DIP Episode 458 - Biases in Screening Tests (Lead time, length time, selection)

Okay, welcome. My name is Divine. This is episode 450 of the Divine Intervention Podcast. In this podcast, I'm going to be talking about biases associated with screening tests. This is actually kind of a high-yield topic. Many people really really struggle with it. But again, I will try to clarify this here. So, there are three major biases we see with screening tests on USML exams. Selection bias is one. Lead time bias is the other. Length time bias is another one. So, let's go through them. The simple one, actually, one of them is exactly easy to understand. But if you really follow what I'm explaining, you get the stuff down. So, first one is screening bias. The thing is, think about it. The people that get screened for diseases. Generally, people that... I mean, think about it if you're going like, oh, doctor, please, I need my bumbogram at 40. People that actively seek out their own screening. Chances are, they have a higher socio-economic status. They probably have higher levels of education. They're probably wealthier. Those are just good at pay more attention to their health. Versus a person that doesn't really seek out screening, it has to be reminded all the time. That person is probably at a lower socio-economic status. Probably not as wealthy. Probably doesn't really pay more attention to their health. So, you can already see that the person that gets screening is probably a person that is able to get access to better medical care, better physicians and whatnot.

So, those people probably do better health wise. So, you can then see, oh, wow, this screening test really improves survival. But it just so happens that the population you're screening is just a population that gets better health care or has better health care habits than a population you're not screening. So, selection bias is something that can certainly happen with screening tests. Now, the second kind of one I want to discuss is lead time bias. So, lead time bias, pretty well, really lead and length time bias. Their biases where you overestimate the survival. You pretty much overestimate the survival. So, let me explain. Let me start with lead time bias. This one is maybe a little bit easier to understand. Let's say everyone that has the shame of scythe dystrophy dies at 25. But let's say that people don't start having clinical symptoms until they are age 15. So, from the time where they start becoming symptomatic to when they die is 10 years, pretty much. Well, then let's say you just say you're like, wow, life expectancy with the shame of scythe disroves 10 years. Because again, from symptoms to death 10 years. Again, I'm just making up numbers here. Well, let's say you design an awesome, awesome newborn screening test. And you're like, oh, okay, I can catch the shame of scythe dystrophy at zero years old. I can't catch it when the person is just a newborn. So, wow, people that get my test, they improve survival by 10 years, by 15 years compared to placebo.

Because if you think about you're like, oh, wow, I'm detecting this thing 20 like early when they're born. No, you're not really improving survival. They're still going to die at age 25. They're still going to die at age 25. You're literally still going to die at age 25. It's just that you just found it, you just found it earlier in the course. You found it before they started having symptoms. But you literally did not improve mortality in any way shape or form. You literally did not improve mortality in any way shape or form. So, that's when classic bias associated with screening tests. Where again, you're catching a disease early, but you're not necessarily altering the disease course. The person is still going to die. Many times, you know, these lead time biases, many times found with many of these incurable diseases. Where there's really not much you can do. So, like, for example, you know, like content in disease for example, for person else, content in there, they're going to die at a pretty consistent age, regardless of when the symptoms start. So, it's just a matter of you caught the disease earlier in its course before it became symptomatic. So, you think you've magically improved survival. But again, it's not a, you know, some screening tests by catching a disease can alter the disease course. They can make you maybe get on medications and whatnot. And then you treat it and then you don't die at that fixed time that others die.

So, again, screening tests are still useful. Don't say, oh, wow, the virus is pretty much telling me that screening tests are not useful. Now, the next kind of bias our discourse is length time bias. Again, length time bias, you pretty much again, overestimate disease survival. You overestimate disease survival. So, let me give you an example. I think this is one of those things that just are easier to explain with some kind of example. So, let's say we're looking at, let's say we're looking at some kind of cancer. Let's call it cancer X. But then we know that cancers, even if they may be of the same type, occasionally, may have differences in aggressiveness. You know, like there are some cancers where it's the same cancer. Let's, okay, let me maybe see invasive doctor of carcinoma. Let's say there are some invasive doctor of carcinomas that are very endolate. Let's say, wow, that these cancers, you know, they don't progress very rapidly. Maybe takes 10 years from when you develop disease to when you die. But then, let's say that there's some variance of invasive doctor of carcinoma that, you know, maybe have some kind of weird genetic mutation where if you have that disorder, if you have that particular genetic mutation, that type from when you develop disease to when you die is three months. So, if for some reason you design a screening test and you're like, wow, man, this is my screening test. I want to use it for invasive doctor of carcinoma.

And you're like, hmm, this test is really awesome because I caught in invasive doctor of carcinoma early. I mean, these people literally are living for like 10 years after I catch the invasive doctor of carcinoma. The problem there is you are overestimating survival. You may wonder, define how. Let me explain. The thing is basically, because it's a screening test. And you know, screening tests are done at certain defined intervals. Chances are pretty good that the people that you're catching are people that have the more indolent forms of invasive doctor of carcinoma as against the more aggressive forms of invasive doctor of carcinoma because if it's an aggressive form where from the time you have the first genetic mutation to the time you die is three months. Chances are, if for example, you're doing a screening test that is done every year, every two years, you're almost never going to see those people that have those aggressive forms of invasive doctor of carcinoma. But you're going to see more of the slow grain forms, the more indolent forms because those things take time. So you're they're going to sit in the body for a longer time period. So you're going to catch them. So if you understand this thing, I've just explained, I think you're being, I think you're being good shape. I think you're being good shape.

Again, our friends at the MBM is they can write questions like this where they talk about like, for example, malignancy where there are multiple, multiple types where there are some types that are indolent and then there are some times that are aggressive. So like, for example, there are certain kinds of AML that are, you know, I mean certain certain kinds of really, let me know me can especially disease, but there are certain kinds of hematologic malignancies where some are pretty indolent and some can have like a very rapid course, right? Some can have a very rapid course. There are certain brain cancers where some are indolent and some can have a very rapid course. So again, just be careful of length time bias, length time bias, you're basically, when you're dealing with length time bias, you're dealing with just one disease. But that one disease has a less aggressive form and a more aggressive, more rapidly growing form. Screen is going to catch more of the less aggressive form. So that can cause you to overestimate the survival of that disease just because you're finding the less indolent, the more indolent forms as against the less indolent forms. So hopefully this makes sense. So again, I'll encourage you, just maybe listen to this again. I think you'll be in good shape if you pay attention. And you know, maybe let me just go a little deeper into this length time bias of a thing.

Because again, I can almost sense as a meek in this podcast that so people may still be a little confused. I think a good perspective to understand length time bias from is to ask yourself, why is a person surviving for longer? Why? Why is a person surviving for longer? Are these surviving for longer because you caught their disease earlier? In length time bias, the answer to that is no. Because again, in length time bias, basically what you're identifying is less severe disease. Like for example, I can give you this situation of invasive doctorate carcinoma. You may say, oh wow. Again, I'm just making this up. My mom has been shown to help people that have breast cancer. But again, I'm just making this up. But if for example, you identify it, maybe I shouldn't even use invasive doctorate carcinoma, let me just use a random disease. Let's call it disease X. Let's say disease X, you say that, wow, by getting screened survival is 10 years. You're in first survival by 10 years. But the thing that's happening in length time bias is that you may essentially just be finding more indolent forms of that disease. So the reason the person has a longer survival is not because they got the screening test. It's because you identified less severe disease than the more aggressive forms. Because if for example, you identify the person that had a more aggressive form of disease, they would still die. They would die very quickly.

So really, the crux of length time bias that I think is the big problem for many people is asking, why did this person survive longer? The reason why the person survived longer is not because you caught the disease earlier. It's because the disease you caught is a more indolent disease. That's the critical thing. Again, some people may be annoyed saying, I'll divine your opinion yourself over and over and over again. But I'm just doing this to make sure you truly understand what's going on. I'm just doing this to make sure you truly understand what's going on. Okay. So the thing is, again, it's not the screening that made the person live longer. Is the rate of disease progression that made the person live longer. Right? Like some of these hematologic malignancies. It may be the same hematologic malignancy, but some have a certain genetic mutation that is really bad. To rate, that's the mutation you have. You're dating like a few weeks versus, oh, wow, you have a different kind of mutation. Even if it's the same disease, but you know, you can live six years. The thing is, since you're screening, you're more likely to catch those people that have the variant that gives that is associated like six-year survival than the variant that's associated with a six-week survival. So the reason, and then you can then not go and see this bridge generalization now. Wow. If you have this particular malignancy, you're surviving by getting screened in improves survival by six years.

No, it's not necessarily doing that. The reason behind that increased survival is because you got, you identified people that had that malignancy with a more beneficial, more favorable mutation than people that did not have the favorable mutation. So again, hopefully this kind of makes sense. Really one of the best ways to deal with these biases, these are lead and length time biases is to do a randomized control trial. Do a randomized control trial where everything is randomized. You get like a large population of people, you run the mice. And then you're going to see that, wow, there's not much difference in both groups because by randomizing, you know, in your experimental group, which at the point that gets screened and your control group which is put at don't get screened, you're pretty much by randomizing, right? So remember the randomization has to be there. You're pretty much putting people like relatively equal numbers of people that have like really bad prognosis, like lesions and less aggressive lesions. You kind of even them out between the groups. So you notice ultimately that there are no significant differences. So again, I think I'm going to go ahead and stop you. I think I've kind of talked about this enough. But again, just listen to this. It's a pretty easy to understand concept if you just kind of take your time and wave through it. So thank you for joining me today. Again, I offer tutoring for step one to three.

I have a bunch of review classes like a testing course for step one to three. I have one in June. I have a four-hour biostatistics class for step one to three. I have one in June as well. I have a social sciences and ethics class for step one to three. I have one at the end of this month in May. I have a 20-hour step two step three review course. I have one actually next week. And then I have a 25-hour step one review course that's in early July. And then I have a 500-question 50-hour step two step three review class taking place from June 1st to June 11th. All these classes over Zoom. If you're interested, shoot me an email. I have these podcasts on the major apps, Google, Spotify, Apple, Apple You Tube channel, Divine Intervention, US Emily podcasts and videos. And then I also have a website called Divine Intervention Lifelessens.com. We're from a biblical perspective. I go over a life lessons. I post two podcasts roughly every week. So thank you for listening to me today. I'll see you next time. Bye for now.

The bias that occurs because the population actively seeking out or able to afford screening tests (e.g., higher SES) is inherently healthier than the general unscreened population.

Overestimating survival benefit by measuring the time from early detection to death, when the underlying natural history and ultimate cause of death remain unchanged.

Length time bias occurs because the screening test preferentially identifies less aggressive (more indolent) forms of a disease, making it appear that survival is extended when it is merely identifying a milder variant.

Performing a Randomized Controlled Trial (RCT).

The rate of disease progression; specifically, that the identified disease variant was inherently more indolent than the average or aggressive form.

The tendency to overestimate benefits because screened individuals are often wealthier and have better overall health habits/access to care.

Overestimating survival duration by measuring time from early detection of a fatal disease, when the true mortality rate remains fixed regardless of detection timing.

When a single malignancy or condition exists on a spectrum, containing both highly aggressive (rapid) and slow-progressing (indolent) variants.

The assignment of participants into the screening group versus the control group, ensuring equal distribution of prognosis/disease severity.

The indolent (slow-progressing) form, because it takes time for the disease to manifest enough to be detected at fixed intervals.

The tendency to overestimate benefits because screened individuals are often wealthier and have better overall health habits/access to care.

Overestimating survival duration by measuring time from early detection of a fatal disease, when the true mortality rate remains fixed regardless of detection timing.

When a single malignancy or condition exists on a spectrum, containing both highly aggressive (rapid) and slow-progressing (indolent) variants.

The assignment of participants into the screening group versus the control group, ensuring equal distribution of prognosis/disease severity.

The indolent (slow-progressing) form, because it takes time for the disease to manifest enough to be detected at fixed intervals.