Hacker News Reader

Comment by jrockway | Hacker News Reader

Tests are not 100% accurate, so running a test without a good reason can lead to a lot of unnecessary medical intervention. A false positive can lead to thousands of dollars of expense and not getting better.

View on HN

This reasoning only makes sense in the context of a dysfunctional system. "We can't gather information because we will react in stupid ways".

Surely the correct response for noninvasive tests is to run the test, and raise the threshold for intervention?

Or if it's cheap, run the test twice. And you're presumably going to have other symptoms that corroborate the results. Maybe the symptoms of X are ambiguous alone, but what are the odds you have a positive test for X and matching symptoms but not X itself? Not zero, but lower.

Running most tests twice won't necessarily give you any useful additional information. Tests generally don't give binary results. Like a hydrogen breath test will give a value within a range, but a high value isn't necessarily definitive for diagnosis of SIBO. It's just one piece of evidence.

https://my.clevelandclinic.org/health/diagnostics/12360-hydr...

There are well-established frameworks in statistics and decision theory for handling this kind of situation, some of them in continuous successful use for over a century. It's not exactly a solved problem, but we aren't fumbling around in the dark either.

Those well-established frameworks in statistics and decision theory are unfortunately mostly irrelevant in complex medical cases like this. There is no reliable source of truth to index against. We often don't have a reliable scientific basis for taking the result of a particular quantitative test and calculating the probability that a patient has (or doesn't have) a particular condition. Estimates of selectivity and sensitivity for particular tests are often wildly off. And when there are multiple tests involved we often don't know the correlation coefficients. Even the definitions of particular conditions are often somewhat arbitrary and not particularly biologically relevant.

Software developers often have this unrealistic fantasy about how things ought to work in medicine and biology. They reality is far messier.

Fortunately I'm not a software developer!

Also yes, you're right about having good estimates for medical tests.

Mainly I take issue with your claim that "Running most tests twice won't necessarily give you any useful additional information." As long as it is directionally associated with the thing you're trying to test for (i.e. it has any medical value at all), then running it twice absolutely should give you additional information. "Not necessarily" I guess is true, but again, probability gives us the tools to reason about situations like this even if we can't estimate population parameters reliably.

In my particular case, I had numerous other more expensive and invasive tests before my gastro finally decided to even offer the breath test. The treatment is a course of antibiotics for a week.

Reply on news.ycombinator.com