Since it's World Antibiotic Awareness Week, let's revisit the recent article in which Llewelyn et al. convincingly argue that it's time to drop the message that not finishing a course of antibiotics contributes to resistance. Because it's incorrect
I once tried to track down the origins of this idea. I found dozens of papers claiming it, citing others, which cited others, and every chain went back to one source: Fleming's conjecture at his Nobel speech. Neither theory nor evidence support it.
From my reading of the sources, much of the original idea was the correct idea that marginally lethal doses of antibiotic could promote resistance (which is certainly true) but that this got conflated with the length of course at a fixed high dose, which is different
One thing that gets confused a lot in the literature on it is patient disease outcome vs antibiotic resistance. There are cases in which longer courses often do give better patient outcomes, but this is not the same as preventing resistance.
We certainly should not give antibiotics when they aren't needed. But we need to seriously question when that includes the latter part of a patient's course of antibiotics.
You wouldn't give abx to a healthy patient, why give them to a patient after the infection is cleared?
(Oh and the above article is behind a paywall, if that's an obstacle for you I strongly encourage searching for other sources that might let you read it anyway. It's worth it)
If we're serious about controlling antibiotic resistance (and we should be), we need evidence on what does and does not make resistance worse, not 75-year-old conjecture. Because only when we know what practices, if any, make a difference, can we change our ways for the better
This is the deeper question all of this raises: if not finishing antibiotic courses isn't meaningfully contributing to resistance, then what other assumptions about the drivers of resistance are we getting wrong? What human behaviors make a difference? Or even, do any?
Since it’s Twitter I want to be very clear I am not saying “ignore your doctor’s advice and freelance antibiotics.” That’s a terrible idea. I’m saying we, as a medical and scientific community, need to reassess guidance and tailor it based on evidence over tradition.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Others have worked on this problem, but there's a fundamental issue with trying to quantify variants: the mutations that define variants are so far apart that they never appear on the same read, and often not even on the same molecule(!) in heavily degraded wastewater RNA 2/
Luckily @jasmijnbaaijens had a critical insight: this is computationally identical to RNAseq quantification! You have an unknown mixture of transcripts (known variants) which you've chopped up and noisily turned into sequencing reads, and are inferring the original mixture 3/
I don't think it's widely appreciated how incredible an achievement this is. Biotechnology has advanced unbelievably in the last fifteen years, but even still, going from new virus to completed phase 3 clinical trials in eleven months is like... I can't come up a good metaphor
Maybe announcing a Mars program and landing a crew twelve months later? It's certainly on par with the Manhattan project.
Though the dissonance of this incredible technical achievement against the tragedy of our utter failure of public health leadership and policy is jarring
Excited to share my latest preprint with @LeeKShaffer and @BillHanage, “Perfect as the Enemy of the Good: Using Low-Sensitivity Tests to Mitigate SARS-CoV-2 Outbreaks” in which we show how the math of superspreading events can improve contact tracing 1/ dash.harvard.edu/handle/1/37363…
The key idea is: if A is sick and has contacted B, B is probably still fine, but if you also know that A has infected C then there's a much better chance that B has been infected. Superspreading (or overdispersion) means that infection _events_ are correlated 2/
Here's where testing comes in: because B and C being infected is correlated, you don't need a test that gets both of them right all the time. Either one testing positive gives you information. So a low sensitivity test on all of A's contacts is almost as good as a perfect one. 3/
I see a lot of motivated reasoning as to why this can't be as bad as serious models predict be without massive societal action. And I know these are desperate attempts to reason why the world must be similar to past experience, but it's hard to be sympathetic.
But all those thinkpieces two or three weeks ago, what did they accomplish? They sowed just enough doubt to slow action (and apparently some Medium posts got the ear of the White House). And now we are seeing the tragic consequences of insisting the world must be as you hope.
The curve is bending, our fates are not fully sealed. Hold the line. Keep distancing, be safe to reduce the background rate of hospitalization. Listen to epidemiologists with the experience to understand the complications and nuance. This will be a long fight.
I just did an updated calculation of what happens to America if we do nothing. And it is nothing short of terrifying.
The current rate of spread is a near-perfect exponential. If we do not change our behavior dramatically and fast, here is what the math says: 1/n
The last eleven days give a remarkably good fit for linear regression on the log cases (R^2=0.9981), that's good enough to project the exponential. Here's what happens:
~March 18th the US passes 10k cases
~March 26th we pass 100k cases
~April 4th the US passes 1 million cases
If 10% require hospitalization:
~April 5th the US passes 1.6M cases, or 10x the number of ventilators
~April 11th we 9.24M cases, or 10x the total hospital beds
If we do not change our behavior, by early April the entire US medical system will be treating critical Covid-19 cases
The reason to cancel meetings and seminar visits is the same reason we have them in the first place: by establishing long-distance connections and high-connectivity nodes, we help ideas spread much faster through our social networks. It's the same for a virus.
More math: Locally, early in an outbreak, the expected impact of a social event scales as the number of people times the number each interacts with. Roughly the attendance squared.
Therefore cancelling a 50 person event is over a 1000 times as important as cancelling a 1-1.
Of course this eventually becomes linear, at a 10,000 person event you can't possible interact with everyone. But the point remains that from an outbreak-spreading perspective large events are disproportionately more important than small ones.