#616: More Testing, Better Health? Harms of Overdiagnosis & Overtreatment - Austin Baraki, MD

On that last point, essentially it's much more important for a screening test to be able to detect false negatives rather than false positives, at least at this stage of what we're trying to achieve.

So I'll go into this, these concepts a little bit more.

Th- this, uh, biostatistics is kind of like the bane of most medical students' existence when they go through the training.

So I think going through it maybe a different way can also be helpful for people 'cause, 'cause ultimately, medicine and diagnosis is really much more a game of uncertainty and probabilities than people realize.

We use a lot of Bayesian inference when we're actually in clinical practice, where the goal is to either increase or decrease the suspicion of a particular condition.

So in other words- If somebody shows up to the clinic and they present with either a particular symptom or they show up with no symptoms at all, I have some sort of a pre-test probability or a pre-test suspicion that they may have a particular condition and I might do a test or obtain some clinical information by asking questions, by examining them, or by doing medical testing that adds information to my like internal kind of calculus of probabilities and that either increases or decreases my post-test probability.

So it can say when they came in, you know, my suspicion that they have active colon cancer was pretty low, but then when they told me they've been losing weight, I found them to be anemic on a te- on a blood test, and then I sent them for a colonoscopy and it found a tumor, now my post-test probability is much higher that they have colon cancer.

So it's all about probabilities and uncertainty.

And this is really important to understand because all tests are imperfect.

This makes the process really challenging, and this is commonly misunderstood, misunderstood publicly.

I think there's a whole lot more faith and confidence put in a lot of tests and imaging modalities and things like that compared to what they actually deserve.

We overestimate the accuracy and reliability of the information we get from biomedical testing.

So this idea, the two fundamental concepts here are sensitivity and specificity.

Sensitivity is like the true positive rate of our test.

It's the fraction of people who have a condition who are going to test positive when you apply a test.

So a high sensitivity test that is positive is gonna catch everyone with it, but can potentially drag a lot of false positives along for the ride.

So in this way, a positive test is not super helpful for ruling in disease, but it can be helpful for prompting further diagnostic evaluation.

Conversely, a negative test is pretty helpful to rule things out for people who don't have it.

For example, you can stop your evaluation there.

On the other hand, this concept of specificity, meaning a test that's very specific for a particular condition, this is-- can be thought of as the true negative rate.

This is the fraction of people who do not have a condition who ultimately test negative for that condition.

So a high specificity test that's positive for a condition is helpful to rule it in.

Now, ideally, a, a gold standard test is gonna have a hundred percent sensitivity and a hundred percent specificity, but of course, this is like pie in the sky.

Hardly ever happens, if ever, 'cause again, all our tests are imperfect to some degree.

They don't-- They, they tend to have a little bit less or potentially a lot less either sensitivity or specificity or both.