 In this lecture I want to talk to you about the different types of research. You might page through your journal, you might have some data in practice, and you're thinking about the different types of research. Now there are various classification systems, but one way to look at it for instance would be just to divide them between observational and experimental. And observational ones, we're thinking about things like case series, you're just going to describe what you find, whether you do that retrospectively or prospectively. You're just describing what you find on the experimental side. If we think about humans, we call those clinical trials, for instance. You're actually going to divide patients or your subjects into groups, and some are going to get one form of intervention, others are going to get another form, and those are trials, experimental trials. Now there's also reviews and meta-analysis. Lots of stuff to talk about. Let's go. So here we are now. iPython notebook. We see the classification system there that I talked about. We're going to have these observational studies where you see them. And then also the experimental studies over there, otherwise known as clinical trials if we involve human beings. So first of all let's start with these observational studies. I'm going to talk about four main types here. Case series, case control series, cross-sectional studies, and cohort studies. Now I said this was one way to classify these. Certainly this is not the only way, and you will come across study designs, and you will design studies on your own that actually have a bit of some of all of, not all of this, but actually mixing one or two of these. You've got to have something in the back of your mind though when you start. So let's stick to this classification system. Let's start with the case series. Now it is perhaps one of the simplest forms. Well I think actually the simplest form of study design. We're just going to grab a bunch of subjects and we're going to grab a bunch of variables and we're going to collect those and just describe them. So one very good example here you can see is the clinical audit. So you might just look back at some surgery that you did and what the outcome was for that very specific type of procedure. So we're not, there's no research hypothesis. We're not trying to really get to an answer. One good way to, one good use for this though is just to get some feel for your data and to help you out with designing future studies. Now the actual data that we collect. This is very important as far as I'm concerned. The actual data that you do collect that can be performed in two way. And you see there the retrospective. There we go, retrospect and prospect. So I don't want those two terms to be confused. They can be used in slightly different ways. The first way to introduce that to someone though is just in this form. And that is when you look at the data, you can grab the patient's files or your subjects files or notes that were taken on the subjects, whatever. You can identify some data points, some variables that you want to capture. You go through those files and you go look for that. That might be that someone wrote notes and you have to pick through the notes to get that data points. And that is called a retrospective capturing of data. A prospective capturing of data was before anyone ever decided on a study, there was a specific form that had to be filled in at each and every encounter. So say for instance we were concerned, we have this unit that deals with hypertension. There's this form that every time a patient comes in that form has to be filled in and in that form there are blank spaces and those blank spaces have to be filled in. So those data points are captured prospectively. It's not left up to someone just to make a bit of notes and hopefully the data that you want out of it appears somewhere in there. So that is retrospective and prospective. So you needn't even have a study in mind when you develop these forms that have to be filled in during each encounter. But that data, you decided on capturing that data. Someone else deciding on capturing those data points at each and every encounter. When you now go back and you can look into that database, afterwards now you decide look I want to do a study. I'm going to go back and grab the patient's files and the data points that I want are actually part of the form or database that existed before. So that data was captured prospectively. So that is one way to look at retrospective and prospective data collection. There are some other uses of the terms, but I think this is the clearest way to look at retrospect and prospective and prospective. So let's look at the case control series. Now, this is a bit different because we're going to have more than one group. It's not just the case series. We're going to have a series of subjects and a different group, which is going to be our control. So what do we do? We start with the presence or the absence of an outcome and then we look rearwards in time. And we collect data from that point and before that point. What do I mean by that? Let's use a little example. I've operated on a set of patients and they have one or two outcomes. Say, for instance, they develop wooncepsis or they don't. So they all had the same interventions. This is not a trial. So they had their normal surgery, some develop wooncepsis and some don't. Those would be my two groups. The ones with the wooncepsis, my case, my control series, those they don't. So there was this event. They got wooncepsis. Now, from that period of developing the wooncepsis, I'm now looking back. And I'm going to jot down different variables for the two groups. I'm going to try and identify differences between the two groups to perhaps look at what's there. Is there something I could identify that would show me patients that would turn out with wooncepsis or those without. So that is what this retrospective. So I'm really using the term retrospective here, but quite in quite a different way from what I used it before. Retrospective here means I have this outcome with wooncepsis without. And from that point, I'm looking backwards in time. But there's nothing to prevent me from having prospective data collection on this retrospective look back. Okay, that sounds funny. I might have a database in my unit. Every patient who gets this type of surgery, this type of intervention has a form filled in during the first visit, during the pre-op visit, during the surgery. So in those forms are blank spaces and certain forcing the user to fill in those blank spaces each and every time. So those data points were decided on prospectively. Many years later, I might decide to do this case control series. But the data was collected in a prospective manner, although the case control series always looks retrospectively. So I hope you understand there the two uses of this term. So don't get these confused. Important thing is I've got some outcome. I might have done some biological experiment and there are, I mustn't use the word experiment yet. Might be something that you do routinely in the lab. There's two outcomes. I'm going to make two groups with those two different outcomes and from that two outcomes, I'm going to look backwards in time. And I'm going to collect data that existed up until that point and that will make it a case control series. These are very common for novice researchers to start off with. They're quite easy to do. Now you get these special types of nested case control study. Cases and controls are taken from prospective cohort studies. So this muddles the waters a little bit. I don't want to go too deeply into that, but you can clearly see one can mix and match some aspects of these different types of. So now you know two case series and case control series. We move on to cross-sectional series. Now, there are many, it's really, you see, they're a heterogeneous group of studies. What the cross-section means though is it is almost like a snapshot in time. We're just looking at something that happens right at one specific time. A beautiful example of this is a survey. You can hand out a survey to a group of subjects. They fill it in. They might have a Leica type scoring system. Totally disagree, disagree, neither agree nor disagree, agree and strongly agree. Those types of surveys, those would be a cross-sectional study. You'll see there that you can also mix and match it with some other ones. Read through there can get a bit convoluted, but definitely if you want to remember cross-sectional studies, remember the example of a survey. You'll also note that there is also a problem with confounding. I didn't mention confounding before with the case control series. Confounding is where you try to make your two groups very clean. In other words, they must be similar in each and every way and only differ in one specific aspect, and that's the aspect you want to investigate the difference between the two groups. Confounding means you might find a difference in that very specific parameter you are dealing with, data point or variable that you're dealing with, but the true cause for the difference between the groups is hidden from you. It lies somewhere else and it wasn't part of your investigation. So that would be called confounding and a cross-sectional series can also suffer from a bit of confounding issues. Let's move on to the cohort studies. Now a cohort, you'll see a cohort is a group of individuals where they come and treat such as a disease or a risk factor and they remain part of that group. Usually of an extended period, it might not be that extended. You might just follow them, follow subjects up for a couple of days, but they form part of this group and you follow them up afterwards. Now a cohort in case control studies, we differentiate those two by the direction of inquiry. Now this is a bit simplistic, but the format of cohort, it starts at a point in time and then looks forward in time. Whereas with the case control series, remember we had this outcome and we looked what happened before that outcome. So that's a nice way to differentiate the two. It's a bit of a simplistic way to look at it. Cohorts can definitely be seen in other ways, but this is a nice way just to remember the difference between a cohort and case control series. Now we move on to the exciting stuff trials, clinical trials or experiments. Now the first one I want to talk about is trials with independent current controls. So what you're going to do is you're going to have two or more groups and you're going to do an intervention in one and you might not do in the other or sham intervention in the other. There's some difference between those, but you want to choose these two groups that they're nearly the same in all other respects. Now that makes it very difficult to see your groups can differ in age, gender, body habit, and many other issues. If that's the case, you can do some subgroup analysis. What makes this concurrent though is you are going to divide your patients, you're going to do some intervention on one group and not on the other group and you're going to collect your data points at the same time for both groups. Both of these groups are going to run concurrently. Now it might be that the subject and or the observer are unaware of which group they fall into and those would then be if both of them are blind and that's your double-blinded trial and also get the blinded. For instance, if only the subject is unaware. So those are the quintessential double-blind trials that we know of dividing a certain set of patients into two groups. You'll have this random bunch of medication to give to them, packets made up between ones with active substances and ones with just a little placebo and at random patients will be put into one of the two groups and they'll either receive the placebo or the active drug and we can look at the outcomes. So the subject doesn't know what they're taking. People who do the data gathering after taking other medication or doing the intervention are unaware of which group the patients in. So they are blinded as to as as to what group the patient belongs to the subject belongs to and they just then without bias or at least less bias can gather that data. So it might also be I've mentioned the word randomized. So we're talking now about the epitome or the or the the highest pinnacle of this top medical research. We're going to have a randomized control trial by truly randomizing similar patients to two or more groups and then do different interventions or just sham interventions on one of the one of the groups that might then be your control. So a randomized double-blinded trial that's that's about as good as it gets. Okay, we move on to the trials of the self controls. Here the subjects are from their own control. So you might gather gather data and do the intervention gather the same data again. So before and after. So you do have the two groups that before and after but each patient is his own or her own control. Now they similar to cohort study. You can see they accepted for the fact that some intervention took place. So this makes it a trial the most elegance up there the cross over study you can see there. So you'll have two or more groups and you'll do the intervention on some of what some groups are not the others and then you'll repeat the experiment but you swap those interventions around. And so you still have the fact basically that the patients form their own their own controls in the various groups. We move on here to the trials with external controls. So there you're going to have your group in which you do your intervention and you're going to compare that not concurrently but to another group. So for instance, historical controls so data that's already available. So you're comparing your intervention to some other intervention from another series from other research that already exists. So it's very common for instance in oncological research there where an effective treatment doesn't yet exist or you're moving on to new therapy and you're going to give that to a set of patients and their controls are going to be external in other words you're going to compare them to a group that already existed. Let's move on to these uncontrolled trials. So still you're going to do some intervention but here there are no controls and the hypothesis really here is that there'll be various outcomes that you can gather from the data but there's no attempt made to evaluate the intervention itself it's not being compared to either placebo or any other form of intervention. The meta-analysis there, very common thing to do these days you're just going to take some pre-existing research and you're going to gather that data combine it and do new analysis on that. It's very helpful when we have studies with say for instance with only a few patients in them and they're quite underpowered or inadequate size that we can combine all of those, we can combine and do new analysis and get more fruitful results that way and then together with that we have the review. So it's basically going to be a discussion around the recent literature that has appeared on a certain subject and we're going to do amalgamate them and we're going to do this qualitative analysis of them very useful if you want to get up to speed with a subject you can go to the literature and look for review article this is going to basically summarize in a qualitative way the research that has happened up to that point you can of course mix the two, the meta-analysis and review. So that's a look at the different types of research that you can do, it helps you to decide when you need something at least what that means or helps you decide when you want to do some research what type of research you want to do. As I say this is a bit of a simplification it's not the only way to classify these and things can be mixed and matched a bit but at least you have some overview now of the different types of research that you can embark upon.