I'd like to thank Dr. Jay for being a good sport and not suing me into oblivion. The man didn't even threat to sue, unlike that other guy... Alright, enough of that. Thanks, Dr. Jay, for being a good sport.
Tonight we're going to discuss cohort studies and why we have come to the understanding that there are certain lifestyle choices that we know are bad for your health, and why we know that vaccines do certain things. Seriously, we don't pull public health recommendations out of a magic hat. We're not wizards. So let's look at cohort studies, their design, their analysis, and why they are the strongest when it comes to studies... All after the jump.
 |
| We're down the stretch! |
EXPOSURE STATUS
If you remember the case-control study from last time, we picked our case and control groups based on their disease status. The ills went to the case group, and the wells went to the control group. There was no randomness to it. Fate, as it were, had already chosen for us since they were already sick. Alternatively, in a cohort study, you begin with one big group (population) of completely well people. In the case of a drug trial, you begin with an unexposed population in that they have not been exposed to the medication/intervention.
FLIP A COIN
So you have this big group of unexposed folks and you want to know if exposing them to something will lead to an outcome. In order to simplify this "lesson", let's say that you're looking at a cohort (a group) of 20 year-olds, and you want to know if giving them a daily multivitamin will lead to less chances of diabetes when they're 50 year-olds. Yes, this is going to take a while, but it will allow time for us to address some of the things that could interfere in the study.
Because you want both groups - the intervention and the control groups - to be as similar to each other with respect to everything, you literally flip a coin. Heads, you go to the control group and will get placebo. Tails, and you go to the intervention group. If you really want to eliminate the most bias from this process, you take it one step further and don't even tell the researchers which side of the coin will lead to placebo vs. the vitamin. That's what is meant by a "double blind" study. That way, there is no chance that the researchers will purposefully give the vitamin to someone they feel is less likely to get diabetes. Conversely, you avoid people finding out in which group they are by observing the attitudes of the researchers. (Imagine a researcher asking intently if you've had any diabetes symptoms if the researcher knows you're on the control group.)
OBSERVE
You would think that all you had to do is sit back and wait for the people in the study to report to you that they had diabetes or not. Well, it's not that easy. Before you begin the study, you need to decide what constitutes a case of diabetes. Is it someone with a high fasting blood sugar one time? Two times? Three? Is it someone with a glycosylated hemoglobin that is 9%, 12%, or 15%? You need to answer all those questions before you start, and you need to make sure it all stays the same throughout the study.
You also don't want people to go from one group to the other without your knowledge. To that end, you need to make sure the rules are clear to the participants. For example, you don't want someone who develops diabetes to think that it was because they are in the group that they are in and begin taking vitamins.
MEASURE THE RR
When your group reaches 50 years of age, or whatever endpoint you choose for your study, you want to make sure that all of them are measured the same way and in a reasonably similar time. You want to make sure that you retained similar numbers on both groups. (It doesn't help if all your controls drop out, does it?) And then you measure the relative risk (RR). What is the RR?
 |
| It's not the railroad, that's for sure! |
Since you're starting with a healthy or unexposed population, you really can't ask them what their odds are of being exposed. After all, you flipped a coin. It's pretty much 50/50, right? Instead, you look at the two groups at the endpoint of the study and then figure out if they were part of the case or the control groups. You then ask the following question: "How much more or less likely were the exposed to becoming a case or developing the condition you're looking at?"
MORE OR LESS THAN 1.0
Like with the OR, the RR is looked at in terms of its value relative to 1.0, with 1.0 meaning that both groups had an equal chance of developing the condition in light of the exposure. If the RR is higher than 1.0, then you say that those exposed to the intervention (the vitamins in our example) were X many times more likely to develop the condition. If the RR is lower, then you say that the intervention lowered their chances of developing the condition (diabetes). An RR of less than 1.0 may also be interpreted as "the intervention was protective for the condition".
 |
| Just like these are protective for... uh... something. |
Of course, just like with the OR, you need to take statistical significance - and thereby chance - into consideration. You need to make sure that the RR you have measured is not due to chance. That's when you look at the confidence intervals. Like the OR, you want to make sure that you're 95% confidence interval (CI) does not include the number 1.0. Otherwise, there is a good possibility that the RR you observed was by chance and that, if you repeat the experiment 100 times, five of those times you will get an RR that is 1.0. That is not acceptable.
 |
| "I do not approve." |
So when you look at a study that is published, ask yourself several questions. Did they randomize the groups? Did they put in controls to make sure that the people did not cross over, or, if they did cross over, were there controls in place to adjust the measurement? Was it a double-blind study, or was there a chance that the researchers could have had a hand in the outcome? Also look at the relative risk (RR) and the confidence intervals (CI). Do they include 1.0? Finally, look at the sample size of the study ad the distribution of characteristics in the two groups. If they were really randomized, you wouldn't see one group made up of all men and another of all women, would you?
 |
| A well balanced, albeit odd group. |
Of course, there are those times when the study doesn't go as planned. For example, what would happen if you start noticing that all the controls in your study developed diabetes (or some other disease) while all the people in the experimental/exposure/intervention group remained healthy? This has happened before. In those cases, the review board(s) overlooking the study will call off the study and just recommend that everyone take the vitamins, for our example. In real life, this has been the case with some vaccines, where the review board - made up of impartial people and not pharma shills as some would have you believe - sees that everyone who is vaccinated is healthy and has not contracted the disease for which the vaccine was intended while the people who got the placebo are falling ill at a statistically significant rate.
The opposite has been true as well. There have been studies where a new medication causes some bad outcome for those on the experimental group and the study gets shut down. (If only Vioxx would have been given more time, for example, the group taking the medicine would have been observed as having more hyper-coagulation events. This was not observed until after the drug went to market based on results which showed it was better than the previous class of drugs against pain.)
IN CLOSING
It has been an absolute delight to write up these "lessons" about epidemiology over the last few months. I thank each and every one of you that took the time to read. If you have any questions, you know how to reach me. I'm always here to answer question, correct anything I wrote in error, and generally just listen, if that's what you're into. Of course, I don't expect any of you who only read these "lessons" to be full-fledged epidemiologists by now. That would be nuts. But I do expect you to at least have a little more of an idea of what we do and why we make the recommendations that we do. I hope that you saw that there is a method to our madness: The Scientific Method. Everything we do is backed up by evidence, by statistical analyses to make sure we didn't see associations just by chance, and by an understanding of how pathogens and other bad things in this world work. Rarely do we go by our gut alone... And even then, we base that gut decision on a lot of experience with these things.
Thank you again for your time, and have a great rest of the summer.
No comments:
Post a Comment
Keep the comments clean.