Thread by @michaelzlin

A lesson on proper trial design: Was thinking, if nAb levels at 3% of convalescent plasma (CP) prevent severe COVID (nature.com/articles/s4159…) then infusing 1U of CP (5% body volume) should prevent death.

And it does! jci.org/articles/view/…

But weren't we told CP was worthless?

Of course it was the @nytimes who wrote the obituary for convalescent plasma. Maybe nothing factually wrong in their stories, but their writers generally lack the training to dig deeper and find the signal in the noise.

nytimes.com/2021/04/17/hea…

"When a treatment fails, which is often, it can be difficult for its strongest proponents to let it go. Eventually, studies did emerge to suggest that under the right conditions, plasma might help."

Sloppy @nytimes. 1st sentence implies CP failed. 2nd says success. Which is it?

Turns out there are two populations in which CP shows a benefit. One is if you give it early to older adults (within 3 days of symptoms). These are at risk for severe disease, and the early CP helps prevent that.
nejm.org/doi/full/10.10…

We now have evidence that there is a second population that benefits: severe patients already in the hospital (this month's study). There is a ~50% reduction in mortality with no other benefit.

This is consistent with even 5% of the mean convalescent nAb levels being beneficial. Note convalescent plasma contains antibodies but no T cells.

The failure to see a benefit in other trials is mostly because most people would have been fine anyway

That is, the nAbs diluted to 5% of their normal levels only tips the balance in the margins; saves some susceptible people from the hospital if given early, and saves 50% of those who would die from dying.

When these people are diluted into a larger study population, the effect becomes statistically insignificant. It's another example where lack of evidence is not evidence of lack.

So despite @nytimes' medical nihilism, early broad use of convalescent plasma was the right call. There are probably many people alive today because they got CP.

And the case of convalescent plasma shows the importance of identifying the right patient population for trials to succeed.

Not all negative results are indication of failure. You have to find the signal in the noise.