Hacker News Reader

Comment by Clewza313 | Hacker News Reader

The trial in question was conducted on younger people who generally don't get severe cases, so they have no evidence either way on that. However, the trial showed that that the vaccine appeared to nearly completely ineffective in preventing infection:

> Nineteen of the 748 people in the group that was given the vaccine were infected with the new variant, compared with 20 out of 714 people in the group that was given a placebo.

https://www.nytimes.com/live/2021/02/07/world/covid-19-coron...

The sample sizes here are small and the results have not been published, so it's too early to draw any major conclusions, but it certainly doesn't look great.

View on HN

I wasn't aware that we thought any of the vaccines provide sterilising immunity. Has this changed? The Pfizer trial was regarded with some scepticism a while back as they didn't test everyone, so we don't really know how good it was at preventing infection, for example.

It appears that J&J does, and there is reason to be optimistic the Pfeizer and Moderna vaccines do, although the data isn’t in yet.

There was news last week that the AZ vaccine could reduce transmission by 76%. However although widely reported in the media I did see some controversy about this (as yet not peer reviewed) study.

https://www.rte.ie/news/coronavirus/2021/0202/1194681-oxford...

A simple way to calculate Bayesian confidence intervals is sampling from Beta distributions. This is using the Beta(1,1) prior (uniform random) for both vaccine and control population.

The data suggests it is unlikely that the vaccine reduces infection by >=50% (<5% probability) and a probability close to zero it reduces infection by >=70%.

    import numpy as np
    import pandas as pd
    
    # number of infections, and obs
    infect_ctrl,obs_ctrl=714, 20 
    infect_test, obs_test=748, 19

    sample_C=np.random.beta(infect_ctrl+1, obs_ctrl-infect_ctrl+1, 100_000) # sample 100k times from posterior Beta distributions
    sample_T=np.random.beta(infect_test+1,obs_test-infect_test+1, 100_000) 
    
    # percentage reduction in each sample
    perc_reduce = (val_C-val_T) /val_C

    # histogram of % reduction in infection
    pd.Series(perc_reduce).plot.hist(bins=20)
    
    # how likely, given this data, is > 50% reduction in infection from vaccine 
    (perc_reduce >= 0.5).mean()
    # c. 3%

   (perc_reduce >= 0.7).mean()
    # <0.1%

700 sounds like a pretty decent sample size to me.

Reply on news.ycombinator.com