transcript.txt

Hello today I’m going to talk about you know my view of study design this is a fairly long lecture but I think it’s time well spent because what I want to do is sort of present. The whole array of study designs that we’re going to eventually go into in more detail in future lectures but I think it’s...

Hello today I’m going to talk about you know my view of study design this is a fairly long lecture but I think it’s time well spent because what I want to do is sort of present. The whole array of study designs that we’re going to eventually go into in more detail in future lectures but I think it’s helpful for students to sort of see the whole collection of study designs all at once so it’s easier to make some comparisons later on. Just quickly the basic design strategies in epidemiologic research can be broadly categorized according to whether such investigations focus on describing the distribution of disease which would be considered descriptive epidemiology or listening determinants analytic So I remember we talked about describing distribution of disease that’s who when and where and then elucidating it’s determinants that’s what what is causing the problem so. It’s kind of keep those terminologies in mind as we go through so descriptive epidemiology is concerned with describing the general characteristics of the distribution of at least particular relation a person place some time and the studies are primarily useful for the formulation of hypotheses that can be tested subsequently using an analytic is aren’t discovering the the what and why. And I’ve mentioned this before but the feature that distinguishes analytic from descriptive studies existence of a comparison group are sometimes called a control group use of a comparison group allows testing of the epidemiologic hypotheses. So just a really quick summary descriptive studies generally have properties and do not have a comparison group analytic studies can test a hypothesis and they have to have a comparison. So I kind of like to put these different type. Studies in this list and starting at the top would be the weakest study to find and at the bottom would be the most rigorous or the strongest spending design so in terms of under the category of descriptive studies we have population studies which are also called correlational or ecological studies sort of depends on the age of your textbook. And then we get into individual level studies which include case reports case series cross-sectional surveys then we bump into analytic studies. And there are two major head in there’s absolutely a tional studies which include case control studies and cohort studies in those court studies can be the retrospective or prospected and then the Cadillac of study design our experimental or intervention studies also known as clinical trials So again sorry in epidemiology a lot of a lot of things have more than one name so again we have a population studies which are also called correlational ecological and we have experimental studies which are also called intervention or clinical trials. So I’m going to start out by talking about these descriptive studies first and I’m not going to spend a lot of time you’ll get information in this lecture that’s going to cover these descriptive types of studies I’m going to also go over the analytic studies quickly today but then in a lot more detail in each one will have its own separate lecture further on in the course. So again just back to our road map to keep everyone on track and in the same place when we’re talking about descriptive studies right here at the first block under descriptive epidemiology hypothesis generation and there are our study types population correlations flashier logic we’ve got the case study the case series and cross-sectional So again really. Person place in time or who where and when and just doing descriptive level work so that’s where we are right now for the next several slides. OK I think I’ve already said this We’re in a start at the top and work our way down and the study design has become more rigorous as we go down the list. So the big thing with an ecological study or population level study is that it involves aggregated data on a population level so what that actually means Has that we’re not collecting individual level data we’re not collecting data from every single person in this population were using only aggregated data on that population so. There you do get a lot of value from an ecological study it can provide information toward a possible causal association or just statistical socio. However the weaknesses that clearly you can’t control for other confounding factors that makes plain some or all of the association and I’ll give you an example here so you’ll have a better idea of what I mean by ecological. So you know and now of analysis for an ecological correlation study is population or a group of people not the individual so a group that are selected for a study might be like the residents of the people particularly geographic area such as a nation or a state or even smaller groups such as a county or such as track so the information about exposure and outcome is collected at the group level we’re not getting individual level that we’re just taking aggregated data from a certain usually fairly large population or group. So this is an example of data from an ecological study so on our X. X. axis we have the average percentage of people who eat five or more servings of fruit or vegetables per. On the Y. axis we have a percentage of people who are overweight so. Each dot here. Is just representing within this. A state because we’re looking at the relationship between percent overweight and percent eating five or more servings of vegetables in a U.S. state so Each dot represents a state so if you look at the DOT one of the lowest people are twenty five percent of people are eating five or more vegetables a day and about fifteen percent of the people in that state are overweight. So you can see that we’re not cutting data on every single person in every single state we’re just taking general average percentages for each state in we’re graphing them and then the black line is just sort of a trend line that can be laid over all those individual that a point so you can see just a rough trend line of where. What direction the line is going so in this case you get an idea that as people eat more vegetables your percentage of people eating five or more vegetables as that increases the percentage of people in that state or population who are overweight decreases so there’s and you can sort of make a general. Draw a general conclusion that if more people in the population. Five or more servings of fruits and vegetables a day you’re likely to have fewer people in that population be overweight so again you can’t draw any major conclusions about any individual person but you can say hey this looks like eating lots of fruits and vegetables is good way to help control overweight. Or your your weight in very rough crude terms. But it still gives you an idea and it may be enough to generate a hypothesis to go on to. Design. More rigorous study to go on and in trying to elucidate this relationship more clearly. Now ecological studies are prone to something called the ecological fallacy and that’s where and this assumption that the characteristic is definitively associated with those experiencing the health related event so if we go back to the prior slide. I can’t I can’t conclude that you know if I eat either more servings of vegetables a day you know then I only have a fifteen percent chance you can’t assume of being overweight so you can’t make that assumption that that what you see on a population level will actually apply an individual level that’s called a logical fallacy. So here’s another example where they did a coronary heart disease and they were looking at that as the outcome and they were graphing it against average income by city so within the cities that were selected it was found that chronic heart disease was higher in the richer cities than in the course it is so can you conclude that chronic heart disease is actually a disease of affluence. Well no because the problem was was within those cities. Within the cities the poor people had much higher rates of coronary heart disease than the rich ones so you can’t again take. Your general population level characteristics and apply them on individual it’s just sort of very crude types of studies. More often and not that you’re just roughly comparing one geographic area with another or one country with another country. To get sort of snapshot looks at what’s going on in very very crude tools but again they’re fast in their fish and in their cheap. So the next level that we’re going to you know rigorous to move on to a case report and then a series of case or court just involves a profile of a single individual that’s just one person with some condition usually this case reports are written up by physicians or clinicians of some sort and that they see an interesting case and they want to write it up and get it published oftentimes you’ll see resonance on the authors because they’re hungry for publication and hungry to prove themselves medical resident. OK series is just a series of case reports so OK series of of a small group of patients with a similar diagnosis so in this case you’ve got a you have a clinician and a some sort of specialty clinic and they keep seeing this very strange presentation you know a certain type of cancer among you know maybe a group that have a similar occupational type exposure so they’re like wow that’s pretty cool I think that’s worth writing up I’m going to going to write it up that’s actually was sort of the first tip off for the age you know on the long pathway to discovery AIDS was you somebody wrote up a series about the car posies The Posies are coma I think that’s the right term Now interesting type of. Fungal long presentation I think in age patients and so somebody eventually said well it’s weird I’m seeing a whole bunch of these cases all the sudden and I haven’t seen any in years so I think the right amount series and they can provide evidence for larger scale studies and so again they’re very good at hypothesis generation. And these are you know again some actual data on this or a series of cases of Legionella you know it may be that. You don’t know a lot of information but you have a diagnosis and so you might look back through your records and say wow we have been seeing more of these cases and. Then usual than expected so I’m going to take a little time and pull out some data and summarize it and see if I can’t get it published and very least you would want to do it and you know hospital rounds share this information call some other hospitals called State Health Department say hey you know this is what we’re seeing here at McLaren hospital or this you know this part of the state are you guys getting any calls like this from anywhere else is this just a freak incident or is a see you know something bigger underlying that we need to look into. Now the last of these descriptive types of studies is called the cross-sectional survey. These differ from the ecological in that the unit of analysis here is the individual I would be collecting information on the eating habits and the body weight of every individual in this group and twenty now. Not just using aggregated data from large populations so this is our little surveys or big surveys that are conducted over a short period of time a few days or weeks you know developed this is individual there’s not a follow up period that’s why it’s called cross-sectional you just it’s also sometimes people call it a snapshot I’m just going to snapshot of what’s going on and. That’s all there is to again pretty straightforward. Has some advantages and disadvantages. The nice thing is it is conducted over a short period of time meaning it a lot of information. You can look at several associations at once you can put as many questions as you want to your survey you. Could be examining several different. Associations that you might be interested in several different exposures or outcomes it gives you some nice prevalence data because again you’re looking at existing cases remember prevalence and you don’t worry about the last follow up because there is no follow up it’s just a snapshot of course there are some problems of you know nothingness there’s always a trade off so for weaknesses of cross-sectional survey. The biggest weakness really put a star by this is that you’re you’re not able to show if the exposure actually preceded the disease I might be asking in my survey do you smoke yes or no how much do you smoke and do you have asthma yes or no. Do you have. Obstructive pulmonary disease do you have lung cancer and so I could get all sorts of information but there’s no way because there’s a single single snapshot there’s no way to show if the smoking actually came before the disease or maybe the duties came before the smoking so you can’t. Establish any sort of temporal relationship meaning a relationship in time so that’s the biggest weakness of cross-sectional survey and of course response bias we haven’t talked a lot about bias yet but you know just instinctively that certain types of people tend to respond to surveys more than other types of people when you get the phone call and we want to just only take ten minutes of your time just quick you know when you answer these questions we really need this information half of you sitting up there you click is going to hang up the other half of your like well I don’t all right all right I’ll answer your questions right but there are some fundamental characteristic differences between types of people who are more likely to respond to a survey than who are in. If I said well you know I’ll give you ten dollars gift certificate to Starbucks if you answer my survey Well again some people are more likely to take that than not but any time you have a fundamental difference in the people who are responding versus not responding to your survey you can introduce bias into your survey. OK so this diagram won’t make a whole lot of sense with the cross-sectional survey we’re going to use similar diagrams for other types of surveys which will make a little more sense but the main point here is that there’s no direction of enquiry you’re not this is just that one a single snapshot in time for which you’re collecting information you select your study subjects you administer the survey then you sort out the ones with the outcome of interest for Says the ones without the on outcome of interest and you can just get some general idea. General associations may be present but you’re not you don’t have the ability to determine any sort of temporal relationship meaning again that the exposure would precede the disease and the reason that’s important is if you take a look at your road map. If what you’re eventually trying to prove is some sort of causation it’s very difficult to prove causation eventually without having established that the exposure preceded the outcome so that’s why it’s important. That sometimes you don’t need that sometimes you just want to do a quick survey you just want to answer you know a certain question or you’re OK with a very limited targeted targeted audience and you don’t need a huge study so these cross-sectional surveys can be very valuable a lot of you know quick market research and satisfaction surveys after you purchase something things like this. Where they just sort of want to know hey just tell us right now how you feel about this then the surveys are great for that they’re quick easy and and relatively inexpensive. And this is just sort of another way to look at similar data these are two maps and maps are wonderful things and you know at the last several years the availability of mapping software has really transformed a lot of the ways that we can present data in a much more easy to understand visual way people get maps when they can’t get you know a giant table full of numbers so this again is just sort of showing that this is a county in Michigan Genesee County the blue map is showing that. People who don’t sleep very much people get less than seven hours a night of sleep of sleep so the dark blue that sort of inner city parcel Genesee County which is where when Michigan has these folks are you know well over half of them are not getting even seven hours of sleep. And then they have a second map here which is. You know the survey from the same people they same you know some of several of the same people are saying they have very high levels of fear of crime so they’re saying this is fear of crime scale average by zip code in the darker the orange color the higher your fear of crime is so again we can say Well generally people who are having you know not sleeping also have a fear crime or people have a fear of crime or not sleeping we don’t really know which came first right chicken or the egg All we know is that they there seems to be an association at least you graphically with these two. These two characteristics. But wait to make served. More rigorous is to conduct the same survey. A number of times so if you conduct a cross-sectional survey routinely then you can start to. Make broader interpretations of the data that you can pull from the surveys a couple of serial surveys that you may be familiar with the first one is the U.S. census or whatever country you’re in you know a lot of times or some sort of general population such as that takes place so as you know if you’ve ever been involved immunity census takers they want to find every single person in the United States and have them complete this survey so it’s very much an individual level data collection another survey that we use a lot in this country is behavior Risk Factor Surveillance System or sometimes you’ll see it is a B R F S S S And that’s a survey where the C.D.C. sponsors the survey and states can docked it and states can also add on to have their own state specific type questions onto the survey it’s a telephone survey there’s a National Health Interview Survey National Hospital the surge survey speak to your health survey these are surveys that are very detailed and well structured surveys and they’re conducted on a regular basis so again you can start to see relationships over time when you conduct a serial survey and this is just an example of a county level survey again this was Genesee County that conducted the same survey in two thousand and three two thousand and five two thousand and seven and these are percentage of respondents who engage in physical activity of at least twenty minutes three times per week or more and so you can sort of see people in the purple line is flint people in Flint in general are not exercising as much as. Overall county or people who are outside of the city so this you know could give you some indication of. You know that there are there’s a lack of education within the city Flint maybe a lack of access a lack of safe places to exercise. You know could indicate all sorts of things but for whatever reason living in the city of Flint folks living there have in general and a declining. Ability or desire or whatever they just don’t exercise as much as folks who live outside the city of Flint So this is information that city planners can use if they want to try to. MAKE HIS of all activity more a part of people’s daily lives in Flint. And again this is a you know this is a two surveys but you can start to see if you ask the same question in the same way then you can start to see whether perhaps if you some of your health education efforts are having any sort of impact this is frequency of condom use and having sex with a new partner survey was conducted in two thousand and five which is the those are the red bars two thousand and seven and you can kind of get a quick. You know a quick idea that whatever program they’ve instituted in Genesee County is having a positive effect because. You know if you look at the category of always you see an improvement in the number of people who are always using a condom and you see the corresponding decrease there and rarely or never category that that folks who rarely or never use a condom that percentage is decreasing so again crude data but you kind of get if you conduct your survey more than once in a scientifically rigorous way you can start to get some valuable information. OK now we’re going to move on to the analytic study. Then again just so people don’t get lost you are here the yellow star is moved over to the second big column which is the analytic studies or hypothesis testing study and we’re going to talk about case control cohort in experimental studies this is where we get try to answer the question why you know is there an association between this exposure and this outcome and what how strong is that and so on. So what is analytic epidemiology it’s it’s an analytic study that attempts to identify causes are risk factors that explain health related states or events and test specific hypotheses that’s often developed. Based on information gleaned from a descriptive study so again descriptive studies generate have proxies analytic studies test them and we’ve mentioned before the key key description of an analytic study is that there is a comparison group so you’re not just looking at. Only people with disease you’re looking at people with disease and without disease or people who are exposed and who are not exposed so the two main categories of analytic study design the first one is observational and the second one is experimental. So when we’re talking about an observational study design the investigator does not have control over the exposure factor and is usually unable to assign subjects randomly into one you know either exposed or unexposed category so literally an observational study to design means that something happens and you as the investigator you observe it right you don’t intervene you don’t create the situation these are natural situations that happen. And we as investigators try to learn from them right we don’t go out there and give people cancer we go out there and we find people who have certain types of cancer and then we try to figure out why they got cancer and or conversely we don’t go out there and cause a chemical spill but if a chemical spill happens we can choose to follow those people who are highly exposed to see if they have an adverse outcome in the future so they’re things happen we observe them that’s observational study design investigator does not control have control over the exposure factor and you work with the people who are naturally present you don’t get a chance to randomly. Assign people into one category or another so that’s observational experimental is or like a clinical trial. The investigator in experimental design does get to control who is exposed to a factor of interest or not and they can assign subjects randomly into study groups so we’re going to have. You know entire lecture just on experimental study design and so I’m just going to give you guys. Quick overview of these three actually there’s three there’s two types of observational that we’re going to talk about and one experimental but I want to just kind of give you guys sort of some basics now let them sink in for a week or two before we go into these and in the excruciating detail that you’ve come to expect. So. We’re going to just talk about observational studies at this point these are the ones they are naturally occurring situations and the most familiar in most feel heard of a case control study right and we’re going to go into this quite a bit of detail have a whole lecture on it so their case Contro. Studies in there are a cohort studies so. It cohort study and put it like this a case control study is always retrospective. Coord study can be either prospect of meaning that the exposure has occurred but the outcome has not occurred so you start your study when you have some sort of exposure you follow people into the future to see if that outcome as yet occurred retrospective study means that both the exposure and the outcome have already occurred at the time you start your study will go over this in more detail Dorie but observational studies can be exploratory meaning there’s no specific a priority hypothesis there’s no specific hypothesis that you’ve already developed based on prior knowledge of prior studies. Or can be analytic where you’re actually testing a hypothesis that you generated from earlier work so. You know again depends on the type of study and the answer the question that you’re trying to answer whether you would have she have a hypothesis you’re trying to test or if you’re kind of out there fishing and looking for a you know I have boxes that would make sense to fit the situation. So we’re going to talk about case control studies it’s control studies involve grouping people as cases and controls hence the name. But these are your cases you choose your study design is based on your choosing of cases so what I mean by that is that you everything revolves in a case control study around your cases So first and foremost you have to go out in find your cases are people who are experiencing the health related state or event that you want to study. Then you go out and you find. Controls people who are in as similar as possible to cases but that don’t have the disease so you start with your cases you find your controls then you ask them all the same questions or and you investigate whether the cases are more or less more or less likely than controls to have had past experiences behaviors or exposures so in this case. The outcome is it’s always identified prior to the exposure you know you have cases of. Some Let’s pick a rare cancer of the pancreas so I’m going to go out I’m going to find a bunch of cases of this rare pancreatic cancer we’re going to find a bunch of controls who are similar to cases but they don’t have cancer then we’re going to ask them all a thousand questions on prior eating history smoking other lifestyle exercise chemical exposures and when to go out there and ask all sorts of questions and then I’m going to compare the responses from my cases and trolls and see if there’s is there something out there that can you know help us understand what might be a risk factors for this particular type of cancer so I’m starting with my cases and then finding controls working backwards to elucidate what risk factors might have contributed to the acquisition of this type of cancer the outcome is always identified prior to the exposure. So a case control study these are just really handy little studies they’re great when exposure data are very expensive or difficult to obtain when the disease has a really long induction or late in period when the disease itself is rare or there’s very little known about the disease when we know we have these rare cases of pancreatic cancer but we really don’t know what causes. So it’s they’re great when you have a rare disease and you need to try to figure out what the risk factors are. And or when the underlying population is is dynamic way you can’t get a nice solid group of people who will stay in one place long enough to follow them into the future so case control studies are kind of quick and dirty compared to cohort studies but they’re awfully handy in certain situations so again when exposure data is expensive or hard to find and so you don’t want to try to find lots and lots of you want to target your your efforts to find exposure data to where you have the highest yield so you would pick your cases and your controls and only find exposure data for them. And again because. For case control study the whole thing basically has happened in the past year your case your outcome is already happened in you’re going back and looking for. Exposure you can study things like cancer that can have you know ten twenty even thirty year induction period or late period because it’s already happened you don’t have to wait thirty years to get the answers when the disease itself is rare so because you are going out and finding the disease the cases it’s a very efficient way. You go find your cases first and then you work backwards so you go to your cancer registries or you go to your specialty treatment centers or clinics and you can find your cases all in one place and work from there so it’s not hard to find them in the general population. Or when you just don’t know this is where you may not have a hypothesis to test and a priority apotheosis you don’t know what’s causing it could be a brand new organism like it was with AIDS or SARS or something like this. Very little is known then you start with your cases because they obviously were exposed and you work backwards from there or when your underlying population is dynamic you don’t need to have a nice cohesive group of people. All living in the same place doing the same thing like you do for your. Your cohort studies. So this is just a little diagram sort of showing that you know there’s a total population of some sort you can be defined in a lot of different ways but basically it could be the whole country the whole world the whole state. But within your total population you you have to somehow define your reference population and it’s from within this reference population that all your cases are selected and your controls also have to come from that same reference population so when you’re done with your study you really depend on how you choose your participants but how you choose your participants dictates how broadly you can generalize your results so generally you can’t generalize much past your reference population I can’t do a case control study and firefighters in Michigan find some results and then try to apply those to the whole U.S. population I really can only accurately apply those results to firefighters in Michigan because that’s where my cases and controls. Arose from. When you’re trying to find your cases for a case control study you can find them from various sources but they need to be representative of all persons with that disease and as much as possible. There are different ways you know if if you have lots of cases then you might want to just take a random sampling of all of cases are available maybe there you know five thousand cases of this rare pancreatic cancer that would that’s too many people for a study like. The so maybe. I mean say I don’t have enough money to track down all five thousand cases and then go find five thousand controls really I only have this much money so I’m going to you know take every tenth case and make a giant list of cases and I’m going to do some sort of random sampling to select the cases that will be included in my study so that’s sort of one way to you know so to narrow down the number of cases or you can do restriction so maybe I’m going to only select cases. I’m going to restrict because I really only want to look at women or maybe it really only I’m interested in this disease and kids I’m going to restrict an age I might restrict and gender might restrict and race I might restrict by saying well we’re really we don’t want to have to deal with the confounding effect of smoking so we’re only going to we’re going to restrict to cases that have never smoked because you know we’re trying to look at lung cancer. And we want to know what’s causing lung cancer in nonsmokers so we might restrict let’s look at Monkey answer cases in people who have never smoked so that’s a form of restriction so you can you know you have to think about all these things when you select your cases but you can reduce potential bias by restricting your cases to certain types but you also limit your ability to generalize that is to take your results and apply them to a larger population. Now however you pick your cases dictate how you pick your controls so your controls need to look and be just like the study subjects the case subjects except they don’t have the disease so you want to select your controls from the same population from which the cases are drawn you need to make sure your controls reflect the same age groups sex or any other significant factors of type only cases that were nonsmokers I need to pick controls for nonsmokers. You want them to be normal healthy and reflect the well population from which your cases were drawn. And control should have the same probability of being exposed as the general population and they should have the same possibility of being selected. Of exposed as to the cases. OK so now I have my cases of pancreatic cancer and I have my controls they’re very similar in age. Geographic location and you know whatever gender race they’re very there’s very similar. But now we have to go back we’ve got a cases and controls now we have to go back and find out about their exposure so how do I find out about their exposure status. And it depends again you know in the example I’m using a sub pancreatic cancer probably I would go back through their medical records if the people are still alive if you interview them you can ask you standardize questionnaires interviews if they’re if they have died than you did Mr same interviews or questionnaires to surrogate such as spouses siblings employers it could be children adult children. And information that you gather needs to be to same as the information that you gather on cases and controls need to be quick to in a similar manner as possible to avoid information bias in this case. So if you’re checking medical records of cases you need to check the medical records of controls because now we’re looking for exposure we’re not looking for outcome we’re looking for exposure what behaviors did they have what food intake did they have what occupational exposure what environmental exposure you know whatever it is that you think might be a risk factor you need to collect information on that for both cases. And controls. And so this is just sort of. You know in this study you start with the cases so you start at the bottom you find your cases first then you find your controls from the same reference population and you work backwards in time and you determined you determine which cases in which controls were exposed are not exposed to this risk factor. And it could be the nice thing about case control studies you can explore a whole bunch of different risk factors you’re not limited to just one exposure you can. Collect information on all sorts of different exposures. You can have only one outcome with a case control study lots of different exposures can be explored. Back to the two by two table I told you you would be bonded with this table. So with case control study the two by two table. Looks like what you see in the top there you have cases and you have controls and then you’d go back for each risk factor that you’re interested in and you say where they exposed yes or no. Now the thing is with the case control study you don’t actually know the total population you know your study population because those are the people you put in your boxes A.B.C. and. But you’ve pulled these people from a general population but you don’t have any idea really how big the general population is so you can’t really fill in the total population category here so when we’re analyzing a case control study we use A B. C. and D. boxes. But we don’t use a total box because our A plus B. doesn’t really equal total population it’s not like we know what the total population is so this becomes relevant fairly shortly here. So when we get into a. Week on case control studies and. Module on covert studies will get more into the how you analyze these studies but for now we’re just sort of introducing the concept so. The key to an analytic comparison study is your measure of association so a disease rate among a population might seem high but it’s actually higher than expected so in this case a group of patients in an outbreak all ate at a particular restaurant but is it just a popular restaurant or have more cases actually eaten there you have to come up with an actual measure of association and we talked earlier about. How you can calculate your risk ratio. Or relative risk it’s what that is one example of a measure of association. But we’ll get into this in more detail soon. Now the measure of association you choose depends on your study design so the way to address the concern of. The how you determine if there’s a true elevation or not it by comparing your your observed group with another group that represents your expected level so for a case control study your observed group or your cases and your other you comparison group is your non-exposed control group or non Disease Control group so for case control study the measure of association is called an odds ratio in a covert study the measure of association is the risk ratio or relative risk and again we’ll get into these when we get spend on each study design but you kind of get the idea that each type of study will have either you you should calculate either an odds ratio or a risk ratio. As you as your measure of association. OK now we’re going to talk about a cohort study Again this. Just enough to to whet your appetite for learning more about these different study designs. And again I’m going to reiterate this enough times you’ll be so sick of case control study you’re starting with your cases. You find your controls and then you work backwards and you can look for a whole bunch of different exposures you’re trying to find out what exposure might be most associated with your outcome. Now the flip side of that is a cohort study so just again very quickly a cohort study you start with your exposure you find you’re exposed to people then you find you’re not exposed to people that are controls and then you look for outcomes and it could be a number of different outcomes. So play this slide over a couple times until you kind of get that fundamental difference locked into your brain. So for this little section we’re going to define cohort studies identify some of their distinctive features we’re going to talk about measures of association. Common measures used in epidemiology for describing cohort data potential biases ways to control for bias and then distinguish between effect modification and confounding and may actually skip that last part we’ll see how it goes. So again with the coord study you’re starting with some sort of exposure and then you’re following them to see if they come up with some sort of disease or other outcome of interest and you’re trying to determine is there an association between their past exposure and their current disease or outcome. So Court literally is a group of individuals who have characteristics that are uncommon So there you hear the term cohort you might hear used as sort of a birth cohort you know all. All individuals in a certain geographic area born in a certain period of time usually for birth cohort it’s you know the birth cohort of two thousand and eleven birth cohort of two thousand and twelve a marriage cohort I’ve actually never heard that term all marriages within a given period of time what we’re talking about is in exposure cohort these are individuals that are Cymbalta as a group based on some common exposure so it could be. Exposure such as radiation that exposed during some sort of testing of a device could be they’re exposed because they have been lifelong smokers could be exposed to. A certain type of virus in childhood you depends on what you want to study you can assemble your cohort but the key is that they’re symbolism group based on a common exposure. So for a cohort study healthy subjects are defined according to their exposure status and then followed over time to determine the intent of their symptoms. The common characteristic for the group being subjects is their exposure level so I think you’ve heard that about five times now hopefully that’s starting to settle in. So again with courts that you determine whether there’s actually an association between a factor. Or characteristic in the development of a disease then after you’ve determined and again this is going back to the road map first you have to determine if your association is a valid association or not and then you look for evidence of a possible causes Association so in that case control studies cohort studies are used in epidemiology to do this they all fall under the heading of analytic epidemiology So this is just the example of a cohort that’s pretty well known but this is called the Ranch Hand study I’m not sure why it’s called that but they picked an exposure group. Air Force servicemen who sprayed agent orange during the Vietnam War So your exposure of interest years Agent Orange they picked a similar very similar group again same number they’re all in Air Force their servicemen and they flew other missions during the Vietnam War Agent Orange was sprayed from aircraft so you could assume that they’re all on aircraft. So the outcome of interest now here we have a single exposure right and a number of different outcomes so opposite of a case control number case control you have a single outcome a number of different exposures but this is a cohort single exposure lots of outcomes that you can look at so they got all these guys they’ve been exposed to Agent Orange you’ve got all these other guys not exposed very similar Otherwise let’s follow them through time and see how many of them come and up with different types of cancer post post traumatic stress adverse pregnancy outcomes cetera so the principle would be if if Agent Orange is not associated with the outcomes under the study then the outcome rate should be the same for both groups that’s basically your no hypothesis that there’s no association between Agent Orange and these outcomes. So this is another type of study design your population of interest is pregnant women so you pick a subset of pregnant women or certain population and then you recruit into your study say five hundred pregnant women who are smokers five hundred pregnant women who are nonsmokers then you follow them through the pregnancy to their birth and you can see how many. In each category have what’s considered a low birth weight which is less than twenty five hundred grams and then you can compare is the rate of low birth weight higher in smokers versus nonsmokers and then you can. Calculates your your risk level in smokers your risk for low birth weight nonsmokers you do your risk ratio and you come up with a measure of association. So again this is kind of repeating You’re looking at an exposure among your your cohort your total population cohorts and divide into expose and unexposed you evaluate to see whether an Association expects this between exposure and the outcome of interest and then you look at the incidence of disease or death rate from the disease compared by exposure group that means exposed versus unexposed. And this is sort of in a similar graphic to one we use for case control but in this case you’re starting to exposure and you’re following them to see if disease develops or not and then you compare the rates of the different diseases in one population exposed for says the non-exposed. OK I’m not sure what you can see all I can actually see here is a black box with a couple blue lines in it but you’re supposed to see it two by two table so basically again you have disease across the top yes or no and then exposure yes or no on the left side the dice thing about a cohort study is that you do know your total population so you can fill in. Part of the table where it actually says A plus B. and C. plus D. because you have you you have an actual cohort you know your total study population so you can fill in A plus B. with the total and then you split them into disease or not disease and expose or not expose the fact that you have a total population for a cohort study makes this study design a little bit more rigorous than your case control study. So for a cohort study if a positive association doesn’t exist between exposure and disease expect the proportion of the exposed group could have dealt the disease to be greater than the proportion of the non-exposed group who developed the disease and this is when you eventually do your measure of association which is your rate ratio risk ratio when your risk ratio is greater than one then you would have a higher proportion with the disease in exposed versus the unexposed So the rate is higher in the expose because you’re dividing by your unexposed. So there is something called a cohort effect that you may run into it can also be called sort of a generation effect and it’s where there’s a change or variation in the disease or health status of the study population as the study group moves through time so some of the. Greatest studies that we’ve conducted at least in the U.S. have been these gigantic long term cohort study where you actually follow a group of people who are you know many years and even generations and you can in fact have variation within a group of people over time and the court effect could it could any exposure influences from environmental effects to societal changes so as you know if you think about if you started your cohort in the sixty’s you know through the seventy’s eighty’s ninety’s you know there are lots of different changes in behaviors and understanding that may influence people’s choices to either to smoke or to exercise or the eating habits those types of thing. So now we’re starting to transition from cohort study into our next topic which is the randomized trials so a randomized trial is an experimental cohort so when we talk about randomized trial. Or clinical trials or experimental studies. They are actually all cohort studies it’s just sort of the next step up where you have an investigator has a lot more control over how people are exposure are assigned and which people end up in which control versus the exposed group. So a randomized trial is just an experimental cohort and ethical and other reasons prevent randomizing people to potentially harmful exposures so in most randomized trial exposure is actually a treatment or preventive measure. So in. For this reason what we use observational cohort study remember when I talked about the difference between X. observational studies and the experimental So generally what we’re working with if we have a negative exposure it’s going to be a natural type occurrence this exposure happened and now we want to learn from it. In the. Clinical Trial randomised trial you actually control the exposure so in both types of studies you compare an exposed group to non-exposed group so they’re fundamentally both cohorts it’s just how much what role the investigator plays. So this is just a graphic to help you sort of understand some the main difference between the cohort study which is observational and experimental or randomized. Study which is experimental So the main difference is that. Number one you. Observational study you don’t choose who’s in exposed unexposed group that just happens naturally whereas in random my study or experimental study you choose who was exposed and unexposed that’s the one difference in the second difference is. That you can. If you’re in an experimental study investigator can randomly allocate people into either exposed or not exposed groups but it doesn’t happen randomly if you’re talking about a natural setting or an observational study. So the main difference between the two designs is the presence or absence of randomization and this is critical with regards to how you interpret the study findings now one of the weaknesses of an observational cohort study design is that because people are not randomized to exposure perhaps it’s not actually the exposure but some other factor that lead people to be exposed that is truly so stayed with the disease so here’s is an example to kind of help you understand this concept I could design a study in my exposed group are certain people who live or who work in a certain factory and Mike can and I think there’s an exposure in this factory that is that is maybe the source of an increased rate of cancer or something so maybe I pick another factory that doesn’t have the chemical or exposure of interest in X. my control group. And I do my study and I find out that in fact people who work in the certain factory do have a higher rate of cancer than. People who work in a different factory but maybe it’s not the factory that’s actually the problem if I picked all my people from one geographic area because they’re they live near the factory where they work it may actually be some other factors not actually something in the factory it could be something in the the well water the aquifer could be pollution from a different factory could be a lot of different things that actually is leading is the true association with the cancer not necessarily the factory. And so now we’re going to talk about our final study design. We’ve been sort of working up in complexity and in scientific rigor So the the top of the heap is the experimental studies in epidemiology. So quickly these are the objectives are going to try to get through for this last part of this rather long lecture and I apologise for the length of the lecture may want to stop and grab a cup of coffee about now to talk about the role of randomization in controlled trials discuss the role of blinding in controlled trials and then talk about strengths and weaknesses and some of the ethical issues. And so the objective of clinical practice and public health both is to modify the natural history of the disease so as to prevent or delay death or disability and to improve the health of the patient or the population so the challenge is really to select the best available preventive or therapeutic measures to achieve this goal experimental studies can help us do that. And so to do this we need to carry out the studies that determine the value of these different measures control of prevention measures in the randomised trial is considered the ideal design to evaluate the effectiveness of different levels of intervention. And so again when we’re talking about the observational studies that’s a cohort in our case control study groups are pre-determined by variables that were beyond the control of the investigator so with the cases case control study. You go out and you find your cases first in those cases already happened they are in existence and you’re trying to figure out what risk factors lead to the development of that disease with a cohort some sort of exposure already happened it was and you natural type situation you know it’s an accident of some sort still or an unknown risk factor people were exposed to. So there’s. No random assignment there people into these groups they’re naturally occurring. Or accidental occurrences. And when these types of exposures are outcomes happen naturally there are a lot of factors that can influence the state of health and stuck the ability to illness and individuals in the groups that are way beyond the control of the investigator those types of things can be past experience lifestyle personal behaviors or training or education level immunization levels exposure to risk factors. Including our other than the ones under study or environmental factors and then these observations are observed but you don’t have the ability to control for a lot of these other factors that can come into play. With an experimental study design you do have more control over all those other factors so it’s just an extension of the cohort design that would be an experiment to see whether the expected outcome is found in a group receiving exposure with a control group not receiving the exposure so it’s just really investigated does get to determine who’s exposed and who isn’t experiment experimental are applied to clinical trials remember I said when things are bad things occur in nature you can still learn from it but when you are a scientist you’re not allowed to. Apply you know expose people to risky things there are ethical issues you know obvious it’s been done and it’s generally a bad idea so when we talk about experimental studies or clinical trials we’re generally trying to expose people to something that is good a preventive measure or a new drug a new vaccine something like that or a combination drug so we think of them more often as. Clinical trials or the drug trials or treatment trials so we’re generally using a study design to assess if efficiency of some type of intervention I can search of gold technique a medication new health service. So experiments are seen as the Supreme Court or the Cadillac of the Epi research study design has they provide the strongest possible evidence of disease causing experimental study design can rule out with greater certainty other factors that may be confounding the potential cause and effect relationship because investigator has a lot more control a studied degrees of studies degree of internal validity depends on the study designed billeted to determine whether an intervention causes an arm or act so I will talk about internal validity. A little bit later on but it’s basically the you know how well the study is actually being conducted would then there’s extra whole village it which is how far how wide the you can interpret the results. And so again investigator in a randomized control trial a clinical trial can control the predictor variable the intervention or treatment the major advantage or observational studies is the ability to demonstrate cause Alan randomized. Randomization can actually help control for a lot of those unmeasured confounding variables and you would only do a randomized control trial or clinical trial if you have a very mature research question you have a very specific thing you’re trying to measure you have a new drug and you need to show that it is actually a much better drug than existing drugs for treating disease X. it’s only for mature research questions a very very expensive and very complicated to run so you really for the most part it’s really at the. Drug trials are intervention trials that you’ll see researchers use in the study design well funded trials. And these are just a couple examples of prophylactic is just preventive of vaccines tested on children populations of children to prove the efficacy of vaccines in preventing disease or prophylaxis with drugs to prevent disease or could be treatment also or impact of health related behavior and coronary heart disease in response to some community wide heart disease prevention intervention. And I think I’ve said it about eight times in Vesta gaiters intervene in the study by influencing the exposure of the study subjects and there are basically two types of experimental studies there’s the randomised controlled trials where you’re dealing with individual subjects and then there’s also community trials where you may do introduce a treatment or prevention effort at the community level not at the individual level. And the unit of analysis is always the individual and it’s referred to as a clinical trial or in a clinical setting with a randomized control trial. When we’re talking about a community trial or population trial the experimental study is where one group of people or a whole community receives an intervention and another group does not in this case the unit of analysis is actually the group or community and it’s appropriate for diseases that have their origins in social conditions that can be influenced by intervention directed at group behavior as well as individual behavior. And on occasion you can have a natural experiment where some sort of unplanned event may actually produce sort of a natural experiment where. The levels of exposure of a presumed. Cause may differ among different populations in in a way it’s an unknown that is unaffected by extraneous factors and it could resemble a planned experiment so an example would be screening and treatment prostate cancer so maybe there are. There’s a community and insurance company decides they want to pilot a program so they start doing some extensive screening just in this one community and want to start aggressively treating prostate cancer so you may be able to use that your as a natural experiment so as a researcher you could say well let’s start looking at rates of. Prostate cancer as the outcome in this community versus another community that didn’t have that intensive. Intervention happening. And so here’s the graphic that goes with the randomized controlled trial where you you have a study population and it’s a cohort so you know everybody in that study population and then you randomly assign your people to either into the current treatment and oftentimes into the new treatment you’re trying to you’re hoping has a better. And then you follow these folks over time and you see how many improve under the current treatment how many improve under the new treatment and then you do your measure of association and see if there’s if there’s more improvement or less recurrence. Or better prevention in your new treatment group versus your current treatment group which is often the existing standard of care among physicians for that outcome. So again you begin with a defined population in a randomized control trial you randomized them to receive either the new treatment. Or the current treatment and which is similar. You can think of it as sort of the exposed versus not exposed to being exposed to the new treatment or not being exposed to new treatment. And then you follow the subject to see how many are approved. And compare them to the your control group which is the ones not receiving a new treatment. And with the theory being that if the new treatment is associated with a better outcome you would expect to find you know better outcomes in more of the new treatment group than in the current treatment group. So the slightest just to show you where we’ve been we started with the descriptive studies the population level correlation and slash ecological studies then we went to individual descriptive studies such as case reports individual case a series of cases the cross-sectional surveys which can yield some pretty good information especially if you conduct a series of surveys then we jumped into the analytic studies where we’re actually testing hypotheses that we generated in the descriptive studies for the analytic studies you have a control group. And we have the two categories observational studies which are naturally occurring situations we have case controls or cohorts and then we have experimental which are also known as intervention or studies or clinical trials. So that wraps it up.