Thread by @Pottegard

THREAD:
Are you a journalist writing about the newest study by Didier Raoult on #hydroxychloroquine in #COVID19?

Allow me to point to a few issues...

TL;DR: One of the worst studies I have ever read. It provides zero evidence of clinical relevance!

sciencedirect.com/science/articl…

There are MANY issues that would individually make this useless. Considering them all, the paper is borderline nonsensical. We’re talking very fundamental flaws. As pointed out by others, they violate core principles that we have previously outlined.

https://twitter.com/lucasmorin_eolc/status/1276452488393129986

Note that this study is OBSERVATIONAL. There is no randomization of patients. Such studies can be useful if, and only if, the study question lends itself to an observational design and if the study is performed correctly. Neither of these two criteria are met in this case.

Main problem: Comparison of those receiving treatment to those that, for various reasons, do not receive treatment. We follow them to see whether they die or are transferred to ICU. Here’s the three main ways you can get into the untreated/comparison arm!

The first route is highly problematic. Those that die or go to ICU early while receiving treatment are RELOCATED to the non-treatment arm. This infers a STRONG bias in the results, making the treatment look good and non-treatment look dangerous!

I could propose a new treatment – I’m willing to bet that this particular treatment will work wonders in #COVID19. Here’s the recipe for a cure that will ensure 100% survival for 10 days following COVID-admission!

The second route is also problematic: Heart disease is convincingly and repeatedly associated to poor prognosis in #COVID19. Channeling those with strong risk factors for a fatal trajectory into the non-treated arm will also create a strong but spurious “treatment effect”.

The third pathway into the untreated group is not problematic either. Which patients are treated contrary to the local guideline of using #hydroxychloroquine? Probably not the average patient. It’s not a random choice.

There are other issues. As an example, the choice of statistical model (Cox regression) that is a so-called ‘time to event’ approach. This emphasize / upweight events that happen early. Yes, exactly right: The same events that we have all reclassified into the untreated arm.

It’s a disaster that this made its way through peer review. Nothing short of a disaster. While 7 (!) reviewers were involved, I’m willing to bet that they have not included (or listened to) anyone with any knowledge of observational studies! The system is failing…

Post hoc: Quite a few people highlight overlap between authors and editors. That is not necessarily problematic. Expert authors are also editors at top journals within their field. I know it looks suspicious, but there is no evidence to suggest foul play (other than poor review).