Hacker News Reader

Comment by spdionis | Hacker News Reader

> 171 were randomized to the Proxalutamide arm and 65 were in the placebo group.

Maybe I'm missing something but, how does that happen? I'd expect a 50/50 split

What split is optimal depends on what effect size you expect. If you expect a very large effect, you get more statistical confidence from allocating more subjects to the treatment group.

On the other hand, if your treatment is very costly, adding subjects to the control group may be an effective way to increase study power without as much increase in study cost.

(Intuitively, I'm not an expert) Apart from the statistical reasons already mentioned, there is also an ethical one. If you expect the study to confirm the treatment to be highly beneficial, you prefer not to withhold it from your patients.

Trials are often run with multiple arms.

It's not treatement/placebo.

It might be 50mg/20mg/10mg/placebo.

Somewhat unintuitively the statistics in trials don’t require a 50/50 split.

Here is an extreme example: suppose you are an HIV skeptic who believes that AIDS is not caused by the HIV virus but rather by drugs [1]. How large a study would you need to falsify your hypothesis?

Answer: one HIV-positive subject, left untreated, would be enough, if that person were otherwise healthy. If that person died of AIDS, that would leave one of two possibilities:

1. HIV causes AIDS, or

2. Something else causes AIDS and that person JUST HAPPENED to have that other causal factor AND a (harmless) HIV infection.

The odds of #2 are low (because AIDS and HIV infections are both uncommon) so that just leaves the first possibility.

As a side-note, that experiment was actually done by an HIV skeptic, on both herself and her unfortunate daughter [2]. Needless to say, both are now dead.

[1] http://duesberg.com/

[2] https://en.wikipedia.org/wiki/Christine_Maggiore

Reply on news.ycombinator.com