[grandalf]

What Gawande describes is more of a sysematic irrationality in the healthcare system than a flaw with a test.

The decision to over-treat is part of the tradeoff we get when physicians are viewed as authority figures. Their inaction (not doing a treatment) is viewed as a delegitimization of the patient's needs, and so there is social pressure to treat, even when harm could be caused. This is a psychological blind spot that equally effects patient and physician.

But with respect to tests, so long as a test has a known false positive and false negative rate, its result can be accurately factored into a probabilistic model of a patient's overall health.

Our healthcare system is biased toward acute conditions and extreme interventions. Things like early disease progression and wellness are generally not even considered relevant to most doctors.

The reasoning approach of an evidence-based differential diagnosis which is taught to medical students is a powerful heuristic, but it is designed to work within the constraints of acute illness and (potentially) urgent intervention. So of course if fails when test results are considered without appropriate measures to improve the signal to noise ratio of the first branch of the decision tree.

With any kind of broad-spectrum, speculative testing, any result would need to be considered over time and in the context of many other factors. It is not a drop-in replacement for any step of the traditional differential.

[cortesoft]

"But with respect to tests, so long as a test has a known false positive and false negative rate, its result can be accurately factored into a probabilistic model of a patient's overall health."

In order to do that, you would also need to know the correlations BETWEEN tests in terms of false positive and negative. And there are a lot of combinations.

[xorcist]

Not only that, many countries have health care systems where the same person who earns their living from testing also earns their living from treatments. That's a tough premise to work with if you're designing health care for Utopia.

[gvb]

> its result can be accurately factored into a probabilistic model of a patient's overall health.

But my health with respect to an illness is not probabilistic[1]. I either have the illness or don't have the illness. Probabilities are not useful when the sample size is one (me).

[1] Pedantic: it is probabilistic, but the probability is either 0% or 100% because the confidence interval sucks.

[TeMPOraL]

It's probabilistic because you don't know whether or not you have an illness. It's like tossing a coin - it's 100% on one side and 0% on the other, but you don't know which side it is on until you check - that's why we say a fair coin has a 50% probability of landing on either side when tossed.

Now the test you use with that (mathematical) coin is 100% accurate. Tests in medicine are not. They're more like "oh I see you sort of seem to have X; X has been known to occur a bit more in people suffering from Y than in those not suffering from Y". Hence the uncertainty.