Randy Neil Biostats 08: USMLE Step 1 - Bias & Confounding with Questions

Cases and controls remember past exposures differently, especially in retrospective/questionnaire studies.

Lead-time: disease is detected earlier, so survival appears longer. Length-time: screening preferentially finds slower-growing disease.

It is associated with both the exposure and the outcome.

Answer: C) Recall bias

Retrospective questionnaires are vulnerable because cases may remember exposures differently than controls.

Answer: A) Lead-time bias

Lead-time bias creates the illusion of longer survival by starting the clock earlier.

Answer: B) Confounder

A confounder is associated with both the exposure and the outcome.

Alright guys, so this video is about the biases. What we did was tried to organize most of the biases that I've seen over the years on the US-Emily exams and the Q-Manks and try to focus it on those because if you look up the biases online, there's just the list just goes on and on. So we tried to focus on the main ones. I want you to know the length time versus lead time. I want you to look at the basically the confounding. So we cover all these in the video. Hopefully it's helpful. And let us know in the comments. Don't forget to subscribe. Alright guys, so here's our attempt at the biases. So let's get started. It says a study was done to assess the relationship between attending a weekend seminar on how to get in residency and candidates' confidence when applying to match. Attendees who matched into their top choice as well as attendees who did not match that year were interviewed using a standard questionnaire. The study showed that people attending the weekend seminar without their chances of getting into residency. I don't know if it's going to show that. Increased after attending the weekend course. So long story short, they went to the seminar and after they felt like they increased their chances of getting into residency after the course. The odds ratio is 2.2 meaning they felt there was a 2.2 likelihood or increased chances of getting into residency after attending. And the P value is 0.05 good, right? Because we want points to P value 0.05 or less. Based on the above information, what would be the most likely bias concern? Alright, so essentially there's a seminar and then they sent a question here to these guys after the fact. And one group, people who matched and one who didn't. So what kind of bias might that be? So let's look at our choices. We have lead time, selection, figmallion, recall, hothorn and procedure bias. Now obviously some of these just

flat out don't even fit, right? Hothorn is that one where you're being observed, right? It's like I know I'm being watched and it's going to influence how I kind of act, right? It's like they say that's the one where if it's a vitamin study, they're chance to other going to start eating healthier and stuff like that. We know that in this situation it's not the hothorn effects these guys aren't being watched. And then the pigmallion effect, well that's just essentially saying if I set high expectations, there's high expectations that leads to an increase performance. Alright, so that's a good one. Good for kids and stuff like that. You set the bar high, people tend to meet that expectation. We know it's not those two. Again, hothorn somebody's watching you and that affects it. Pigmallion effect, high expectations lead to increase performance. So now we're down to these three, lead time selection and recall and procedural. Well the lead time, I'm going to explain this one here in a minute for lead time, for the step one and two and all that stuff. I want you to differentiate lead time from length time bias. Those are two that you got to differentiate. They're usually coming in the same answer choice. And for right now, lead time and again lead time is not for this one. But lead time has an E in an A, right? So I want you to remember that E in the A and lead is going to be used for early. Okay, that's early detection. Early detection is leading to a bias that people are living long. But note you're just finding out there's a positive sooner. So lead time bias E A is for early. And then length time, I just from you're going to associate that with the word slowly progressing. They got to give that to you. And again, the meaning that the length time bias is that whatever disease it is, it's a very slowly progressing disease. It gives the illusion that people are living

longer. Well, no, it's just a slow progressing disease that you caught. So it's kind of a bias that shifts the results a little bit. Okay. And just remember, big picture here is a bias is just it's kind of a slant. You know, it's a slant on the facts per se. Okay. So back to this, a selection bias. Okay. Slection bias means you're only choosing certain demographics or certain people. Okay. And this situation, they actually interviewed both, both people who matched and didn't match. So it's not this one, but again, selection bias is the only chosen certain people. Not that one. Now a recall bias, okay. Recall bias could could be affected here because this, this questionnaire may be influenced, right? The people who matched maybe like, hey man, I remember everything about that. Some of them are not really helped me get the interview blah blah blah blah. But the people who didn't match may not be so inclined to remember that, that, you know, they paid money for some. We can summon on that didn't effect how they got into residency. So in this situation, it's the recall bias. Anytime that you do a questionnaire that's done after the fact, there's a chance that you could have a recall bias in that situation. And procedural bias, this is the one, think of like this. At the end of a rotation, you know, you're supposed to fill out, you know, the attending does one on you and you're supposed to fill out some evaluation on the attending. But the fact is, you're under pressure, right? You're thinking, well, what if this attending reads this? And I know they say they don't read it for a little bit of time, but what if they do read it? You're thinking, this could come back to haunt me. So anytime that you're under pressure and you got to fill out like say, You know, evaluation or some type of, give some type of feedback that's influencing how you could fill it out, that's called a

procedural bias. Okay. And I feel like I do residency. I felt like I was one that we always kind of suffered from. But in this situation, recall bias questionnaire done well after the fact. This one says, a student researcher is designing a project that will hopefully minimize the use of confounding. Okay. Now, I'll say this, confoundings huge. They like to use that word, confounding bias or confounders on step one in an incidentally, in on step one exams. Confounding in a drug with exceptionally long half life. They report that using a crossover study would have some advantages and reduce confounding true, which is the following supports the students understanding other reducing compounding. I should say, reducing confounding. You think I do a better job rereadings. So what would reduce confounding in a crossover study? Well, first of all, we got to understand a crossover study. You know, we start out like this, you know, you got one group up here, one group up here, one might be getting placebo. We don't know. And then all of a sudden, they have this thing called a washout period where they just give them time for the drug to get out of the system. So if these people were on the drug, we need them to cross to go through this washout period. And they can start the next phase, whether they're going to be in the placebo arm or maybe the medication arm, nobody knows, right? So, but this is there's a washout phase with that. So that's a crossover study. Now, and it does reduce confounding. But in this situation, let's see what it says. Is it A, as patients proceed to the next phase of treatments, there may be carryover benefit from the prior cycle? It's kind of true, but we don't know if that really would help reduce the confounding here, right? Because if someone's kind of influenced already, then chances are that might impact them on the backside of this on the

backside of the study. Because they're already kind of, you know, trained per se, and that might actually affect the results. So I'm not too keen on that one. A short washout period? Well, this was a long half-light drug. So if we had a real short washout period, then the people who were taking the drug, if they don't allow that drug to wash out of their system, then it could carry into the second phase of the study. So in that second phase, maybe influence for that drug still in their system. So we don't want to really, in this situation, a short washout period with the drug with a long half-light. So I know it's not that one. Patient acts as her own control, okay? Yes. And across over study, that's kind of the, you know what I'm going to say, the genius of the study, is that the patient, you know, if they're the placebo right here, and then they cross over and now get the drug or vice versa, they act as their own control in that situation. Okay? So that's one of the benefits of being across over study. Now, and as these as the patient moves to the second phase, they may be educated and not have to be rechained. No, you know, that that goes back to what I was probably extending for A is that if they carry over anything from it, they could be influenced from one phase to the next phase, and that could impact the results over here. You know, again, with the same thing with that short washout period, and that's maybe what A is really stands for, is that the drug stays in the system. They may be carry over benefits to the next cycle, from the prior cycle. The benefits of an elevated type one error, a way to second. Remember, we remember all from our basics that a type, a type one error is often as alpha error. It's also known as the p-value, right? There's no benefits of having an elevated. We want that guy to be a low, you know, low p-value, right? So having an

elevated makes no sense. So across every study, they act as their own control. Now why did I even give this question? Because you're like, wait a second, and you're supposed to teach biases here. Well, the purpose of this question is to kind of educate on how do we reduce, you know, how to reduce the biases. And what you need to know is we can reduce biases by doing double blind studies. Okay, double blind studies. We can make our studies randomized. Okay, that reduces. And that, basically, that will limit the selection bias we talked about, right? We're only selecting a certain group of people if we do it randomized. That limits the selection bias. And, you know, in the double blind thing, that limits the observation bias. Okay, and then we can do a crossover study. And the benefits of that is the subject or the patient is their own control. Okay, all of these things will reduce the biases. Okay, now it's purpose of this question. But no, the whole kind of confounding we're going to talk about here just a second as well. Okay, all right. So it says, when a group of researchers and students are trying to assess the relationship of birth order, whether it's for a second or third or to the association of having a child with Down syndrome, a student argues that the age of the mother needs to be taken into consideration. The researchers praised the student, reporting out the confounding bias, right? Based on this information, which of the, which of the following would be the confounder in this situation? Okay, so we have to understand what this whole confounder stuff is. Okay, and think of it like this. If I do a, which is say, for our purposes, let's just think I give a drug, and I'm looking for some outcome, okay, which is b. All right, and, and, and more, we call this exposure. A is kind of the exposure, and then b is the outcome. Now, what this is saying is that I

can't just say, a, I'm looking at, I got a, and I'm looking to get b, is that there could be this one thing out there. It's one thing, or more, that affects both of these, and we have to make sure that we understand that. Because if it affects both, then we have to kind of, you know, when we look at the data, we have to kind of, you know, tease that out, you know, and you can tease it out. They do that through that stratification, okay, that's kind of reducing that. But in this situation, what's going on? They're saying birth order, they're looking to do a study that's that talks about birth order, and how it's associated with, down syndrome, okay, but we have this thing that, as the student says, what about the mother's age? Now, the thing is, does this affect both down syndrome, the other's age, and the birth order? Are they, is, is the mother's age associated with both? Because that's the real question, because that's what a confounder is. It has some type of correlation or some type of, it affects kind of both the exposure and the outcome. So, just by general knowledge, we know that the older the mother is that there is going to be some type of potential increase risk for down syndrome. Yes, it could be associated with that. And we also know that when they're doing these calculations, that the mother's age does impact birth order, right? Because obviously the mom's going to be older when she's having the second or third kid than when she'd had the first kid. So it is associated with that. So this question says, based on the information, which the following would be the confounder in the situation, is it the birth order? No, birth order is going to be the exposure, kind of, that's the exposure to this, okay? Is it the genetic loading of the mother? Well, I mean, it's good to know that, right? Genetic loading is kind of the, the history, okay? Sometimes we, in

psychiatry, it's like, okay, we say there is no genetic loading, other is genetic loading, so someone else has that condition back in the family tree. But they don't, we don't have an information here. Is it the mother's age that, look at our confounder, which would be the mother's age, it affects both the exposure and the outcome. So it's a confounder. How do we reduce it? We can reduce it by stratification, okay? And that's the story from the other day. The researcher in the students preconceived opinion on the outcome. No, what kind of bias is that, like when someone already has an opinion about it? Well, that's a confirmation bias, right? That's like, you know, it's just saying, hey, look, you know, I'm going to do my own research on this, I know I'm going to get this outcome before I even start. That's a confirmation bias. The outcome of Down syndrome, nope, that's just also, that's just known as the outcome, okay? And then alcohol use or smoking, but it weren't mentioned. Well, they weren't mentioned. So I can't really say that wouldn't part of our question, but that's another example, right? You could say they do a study where they say, look, I'm going to study with their alcohol is associated with heart disease. And then the student would say, wait a second, what about smoking in this? We got to take an account for smoking, right? Because it's smoking associated with heart disease. Probably, right? There's some data in the sport that is smoking associated with alcohol. Well, if someone drinks alcohol, chances are they may smoke as well. There's probably a higher chance. So it's associated with both. So what's the confounderness in this study? Smoking, how do I reduce it? Stratification. And this one, it says, a group of students is working on a research paper about a new detector for brain tumors and adolescent in adolescents. It is reported that these types

of brain tumors are very slowly progressing. Remember I talked about that earlier, the students are concerned that this new detector may be impacted by the nature of the tumor's progression, which of the following biases are the students most likely concerned with. Now, anytime you see slowly progressing, you better be jumping all over, late time bias. Okay? Now, remember, for step one, you better be differentiating lead time versus length time bias. Okay? And what do we say? We said lead time has an ENA, so that's just detecting something earlier. You know, it's not saying, oh, if people are living longer, no, we're just detecting it earlier. That's all. They're probably they're probably maybe living the same amount of time. We just catch it earlier. Length time, they got to have something in there that says it's slowly progressing versus versus something else. So they got to give you that somewhere in the question. The sampling bias is just filling, it's a whole deal where you fill out the filling out a questionnaire. Okay? Got to make sure that you're getting a true representation of the population. Okay? And sometimes it's done new, not if you're, you know, if you're in a very upscale neighborhood dealing questionnaires, it might not be indicative of your entire city. Okay? That's a sampling bias. I don't know why I put the question here thing, because mainly of that, it's just, it's just not indicative of a population. Okay? That's your take home point with the sampling bias. The procedural is when you're filling out, you know, filling out an evaluation and you're under pressure per say, and that influences you about how you're going to fill that out. That would be considered more of a procedural procedural bias. Again, you're under pressure and not influence you influences you. The recall bias. Again, that's just, it's kind of like, as time, time from the

event. You know, there's a question, I've seen out there where they talk about, you know, moms who delivered, had trouble with delivery. Well, the longer you go off, not delivery, moms are not going to remember as much. And then the confirmation bias is like when you do research, but pretty much when you start to look things up, it's, you're looking at data. Is, are you interpret data upon your own beliefs? Okay? So, you know, if I, if I, if I, if I, if I had something of intent in mind, I can kind of Google certain keywords. And so I do that and I do that and get my results at I expect that would be a confirmation bias. And of course, the late time bias, we got to go back slowly progressive, slowly progressive, lead time early. So again, lot of these have just somewhat memory memory. But if I were you, I would understand what the confounder really kind of means reduce it with stratification or stratified analysis. I think it's called. And remember, the crossover study have a good washout period and the patient acts as their own control. And those, those are kind of your, your take home points guys. So I hope this helps. Thank you.