Thread by @sudokuvariante

Handbook to conduct fraudulent clinical research on repurposed drugs in the context of covid-19, in order to show that they do not bring benefits (updated jan 2023).
⤵️
1/n

https://twitter.com/sudokuvariante/status/1609555606070763520

1/ Randomized trial protocols:

➖Choose the wrong drug dosage, too high (HCQ – Recovery, Solidarity) or too low (IVM - Together) depending on the safety and efficacy of the drug,
2/n

➖Choose the wrong treatment duration (IVM - Together, etc...),
3/n

➖Choose the wrong timing for intervention: start treatment late when your trial aims to study an antiviral (HCQ – Recovery, Solidarity, Discovery),
4/n

➖Modify the inclusion criteria concerning the time between symptoms and recruitment (Principle: 7 days, changed to 14 days),
5/n

➖Do not exclude young patients in good health, it will be difficult to see a difference between the groups (IVM - Lopez-Medina et al.),
6/n

➖Do not exclude patients who took the drug tested in the trial before recruitment, (IVM - Together, Lopez-Medina et al.),
7/n

➖If it is recommended to take the drug with a meal, prescribe it on an empty stomach (IVM: Togtether, Activ6, Bramante et al., Vallejos et al.),
8/n

➖Choose a 'soft' outcome, such as 'resolution of all symptoms after 21 days' (Lopez-Medina et al.),
9/n

➖Do not measure viral load if you are studying a potentially antiviral treatment (IVM: Lopez- Medina et al., Activ6, etc...), measure viral load only when you start treatment AFTER the viral phase (HCQ - Discovery ),
10/n

➖And argue that not seeing a difference in viral load is a sign that the drug is not working,
11/n

➖STOP trials with good protocols if blatant fraud (#lancetgate) is published, but continue trials with bad protocols arguing that everything is fine (HCQ - Recovery),
12/n

➖ Argue that your (well done) study showing a 70% non-statistically significant mortality reduction contradicts a (poor) meta-analysis showing a statistically significant 70% mortality reduction (Lim et al.),
13/n

➖Do not test multi-therapy (remember that it is forbidden to save lives by using more than one drug), but only one drug at a time,
14/n

➖Prescribe a macrolide at 20% of the control group when your trial is testing azithromycin, a macrolide. Don't talk about it in the study, hide it in the supplementary data (Recovery),
15/n

➖Include patients who have already recovered (!) in trials whose primary endpoint is "duration of symptoms" (IVM - activ6),
16/n

➖Give an active placebo (Vitamin C) to the control group, arguing that it is not an active treatment (HCQ - Together),
17/n

➖Overestimate the number of events expected in the control arm, so that the results risk being "not statistically significant" (HCQ: Skipper et al., Boulware et al., IVM: Lopez-Medina et al.),
18/n

➖Fail to correctly take into account the shipping time of drugs in the analysis (PeP-HCQ, Boulware et al.),
19/n

2/ Meta-analyses:

➖All studies satisfying one of the above points are studies at "low risk of bias",
20/n

➖Do not include trials testing multi-therapies (remember that it is forbidden to save lives by using more than one drug),
21/n

➖Download data and do a quick meta-analysis before you register on prospero (Fiolet et al.), make a youtube video and argue that it was only educational,
22/n

➖Include observational studies, until large positive observational studies are published, then STOP doing that! (HCQ),
23/n

➖If results for outpatients and inpatients are different (HCQ), include both in a single metaanalysis to mask positive results (Cochrane: Singh et al.),
24/n

➖On the contrary, if the results for outpatients and inpatients are positive (IVM), separate both in distinct meta-analyses to hide the positive results by decreasing the statistical power (Cochrane: Popp et al.),
25/n

➖Include trials that do not meet your inclusion criteria if they gave negative results (HCQ: Fiolet et al, IVM: Popp et al.),
26/n

➖Assess studies at “low risk of bias” if published in a high impact factor (IF) journal, and all preprints at “high risk of bias”. Oops : high IF journals are only interested in publishing negative studies on repositioned drugs😇,
27/n

➖Trials with positive results are "high risk of bias", because, you know, we know it's impossible...
28/n

➖Choose the wrong statistical model, depending on what you want to show, remember that the fixed effects model will produce a smaller CI (Shankar-Hari et al.)
Argue that it's ok, because you wrote it in the protocol, nananère!!!
29/n

➖If a trial shows a positive result, JUST REVERSE THE TWO GROUPS (Roman et al. preprint) 😉
30/n

➖Create your own imaginary data on the duration of hospitalization of a trial (Roman et al.),
31/n

➖Write a conclusion opposite to what the data show (Hill et al.),
32/n

➖If someone says you made mistakes and if you fix mistakes it changes the result, argue that you find exactly the same results as other published meta-analyses, so it's ok to not fix them (Fiolet et al.),
33/n

➖Create new subjective criteria to exclude studies from the meta-analysis, but do not apply these criteria to other drugs such as paxlovid (Cochrane: Popp et al., Reis et al.),
34/n

➖Modify the protocol of your meta-analysis before the update to mask a reduction in the need for invasive mechanical ventilation by excluding this outcome (IVM: Popp et al.),
35/n

➖Decrease the level of evidence for "imprecision" when the confidence interval is small, in contradiction with the GRADE recommendations (IVM, WHO)
36/36