Hacker News Reader

Comment by leoplct | Hacker News Reader

I read the method used to measure efficacy: They have tested 44,000 people. 50% got a placebo, 50% got a vaccine. Then has been giving them the freedom to get the covid. 95% of the 50% who got the vaccine haven't contracted the virus.

Correct me if I am wrong. Supposing a random person has a 5-10% chance of contracting the virus (according to average cases/testing worldwide), should the efficacy be 5-10% * 95% = 0.7%-0.5%?

View on HN

You misread. You write:

> 95% of the 50% who got the vaccine haven't contracted the virus.

Much more than 95% of the 50% who got the vaccine haven't contracted the virus. They looked at the first 95 people in the entire test to get the virus. Of these 95 people, only 5 were in the treatment group.

(A) Did they have someone with Covid-19 sneeze or cough or spit on them?

(B) Or did the candidate just go about their daily lives afterwards?

Because, unless path A was taken, then this 95% efficacy sounds very misleading.

Nothing is misleading. Half the participants got the vaccine, the rest placebo, before going about their usual lives. Now, a few months later, 95 cases have been confirmed across both groups. Of these, 90 had received the placebo and only 5 the vaccine. If the vaccine didn't work at all, the split would be roughly 50/50. A somewhat effective vaccine might result in a 70/30 split, and so on.

This sounds like a terrible way to test the vaccine’s efficacy.

A better test would be to have all the participants walk through a minefield of people with the virus. Breath in all that infected air.

Then run the numbers and see who got infected.

It's the standard method, and I see nothing wrong with it. What you suggest (deliberate exposure) is called a challenge trial. This is normally considered ethical only when an effective cure (e.g. antibiotics) is available. Even so, there are plans to conduct such a trial in the UK, and there's no shortage of volunteers. We'll see if they get approval.

Reply on news.ycombinator.com