3 ms·

having many data points over the past month is arguably better for a physician than having a single data point from a visit

by _cenw 4y ago

vidarh 4y ago
But only if the false positive rates are low enough, and/or the interventions required to follow up a false positive is small enough, and the benefit of earlier intervention in the case of true positives outweigh those risks. Large scale testing is highly dicey and requires careful consideration because it often takes very low risks from false positives to outweigh the benefits.
- amalgamated_inc 4y ago
  We seem pretty great about not having false positives with blood glucose and diabetes. There are of course lots of false positives in other areas. But I'm sure we could dial in many of those type-1/type-2 errors, or just look at the ones where we're confident in the ratio and that there's a net benefit.
  - vidarh 4y ago
    Of course there are false positives. Any time you do anything at scale there will be, be it due to precision of equipment, equipment failure, or user error. When a true positive is rare, the false positives can easily swamp them. E.g. one Canadian paper claims a glucose test for diabetes gives a false positive rate 6 times higher than true positive in the general population [1] (the numbers will of course be much better in a group where risk factors are present). A challenge is that becoming confident of the ratio is difficult. What do you count? Routine mammography has been scaled back some places because even miniscule false positive rates seemed to rack up enough downsides alongside surprisingly low benefits (you caught things earlier, but the difference turns out to have less impact on survival rates than you might think) when the true positive rate is also tiny. The end result is that the candidate pool for large scale screening programs where we can be confident of a net benefit is surprisingly small, and it may well be we'll see further declines in use of large scale screening rather than increases, as improved treatment often further cuts into the net benefit of early detection. [1] https://pubmed.ncbi.nlm.nih.gov/9368868/ https://pubmed.ncbi.nlm.nih.gov/9368868/
dekhn 4y ago
There's a whole literature and backstory supporting my statement, if you want to engage do the backing research.
- usea 4y ago
  It's not reasonable to demand people do legwork if they want to reply to your naked claims. This is not a forum exclusively for people who are well-versed in the research you're familiar with.
  - dekhn 4y ago
    it's not a naked claim; it's a well understood problem in the field that is trivially researchable and this comes up over and over again on HN.
    - usea 4y ago
      HN is not in "the field," and if it's trivially researchable then you can include links to it yourself. It's not constructive to demand other people are read up on research before interacting with you here.

having many data points over the past month is arguably better for a physician than having a single data point from a visit

by _cenw 4y ago

having many data points over the past month is arguably better for a physician than having a single data point from a visit

vidarh 4y ago

But only if the false positive rates are low enough, and/or the interventions required to follow up a false positive is small enough, and the benefit of earlier intervention in the case of true positives outweigh those risks. Large scale testing is highly dicey and requires careful consideration because it often takes very low risks from false positives to outweigh the benefits.

amalgamated_inc 4y ago
We seem pretty great about not having false positives with blood glucose and diabetes. There are of course lots of false positives in other areas. But I'm sure we could dial in many of those type-1/type-2 errors, or just look at the ones where we're confident in the ratio and that there's a net benefit.
- vidarh 4y ago
  Of course there are false positives. Any time you do anything at scale there will be, be it due to precision of equipment, equipment failure, or user error. When a true positive is rare, the false positives can easily swamp them. E.g. one Canadian paper claims a glucose test for diabetes gives a false positive rate 6 times higher than true positive in the general population [1] (the numbers will of course be much better in a group where risk factors are present). A challenge is that becoming confident of the ratio is difficult. What do you count? Routine mammography has been scaled back some places because even miniscule false positive rates seemed to rack up enough downsides alongside surprisingly low benefits (you caught things earlier, but the difference turns out to have less impact on survival rates than you might think) when the true positive rate is also tiny. The end result is that the candidate pool for large scale screening programs where we can be confident of a net benefit is surprisingly small, and it may well be we'll see further declines in use of large scale screening rather than increases, as improved treatment often further cuts into the net benefit of early detection. [1] https://pubmed.ncbi.nlm.nih.gov/9368868/ https://pubmed.ncbi.nlm.nih.gov/9368868/

dekhn 4y ago

There's a whole literature and backstory supporting my statement, if you want to engage do the backing research.

usea 4y ago
It's not reasonable to demand people do legwork if they want to reply to your naked claims. This is not a forum exclusively for people who are well-versed in the research you're familiar with.
- dekhn 4y ago
  it's not a naked claim; it's a well understood problem in the field that is trivially researchable and this comes up over and over again on HN.
  - usea 4y ago
    HN is not in "the field," and if it's trivially researchable then you can include links to it yourself. It's not constructive to demand other people are read up on research before interacting with you here.