Computer generated transcript

Warning!
The following transcript was generated automatically from the content and has not been checked or corrected manually.

He went off live. Oh, it's live back now. OK. Fine. So um hopefully you can hear, hopefully you can all hear this. So beware, it can be very sudden and you need to be ready. So usually they will tell you the interview while I was informed a week before the actual date. So just be warned. OK. Um Next slide. So this is my little bit. So next time, right? So the academic station, they will ask you questions surrounding study designs methodology, how to formulate a research question, talk about a paper you've read et cetera, but it's all surrounding basically critical appraisal and you need to know when c with critical appraisal is how to formulate a research question by means of pot model. So the po model is used really to formulate a research question for like a systematic review in particular or it's really to frame your research around it. So pot stands for population intervention comparison and outcome population is what you, you know what you'd expect. You've got to predetermine your population you are studying. Now, this will depend on the study design, which I'll get on to a bit later. But a population really would be, you know, a certain demographic, a certain individuals with a history of said what? And you have to then clarify specific characteristics within that to conduct your study. The intervention is the intervention you go. If you are introducing, say a program or a drug, you'd have to really specify the course of the intervention, the location of said intervention, you just have to be specific the comparison. This is where, where you are either comparing adhering it to another intervention, comparing it to whatever you want to as your comparator. Now, if you were doing a randomized control trial, this would be compared to something you want to study against. So say you are introducing a drug compared to it's placebo, that would be your comparator. Now, in most studies, that would be your control. So your outcome is what you want to um study what you want to evaluate. So if, for example, you were doing a randomized control trial and you wanted to see if one drug has more efficacy for lowering, say cholesterol than the other than a placebo, then that would be your outcome. But the outcome is determined on what the study is, which I'll go to in a bit. So that is your Pica model. You need to have that in your mind. And when, for example, you'll ask a question saying what sort of project or would embark on when you're doing an SFP um program or post, you would need to say, well, I have formulated this question according to a PICO model and got in touch with Xy and Z that showed formulating said research question and knowing and also showing that you did, you have some research acumen behind you uh next slide. So study designs. I would if, if, if I were you looking back on it at the advantages and disadvantages of most study designs, if you've got a background in stats or epidemiology, then this would this should be second nature. But if you don't, as medical school really doesn't prepare you much for this. I would say the main studies you definitely out is according to either an experimental study or an observational study, experimental, just think randomized control trial or a non randomized control trial. Basically, you know, these are your ones which you're wanting to see and investigate an intervention or something which you've introduced, which is not known in the literature and cannot be observed, you're actually intervening. Hence why it's called an experimental study, randomized control trial would be your gold standard really for this um really is based off of the randomization process. So if um and you know, for more quality evidence, which has a lack of bias, you need to randomize and there are ways of doing that. But for your knowledge, you should just know that RCT is probably your gold standard for experimental studies, then you have the opposite. So you know observational study. This is where you don'tt assign exposures and you're just comparing with an observing against the comparison group. So this can be either a descriptive or analytical study. Now, the analytical studies you really need to know is the different cohort studies which can either be prospective or retrospective. So prospective is when you're looking, you're evaluating, say a cohort of and you're looking for the outcome and you're waiting for that outcome to happen through the study. Whilst retrospective, you sort of know the outcome and you just retrospectively looking back on it a case control study. So this slightly differs, this is where you know the out um know the the outcome of, you know, certain cases, but you really want to have get a case a case load, then you control it with an, you know, a healthy population you compare in which and then it's really important to know the differences between a cohort and case control because that might come up in your academic interview, you know, the cohort study exposure to outcome whilst a case controls outcome to exposure. So it's the other way round. And in cross sectional, you're just taking a snippet of a population or exposure and outcome at the same time. So for example, a survey is the most common type of cross sectional study. You get that one population, you study them at one specific time and you see a cross section of the study of the data at that time, then you have your cases, editorials, ecological studies. So just have a basic understanding of, you know, the different types of study designs. You have now advantages of say an OK. The best thing about randomization, it causes low selection bias and it has low confounding factors. But at the same time, it's a lot more expensive, takes a lot more time you need equipoise, which means you have to believe the intervention is beneficial and in and control is current practice. Now, it sounds like a big word. But basically, if you can explain that in interview then asks plus marks, some RCTS cannot be Generali meaning that it's only for a specific population. So therefore, that could be a disadvantage. There's funding biases, there's ethical issues and answer it. So with a cohort study, um you really need to be aware of the advantages being as it starts with its exposure, it has multiple outcomes which you can then study disadvantage. If you're doing a retrospective cohort study, you are bound to have some missing data, meaning that you're not, you know, you're not gonna get the full data set. Therefore, it's going to be sort of, you know, the the results will have to be extrapolated, sometimes results can be uneven. Um And to be found in variables, you really do need to introduce randomization, case control. So a lot of people get confused between case control and cohort, but case control really is a comparison of two or more groups but divided by the outcome of whether they had a condition or disease or ever had an intervention, et cetera. Now, these are less expensive, less time consuming and really, you know, there's only, they only have two outcomes. Thus can have a look, multiple risk factors whilst disadvantages is it can be quite difficult to choose these control groups. And it relies on odds ratios as statistics wise. So it's not as when it comes into the hierarchy of evidence, it's not as qu um quantitative as a cohort study. Now this thing on the side, this um this is called the Prisma Protocol. The reason why I put that is systematic reviews follow this and meta analysis. If you know how to do a Prisma Protocol and you know how to say this is what I'm doing from search strategy down to getting your studies included. Then that is basic bone to you know, backbone to your other studies design because most studies you will have to do a comprehensive review at some point. It doesn't have to be systematic, but it has to follow the sort of the rhythm of conducting um a systematic review and this is the gold standard. Now, can you go to the next slide please? Now the hierarchy of evidence, the reason why I've included Prisma in these different studies is you might be asked about what is the gold standard of you know evidence when it comes to the hierarchy of different studies. Now believe it or not randomized controlled trials is not top of the list at all. It's actually metro analysis. The reason being is with a meta analysis, you are systematically going through different databases to re with a strict inclusion exclusion criteria through results from a vast amount of sources. I only include pretty much our CT S to be searched within. So what I mean by that is you could do a meta-analysis on just best randomized control trials, which would be better evidence than just one randomized control trial. So you're getting a scope of all these RCT S and basically using some fancy stats to tell if it's an effective treatment or not or its effective intervention or you know, or EP factor. So that's why Metro Analysis and systematic reviews are actually up the top when it comes to hierarchy evidence according to the Oxford Center of um of Center of Evidence Excellence um which I used in my interview. So just under that, then you have your obviously your critically praised sources, but your RCTS cohort studies, case control, case reports, case series and then basically expert opinion, which is your editorials, commentaries, textbooks, et cetera. Um So if I were you, I would learn this pyramid of evidence off by heart because for example, when I had my Oxford interview, I was asked about, um I was asked about basically to interpret some data. And then I was asked about the confounding variables, biases associated with different study designs and then, you know, to reel off the hierarchy of evidence, et cetera. So it's really important that you learn this off by heart and knowing the advantages and disadvantages of each and every single one and to elaborate on them as well. Um Next slide, please. Wonderful. I think it's someone else's Tenner. I'm going to be talking about uh the common statistical terms that we come across when we're critically appraising a paper or just um or reading any kind of excerpt from a paper. And the main thing is to understand what these statistical terms mean and how we can uh use these statistical terms and understanding the results and uh coming uh to a fair conclusion and whether the conclusion that is given by the researchers of the particular research study matches with the actual results that they have uh presented as well. Uh This is going to be a very quick whistle stop to, of just highlighting the different possible statistical terms that I feel are commonly presented in most papers, especially uh RCTS. Um I'm just going to provide definitions and brief meanings for each of these uh statistical terms. It's not going to be an in depth analysis of what they mean and how to calculate them. And I don't think you need to know them as well, but it's just what they mean, what the definition is, and how do you interpret these uh in the context of uh the study? So this is all that we're going to be covering today. If anyone has any questions or wants to know um about any of these uh terms in a bit more detail, you can contact me and I can possibly help you with that uh later on. So, firstly, what is the population and what is a sample? Uh So Jack briefly touched upon this um when he was going through the PCO framework as well, I'm just going to briefly touch up on this as well uh in the context of statistics and what it actually means. Uh So the target population is the population to whom the findings are Generali via uh via explanation, obviously. Uh So this could if you're, if you're doing a study uh looking at uh smoking and lung cancer. So all the patients that smoke would be your target population. Now, a sample population is a proportion of that target population available to the researcher. Uh And this is derived from a sampling frame. So obviously, not all the patients that are there in the world are going to be accessible or available to that particular researcher. And that this is why we need to kind of narrow it down. And that's where you get your sample population from. Now, this is derived from a sampling frame which means uh which can be a list to register anything uh that the uh study population can be derived from. So if we are conducting a study that looks at patients with lung cancer, so you would want to go through the cancer registry to look at all the patients with a particular type of lung cancer or with lung cancer in general. And that becomes a sampling frame. And that, that's where you derive your sample population from. Now to further narrow it down, you get your, when you further narrow it down, you get your sample. And this is the proportion of the sample population that is used in your study. And the reasons why you further narrow it down can be mainly because of cost effectiveness and convenience. So again, if you were, if you, if we were to take the example of the lung cancer uh patients over here as well, you would not be able to do to analyze uh patients with lung cancer all over the world. Rather, you would want to stick to one particular region or one particular nation. Uh So if you're doing a randomized controlled trial, this would be one hospital or a few hospitals in uh the particular area. Or if it's a multicenter study, then the centers that are involved can be included, but you're not going to include all the patients that are available to you as well. The next thing is a power of a study. Uh So a power the, so the textbook definition of this is the ability of a study to detect uh the smallest possible true difference between the group if a difference does exist. So power calculation is usually made to figure out how big, how uh how big a sample size is needed to detect a true difference between, between the group. So if you have an intervention and control group, you want to figure out how big your sample size should be. So that you, so that the difference that you actually get from your results is actually a true difference. It is also the probability that a type I error will not be made. We'll talk about what a type two error uh later on in the pre presentation. But this also predicts uh the probability that a type I error will not be made. A power of 0.8 is is widely accepted as being adequate in most research studies. Uh which means that it there is an 80% probability of finding a statistically significant difference between the study groups if difference does exist. So you're doing this power calculation before you actually analyze your results. So you're not actually seeing if a difference exists or not, but you're actually seeing what the probability is that the difference that you're gonna get is a true difference if that makes any sense. And to determine the sample size, pilot studies are often done, interim analysis can also be done to inform a study inform whether a study is adequately powered. So if uh if you're reading a paper and they've done a pilot study or if they've done uh an interim analysis is uh generally considered to be. Uh so the, the study is generally considered to be considered to be of a higher quality, uh just because they have taken this into consideration and done a power calculation. So that, you know, the, the results that they've got is that uh so if, if they've got positive results, the the results are actually uh positive if that makes any sense. So we talked about type two error. This is how power um predicts the probability of a type two error as well. So we're talking about the different types of error over here. So type one error generally generally relates to false positives. Uh So we can, so if a type one error has been made, it concludes that the difference is significant when in fact that this is because of a sampling error. So when your results are statistically significant, um but there, there are underlying errors, especially in terms of sampling. Uh Then you would say that you have a type one error and the main reasons for why you would get a type one error is due to uh exclusion, uh overuse of an exclusion criteria, extensive exclusion criterias, too many sub analysis. So they're trying to find a result uh rather than they're trying to probe after the result rather than just find one. Uh If there's confounding variables, which we'll talk about later uh in the presentation as well, if there's multiple hypotheses again, when they're probing too much into finding a positive result, uh And if there's selection bias as well, which we'll talk about later in the presentation, so you need to think about the type one error, especially when you have a statistically significant finding. And you need to go through all of these uh sections in terms of exclusion criteria, sub analysis and all of that, that I've mentioned earlier and see whether uh the results are actually true or it's because of some bias that exists within the study moving on to type two errors. This essentially means false negatives. So you're concluding that the intervention has no effect when it actually does, which is exactly the opposite of type one error. And this is mainly due to inadequately powered studies. So studies with a low sample size. So if you have only a couple of patients in one of the interventional control group, you're not going to actually find a difference between the two groups. If you, if your, if your sample size is that that low, uh because each patient is gonna, even if you have one negative result, that's going to skew the overall results massively. Therefore, you need to think of whether a type two error has been made when you have a statistically nonsignificant uh finding. Next thing is P value and confidence intervals. And this is something that uh most people place emphasis on. Um I was definitely asked the definitions uh for both of these in my interviews. And I, and I doubt that that um no one else would be asked. I mean, ii I highly doubt that uh no one else would be asked this question. So definitely going to get um a question on the definition of P value on how to interpret it. And the same thing with uh 95% confidence intervals as well. So the definition for P value is the probability that the association uh detected is due to chance or no. Another definition is the probability that the observed results occurs when the non hypothesis is true. So you're seeing what the chances are, what is the probability that the results that you've observed is due chance. So, um and this is normally set at less than point naught five because you want to uh because, and this is widely accepted, which means that only 5% there is only a 5% probability that the observed results uh is a, is a false result or has occurred to uh to John and 95% of the time that this is the true result moving on to 95% confidence intervals. And this is a range uh between which the true value lies uh between 95% of the time. So, if we were to do the same study 100 times. Uh The true value would lie between that range 95 times. And again, this is widely accepted. Uh And this is uh and this is considered to be a significant risk. Next is the type of analysis, sorry. Um So there are two types of analysis that can be done. And this is based on um how many patients or the type of patients that are included within the analysis. So the intention to treat analysis is a more pragmatic approach where all partici all participants in the study, including those who dropped out are included in the analysis. And this is generally uh more representative of real world scenarios. Um and is more advantages as well because there can be systemic differences between those who completed uh the study and those who didn't complete the study. And we want to take those into account when we are actually analyzing these results. Uh And therefore, uh this provides a more holistic and more representative um picture of um of the study. Whereas on the other hand, you have the core protocol analysis, which is a more, which takes a more explanatory uh approach. So this only includes uh those participants that follow the study study protocol. So if you had a few pa participants that actually dropped out uh before finishing uh the study protocol, this type of analysis would not include them. And this then include introduces biases due to exclusion of participants. So this attrition bias, the selection bias and other types of biases as well, which we will touch on later and therefore can lead to overestimation of the effects uh of the intervention. But again, both of these have their advantages and disadvantages. The uh the main disadvantage uh of the intention to treat analysis is that you have to actually go um and find those patients that have dropped out and include them in the analysis and have their uh data as well. And that becomes an advantage for per protocol analysis because sometimes it can be difficult to find those patients or participants and get their data because they may have uh gone to a different country. There may be various reasons um as to why uh that that may be difficult and that's why sometimes per protocol analysis is preferred over the intention to treat analysis. But the gold standard still remains to be intention to treat. Next thing is relative risk. Uh So this again is a hot topic. Uh The definition is risk of an incidence occurring in the intervention group compared to the risk of the incidence occurring in the control group. The formula to calculate that uh is uh as on the screen. So the probability of event in the treatment group versus the probability of the event in the control group. Now, very popular question uh after the definition of of relative risk is how do you interpret the relative risk. So if the relative risk is less than one, then the risk of the event occurring is less likely in the control group compared to the intervention group. So that means it's more likely to occur in the intervention group. If the risk, uh relative risk is equal to one, the risk of the event occurring is equally likely in both groups and the relative risk is greater than one, then the risk of the event occurring is more likely in the control group. Oh, hopefully that makes sense. Another hot topic is uh odds ratio. So the odds ratio is defined as an odds of an event occurring versus the odds of the of the event not occurring. Um And as Jack had mentioned earlier, this is often used in cohort studies and case control studies. Um there is relative risk is used in octs. Again, the interpretation of this is if the odds ratio is less than one, the odds of the event occurring is less likely. If it's equal to one, the odds of the event occurring is equally likely to not occurring. Uh And if it's greater than one, then um odds of the event occurring is more likely. So if we take a look at this table, the small table over here, and if we divide our patients in the cases and control versus exposed or not unexposed, we get this, uh we get four sets for different uh strata of patients ABC and D. So the odds of the exposure in the cases group is gonna be the number of patients, number of cases with the exposure versus the number of cases without the exposure or in the unexposed group. So that's gonna be a divided by C um and the odds of the exposure in the control group are gonna be the number of uh patients in the control group with the exposure versus the number of controls in the uh number of patients in the control group without the exposure exposure, that is B by D. And therefore the odds ratio would then become the odds of the exposure in the cases group versus the odds of the exposure in the control group. And therefore we get this formula, this is just so that you understand as to how the uh uh calculation is um done. Uh You, you won't be asked to calculate the odds ratio. You won't be asked to calculate the relative risk. But just to understand um the background uh formula just included the that in here as well. The next thing is hazard ratio, it is very similar to relative risk. Um that describes the ratio of the hazard in one group compared to the other. Now, the term hazard is used when the risk is not constant with time and therefore, is mostly used with respect to survival over time. Uh And this is of often associated with the Kaplan Mayor survival curve. As well, which we will talk about in one of our future uh sessions as well. So the the definition of a hazard ratio is very similar to that of relative risk. Only that we use the word hazard instead of um risk. Because the hazard, the term hazard is used uh when the risk is not constant if that makes any sense, and then we have number needed to treat and number needed to harm. So number needed to treat is the number of subjects that need to be treated with the intervention compared to the control for one additional participant to receive uh a benefit. And this is often calculated by uh this formula, one divided by A RR. So A R is um the absolute risk reduction which is calculated by the probability of the event occurring in the treatment group minus the probability of the event occurring in the control group. Uh And that again, uh is the risk. Uh the absolute risk difference between the two and it's not a ratio as previously mentioned, number needed to harm is pretty much the same concept. Uh But it's the number of subjects that need to be treated with the intervention compared to the control for one additional part participant to suffer from an adverse outcome. And the main advantage of uh so the, the main um advantage of using a number needed to harm is to see whether your treatment does not have any uh significant side effects and you, you've taken into account some of the safety outcomes as well. And this increases the quality of uh the study as a whole. And it's, it's something worthwhile if you mention it when you're critically appraising uh study as well. And then finally, we have the sensitivity and specificity and these are often used uh with respect to a new diagnostic uh test. Uh sensitivity is the ability of a test to correctly identify patients with the disease. So a high sensitivity means low false negatives. Uh So you're not uh ruling out a disease in a patient that actually has the disease specificity is the ability of a test to correctly identify patients without the disease. So, again, high specificity means low false positives. So you're not inaccurately um diagnosing someone with the disease. Uh Even though they don't have it. If that makes sense, then we go on to the next section. All right. Thanks Andre. Um I'm gonna be covering the confounding and bias. Uh Part of the talk. This is the final part of the talk. Uh We'll go through it in the next 15 minutes. Um Why is confounding and bias important? Well, well, it is important because when the examiner asks, you tell me some strengths and weaknesses of the study, this is the majority of the stuff you'll be talking about next slide, please. So, what is confounding? Well, a confounding variable is a third variable that distorts the relationship between the independent and independent variables resulting in incorrect conclusion. Next. So the most famous one, I'm sure you're all aware of is an increase in ice cream sales leads to increased rates of skin cancer. We know this is not true because there's a confounding factor here. The confounding factor is the hot sun and warm weather which increases both the ice cream sales and the skin cancer rates. And so that results in a positive confounding factor where we think there's a relationship between the ice cream sales and the skin cancer rate, but there really isn't next. But there's something else called a negative confounder which hides a relationship with, with between an independent and independent variable. Another good example is a poor diet, increases the risk of heart disease. However, if you have a poor diet, but you also exercise a lot, you mask that effect, which doesn't. So it doesn't mean that poor diet does not increase heart disease. But the fact that you're exercising is hiding that. So this is a negative confounder. Next, there's a lot of ways to prevent confounding. And there's some listed below restriction randomization matching and statistical methods will cover some of these in the next slides. Next. All right, bias is something entirely different which people often confuse with. Confounding. A bias is a systematic error which is introduced when sampling or testing by selecting or encouraging one outcome or answer over others resulting in incorrect results. So that's a fancy definition. But normal definition is simply that there's methodological errors in the study that are resulted in results that are not so true, they would be different if those methodological errors did not exist. And we usually classify bias into two main groups, selection bias and observe bias. There's plenty more there, there, there's far too many biases out there to think about. But these are the main two that we classify them into selection bias is when you have errors, when you're recruiting and allocating patients to groups, observe bias is when you have errors when collecting and analyzing data and that slide. And all this comes into play when you're just, you're doing your critical appraisal when you have that abstract in front of you, when you have that paper in front of you, and you're being asked about the strengths and weaknesses of the paper, critical appraisal will be discussed next week and as well as well as its three components, intent and validity, extend validity and ethics. But today, we'll focus on intent and validity next, which is circled right here. Sorry. Can we go back again, please? All right. So in inter the circle right here, it's essentially looking at the methodology, looking at how the patients have been pre through it, how they've been recruited, how they've been selected, what's happened to them during the study. This is an a full appraisal includes this mostly because this is looking at how well the study has been performed. And this is where a lot of your biases and confounding come in next slide, an easy mnemonic to think about intend ability and to know what to assess when it comes to intend ability is Rambo or if you have a statistics spot, it's Rambos. Rambo is a Rambos, whatever you wanna say. And that consists of recruitment allocation, maintenance, blinding outcome statistics. We will discuss this in more detail next week. But I want to go through these with you to tell you where different confounding factors and biases may occur. Next slide. So we'll start off with recruitment next slide recruitment. When it comes to recruitment, you think about some of these factors, you think about inclusion criteria. So the characteristics of the ta target population that have been included in the study, you think about the exclusion criteria, which is the character characteristics of the target population that have been excluded from the study, you think about sampling and recruitment. So this is how exactly has the study been advertised? Where has the sample population been uh sample population come from? Has it come from the hospital? Has it come from the community? You also want to think about whether it's an explanatory trial or prag pragmatic trial. The definitions are on the slide as well and also whether it's a Multicenter trial. But the main components where you wanna consider, whether there's confounding or bias is the top three next slide. And that's inclusion and exclusion criteria. Exclusion criteria comes into play when you want to avoid confounding. For example, if, if you have a population, if you have a study and you don't want age to be a confounding factor for that study. Because of course, as you get older, you get more confan, uh you get more uh comorbidities, you become more frail and that can cause confounding. If you don't want that to be a factor, you can stay in your study. Well, I want to exclude all patients above the age of 60. And therefore you have avoided confounding from all those factors including frailty comorbidities, et cetera. There's also an element of bias that comes into play here that we will discuss in the next slide next. And also there's bias that comes into play. What do you think about sampling and recruitment next? And this is all shown here. This all comes under selection bias. Selection bias itself is a big tree term for all the biases that come under recruitment uh sampling and all these things. And the definition for this is recruitment of sample population that is underrepresented of the target population. And within this, you have sub biases, for example, diagnostic pity bias. So when you have so many exclusion criteria that you're excluding every single co morbidity under the sun, you're excluding a range of different factors which are simply not representative of the actual population. People tend to be complex. Patients tend to be complex. If they have heart failure, they will have a whole load of bunch of other problems including diabetes, hypertension, et cetera. And if you exclude this in your study, you're getting a very small component of the target population. So that means if you do have a new treatment that you wanna test, it's not gonna be effective or it's not gonna be appropriate for the majority of the population. And that's where this diagnostic majority bias comes in. Another bias is the membership or response bias. So where are you recruiting the patients from? If you're recruiting them from charities, organizations, media outlets, you're gonna get a very small cohort of patients and often patients that are part of charities, organizations are tend to be very active, highly motivated patients. So they're more than likely to stay within the study, they're less likely to drop out, they're less like to rate the treatment negatively. And that all makes a difference in how your study goes forward because the general population, you're gonna have a lot of people who are very lazy or patients who are, you know, not happy with their treatment and that affects the results as well. You also have other biases like the bein bias, Nieman bias, which I won't cover in this talk, but it's important that you have a look at them as well because when it comes, when, when you get that question in your interview, the more you know, the better your answer will be and more intricate points you'll be able to pick up from that abstract or paper they can talk about and press the examiners with next slide. OK. So the next component of internal validity that we wanna look at is allocation. Next, when it comes to allocation, we think we think about how the uh cohort has been randomized. There's usually three types of randomization, simple randomization, block randomization and stratified randomization. Next slide, we're all aware of simple randomization. You flip a coin, you use a random number generator, you, you use a computer generator sequence to put the patients into a case group and a control group. The issue with this is if you have a small sample size, there's no way you're gonna be account for, you're gonna be able to account for the confounding factors because for example, if you have 10 patients, five of them or six of them have a confounding variable like they're very old or something, there's very, there's very high chance that all of them will be allocated to the case group or the control group. And therefore your case and control groups aren't similar enough to be compared next line. Another type of randomization is block randomization. This is a bit different. You do this when you're recruiting patients on a rolling basis or you're recruiting patients during the trial trial as it goes on what you do here is you randomize patients to groups. So on the screen, you have group A B and C. So patients are randomized to groups three different groups here and then they're allocated to either the case or control groups. What this ensures is that there's always an equal number of patients in both the case and the control groups. Next slide. Another type of randomization is stratified randomization and this is really effective to avoid confounding variables. As you can see here, you have a bunch of people, you have some uh black colored stigma and some red color stigma. Let's imagine that the black colored stigma are patients with a mild disease. The red color stickin are patients with a severe disease. For our case and control groups. We want to allocate them equally into the two groups. We don't want all the severe patients in one group and all the mild patients in one group. So what we do here is we first stratify the patients based on the severity. So group A has the mild disease patients and then group B has the severe disease patients and then we equally allocate them to the case and control groups. Therefore, preventing confounding from taking place. Another way we can do this is called matching, which you can see on the right side of the slide matching. Basically means for every case you have, you have, we find a patient that's very similar, you find a patient that's very similar to that case patient and you do that for every single case patient. So you have very similar patients all the way through the issue with this is very time consuming, especially if you have a large sample size, although it does work very well, very well. Next slide, the next part of interval if we can think about is maintenance. Next, when you think about maintenance, you want to consider whether the study is a fair test, were both arms of the study treated equally. What analysis was used? Was it a per protocol analysis or intention to treat analysis which did drew covered? Um And you also want to think about what the dropout rate was. Was it a high dropout rate or a lot of patients lost a follow up because the more patients that lost a dropout or uh lost a follow up that increases the risk of bias. And this is a very particular type of bias called nutrition bias. Next slide which is shown here. Sorry guys, if I shake a little bit, it's a bit cold in my room. Attrition bias is a systematic difference between the patients who leave the study and those who remain. What this essentially means is that patients that tend to leave studies are usually more unwell, less tolerant to side effects or not as motivated. And so the the final group of patients you have are not very similar to you to what you originally had when you recruited these patients. The other thing is as when patients leave the study, they don't, equal numbers don't lead both the case and control groups. More people tend to leave the case groups because that's an active treatment and it often leads to more side effects. And so you have unequal groups that therefore you can't compare very well. And this again contributes to that attrition bias next line. The next part of inter validity we can think about is blinding. Next. So there's three types of blinding. Single blinding is where the participants don't know whether they're receiving the uh active treatment or the placebo or gold standard. Next, double blinding is when both participants and researchers don't know what group the patient is in next slide and triple blinding is when the data analyst, patients doing the data work or analyzing data don't know what group the patients are in. And you might wonder why, why is triple blinding even relevant. Well, if you think about uh histopathologist, if they're given slides of cancer patients and non-cancer patients to interpret what the slides are showing, they might be more critical of cancer patients with known cancer, they might look for the uh you know, the cancer a bit more. And so disrupting what the results show next. When blinding doesn't take place, it leads to observe bias. Again, observed bias like selection bias is a branch which consists of many different types of biases observed bi observed bias itself means failure to measure or classify exposures or outcomes correctly. And when this, you have interview wise. So this happens when the researchers are not blinded. If the research are research is aren't blinded, they may modify their approach to collecting and recording results from participants. So say a questionnaire is part of the study protocol. The researchers might phrase their questions in a positive way to get positive answers from the pa from patients response bias is when the participants are not blinded. And what this what happens is participants are more inclined to give favorable answers or ratings during interviews and questionnaires. If they're aware, they're in the experimental group, they do this. So they don't offend the examiner or the interviewer. The next bias is a very interesting one. It's called the hawthorn effect. And what this means is patients alter their behavior usually positively when aware that they're being observed and monitored. A good example of this is if the patient knows that their weight is gonna be taken or their cholesterol is gonna be measured, they might try to eat healthier. And again, this can disrupt what the results of the study show. Finally, we have something called recall bias. And this comes into play when you have case control studies and cross sectional studies, people are terrible at remembering things. And if you, if you do get a case control or cross section study, just say there's going to be elemental recall but I say because people are very terrible at remembering things, they have poor memories and that will always get you the points next, outcome of internal validity next. So outcome usually consists of three end points that researchers tend to measure. There's clinical endpoints, this is direct clinical outcomes like mortality, morbidity, survival, and there's also surrogate end points. So biomarkers the physical signs, that's a substitute for a clinical endpoint like LDL for cardiovascular disease, you're not gonna wait for someone to ha have a heart attack that can take some time. So to assess if a treatment has worked, you can just measure things like LDL. Obviously, it's not as effective, but it can give you some sort of indication of whether your treatment has worked. Finally, you can have composite end points where you combine a different number of end points together. And this is again a very effective way to get immediate results instead of waiting for various clinical endpoints to come into play next. Again, with outcomes, you can have outcome reporting bias. This means selective reporting of pres specified outcomes in published studies. So usually the big studies will publish their protocol a priority. So before the study takes place, they'll say what they wanna measure, they'll say what they want to look at. But when it comes to publishing things change, OK. If certain outcomes that they decided on didn't do so well, they might not report those. If certain new outcomes that they hadn't thought of do really well, they might publish those because that puts the study in a positive light. However, this is again biased because they're not sticking to the original protocol. Next, the final part of internal blood, you want to think about statistics next. So this is all the stuff that need to have covered. It's very important to know the definitions of these and to think critically about them as well. But coming on to confounding how confounding comes into play here. Well, there's actually some statistical methods you can use to count for confounding, including standardization and Multivariate analysis. Standardization essentially means if you have two groups, your case and you controls, you can sort of standardize or statistically manipulate the data so that certain confounding factors no longer come into play. So you can standardize the data for age, for gender, things like that, the more confounding factors you standardize the data for the more difficult it becomes Multivariate analysis is actually a good way to standardize for a lot of different confounding factors. Um I used this for my uh study back in, back when I, when I was in my I BS C and it's a very effective way and it took takes a lot of clever maths and statistics to do which I don't have the knowledge to speak to you about. And this is also not something you will be quizzed about. So it's not to worry about. But just to, just to have something on the top of your head to know next. All right, what I recommend for all the statistical stuff that we've discussed is this book. Um It's quite small. It's a very easy read. And I really recommend that you guys read through it between now and when you have your interviews, this will explain everything very well with examples that we don't have time to speak to you about in the talk. Um And this is in my opinion, the best resource out there, you can use to really be a very good person who can critically praise um in terms of critical appraisal itself. Now that we've discussed some of the core stuff next week, they're gonna go through some work examples with you and also go through what internal validity is, what standard validity is and what ethics is and then speak to you about how to assess each one and present it to an interview when asked about it next. If you guys do have any questions, uh please let us know otherwise we'd really appreciate if you uh you know, fill in the feedback for us so we can carry on improving uh and give you guys what you want. It helps us out and it helps you out in return. Um And if there's any questions now, we're very happy to take them, just send them in the chart. If you have any questions, doesn't have to be about statistics. Any questions about interviews, about our personal experiences during those interviews and any tips we can give you happy to let you know anything at all? Ok. Ok. I guess not. We can stay here for, for 555 minutes, uh, 5 to 10 minutes if we can stay live until um, you guys are sure you don't have questions. Do you have any questions, Jack? No, you knew everything that was discussed today. It's pretty funny. No, I think there's only four people on the thing now. So, and yeah, and then they have you um feel free to message us as well if you have any questions. I think you have Jacks Linkin. Sure. Oh. Any other resources besides the past test book? Um There was a book which I had and I've forgotten what it was called. It was like how to Prepare for AC F interviews or something like that. I remember um I remember my friend's, I think cousin had wrote it or something. So um it was really, really good. I forgot what it's called though. It was like how to prepare for like academic interviews for medicine or something like that. I think my, my main ways I revise was that book. And also there's plenty of youtube videos as well explaining certain er complicated statistical terms like hazard ratio odds ratio. So you'll um s so, so you sort of understand when and how to use them and what they're talking about because they are all very similar and it can get confusing. Um So try youtube videos and that book gives you examples of where they've been used. So it's, it's, it's very good in that respect. Um And I think the, the there's another book wasn't that um had it somewhere. It's called the Medical Interviews Book by Oliver Picard. And that's the one we talked about. Last time. There's also um first choice A FP uh which is a good website um that talks through these things. Um Yeah, I think those are the main ones for academic stuff. It's, it's often very difficult to find resources for academic stuff because they can literally ask you anything. Um Which is why books are books tend to be the best for it and also youtube videos to understand specific statistical terminology. Yeah, that's true. Yeah. The guides ac TST and reg infuse. Yeah. Mm But it's not like again, yeah, it's not very academic but um but yeah, I'm trying to remember that book. But yeah, just those books which we've already mentioned is probably the best bet. And there's just a whole lot of critical appraisal resources out there from universities that you can have a read of. That's how I started. I think it was like some weird imperial um critical appraisal document that I was reading. That's the best way to start to introduce things and you don't have to pay for these ridiculously expensive courses either to learn how to critically appraise. It's a waste of money and it's just like con. So I wouldn't bother. What do you guys feel about going to courses for, for a FP and SFP? Did you, did you attend any? I know, I just thought they were very expensive and just, it was very much based around the London SFP interview and I didn't apply to London. So, um, I thought I was very biased on what they wanted so I didn't find it very useful to. Um, well, I heard that it wasn't very useful. I didn't intend so. Yeah, I think there's always, there's always a big, um, group of people that really want to attend these courses and stuff and you still go along, er, just because everyone else is doing it. But I think books and online resources and also practicing these things with a friend if you just, um, pull up abstracts, um, and, um, you know, critically appraise them while speaking to a friend or critical braise them to, you know, in your parent, for example, just speak through it and learn those presentation skills. That's enough. You don't need to attend courses. It, it, it's, it's as much as it is about knowing all the academic and statistical stuff. It's also about how you present it and how organized you are at presenting, which is what, you know, we'll be focusing on next week, how to, you know, systematically express what your answer is also. Really strange. What's the likelihood of getting a research station for the? Ok. So, no strange question at all. And the likelihood is very slim, you will not get a research station for the leadership domain. Very, very unlikely. Um, you'd probably get more of a, to be honest, you're probably gonna get more of a personalized station more about, like, explain what your leadership, what you've done to, you know, within clinical leadership, et cetera. You're not going to be like, asked about critical appraisal at all. Hm. Would you agree? Yeah, I, I definitely agree with that. Um From people I've spoken to have done leadership in medical education. They, they, they haven't had any research. Um and it wouldn't make sense either because there's, there's a dedicated research track, the skills they want from. You are, are very different that, you know, you, you leadership and me medical education. They want you to hone in on those sort of skills instead of research because that's not what your sap will be based around. Yeah, I remember some people actually liking the leadership SFP interviews. They thought it was fine. Like it was very much personalized. The med ed one was very much personalized as well, depending on where you reply, it was just a clinical station which was still, you know, still that's gonna be the same. But other than that, um no, I wouldn't worry about it because whether you're a leader educator or a, er, or a researcher, you will need to do your clinical stuff. You're, you're, you're f one first who, who have responsibilities and they'll be in very tricky situations when you start. Like Jack is on his psych ward. Sometimes my ward seems like a IOR as well. You pick up different skills. Trust me. Yeah. No, there's, there's plenty of sight happening on my ward. I can imagine. I don't think I'm missing out. Right. I think we should call it. All right. Thanks, sir. Thanks for coming guys. Fill in this, fill in the feedback form for us. Thank you. All right. I need Groove. See you later.