Prediction: Astra Zeneca approval date
post by Tim Liptrot (rockthecasbah) · 2021-01-19T21:59:01.555Z · LW · GW · 5 commentsContents
Models Institutional culture Congressional dominance President Trump made a poor principal Final prediction None 5 comments
It's widely publicized that the Trump administration pressured the FDA to release the vaccines earlier by reducing our evidentiary standards in light of the thousands of daily deaths. While not fond of Trump generally, I heartily agree that the FDA should end its vaccine bans earlier than it does. For arguments in favor of earlier authorization, see MR1 and MR2.
The FDA has refused to authorize the AstraZeneca vaccine in the US until the third stage trials are redone, unlike the UK which is already using them. The stated justification is that some participants were given a half dose in the first stage, and therefore the results are ineligible. The half-dose receivers actually had better protection than full-dose receivers, but the FDA rejected the trial anyway. They have said they will consider the new results in April, possibly later. Under a cost-benefit analysis the FDA's decision is pretty crazy, as thousands are dying everyday and vaccine supply is so constrained we've resorted to very inefficient rationing schedules. The potential downsides are just far smaller in expectation than the tens of thousands of lives that we could save. But if the FDA isn't using a cost-benefit calculation, what explains their behavior.
Models
Institutional culture
One explanation is that the FDA just has strange epistemology and ethics. They profess belief that claims unsubstantiated by a high-quality RCT should be treated as false until proven otherwise. That sounds like a strawman, but actually fits their justification for second-doses-fist to Alyssa Vance. Ethically the believes sins of omission do not count but commission does. Killing thousands by banning a vaccine is not immoral, but killing one person by authorizing it is immoral. Here's Tyler Cowen's best guess
At the risk of venturing into psychoanalysis, it is hard for me to avoid the feeling that a lot of public health experts are very risk-averse and they are used to hiding behind RCT results to minimize the chance of blame. They fear committing sins of commission more than committing sins of omission because of their training, they are fairly conformist, they are used to holding entrenched positions of authority, and subconsciously they identify their status and protected positions with good public health outcomes (a correlation usually but not always true), and so they have self-deceived into pursuing their status and security rather than the actual outcomes. Doing a back of the envelope calculation to support their recommendation against First Doses First would expose that cognitive dissonance and thus it is an uncomfortable activity they shy away from. Instead, they prefer to dip their toes into the water by citing “a single argument” and running away from a full comparison.
This model would predict late release. Biden ran on giving "the science" more autonomy. The FDA would happily use that autonomy to delay AstraZeneca approval (I put "the science" in quotes because understanding vaccine biology and doing cost-benefit analysis are two different skills).
Congressional dominance
Nancy Pelosi publicly called for stricter vaccine approval standards prior to the election. The congress does control the FDA's budget, so she has influence. That comment could have given the FDA bureaucrats the cover to slow EUAs.
Under this model a clever FDA delays authorization when a challenger is about to win big, but authorizes under a durable incumbent.
President Trump made a poor principal
In presidential systems, many voters just use elections as referendums on the president. Therefore the president has a strong incentive to improve people's welfare through all available means, while most state actors have weak or conflicting incentives. Since granting EUA's are positive in expected utility (make people happier), the president has good reason to push for them. It was no coincidence that Donald Trump suddenly became fascinated with FDA approval schedules. If anything, Trump was unsuccessful at pushing his agent (the FDA) to do what was best for him and us. Hat tip to Nate Silver
Hot take, presidents seeking reelection is good. We control the votes, so the president must compete to please us more than the challenger. Thus elections align the presidents incentives with ours and representative democracy works! Hooray for competitive elitism!
Under this model, Biden should realize EUAs are great for the Democratic Party. If the FDA releases AZ after he takes office, Biden gets credit/avoids blame. Biden need only incentivize the FDA successfully. I expect Biden would be more successful than Trump given Biden's greater experience.
Final prediction
Metaculus has a 25% chance of AZ EUA by March 1st. The second and third models make me more optimistic. I give a 50% chance by March 1st.
5 comments
Comments sorted by top scores.
comment by ChristianKl · 2021-01-21T19:23:48.056Z · LW(p) · GW(p)
One explanation is that the FDA just has strange epistemology and ethics. They profess belief that claims unsubstantiated by a high-quality RCT should be treated as false until proven otherwise. That sounds like a strawman, but actually fits their justification for second-doses-fist to Alyssa Vance.
I don't think that's surprising. It they would allow people to approve a drug based on a single high-quality RCT a company could simply run 20 RCT for different outcomes with the hope one of them being statistically significant and then get their drug approved based on the statistically significant trial.
Replies from: rockthecasbah↑ comment by Tim Liptrot (rockthecasbah) · 2021-01-22T17:08:31.822Z · LW(p) · GW(p)
I agree that this is the FDA's model. It implies a "strange" unawareness of the underlying statistics. The p-values of the vaccines are really really small because their effect sizes are may orders of magnitudes above most drugs. Also the drug companies in this case could not run many RCTs. So if the FDA were maximizing lives they would authorize them. If they are making an incentive system that requires no deep understanding of statistics, it's not "strange".
Importantly, the FDA is "strange" relative to the calculus of politicians, who to survive must be utilitarian and use induction.
comment by Ericf · 2021-01-20T01:59:00.399Z · LW(p) · GW(p)
Where's the interactive prediction plugin for this post? :-)
If FDA is saying April, I would say 10% by March 1. 75% by April 1. 99% by May 1. 1% some freak event delays it until late in the year or never
Replies from: rockthecasbah↑ comment by Tim Liptrot (rockthecasbah) · 2021-01-23T19:57:47.676Z · LW(p) · GW(p)
I do not know how to add a plugin, I would have done so. It is active on Metaculus currently.
Seems like we could bet about it being approved before April 1st profitably. I am comfortable with this bet because I assumed the FDA/CDC conservatives would win on vaccination priority and first doses first, but Alex Tabarrok won the first and is starting to win the second.
Replies from: Ericf↑ comment by Ericf · 2021-01-23T23:27:35.006Z · LW(p) · GW(p)
Sure. Just for fun/norm setting, I'll bet $10 against your $10 that approval does not happen before March 15, assuming you accept the bet before Feb 1, and it wasn't approved at the time you accepted. That's your 50% by March 1, with 2 weeks of leeway for you since a week has now passed without approval, and I don't feel like doing complicated math.