Hacker News Reader

Comment by gus_massa | Hacker News Reader

Do they have a control group? From the press article it looks like they are comparing with the global average that is meaningless because there is too much variation.

This is probably a good hospital with good staff and good equipment. Globally many hospitals are not so good.

There may be some involuntary selection of the patients. For example here in Argentina the retired people have a special health plan that include only some hospitals. That may change the population that goes to each hospital and change the expected death rate. Some hospitals get more rich patients. Some hospitals may get more athletes. It is very difficult to be sure, so you need a randomized control group.

View on HN

No, it's yet another observational study. The two RCTs that have been done -- for prophylactic use and for severe patients were both negative, although the studies were small, so there could be a small benefit from HCQ. No one to my knowledge has published an RCT for HCQ given early in the disease.

They also state that most of the patients that died had pre-existing conditions, but no mention of if that would have prevented them from taking HCQ, leading to the majority of the "no HCQ" cases being people who were already more likely to die. If that were controlled for, we might see the same results we've seen in every other study, that taking HCQ actually led to increased mortality rates.

> The study analyzed 2,541 patients hospitalized among the system’s six hospitals between March 10 and May 2 and found 13% of those treated with hydroxychloroquine died while 26% of those who did not receive the drug died.

Worthless without control group

Technically people who did not take the drug are a control group. It's just not placebo controlled. (Which is quite a big no. Unless there's a known working treatment, which should be used as the control then.)

The big question is, how well randomized are these, and what's the effect size - could be suspiciously small.

If it's not randomized at all, the results do not necessarily show the drug is good - just that it's given to healthier patients. You can try to reduce such artifact by considering matching reference subgroups.

Did they have the same inclusion criteria? The people in the treatment group probably have to sign a conformity form, this implicitly exclude people that is too wasted to even hold a pen and that probably have a worse prognosis than the average. Did they include people with more than 90 years old? People with a recent previous heart attack? Small children (parents are annoying)?

Without the same inclusion criteria it's not a control group but a unrelated bunch of guys.

Also known as randomized control group. That said, it should be equivalent to treatment group. Which is exactly what I said.

That's the reason why in RCTs treatment and control groups are split from the same pool - less need to make sure they are.

I think we agree.

Just remember that the first study by Raoult claims to have a control group, but it's better described a a unrelated bunch of guys in another city. And there are other studies with similar tricks.

They're all wildly different populations within different environments and infection rates. COVID-19 infects with very stochastic clustering, making controlled experiments difficult. So one may even need global meta-studies with local control groups, for meaningful results. During an uncontrolled pandemic, you just don't have that luxury and enough degree of control.

What do you mean? The 2,451 patients are partly made up of control group, namely, those who didn’t get the experimental treatment.

Reply on news.ycombinator.com