4 ms·

> 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). G

> 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

klmadfejno 6y ago
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.
arcticbull 6y ago
> 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 https://www.bbc.com/news/health-52301633
- klmadfejno 6y ago
  > 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...
  - arcticbull 6y ago
    That is what I meant, yeah. Edited for clarity.
akiselev 6y ago
> 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.
- zeku 6y ago
  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.
  - SiempreViernes 6y ago
    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?
  - akiselev 6y ago
    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.

klmadfejno 6y ago

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.

arcticbull 6y ago

> 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 https://www.bbc.com/news/health-52301633

klmadfejno 6y ago
> 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...
- arcticbull 6y ago
  That is what I meant, yeah. Edited for clarity.

akiselev 6y ago

> 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.

zeku 6y ago
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.
- SiempreViernes 6y ago
  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?
- akiselev 6y ago
  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.