well, very much to all dimensions, also to especially to any other person who prepared an excellent presentation. We look forward to hearing the next one. We'll stop the recording now.
Viva sessions will start soon after this. These are not recorded, so please do stay on.