Hacker News Reader

Comment by klmadfejno | Hacker News Reader

As far as I can tell, this is not a controlled study, but just an analysis of a bunch of people who took the drugs at a pretty low dosage. The analysis doesn't (and couldn't) make any conclusions about HCQ improving patient outcomes vs. not using HCQ; but rather just tries to assert that it didn't make things worse and we should therefore do it because China says it's a good idea. Without any baseline it's not clear to me how one could evaluate these claims. Sounds just about as useful as the previous paper.

As an aside, I do feel like the mean age of poor clinical outcomes, 69, feels a bit low relative to my expectations.

View on HN

"Administration of the HCQ+AZ combination before COVID-19 complications occur is safe and associated with very low fatality rate in patients."

sigh COVID-19 is associated with low fatality rate in patients. It's not clear this combination of drugs does anything other than not kill people any faster.

> As an aside, I do feel like the mean age of poor clinical outcomes, 69, feels a bit low relative to my expectations.

I'm confident that has to do with top of funnel.

> I'm confident that has to do with top of funnel.

I think you're probably correct. But, frustratingly, it's equally possible that the HCQ increased the likelihood of negative outcomes for younger patients. Or, that HCQ decreased the likelihood of negative outcomes without regard to age, resulting in fewer bad outcomes for really old people.

Ahh I see what you're saying. That's a really good point.

"but rather just tries to assert that it didn't make things worse and we should therefore do it because China says it's a good idea. Without any baseline it's not clear to me how one could evaluate these claims".

China never claimed HCQ was very effective. Chloroquine was considered an experimental viral treatment and the side effects were notable.

HCQ was used more commonly in South Korea.

I don't disagree, but the paper explicitly states:

> As a conclusion, based on our experience, we consider reasonable to follow the recommendations made in Asian countries for the control of COVID-19, notably in Korea and China

I think it states that in the sense of "well based on our study it won't hurt, so there's no reason not to follow the advice of countries that have been dealing with this longer"

Patients should wear party hats too. It won't hurt.

In case I didn't do a good job of expressing myself, I'm on your side haha.

> any conclusions about HCQ

did you read this "A total of 1061 patients were included in this analysis (46.4% male, mean age 43.6 years – range 14–95 years). Good clinical outcome and virological cure were obtained in 973 patients within 10 days (91.7%). "

> l, this is not a controlled study,

You can retrospectively get the control data for same time frame. This won't br same as placebo but still good.

AND, death rate in france for closed cases is around - 15-25% .

Death rate with HCQ+AZ is 0.75%

If I got crona. I will take the HCQ+AZ

91.7% for a sample of 1,061 people with a non trivial subset removed because of pre-existing health issues isn't that noteworthy. Given only the mean age, it's very difficult to know what the baseline levels should be. Averages are not realistic. They depend on the age and health of individuals. The concern about placebos is a thing, yes, but the real problem is that the rates of poor clinical outcomes are not standard. You could get a decent estimate by looking at other results coming out of this hospital from a similar age and health cohort, but that's not here in this paper. The baseline mortality rate for an average individual below age 60 is nowhere close to 15%, but of course, it all depends.

It's probably fine if you want to take the drug, but accurate statistical reasoning is not the driver of that decision.

> Good clinical outcome and virological cure were obtained in 973 patients within 10 days (91.7%).

I think that's pretty typical in untreated patients. Average recovery period is about two weeks. [1]

> Death rate with HCQ+AZ is 0.75%

The death rate of COVID-19 in the general population is less than 1% (0.25%-1% depending on the study). Even New York, the worst city (by case count, and fatalities) in the worst country in the world is ~0.7%. All this study shows is that HCQ+AZ doesn't kill people faster than not taking HCQ+AZ.

> AND, death rate in france for closed cases is around - 15-25%.

Not worth re-hashing CFR vs. IFR (and the associated adverse selection risk) but you won't get anything approximating your risk without referencing serological studies. I can't stress how this number is not representative, and treating with HCQ+AZ does not by any means get the number from 15-25% down to 0.7%

[1] https://www.bbc.com/news/health-52301633

> Even New York, the worst city in the worst country in the world

Pretty sure this is inaccurate, except perhaps by total number of deaths...

That is what I meant, yeah. Edited for clarity.

> You can retrospectively get the control data for same time frame. This won't br same as placebo but still good.

No you can't and no it won't. Clinical study protocol documents are often thousands of pages long because they need to describe mundane procedures down to how doctors and nurses are supposed to draw the subjects' blood so that there aren't significant variations between participating hospitals. The point of a control group is to control as many variables as possible and placebos are only a small part of that. Selecting control groups is an entire subfield of modern medicine.

Many drugs with much more promising early results that were gathered under nearly ideal conditions (instead of the chaos of a pandemic) have failed hard in third phase trials.

You actually can. I'm a Software Engineer/Data Scientist whatever--and this is a part of my job.

You can take the entire case cohort and filter out all of the people who we don't have robust data for. Then find a control cohort who do have all of the things we're looking at and then analyze the data. Yes it would've been better to do a grade A trial from the get go, but these kinds of studies can be done.

Ok, so are you saying you could find a control cohort for the patients in this study?

Or are you making the much less useful statement that a proper study is possible?

I've actually worked on clinical trials. Most of the industry has been moving towards preregistering clinical trials so that statistical analyses like the one above can't be used to demonstrate efficacy by using retroactive patient selection criteria that only keeps the most favorable data points (like this paper does, in effect).

I'm not arguing with the validity of the statistics, I'm saying it's not enough to fulfill the medical standards for demonstrating drug efficacy. An actual clinical trial protocol needs to be properly designed and coordinated across participating facilities - otherwise it's simply garbage in, garbage out, no matter the quality of the statistical method.

Edit: The exception is cases like this where the drug is approved for some uses and doctors in some countries are legally allowed to prescribe it off label using their own judgement of statistical analyses like the one above. However, the issue becomes liability and most doctors will fall back to FDA recommendations that are based on controlled clinical trials.

Yeah, I agree you need a full up to standard clinical trial to really get something like this into the US medical system under normal circumstances.

I was just mentioning that things can be learned from this study retrospectively by creating a control group from an EHR.

It has validity in the short term for MD's who might begin using this treatment under our fast track system we're using for COVID stuff.

If this trial was done at my institution and my team already had all of the paperwork out of the way with regards to the data access we could have something useful in a week or two out the door. So I hope this will be done by someone somewhere, just to save lives in short term.

My apologies, I misunderstood your previous reply. I'm just afraid that loosening standards in this specific crisis is walking a really fine line. It makes sense during short burn epidemics like ebola where the fatality rate is so skewed, but exposing millions of people to HCQ during an extended highly infectious pandemic, especially without proper screening, is dangerous. Small price to pay if we're talking about something safe like aspirin or if HCQ cuts the CFR in half but if it's only a 10-20% difference, or the effect disappears with a larger sample size, or the mad rush for the drug causes a breakdown in quality control, or the demand deprives people who need it for other reasons - there's a very high chance of worsening quality of life for many people and creating a net negative.

What is the fast track system for COVID? I've seen the emergency use authorizations for clinical testing but not any for new treatments other than the usual off label use.

> As far as I can tell, this is not a controlled study

Right there in the discussion section:

> In our analysis, which is not a RCT

> make any conclusions about HCQ

AFAICT, they suggest taking HCQ early on during the course of the disease, assuming the patient can take it (they excluded quite a few that might have got bad side effects from it).

That is what they suggest, but a suggestion isn't the same as compelling evidence that HCQ improves clinical outcomes versus not doing it. They don't do that here, or even claim to have shown this.

>but rather just tries to assert that it didn't make things worse and we should therefore do it

Rheumatologist with 30 years of experience who, while skeptical of hydroxychloroquine's benefits for COVID19, has no problem with Trump's "what have you got to lose?" statement: https://www.youtube.com/watch?v=hRCG9RtFJ_Y

Is that a framed picture of Mike Pence in the background?

Reply on news.ycombinator.com