Tuesday, 7 July 2020

Crap detector needed

Anybody there? there? there?? There is no "better" time for inhabiting an echo-chamber. It is certainly an extremely fashionable domicile occupied by all sorts of big name Twitterati, Celebs and Influencers [TCI].  Now whatever your virtues, you really don't get to be a major TCI if you are a shrinking violet with a strong sense of your own humility and fallibility. When it comes to public health, it's not a problem if, say, David Beckham or Justin Bieber (who are rich, fit, personable and good looking) make an announcement about diet, mental health or medical treatment. Because you'll discount their opinions with the realisation that neither is, nor claims to be, super bright.  It is of course mildly annoying if you see Gwyneth Paltrow making millions on her goopy "health" products but you can take comfort that it's only money that the buyers-in are losing . . . not their lives.

It's much more worrying when scientifically qualified people start spouting their thoughts on matters medical when those ideas are based more on scienciness than science. Linus Pauling and Vitamin C is the prime example. Life is too short, and brain too small, for us to evaluate, from first principles, every fact and factoid that comes lollopping over the horizon on the TV news or Twitter. Rather than re-invent the wheel again we turn to some authoritative source to see what they think of the new data.

I am a big fan of John Ioannidis, the iconoclastic number cruncher from Stanford [bloboprev I - II - III]. In April he suggested that the Covid-19 death rate was getting exaggerated because not all the undead covid-positive people were being identified. He lamented that there was no good data on this population prevalence. The thought being the deed, within a few weeks he'd recruited more than 3,000 people in Santa Clara County, California and given them a test. That study found that  the rate of infection in Random County USA was maybe 50x higher than the numbers who were known to the authorities. And that meant that the death rate was 50x lower, indeed about the same as seasonal 'flu. Partly because that was putting it to the man and partly because of confirmation bias in my mind that the lock-down was an over-reaction, I bruited the news abroad on The Blob and elsewhere. Finding out the rate of [sub-clinical] infection is really important to develop scientific evidence based policy . . . but the Santa Clara County study is cutting no mustard for addressing the question. So I hereby, far too late, withdraw my endorsement.

You see, sounding off in a we-print-everything-submitted blog or with more brevity on Twitter is easy. Doing the heavy statistical lifting is not. Before I'd even given Santa Clara my imprimatur Andrew Gelman, a competent statistician from Columbia U., had skewered Ioannidis paper as fatally flawed. “I think they need to apologize because these were avoidable screw-ups. They’re the kind of screw-ups that happen if you want to leap out with an exciting finding and you don’t look too carefully at what you might have done wrong.” What Team Stanford had done was assume that the false positive rate [people who test positive but are in fact covid-free] was 0.5% and took their calculations forward from there. What Gelman noted is that 0.5% is the best guess for false positives but they could be as low as 0.1% or as high as 1.7% because the analysis establishing the rate was carried out on only a few hundred people. This is real worrying because Ioannidis had found a rate of 1.5% among his random street sample . . . they could all be false positives! Maybe not all, probably not all, but we are all at sea about what the actual rate of infection is in Santa Clara County or anywhere else. Check out Stephanie Lee at Buzzfeed for a nice Exec Summary of the statistical niceties, written in normal speak. And while you are there check out Stephanie Lee at Buzzfeed noting that there was probably a conflict of interest in the scientific research because a major funder of the Stanford study was running a lockdown-impacted airline.

They say that Justice delayed is Justice denied but the opposite might be true in science. Too much competition means that scientists are all in a rush to be firstest with the mostest. But rushing results into print in peer-reviewed journals or the biorxiv pre-print server, serves science poorly and may result big red faces when retractions are called for.

No comments:

Post a Comment