The Lifeboat News
[ Message Archive | The Lifeboat News ]

    I'm a fan of Bayesian statistics - better than standard statistics for clinical trials...probably Archived Message

    Posted by Dan on October 11, 2020, 12:07 am, in reply to "Lies, Damned Lies and Health Statistics – the Deadly Danger of False Positives"

    It's ironic that you should fairly castigate GSK and pharma in other posts and then use an article by someone who uses his position in the pharmaceutical industry as a call to authority.

    This article is notable for its omissions. He must know that he's missed half of the Bayes position out. Reminds me of Pinter's description of writers who lie. They stop being writers and become politicians.

    For anyone who thinks this article usefully takes down testing.

    All tests have false positives and negatives. Briefly, Bayes says that the probability of a result being false depends not just on the reliability of the test itself but also on the conditions of the experiment, so for a situation in which there are genuinely no positive cases, any positive test must be a false positive. What's missing from the article is the corollary - in a situation where all the tested are genuinely positive then the chances of a false positive are zero.

    For a test with a 1% false positive rate when 10,000 people without the disease are tested then 100 will be positive. This rate is constant so if the number of tests being performed stays the same, and there is no covid, then the number of (falsley) positive tests should also remain static..... or should increase proportionally with increased number of tests.

    The government has stopped publishing the number of daily tests so it's not possible to know the denominator. There is weekly data which suggests that the testing capacity has increased by 40% since August. The number of positive tests has increased many times more than this.

    As we are seeing more admissions with respiratory failure I'm sure the proportion of false positives is going down.

    He also assumes that the people being tested over the summer had the same low incidence as the whole population. Given how hard it was to get a test in the last few months I would assume that it was people with viral symptoms who got tested and had a higher incidence of covid the general population. Therefore his assertion that the pre test probability is low is also an assumption requiring proof.

    Bayes is a bit mind boggling, so no surprise that the article is bamboozling but this guy from the pharma industry is lying by omission. I don't know what the incidence is, or will be. Without transparent testing and data publication which includes the number of tests being performed, it's still not possible to know where we are. To claim that that invalidates a standard medical test is flam.

    Cheers,
    dan

    Message Thread: