Thread by @StevePhillipsMD

Designing valid retrospective reviews is like putting lipstick on a pig. But RCT's have no porcine roots. They start out with racehorse potential, which made it even more sad when the major design flaws in this study rendered this horse lame. Analysis:👇 acpjournals.org/doi/10.7326/M2…

To test whether #HCQ works for #COVID19, subjects must have had proof that they actually had it, not just clinical suspicion. Without this evidence, we can't draw reliable conclusions. But only 1/3 had positive COVID tests. And maybe that's why the following didn't add up:👇

Case fatality rate (CFR) for the placebo group was about 0.5%, but the observed US COVID19 CFR at the time of study enrollment was 5.58%. Could this discrepancy have been in part due to many in the study never having had COVID19 in the first place?👇cebm.net/covid-19/globa…

The initial plan was to study a primary end point of hospitalization or death, estimated at 10%, but when patients were less sick than anticipated, they changed plan on the fly because they thought detection of a #HCQ benefit would need a larger sample size. Then, yikes:👇

They decided to study if #HCQ improved symptoms based on a visual analog scale, which can be fine if done properly, but also has clear disadvantages. In particular, there can be "negative effects if questions are unclear or patients feel ambivalent." And:👇ncbi.nlm.nih.gov/pmc/articles/P…

Tucked away in the supplemental material is how the authors judged improvement on #HCQ. They asked patients that they had never met, “Are you experiencing COVID-19 symptoms?” Asking the patient to remotely self-diagnose? And if that wasn't bad enough:👇acpjournals.org/doi/suppl/10.7…

The authors didn't ask about specific symptoms. They asked, "Severity of overall symptoms?" So #1, asking the patient to self-diagnose if symptoms were related to #COVID19, & #2, getting no specific info as to what those symptoms were. And as if it couldn't get more disturbing:👇

While at the same time, they prescribed massive doses of #HCQ, which in my opinion was incredibly irresponsible & potentially dangerous. It's a testament to the safety of HCQ that there were no cardiac arrhythmias reported in the treatment group. And it gets even more tangled:👇

There are side effects from massive doses of drugs, including #HCQ. It can't be known if these were interpreted by the test subjects as symptoms of #COVID19, thereby obscuring potential benefits from HCQ. It's too bad, because their initial plan would have been clear: 👇

A pooled endpoint of hospitalization or death is an unambiguous parameter. I think they should have kept their initial plan. Rates of pooled endpoints of hospitalization or death were markedly different between groups: About 2.5% for HCQ group & about 5.2% for the control group.

How can hospitalization rates be double in the control group vs the #HCQ group but symptoms not be worse? This likely indicates a material flaw in how symptoms were assessed. And how come @boulware_dr has protected his tweets? What kind of science isn't open to scientific debate?

Steven Nissen, clinical trialist at the Cleveland Clinic said, “The study was of such low quality that it was fundamentally uninterpretable.”

I think its design flaws are egregious, its conclusions false, & its implications a danger to public welfare. statnews.com/2020/07/16/new…