There has been a lot of excitement – and even a presidential tweet about a recent paper from the lab of Didier Raoult, an infectious disease specialist in Marseille, France. But although this study might offer a glimmer of hope, there are some serious problems with the paper too.
The paper, by Gautret et al., first appeared on 16 March 2020 as a preprint on medRxiv. A “preprint” is a raw version of a scientific paper that has not yet been reviewed by external scientists. Because the COVID-19 pandemic is so severe and treatment options are so limited, it makes sense that the authors chose to publish this small study as soon as possible. Peer review usually takes weeks or months, so it was good that they shared it as soon as the first results came in.
On the same day as the preprint appeared, 16 March, the manuscript was submitted to the International Journal of Antimicrobial Agents, where it was accepted within a day, on 17 March, and published online on 20 March. That suggests that peer review was done in 24h, an incredibly fast time.
A summary of the study
In the study, a total of 42 COVID-19 infected people admitted to hospitals in Marseille, Nice, Avignon and Briançon (all in South France) were enrolled. Of these, 16 patients received the normal care (“controls”), while 26 other patients were treated with Hydroxychloroquine (HQ), a drug normally used to treat lupus and reumatoid arthritis, which is related to Chloroquine, a drug used to treat or prevent malaria infections.
Of the 26 HQ treated patients, 20 completed the study. Of these, 6 also received Azithromycin (AZ), an antibiotic. During the 6 day study, patients were sampled for the presence or absence of the COVID-19 virus (PCR positivity). On day 6, most of the 16 control patients, about half of the 14 HQ treated patients, and none of the 6 HQ+AZ treated patients were PCR positive. The conclusion was that HQ treatment, but especially HQ+AZ treatment is very effective in treating COVID-19 infections.
This study gained an enormous amount of attention because it reports very encouraging results of a small study in which COVID-19 patients were treated with hydroxychloroquine. This is great news in the midst of a pandemic outbreak for a viral disease for which there is not good current treatment, which is why this paper was quickly published.
Unfortunately, there are many potential problems with the way the data and the peer review process were handled. The discussion is still ongoing on PubPeer, with 2 posts on either the preprint (40 comments as of today) or the published version (3 comments). Let’s take a closer look at the paper and discuss some possible problems. Note that part of the text in this blog post has been previously posted by me on Pubpeer and on Twitter.
In this study, people who had COVID-19, a coronavirus infection, were treated with a drug designed to kill the malaria parasite. Studies like these, where people are being treated with a drug that is approved to treat a different disease, need to be approved by ethical and drug safety committees before they take place.
The protocol for the treatment was approved by the French National Agency for Drug Safety on March 5th 2020. It was approved by the French Ethic Committee on March 6th 2020. The paper states that patients were followed up until day 14. The paper was submitted for publication on March 16th. See screenshot.
But, how does a 14-day study fit between March 6th and March 16th? Could the authors have started the study before ethical approval was obtained? Something does not seem quite right.
Strangely, although the text states that the patients were followed-up for 14 days, the figures and tables only show data for 6 days, so maybe the timeline is actually OK. But it is not clearly formulated in the Methods.
A change of plans
In the EU Clinical Trial Register page, the study was described as evaluating PCR data on Day 1, Day 4, Day 7 and Day 14. However, the study show the data for Day 6, which is different than planned. Why did the authors not show the results on Day 7? Did the data on day 7 not look as good?
In an ideal clinical trial study, the control group and treatment group should be as similar as possible. Patients should be of similar age and gender ratio, be equally sick at the start of treatment, and analyzed in the same way. The only difference between the 2 groups should be whether the patients received treatment or not.
Here, however, the control and treatment group differ in various other ways from each other. Scientists call those confounding factors, factors other than the treatment that are different between the controls and treated patients, and that might explain the different outcomes.
The patients were recruited at different “centers”, but it is not very clear which patient was located in which hospital. The HQ treated patients were all in Marseille, while the controls were located in Marseille or other centers. One can imagine that hospitals might differ in treatment plans, ward layouts, availability of staff, disinfection routines, etc. It is not clear if controls and treated patients were all recruited and treated at the same hospital? This should be added to Table S1.
The control and treated patients were also not randomized, as is usual in a trial like this. Some patients were chosen to be treated, while other patients were chosen to serve as controls. That might bring along all kinds of biases, in which e.g. researchers might be tempted to favor e.g. sicker patients, or patients that have been sick longer to be chosen in one of the different groups. Unfortunately, the lead author of the paper, Didier Raoult, does not believe in randomized clinical trials.
Then, the patients in the control group included some very young patients (ages 10-20), while the two treatment groups only had patients 25 years and older. See the lime green box in Table S1 above. This is strange, because the paper also states that patients could not be included if they were 12 years or younger. Control patients 1-3 do not fulfill that criteria.
Of particular note, control patients 6 and 8-16 appear to have been analyzed differently. Their Day 0 PCR values are not given as CT values (the number of cycles after which a PCR becomes positive, the lower the number, the more virus is present) but as POS/NEG, suggesting a different test was used. Highlighted with red boxes in the Table S1 screenshot above.
Update: Another possible confounding factor is that the patients who did not want to receive the HQ treatment or who “had an exclusion criteria”, such as a medical condition that would put them at risk for treatment with the drug, were assigned to the control group. This might mean that they were even more differences between the treatment and control arms of the study.
Six missing patients
Although the study started with 26 patients in the HQ or HQ+AZ group, data from only 20 treated patients are given, because not all patients completed the 6-day study. The data for these 20 patients looks incredibly nice; especially the patients who were given both medications all recovered very fast.
What happened to the other six treated patients? Why did they drop out of the study? Three of them were transferred to the intensive care unit (presumably because they got sicker) and 1 died. The other two patients were either too nauseous and stopped the medication, or left the hospital (which might be a sign they felt much better). See this screenshot:
So 4 of the 26 treated patients were actually not recovering at all. It seems a bit strange to leave these 4 patients who got worse or who died out of the study, just on the basis that they stopped taking the medication (which is pretty difficult once the patient is dead). As several people wrote sarcastically on Twitter: My results always look amazing if I leave out the patients who died, or the experiments that did not work.
Problematic PCRs and outcomes
As you can see on the Table S1 screenshot above, patients’ PCR results were a bit variable. Patients 5, 23, and 31 in particular show days in which their PCR is negative, followed by days in which their PCR is positive again (dark blue boxes). This might be because the PCR were done using a throat swab, and the virus might be just around the detection limit in the throat, leading to variable results from day to day.
Several patients in the control group did not even have a PCR result on Day 6, so it is not clear how they were counted in the Day 6 result. E.g. look at patients 11 and 12, who were positive on Day 4, but not tested on Day 5 or 6. How sure are the authors that these patients did not convert to PCR negativity?
It would have been better if the authors would use clinical improvement (e.g. fever, lung function) as the outcome, not a throat PCR. The virus could still be rampantly present in the lungs, and the patient could still be very sick, while the virus is already cleared out of the throat. If PCR is an outcome, it would be better measured as e.g. at least 2 or three consecutive days of PCR negativity.
Troubled peer review process
The paper was submitted on 16 March and accepted, presumably after peer review, on 17 March. Assuming the paper was indeed peer reviewed within 24h, that seems incredibly fast. But there might be a reason why this process was so fast.
One of the authors on this paper, JM Rolain, is also the Editor in Chief of the journal in which the paper was accepted, i.e. the International Journal of Antimicrobial Agents. This might be perceived as a huge conflict of interest, in particular in combination with the peer review process of less than 24h.
This would be the equivalent of allowing a student to grade their own paper. Low and behold, the student got an A+!
The journal should make the peer reviews (usually reports from 2 or 3 other scientists who are not authors on the paper or work at the same institutes) publicly available, to prove that a thorough peer-review process indeed took place.
Obviously, in the case of a viral pandemic there is a great need for new results to become available as fast as possible. It is understandable that the scientific peer review process might be a bit less polished in this case than for other science papers.
However, the preprint was already publicly available for all, so it does not make sense that this peer review was done in such a rushed manner. It would have been better if the raw data were shared by preprint, with a more carefully written and analyzed, peer-reviewed version being published a couple of weeks later.
Other posts about this paper
Here are some Twitter and blogs post written by others that have raised similar or additional concerns about the Gautret et al. paper. I will update this list, so leave comments below if you have seen a good one.
- Oliver Hulme et al. did a reanalysis of the data adding the 4 dropped out HQ patients back in and found the HQ treatment group is still significantly lower in PCR detection. See their preprint here: https://osf.io/96bce/
- Gaetan Burgio: https://twitter.com/GaetanBurgio/status/1241201751916568576
- Leonid Schneider: Chloroquine genius Didier Raoult to save the world from COVID-19
- Krutika Kuppalli: https://twitter.com/KrutikaKuppalli/status/1241457057976315904
- Nick Brown: https://twitter.com/sTeamTraen/status/1241330562830422017
- Andrew Lover: https://twitter.com/AndrewALover/status/1241735899546890242
- Podcast with C Drosten (in German): https://www.ndr.de/nachrichten/info/17-Coronavirus-Update-Malaria-Medikament-vorerst-kein-Hoffnungstraeger,podcastcoronavirus144.htm
- Matthew Herper for STAT news: https://www.statnews.com/2020/03/22/why-trump-at-odds-with-medical-experts-over-malaria-drugs-against-covid-19/