Computer generated transcript

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

Hello. Hi, everyone. Can you all hear me? Well, is everything okay? Okay. Good view says yes. Ok, cool. Hi, everyone. Thank you for taking your time to come to the ledge today. I know that's football. So, um, a lot of you might not be able to make it. So if you're on the recording, hopefully this is useful to you as well. Um, but I'm going to be talking a little bit about epidemiology today. Um, just a brief overview about l map in general. Um, last year, I think most people struggled with epidemiology the most. Um, because it's just a lot of concepts I feel rather than just memorization, like a lot of other things. So I would say, when you're revising l map, I would mainly focus on epidemiology just because everything else, um, you can write down in your notes, which you can bring into your exam. So, yeah. Um, so if we move on, so this will be a guide for today. So we're gonna spend about 10 minutes on measuring and describing diseases, and it's on designing epidemiological studies and 10 minutes on interpreting epidemiological findings. Um, I think it's very good the way that Imperial have kind of structured it this way. They also do this on on Sunday, and to learn it this way means that it kind of like makes more sense in your head because there's a lot of, like, phrasing and a lot of similar words that can get job build up in your head. So just learning it by the categories makes the biggest difference in my opinion. So yeah, so we'll start with measuring and describing diseases. So, um, the first thing that they introduce you to is transition models. So these are basically what populations go through over, um, the years. And, you know, everyone started out at basically pre slash stage one of the transition model. So, um, on the left hand side, you have stage one to stage four. And on the right hand side, you have pre early, late and post. They're basically the same thing describing the same thing. Except that, um, the kind of right hand side doesn't include stage five. So what exactly are these different stages? So Stage one, um, is to do with pestis pestilence and famine. Um, and there's a high birth rate and high death rate because, um, you you don't have, like, development in healthcare. Um, and you also have a high birth rate as well. But on stage two, this is where you have development and healthcare, and that's why the death rate sort of decreases. And you also get agricultural development and vaccinations. And the vaccinations also help to reduce, you know, illnesses and things. Stage three is where you get degenerative and man made disease is So this is where you get non communicable diseases like diabetes. Um, and you know, things such as obesity start to rise. Addiction, violence and other issues start to emerge, As you can see on the graph, um, the birth rates significantly decreased at this point as well. Um, and the death rate sort of plateaus, and stage four is delayed. The gender of diseases and emerging infections so includes zoonotic diseases like very relevantly covid. Um, the past two years, um, and you know, there are more emerging technologies, so, yeah, these are the four stages. Um, one other thing that, um, you need to know is about population pyramids. So you just need to know that three different types of population pyramids So the triangle is for the developing countries such as Yemen. They're rapidly growing. Um, the UK and New Zealand are more like housing kind of shapes because, um of immigration. So even though we're coming towards an aging population, there's still a lot of young people who are immigrating into the country and immigrating out of the country. So it's more shaped like a house because of that. And Spain is shaped more like a coffin because, um, there's less so much immigration and immigration and immigration. So the UK, you could say, is more of a stage four. And Spain is more of a stage five, um, in terms of the transition models. So yeah, so when talking about epidemiology, you always talk about the, um, words, exposures and outcomes, and it's very important to know what they are. So exposures are the variables associated with change in health, and it's your independent variable. Um, and your outcome is your dependent variable and the change in health status. So example of this is, um, um, the effect of educational posters, which is what is changing the health status on the proportion of patient's clerked for the A. M. t A m T T s score by the doctors in the hospital, which is what your actual outcome is. So yeah. So in terms of prevention, um, there are three different types of prevention that you can have, um, to, you know, kind of prevent disease. So you have primary, um, prevention, which is before the disease. So this is where you kind of control your risk factors. So, for example, giving very brief advice to encourage patients to stop smoking. Um, so you're basically trying to lower as much of the risk factors as possible. Secondary prevention is where you slow your disease progression. So, um, this is when they already have the disease. And you basically detect this and introduce appropriate treatments and interventions. For example, controlling hypertension with anti hypertensive drugs and tertiary prevention is where you return to normal function. So after you've had a stroke, for example, or heart attack, um, you basically want to minimize the suffering and, you know, improve their quality of life. So you would give cardiac rehabilitation for an MRI patient, for example to get them as close to their baseline as possible. So yeah. Um, so there are four different ways of kind of measuring, you know, um, epidemiology. So you have odds. Uh, I would recommend for all of these, by the way, that you kind of go through the different formulas for each one and do some, like, practice questions on it. So you kind of get familiar with what's what. But you have odds, for example, which are the ratio of the probability of event um, a k the number of people with the disease to its complement, which is the number of people without the disease, and it can range from any value from 0 to 100 2000. You also have prevalence. Um, and this is the proportion of individuals in the population who have the disease or attribute of interest at a specific time point. Um, so this is where you know you have a snapshot and you're just trying to figure out how many people have that disease at that time. So that is, um, only from 0 to 1 or as a percentage, because it's only a snapshot at that time. The problem with prevalence is that it doesn't provide any information on any new cases of a disease and does not reflect the occurrence or duration of the disease. Now you also have communal, communal acttive incidents and incidents. Proportion slash risk. Now they they're all kind of the same thing. They mean the same thing. So it's important that you, like, remember that they all mean the same thing. So this is where the number of new cases during the period of interest, um, over the number of disease freeze individual at the start of this time period. Um, now, this is obviously better than prevalence and odds because it's, you know, it gives a communicative kind of thing, but it needs follow up, and subjects often enter and leave the study. Um, and the most kind of robust way to measure, um, is incidence rate. So this is the number of new cases per unit time. So it's a number of new cases during the follow up period over told to a person time by disease free individuals and person time is basically the amount of time participants spend in the study. Um, on in Sandy, there are loads of actual videos of people going through, um, natural calculations for everything, but because of time, I can't actually do that. Um, but yeah, moving on to standardization. So, um, in standardization, we have two different types of standardization. Have standardization, which gives, um, un comparable incidents. Um, and it's used when specific data is present. So what standardization actually is. It's basically you want to standardize your whole data set to make sure that, you know, different data can be comparable. So, for example, in one in one city, you could have a really, really aging population and really, like all the people that are over 60. So obviously the chances of you having a stroke on MRI is probably a lot larger than if you have a, um, city with those of young people in it. The direct sanitation is where you standardize the data so you can compare it to each other. Um, the difference between direct and indirect. It's a direct. You have specific. You can use a standardized, whereas indirect, you don't. So you normally use national statistics. Um, so it's a bit more inaccurate. Basically, um, So yeah, so we're gonna come onto a quiz. Um, we're not gonna use Manti. We're just gonna put in the chat. So, um, first question is, what stage of the epidemiological transition model do you think the UK is in? Leo says D Yeah, that's correct. Yeah. So a lot of the countries that you that like UK for an example are Indeed, most countries aren't the only There are very few countries that are in stage five, which, um, includes Spain, for example. And Japan can be considered in that as well, but it's very few countries. Um, so, yeah, um d is correct. Next question. Yeah. So increasing vaccine uptake, um, decreases covid mortality. What is the exposure? So it's exposure is what's causing the outcome. So it is the vaccine uptake, which is correct out. So it's a Yeah. So, um, as mentioned before, they're different types of prevention. Secondary prevention is where you're basically treating the disease. So giving oxygen fluids and IV antibiotics is treating a patient with sepsis, so you're not preventing their risk factors, which is primary. Um, and tertiary is where you are returning them back to normal. So this is where you're actually treating them. So yeah, secondary is the right answer. Um, now, this is an odd question. So it might take a while to calculate if anyone remembers the odds ratio. Okay, we've got one on so of three anyone else? Yeah. So three is the right answer. So if you take a look at odds ratio, Um, so probability of student who studied using active recall is 15/20 which is 0.75. And you wanna calculate that compliment as well? So students studied without using active recall, Um, which is 0.25. So the odds is your probability over the probability of the compliment, which is three. So, yeah. Um, so that is measuring and describing diseases. Does anyone have any questions at this point? Okay, I'm assuming now, so we're going to move on. Okay, So, um, the second part of, um, this three part kind of epidemiology thing is designing epidemiological studies. So, um, in this part, we're We're just going to talk about a few different types of studies that you can do, uh, to investigate certain things. So you have descriptive epidemiology and analytical epidemiology. So descriptive epidemiology is where you, um, describe a problem. Basically. And you provide measures of frequency as mentioned before you know, your odds, your prevalence, your communicative incidents. Um, and the one thing that differs this from analytical or one of the many is that it does not test a hypothesis. And so this is more common in everyday healthcare and practice analytical. I mean, the epidemiology is often passing level, and it tests hypothesis. Um, so it can measure, you know, association causation, all these different types of things which will come to later, Um, and it's common in published literature. So, you know, in your journals and stuff. So, yeah. So these are the different types of descriptive investigations. So you've got case reports where you know you have this is just one single report on a specific disease or finding, um, a case series. It's just a multiple. A collection of case reports. Basically, um, a cross sectional study is, um, you know, your it's a prevalence because, you know it's a single snapshot, and it's basically your looking at the prevalence of something in a single point in time. Um, and it's commonly like a survey. For example, um, the longer traditional study is a prevalent prevalence over time. So it has a follow up, Um, and an ecological study is basically groups of people you're studying rather than individuals. Um, now, the one thing about ecological studies which you'll need to know is about ecological fallacy slash aggregation bias. So this is basically where, um, you might have a bias because a group of individuals is not representative of a single individual. Um, so that might cause bias because, you know, um, a group can never be representative of everyone, so yeah, So there are different types of data which you can use to, you know, kind of complete your studies. So you've got primary data, um, which is more accurate, but obviously, it's going to take a longer time, and it's more expensive. And you have secondary data, which is more convenient but less accurate. Now, in practice, you've got routinely collected data e g a census. Now, I think the UK does a census every 10 years, so that is something that they always do. So that's, you know, public knowledge, public information, and you have non routinely collected data. So this is your primary data now, um, in terms of data linkage, it is when you're looking at study is useful to link data together, such as hospital records and GP records. However, this is a major, major privacy issue. And, you know, if you go into the wards, even you often notice that, you know, the hospital cannot access to GP record, for example, um, but because of so many things, but mainly because of primary and technical issues. So even though data linkage in theory should be, you know, used throughout it can't be because of these specific things. So, yeah, in terms of study designs, you have observation, allow and interventional, um, study design. So this is more of your analytical study designs. So you have case control studies and cohort studies for observational designs. So case controls are looking previously at the exposures and outcomes. So the exposures and outcomes have already been, um, you know, carried out, for example, whereas a cohort study is you're exposing someone something, and, um, X looking at their outcomes, basically, So case control is history. Cohort is future. Um, an interventional, um, study designs are basically randomized control trial. So you're giving someone a placebo? You're giving someone a drug, and you're basically trying to figure out if you know the drug is useful or not for example. So yeah, so a little bit more about retrospective control studies. So first of all, you start with the disease status. So you have in individuals with the particular condition and you have controls. Um, and then you check for the whether they have had the exposure. So, for example, I think plumbing exposure is associated with asbestos, which is the outcome. So you get a list of all your asbestos, the patient's and all your non asbestos patient's, and you assess them for the exposure. And then you can calculate something called an odds ratio. So this is your odds of exposed among cases over your odds of exposed among controls. So this is another kind of formula to learn. It's different from straight up odds. Odds Ratio is different. Um 23 things to be worried of is that you need to be wearing a recall bias. So some people, um, may recall differently because of certain reasons. And you also need to be worry your selection bias so some people might be selected more than others. So, for example, if you're you know, in an area where there's a there's loads of plumbing work to be done. You know, the people who, um, are going to be selected, I'm more likely to be plumbers and, you know, more likely to have asbestos, for example. Um, so, yeah, just be wary of those two different types of bias nous, Um, and also, retrospective case control studies don't provide evidence of causation. So even though you know, it could seem like your exposure is linked to your outcome, this might not be the case like it could be just two separate things. But because because you haven't followed them through, it's hard to provide evidence of causation rather than a cohort study which you follow them through, kind of. So, yeah, a prospective cohort study is where you assess the exposure of the target population and you follow up, so you check for that outcome. So an example is, I think plumbing is associated with asbestos. So you choose the people, you choose the plumbers, and then you check 20 years later. Uh, so for the previous one, we use odds ratio For this one. We use relative risk. So this is the probability of an outcome in exposed group over probability of outcome in non exposed group um, And this provides a better evidence of causation. Basically. So Yeah. So, um, to calculate your odds ratio and relative risk I have mentioned before These are your, um, formulas. What I recommend is you kind of draw out this graph, um, with disease along the top hand and exposure along the left hand. Um, and then you write out the numbers because sometimes they give you only one number, and you have to basically fill in and figure out the other numbers. So to visualize it like this is a lot easier. Um, this is just a snapshot taken from incentive. They have videos of people calculating it as well. So yeah. Um, so now a bit more about Rand randomized control trials. So? So this is where you conduct an experiment? Basically. So you have a trial group and a control group, So first of all, you want to randomize your cohort. So you want to generate an allocation allocation sequence and you want to implement that allocation so you generate, you know, a certain number of numbers to a certain number of people, right? And then you would wanna There are many ways to do it in the way of doing it is giving them an envelope. Um, each person an envelope and they open it themselves and see what they get, or they don't even have to open it. They just get assigned a certain placebo or certain drug without knowing, Um so, yeah, this is what blinding is. Basically. So a single blinded test is whether participants don't know what they're getting, so they don't know if they're getting a drug or they don't know for getting a placebo. A double blinded test is whether participants and the helpers don't know. So the people actually, um, carrying out the investigation don't know. And a triple blinded test is whether participants, helpers and analysts don't know. So the people analyzing the results sometimes may also have bias. So you want to kind of eliminate that. So are there are two other biases that we need to know of performance bias. Um, where there's a systemic difference in treatment of care. So, for example, if you know that you are getting the drug or the the thing, that's you better they, you know, try and make do whatever they can to get better. Basically, and you also have detection bias. Um, where there's a system, a systematic difference in determining outcomes. So, for example, you could do more tests on the trial group till you get the outcome you want. So this is why it's important to blind your test. So you kind of eliminate all these biases. So in terms of when you're doing a study design, you need to know what your sample size is gonna be. Um, Now, there are different things that will affect your sample size here, three factors. So the first factor is your power. So what power is is your ability to be able to find a difference if it's if it exists, so you want to aim for about 80%? Um, you don't actually know how to calculate. You don't need to know how to calculate power, but just know that you need to aim for around 80%. And for higher power, you need to increase your sample size. So if you have more people, you will more likely be able to find a difference than if you have less people, because there's less chance you'll be able to find difference. Basically, the second, um, way that you can determine your sample sizes by difference of interest. So if you have a small difference that you want to pick up, you need to increase your sample size because otherwise it will be less likely for you to pick up that difference. So if you increase your sample size, it'll be more sensitive. And you've also got your alpha, which is basically how much do you want to rule out a chance of causing a positive finding. So alpha is basically your chance of getting a false negative kind of threshold. Basically. So people normally consider your alpha two B 0.5, and it doesn't change, really, If your sample size increase. Uh, if your sample size increases, your alpha will, uh, is decreasing basically. So, yeah. So, um, the two different types of errors that you can get with your epidemiological studies so you can get type one error, which is called your alpha. So this is your false positive as mentioned before. Um, so this is the probability of getting your results. If there was actually no real difference, 0.5 is usually the threshold. So if your P value is less than 0.5. That means that there's less than 5% probability that the results are due to chance. So if your P value was 0.1, that means it's statistically significant. Um, enough to say there is a difference and in your values and that there is a reason for this. Um, Type two error is your Byetta, and this is your false negative. So this is your probability of not getting any results when they're actually was a difference. So the opposite of a type one error, Um, so in order to reduce your type error, you want to increase your power. So increasing power increasing your increases your sample size. So if you increase your sample size, you're more likely to find a difference. So just to be aware of two different things multiple analysis and clinical significance So you have multiple analysis of, uh, investigation or study. You're gonna get more chance that you're going to get something clinically significant or a P value of less than 0.5 because you're doing multiple studies. So just be aware of that. And the other very, very important thing is clinical significance. So let's say you're running a, um, kind of study to do with changes in blood pressure medication. So you're giving, you know, a certain cohort, a BP medication. Um, and you find that that, you know, it's only diff, like changed by 0.5 or something like that. Um, even though it could be statistically significant, so this could be a statistically significant change. So your P values less than 0.5. It's not clinically significant in terms of that is not enough to kind of implement that drug because, for example, there could be so many other side effects for such little change. So there's a difference between clinical significance and statistical significance in this case, so just be worried with that. So we're moving on to the next quiz. Um, so this is the next one. And if you have any questions, also reply to them in the group, chat or text or write them out in the group chat. Yeah, okay. No one said anything. Oh, someone's saying they say, I don't know. Ok, right. So I'll explain this very briefly. So, um, basically Oh, sorry. At the end, I've put ecological study, but ignore that. That's not meant to be there. Um, so basically, this is, um, the CDC workers. Sorry, I'm just picking up the key information. But CDC workers checked the food histories of 20 infected patient's without break strain and compared them with the food histories of 20. Patient's infected with other hysteria strains. So basically, it's testing a hypothesis. So, you know, for a fact that this is not descript is this is gonna be analytical, Okay, because analytical, um, test hypothesis. Now you want to know if this is observational or experimental? So experimental is randomized control trials, observational czar, case controlled trials and cohort trials. So they're not, um, doing a randomized control trial because they're not, you know, giving something to the to the patient's. So it's between b and C. Um, and also it's basically you're trying to figure out if this was in the history. So if the exposure and the outcome had already been determined, or that you have you have found the exposure and are tracking the outcome in the future. So in this case, that sorry, the exposure and the outcome had already been determined. So this is a case control analytical observational question. So the answer is B. Sorry. Okay, Question six. Mm. Yeah. Yeah. So the correct answer. See? So, um, a clinician is conducting a randomized trial, but delivery does more screen tests in the group receiving his target intervention. So this is to do with detecting the outcome. So that's therefore it's detection bias. That was an accident, but yeah. So a type one error is a false positive, false positive, and Oh, great. Someone's put on the group chat. Ready, right? Yes, it's a, um it's Oh, I've put here a false positive. It's going to be see false Positive. I'm so sorry. Ok, um, what about the next one? What is the type two error? Yeah. Yeah. So it's a false negative era. Uh, false, false, negative. So any questions before we move on to the last part? All right. OK, we'll move on to the last part. So this is after you've done your study. You know, after you've done your measuring and describing diseases, you go onto interpreting your findings. Okay, So this is evaluating your findings after you've done your study. Okay, So you've got systematic and narrative reviews where you can kind of evaluate your study So systematic reviews are basically you have a research question and you basically look through all the you know, kind of studies that's been happening, and you do a structured search and find the most relevant ones and basically form a review of it. Now this is obviously going to be very robust evidence. And because it's a structured such you have specific criteria that you need to meet in order to create this review. However, it is very time consuming to do it takes approximately two years, and you can understand the the difficulties with this if someone wants to look at a review and it's two years outdated, with medicine being such a changing, you know, um, environment. And you've also got narrative reviews, um, which are a lot easier and faster, too, right? They just bring all the published literature together. However, there's no structure to it, so there's potential for bias all right and last letter analysis you can use. So this is literally your most robust evidence that you can use. So you combined the findings of multiple studies into a pulled estimate and you communicate your findings using a forest plot. Now the image on the left is your for is an example of a forest plot. Um and and certainly they have a whole way of doing a matter analysis, but I can't included in this session here, but you can go and look at it. Um, but I will talk about a little bit about the limitations of meta analysis. So obviously with anything that you're combining together, So matter analysis combines so many different things together that may not have been intended for mass analysis. So studies are going to be different. OK, there's heterogenic tea. So, for example, clinical means that there's different patient groups. So the different way of choosing the patient groups might be different. You know, one study might be recruiting patient's, you know, from a different side of the world, which can severely, like, you know, impact your results and mythological um, differences as well, such as, you know, you're blinding. You know all the different methods that you go through your generation of your allocation and stuff. And you also have statistical where you have different reporting methods as well. Sorry. So another limitation is that with meta analysis, you need to figure out which publications are going to have more waiting's. So, um, there's this thing called fixed effects versus fried, um, effects. But you don't actually need to know this for your exam. Just know that limb waiting is a limitation of meta analysis and your publication bias, Um, which is mitigated with the publication funnel plot. So this is where, um, some publications may have really high, you know, bias bias, NUS. So you can mitigate that by, you know, having a plan, a funnel plot. After you know, you've done your whole experiment, You also to figure out what the endpoint is. Okay, so, ideally, you'll have a primary endpoint. So this is what the study is originally meant to find out. So, um, then you have your secondary endpoint where your endpoint is slightly different to your primary endpoint. Um, and this could be because, you know, you've realized in the end that you know your endpoint is not feasible. So, for example, if you're trying to measure, you know, mortality and you know you've tried to measure that and it's not feasible you can measure, for example, you know, other complications of a disease, for example. You also have composite end points where you have multiple endpoints added up together. So an example of this is you could be trying to find out if, um, someone had, uh, an ischemic stroke or something, or your control group or your cohort Study hasn't ischemic stroke. Um, but you can combine this alongside, you know, an m I or other kind of cardiovascular kind of related incidences. Right? And you also have safety. Um, so an endpoint, if you're if you're study isn't safe, you're going to have a huge number of adverse events. Um, this would require investigation, because obviously, if you're study was not safe to begin with, why did it go through in that kind of way? Um and so if you know, it gets to that point, you need to make a decision. Whether you know, carrying on with that study is going to impact or like, it's going to have a offset. Um, the adverse effect will offset that efficacy. Basically, so yeah. So, um, this is one way of basically plotting your results. So this is a survival analyst. Uh, analysis. Sorry. So, um, all you need to know about this is that you plot with a Kaplan Mayer plot. So it is shown here where you have your X axis is time and your Y axis is overall survival. All right. Okay. Um, now, this is the Bradford Hill criteria. Um, this is taken from the l map generalized notes that we all got last year for exams. So I would suggest you have a look at that, because it is quite comprehensive. But basically, the Bradford Hill criteria is a criteria used to basically, you know, try and figure out if something is a cause of something else. Um, and you don't need to have every single one of these, um, for an example. Experiment is where you basically experiment to see whether one thing is the cause of another. You can't really do that with everything, because it would be unethical to do it with certain, you know, investigations and stuff. So, um, this is your the 123456789. They're nine Bradford Hill criterias. Um, just have a look at this, um, table in your own free time. Okay, So, um, we've kind of talked a little about association about a correlation causation, but they're all kind of different things. So just to clarify so association is the relationship between two variables. Correlation is the linear relationship between two variables and causation is your bad for Hill criteria. So that's one thing, cause the other. All right, so a correlation, you know, it's not causing it doesn't mean causation. It just means there's a correlation between things. Um, there are loads of graphs of mine that show correlation between two completely different things. Um, but that doesn't mean causation. So you've also got this thing called validity. So all of these things are basically trying to evaluate your experiment to see how well it's been conducted. So internal validity, um, is where your association between your explosion outcome truly exists in the study group. So this is within the study group. An external validity is where it can be applied to external population to um so this is external validity is very important because, you know, if you're doing a study design and you want to kind of try and relate it to the rest of the world or the rest of the patient cohort, it is really important that you have external validity because it determines a real life impact of a certain finding. It's also called generalized generalized ability. All right. Okay. I know I've banged on about bias nous, but it is actually really important. Um, so we're just gonna talk a little bit violence as well, Because you need to evaluate for bias nous in your, um you know, when you're evaluating a study and interpreting it So, um, a selection bias nous is where your chance of inclusion in the study is accidentally related to both explosion outcome. So, for an example, a case control study to investigate association between eating beef and cardiovascular disease. Um, so if your controls are selected from people who visit a steak shop, you have if you eat beef, you have a higher chance of being part of that control group. Um, so you underestimate the effects of beef or cardiovascular disease, and if you're controlled, are selected from a Hindu population, if you be, if you have a lower chance of being part of the control group, so you overestimate the effects of beef on a cardiovascular disease. So that's what selection bias is. You know, under estimating and overestimating your exposures and outcomes, there are also three different special types of selection. Bias is which you need to know barks and bias. Um, where basically this happens when you do educates control study in one single hospital and that single hospital might not be representative of, you know, every hospital basically, um, and healthy worker effect. Um, where active workers are more likely to be healthy compared to those who are retired and stopped working. So the people who you kind of choose for the study are more likely to be active and non response bias nous where people, um who do not respond are different to the people who respond. So, for example, if you're doing a survey out on Oxford Street, I mean, this is a very big generalization. But you know, people, some people who you go to interview are probably very different to the people, don't want to be interviewed and want to just move on, move on with their day kind of okay. And then you've also got information bias. So this is Ms classification of the exposure of the disease or both. So you've got interview a bias where the interviewer kind of has a bias towards, you know, the drug. So this is why it's important to double blind. You've also got recall bias because if people you know know that they're taking a certain type of drug, they might be more conscious of their health and, like, report every single little health condition or like, kind of symptom that they have. You've also got re response bias, um, which is a socially acceptable response. So, for example, um, you might be asking someone a question, and they might think that, you know, society would see one specific answer as the more socially acceptable answer. So they go with that one, even though they think it's something else. And diagnostic bias take a diagnostic. Suspicion, Bias, Um, is where you think that you know, you've got something, and therefore you're more likely to alter your perception on what you're going through. So yeah, um, so, Ms Classification, you have differential and non differential misclassification. So misclassification is a type of bias, basically So differential misclassification is where your exposure or outcome are unequally misclassified. Um and it means that your result is biased towards all the way from the null. It doesn't. It can be either way and non dysfunctional misclassification is where your exposure or albums are equally misclassified. So your non differential misclassification is always biased towards null. Um, it's a bit of a confusing concept, but just remember that differential it's always biased towards or away from the null and non differential is towards the null. So you've also got other types of bias. So you've got like, Whoa, I'm not because I order it or Bigan effect. Um, this is where you know, someone thinks that they're better than average, so they have kind of maybe a security complex or something. So they, you know, kind of big themselves up a bit, too. You know, the their outcomes and stuff. Um, and you got cause and effect. But people participants behavior based on what they are given, basically, So you've also got confounding, which is another type of bias. So this is quite a busy slide. But basically, if you look here at the, um, bottom part, um, where it says all three conditions and then three bullet points after that, confounding is basically where, um, the effects of an of an external variable that accounts for the effect of the study of the exposure or that mask and underlying true association. So it's basically where there is something that, um, effects either the exposure or the outcome, which causes the results to kind of change in a way, um so in order to assess confounding you need to see is the factor associated with the exposure is the factor not caused by the exposure. And this is the factor associated with the outcome in the absence of the exposure. So you need to ask these three things to assess confounding variables. Um, so this is very important and you've also got something called effect modification. So, um, um, the difference between confounding an effect modification is that effect model, uh, effect modification is an explained confounding that you can put in your report and you can be like this happened and this changed my results. Whereas confounding is kind of more of a bias. So there are three tests that you can take the breast low data ask you test and interaction terms and regression models which you don't really need to know about. But what you do need to know is synergy and antagonism Where, um if the effect model, if I it increases the effect of the exposure, it's synergy. And if the fact modified decreases the effect of exposures antagonism. Okay, so I think I've run over, so I'm gonna end it from here because the last question is really long. Um, but if you have any questions, um, please feel free to put it in the group chat. Otherwise, thank you for coming, everyone. Hopefully this was useful to you guys. Yeah, yeah.