I understand that it is extensive data (for about 44,000 patients), but taking 3 weeks to compile/analyze seems excessive. They should have put in an army of analyzers into it...
There are about 3,000 people dying every day.... it is like 9/11 every day, or 9 jumbo jets falling....
The FDA should be treating this as a war effort, and not drag their feet into pointless bureaucratic 'safety' theater.
This is wrong. Median expected remaining healthy lifespan of a 9/11 deceased was ~35 years, whereas for a COVID decease it's ~3.5 years. So 9/11 is still 10x worse.
Don't get me wrong, COVID is still terrible, but it is not a "9/11 every day".
https://www.pnas.org/content/117/36/22035
> We calculate, using these same cohort life tables, that the average person dying of COVID-19 had 11.7 y of remaining life expectancy
From the paper:
> Both of the above calculations may overstate the loss of remaining life in that they assign the remaining life expectancy based only on age, without taking into account that COVID-19 deaths are disproportionately occurring among those with compromised health status.
So their 11.7 average years remaining ignores the critical bit of information that we know this person died of COVID. It is extremely unlikely that if you took 2 80 year olds, one who died of COVID and one who did not, that if neither got COVID they would live to be the same age. I wouldn't be surprised if it was a 10x difference. While they do mention this and add some caveats (a good thing), I think because it completely changes the conclusions of the paper, they should have gone with a range and not a 11.7 number, which seems very improbable.
> Avoiding 1.75 million deaths or 20.5 trillion person years of life lost would be valued at $10.2 to $17.5 trillion
Then they take the 11.7 figure from above and multiply it by the estimated average value of a year of life. Again, there is at least an order of magnitude difference between a year of life at age 90 and a year of life at age 30, for most people.
Finally, although my ~3.5 number is much closer to the truth than 11.7, there is a difference between life expectancy and "healthy lifespan", aka "healthspan" (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6136295/). Healthspan is a newer concept but I expect will be the primary term in a decade or two.
Thank you for providing the link, a good read.
The next sentence:
> Hanlon et al. (10) estimate that those dying from COVID-19 have only about 1 y less of remaining life on average than those at the same age in the general population, which would mean that the overstatement is not very large, around 8%.
Your own link states that healthspan is not quantifiable, so your claim that it is 3.5y doesn't make a whole lot of sense.
> "used to construct a plausible scenario for the prevalence of combinations of LTCs among people who died from COVID-19 for the modelling presented here."
Second, the dataset used at the time to generate their synthetic data was weak. You can see it immediately in their bar charts, where they have mostly healthy 50-69 y.o. dying in their synthetic data, skewing their YLL to the right.
> Your own link states that healthspan is not quantifiable, so your claim that it is 3.5y doesn't make a whole lot of sense.
I provided a link to a "conservative" description of healthspan, but in the field people are more optimistic about it.
Very good questions you brought up though, thanks!
The calculation you reference was done using all people in the US. Just two sentences later they wrote, "... would on average lose about 0.2 y of remaining life ..." You need to compare to the groups of affected people, which is generally not 100% of people in the US.
average remaining lifespan across the entire population:
> we calculate that the 2020 American population of 330 million people has on average 45.8 y of remaining life expectancy, totaling 14.9 billion person years.
reduction in remaining lifespan in the 70-89 age group:
> Older individuals ages 70 to 89 y, taking those who die and those who survive together, would on average lose about 0.2 y of remaining life, and younger individuals would lose far less.
I did not include these calculations because they are not comparable to the number to which I was responding.
Ok, math is off, but covid is going to last for at least 2 more months with death rates like this and that accrues.
Also, do you know what they call it when I kill your grandmother 2 days before she was supposed to die of old age?
Murder.
Just because someone doesn't have much life left makes it no less of a tragedy when they die early.
At the risk of being cold hearted, isn’t this exactly what it does?
https://www.worldvision.com.au/global-water-crisis-facts#:~:....
No one gives a shit.
You can't make a baby in 1 month with 9 women.
If you managed to give 5 million doses a week earlier, you save about 200 lives (about 40 deaths per million per week).
> If you managed to give 5 million doses a week earlier, you save about 200 lives (about 40 deaths per million per week).
No, that’s wrong.
https://marginalrevolution.com/marginalrevolution/2020/12/th...
No we didn't. :P The US has 6.4 million doses of the Pfizer vaccine now. Projections of up to 50M delivered worldwide by the end of the year. We're on a crazy ramp.
If you want to complain about a misallocation, a key complaint would be that they're holding slightly more than half the doses back to ensure everyone dosed can get a second dose... but with rolling production such a large holdback isn't necessary.
> What could matter is if delay increases the speed at which you can ramp up.
All the jurisdictions I know will blow through the entire supply of existing doses allocated to them very quickly. It doesn't go on a slow distribution ramp gated by approval like your ridiculous source suggests. What governs the ramp is production rates, which have been running flat-out irrespective of this decision.
You misunderstand. "Blowing through entire supply" and being left with nothing is exactly the "slow distribution" Alex Tabarrok, a professor at GMU, talks about in the link I gave. He contrasts it with vaccinating entire population on day 1. Read it carefully.
> What governs the ramp is production rates, which have been running flat-out irrespective of this decision.
I consider the vaccine production regulation people to be just as incapable at cost-benefit analysis as vaccine approval people, so I assume that most production constraints are just as artificial as the approval constraints.
>You misunderstand. "Blowing through entire supply" and being left with nothing is exactly the "slow distribution" Alex Tabarrok, a professor at GMU, talks about in the link I gave. He contrasts it with vaccinating entire population on day 1. Read it carefully.
Vs.
"Now assume that the vaccine can’t roll out to everyone immediately. For the sake of this simple model let’s assume that on day one you can only vaccinate half the population. By doing so you save 1000 lives on day 1 and 2000 lives every day thereafter for the length of the pandemic. That’s the fourth panel. Now suppose we delay the vaccine rollout by one day. 2000 people die on Day 1 but you save 1000 on Day 2 and 2000 on Day 3 and every day thereafter for the length of the pandemic. How many people were killed by the delay? Compare the 4th and 5th panels. 2000 exactly as before! "
This compares vaccinating 50% on day 1, and the remaining 50% on day 2, to vaccinating 50% on day 2 and the remaining 50% on day 3.
But in blowing through the supply, we're comparing vaccinating 1% on day 1 and another 1% on day 7 and another 1% on day 15 to vaccinating 1% on day 7 and 1% on day 8 and 1% on day 15. Not at all the same scenario.
https://marginalrevolution.com/marginalrevolution/2020/12/th...
The rigor is to give the public confidence that we're better off getting the vaccine than not.
The first two phases are the safety tests, the third one is efficacy (so we knew the vaccine was "safe" after phase 2, we just didn't know the efficacy - that's what the poster above is alluding to).
And are you suggesting that it’s easier because the data isn’t as extensive as what the rest of us deal with? That only makes it harder.
Sure if it’s some random A/B test with data from your own systems from an experiment you designed, but this is a vaccine the whole fucking world has been waiting for for most of the year. God forbid people treat it a little more carefully.
I would absolutely need to start a performance improvement process. They are definitionally 1/3 as productive as they should be. Also in terms of stakes, please show me even a modicum of work that suggests the downside risk of approval is greater than the delayed vaccinations. What reason have I to believe there is great risk in approval? Again, the UK approved this over a week ago.
> And are you suggesting that it’s easier because the data isn’t as extensive as what the rest of us deal with? That only makes it harder.
No, I'm mocking the idea of it being extensive data. It's not.
> but this is a vaccine the whole fucking world has been waiting for for most of the year.
And that's reason to go slower!? There's no need for this delay!!! https://thedispatch.com/p/fda-career-staff-are-delaying-the
As I understand it, medical ethics isn’t about utilitarianism, or else we’d skip the rats and monkeys and just go straight to testing on the poor for money. They don’t seem to prioritize minimizing the opportunity cost of avoiding bad outcomes. Their guiding principle seems to be avoiding giving harmful treatment.
Meanwhile some healthcare workers at my partner’s facility are skeptical about whether the vaccine was rushed (I think they’re crazy, for the record). But there’s real risk if enough of the public doesn’t trust the vaccine. I’d bet that kind of angle is part of why they focus on “do no harm” instead of “minimize cumulative regret.”
Answering the question of whether the experiment harmed anything important is always much harder, as I’m sure you know. You see some metric is way down, but after a multiple comparisons correction it might just as well be noise, so then you have to start digging, doing a bespoke analysis for the fact that these non iid metrics point in these particular directions. You can’t pre-automate it because with N metrics that are practically significantly up/flat/down, that’s at least 3^N possible investigations to convince yourself whether it’s enough to stop rollout.
That’s the part that takes time, in any analysis I’ve ever done. Maybe literally zero time to check efficacy because that part is automated, then sometimes days to check whether there’s anything else worrisome in the probably-just-noise on other metrics, spending time proportional to the cost of getting the harm part wrong. Some of the hard questions aren’t statistical. You see some KPIs definitely up and some KPIs definitely down and deciding whether to roll out becomes a bitter fight between competing interests. Maybe you don’t roll it out yet for client segment X (like this trial and kids and pregnant women, as I understand it).
Re US vs UK, no idea. I’m not saying 3 weeks is necessarily the right amount of time here. I’m just so taken aback at the idea that 3 weeks is necessarily too long to the point you “would probably have to start firing people” and would “absolutely” have to start putting them on PIPs.
I'm glad we have actual specialists checking the data and not people thinking it's only putting stuff on excel and writing a function
Why is the UK able to approve it over a week ago but the FDA needs longer?
Where you think processing the data just means loading up an excel table (or running a CSV through R) and that's the only data relating to the studies. Patient data is not the only thing being analyzed and it includes things like manufacturing, transport, etc
> Where is this MD, MPH from Johns Hopkins wrong?
He's not wrong but his worries as a clinical trial expert are a subset of the worries of an agency like the FDA. And I'm sure his workflow as clinical study expert is more fine tuned for that.
Or as in a developer analogy: making a code change is a 5 minute job. Actually making the change, testing, deploying, etc may take a whole day.
> Why is the UK able to approve it over a week ago but the FDA needs longer?
Pfizer sent the last info to the FDA on 22th of November. The MHRA had started a rolling review and received the last batch of info they needed on the 23rd of November.
I am certainly frustrated with the speed as well, but I don't think it's too much of a difference. Yes, maybe it could be faster.
Is it possible you could construct an even more ridiculous straw man?
> I am certainly frustrated with the speed as well... Yes, maybe it could be faster.
Oh, so you agree with what I'm saying and your problem is the straw man argument you created.
Sure, you can come up with some standard numbers in an hour. But it wouldn't be the most thoughtful analysis, and the decision isn't really about a single efficacy estimate anyway.
I guess I figure if you're going to go down this route of pissing matches over how fast you can do an analysis on how big of a dataset, why the hell even have an FDA anyway? Just require the company post all their information and get rid of the regulation entirely. I actually think this isn't an unreasonable argument, even if I don't necessarily agree or disagree with it. It certainly seems more reasonable to me than "we should have a regulator that just rubber stamps things."
I'm fine with the FDA taking their time. The problem with risk in decisions is people forget in high stakes high risk decision making, it's about the process. This whole discussion about the FDA taking too long will look a lot different if a year from now the vaccine turns out to have some horrible long term effects (which it could, even though I think it's unlikely, just, you know, 'cause).
No one is saying don't be thorough. What I'm saying and Makary is saying and the UK is saying is that the required thoroughness shouldn't take this long and when it does, real human lives are lost and sustained invasive measures in the freedoms of individuals are required.
> I guess I figure if you're going to go down this route of pissing matches over how fast you can do an analysis on how big of a dataset, why the hell even have an FDA anyway?
What a ridiculous leap. That's so many logical steps away from "the FDA moves way too slow" it's an absurdity. Who is arguing against the FDA's existence other than I guess you now sort of???