Let's try reading this paper the proper way, starting at the end:
>the findings from this retrospective study suggest caution in using hydroxychloroquine in hospitalized Covid-19 patients, particularly when not combined with azithromycin.
So that's not saying HCQ can't be helpful, or is a waste of time as you implied. That just says it requires caution, which is totally fair -- this is uncharted territory. What else do we have?
>Our study cohort comprised only men whose median age was over 65 years. Therefore, the results may not
necessarily reflect outcomes in women or in younger hospitalized populations
OK, so we know the paper itself is a lot narrower in it's prescriptions than the many headlines that will be generate from it. It certainly didn't "kill 20 patients".
>hydroxychloroquine, with or without azithromycin, was more likely to be prescribed to patients with more severe disease, as assessed by baseline ventilatory status and metabolic and hematologic parameters. Thus, as expected, increased mortality was observed in patients treated with hydroxychloroquine, both with and without azithromycin. Nevertheless, the increased risk of overall mortality in the hydroxychloroquine-only group persisted after adjusting for the propensity of being treated with the drug.
So the people who received HCQ+AZ were worse off from the jump, but they adjusted for propensity of treatment, now how was that done?
>we created propensity scores for hydroxychloroquine use alone and hydroxychloroquine and azithromycin use during the hospital stay. Propensity scores were estimated via multinomial logistic regression of treatment group. All baseline covariates were included in the propensity score models. The propensity scores were entered into the outcome models with restricted cubic splines.
Well, I don't know what that all means, but I can Googles -- here's a paper called "A Tutorial on Propensity Score Estimation for Multiple Treatments Using Generalized Boosted Models" , what does that have to say about this method?
>If there are unmeasured variables that predict outcomes and differ among treatment groups then the estimates can be biased. This limitation, however, is not specific to the methods we present; indeed, all causal modeling strategies that use observational study data must contend with this limitation in one way or another.
Hmm, so that's a weak point in this study for sure. How did the VA study contend with that?
>We did, however adjust for a large number of Covid19-relevant confounders including comorbidities, medications, clinical and laboratory abnormalities. Despite propensity score adjustment for a large number of relevant confounders, we cannot rule out the possibility of selection bias or residual confounding.
OK, so what did this study really find? That for men aged 65+ and above hospitalized with COVID-19, based on an adjusted and possibly confounded model of when HCQ/+AZ would be prescribed, those who received HCQ alone may have a 1.1-6.2x increased risk of mortality, but when combined with AZ did not lead to an increase in overall mortality, and in fact could reduce it by nearly half.
But of course, that's not a sexy headline, and it certainly won't rally the right people.
Well, the HC adjust factor was 2.61, i.e., it was expected to have 2.61 more deaths than the no-HC group. So 11.4% x 2.61 = 29.75% and the final death rate in the group was 27.8%. It might not be an improvement but it doesn't seem to be something that killed the patients. Or where I'm wrong ?