There is still far too much uncertainty in how effective Paxlovid is, due to the trial being halted early – the idea that we know what we need to know here already is absurd.
The chi-squared statistic (df=1) on hospitalization is 20.23, p<.00001. This is strong evidence against non-efficacy. What’s your prior on non-efficacy? Or how unstable do you think these sample proportions are at this N? I’ve got a (non-Bayesian) 95% CI on treatment efficacy against hospitalization at (64%, 97%), so sure there’s uncertainty in how effective it is, but I think we know what we need to know here—it easily meets the efficacy bar for approval, and it’s highly effective. Would it be nicer to narrow our confidence interval more? For the sake of basic knowledge, sure, but meh. If there were other treatment options hitting in the same ballpark pre-hospitalization and we wanted to be choosy among them, then yes, but there aren’t (correct me?), so what’s the complaint here?
it should mostly replace prevention efforts other than vaccination.
It’s a treatment, not a prophylaxis. This prevents hospitalization/death, not infection. So for preventing infections (if that is a goal), NPIs are still relevant.
it should mostly replace prevention efforts other than vaccination.
It’s a treatment, not a prophylaxis. This prevents hospitalization/death, not infection. So for preventing infections (if that is a goal), NPIs are still relevant.
The treatment means that it’s less bad if you get infected. Therefore it decreases the benefits of preventing infection. Therefore, infection-prevention efforts that were previously justified in a cost-benefit analysis might no longer be justified (on the margin).
For sake of illustration, if the treatment had negligible cost and total availability and was 100% effective at eliminating all COVID symptoms, then it would be essentially pointless to make any prevention efforts anymore.
On the effectiveness question, I think 64% and 97% are hugely different numbers—they are an order of magnitude difference in remaining risk. So a CI that wide to me very much screams not enough data in terms of deciding how to act. I expect tons of expensive prevention efforts to be justified by that 64% number, that would be much harder to justify if we could be confident in 89%, and if it was known to be 97% would be much harder still.
On treatment versus prophylaxis, yes I understand that, but treatment and prophylaxis can be either complements or substitutes depending on the situation—if you have a good enough treatment it makes it less important to do prophylaxis (and vice versa).
The chi-squared statistic (df=1) on hospitalization is 20.23, p<.00001. This is strong evidence against non-efficacy. What’s your prior on non-efficacy? Or how unstable do you think these sample proportions are at this N? I’ve got a (non-Bayesian) 95% CI on treatment efficacy against hospitalization at (64%, 97%), so sure there’s uncertainty in how effective it is, but I think we know what we need to know here—it easily meets the efficacy bar for approval, and it’s highly effective. Would it be nicer to narrow our confidence interval more? For the sake of basic knowledge, sure, but meh. If there were other treatment options hitting in the same ballpark pre-hospitalization and we wanted to be choosy among them, then yes, but there aren’t (correct me?), so what’s the complaint here?
It’s a treatment, not a prophylaxis. This prevents hospitalization/death, not infection. So for preventing infections (if that is a goal), NPIs are still relevant.
To end on a positive note: good post, as always.
The treatment means that it’s less bad if you get infected. Therefore it decreases the benefits of preventing infection. Therefore, infection-prevention efforts that were previously justified in a cost-benefit analysis might no longer be justified (on the margin).
For sake of illustration, if the treatment had negligible cost and total availability and was 100% effective at eliminating all COVID symptoms, then it would be essentially pointless to make any prevention efforts anymore.
Good point!
On the effectiveness question, I think 64% and 97% are hugely different numbers—they are an order of magnitude difference in remaining risk. So a CI that wide to me very much screams not enough data in terms of deciding how to act. I expect tons of expensive prevention efforts to be justified by that 64% number, that would be much harder to justify if we could be confident in 89%, and if it was known to be 97% would be much harder still.
On treatment versus prophylaxis, yes I understand that, but treatment and prophylaxis can be either complements or substitutes depending on the situation—if you have a good enough treatment it makes it less important to do prophylaxis (and vice versa).