Ivermectin seems to have benefits
but is scorned, while snot-filled masks and expired doses of Tamiflu maintain respectability.
I've been on the fence regarding the effectiveness of ivermectin to treat or prevent COVID-19. Every disease attracts its share of quacks and miracle cures, and medicine is not my field. On the other hand, tens of billions of profit dollars depend on generic medication being shunned. The societal context is that claims supported by huge pharmaceutical companies will have the best promotion, and we can expect their interests to dominate academic, clinical, regulatory and press-reported medicine. Establishment medicine has been so highly critical of ivermectin that it would be difficult now to climb down.1
So we have some awareness of the background for the ivermectin discussion. Fortunately, claims about treatment effect are largely claims about probability and statistics that are also understandable by non-medical scientists. Recent posts by Alexandros Marinos on the ACTIV-6 trial illuminate both the statistical issues and the broader context.
Even though the ACTIV-6 authors support the mainstream narrative against ivermectin for COVID-19, their study shows ivermectin significantly (99% confidence) reducing "Time unwell" (Table 2). The reduction is small (0.49 days), but within Tamiflu’s 8.4 to 25.1 hours range for influenza. Tamiflu remains widely stockpiled, even past its original expiration date.
Scientists often search for hard-to-detect signals in a sea of noise. Once even a small signal shows up with high confidence - a rare Hallelujah! moment - work shifts to make the signal stronger and understand it better. Could ivermectin’s benefit be enhanced?
Because the mean delay between symptom onset and ivermectin treatment was 6 days (Table 1), we can hope that earlier treatment is more effective, as claimed by ivermectin proponents. Need for earlier treatment is specified for other antivirals, and delay-dependent efficacy would be expected for ivermectin. In contrast to ACTIV-6’s average six day delay, Tamiflu’s approval was based on studies with a maximum 36 hour delay, and Paxlovid had a mean of 3 days.
Efficacy for different ivermectin delays are listed in the study’s online supplement (eFigure 3). Data points are reproduced below with an overlayed linear regression that shows higher Hazard Ratio (HR) for shorter delay. In the study language, HR above one favors ivermectin. The linear model has ivermectin raising a measure of benefit (HR minus 1) by several-fold if treatment begins immediately, with a reduction for every day of delay. Some caveats beyond the question of data integrity are that behavior may be nonlinear, zero slope is within the realm of possibilities, errors may be correlated, and prophylactic treatment is said in some cases to be even more effective.
I hope ACTIV-6 isn’t the last word on ivermectin because the small benefit could also be explained by effects like partial unblinding. According to clinicaltrials.gov, ivermectin pills were labeled "123". Placebo pills were supposed to "look similar", but were they labeled "123"? Maybe they weren't, and some enrollees found this labeling info on the web.
Still, the above linear regression with shaded 95% confidence interval contradicts the ACTIV-6 study’s claim about treatment delay:
However, there was no evidence of a differential treatment effect based on the median time of symptom onset to receipt of study drug.
While the regression does not rise to bomb-proof levels of certainty, it points towards improved benefit for shorter delay. In the Bayesian probability language of the paper, this is indeed evidence that earlier ivermectin treatment would be more beneficial.
On the topic of treatment delay, the study’s appendix lists an HTE (heterogeneity of treatment effect) p-value of 0.386 (eFigure 3), among many other potentially p-hackable model variables. According to the study’s statistical analysis plan (Section 3.4), updated mid-data-collection, this comes from comparing model posterior probabilities with and without terms for Duration of symptoms (time from symptom onset) as a restricted cubic spline with 3 knots. But there is no justification for a complicated 3-knot-cubic-spline model or clear explanation of the HTE p-value’s meaning. Extraneous model variables easily lead to statistical errors.
Code in R
# hazard ratio and low/high limits for
# "symptom onset, days" in eFigure 3.
hr <- c(1.07, 1.16, 1.13, 1.00, 0.87, 0.76)
hrl <- c(0.86, 1.01, 0.97, 0.83, 0.61, 0.44)
hrh <- c(1.34, 1.33, 1.32, 1.21, 1.23, 1.29)
days <- c(3, 5, 7, 9, 11, 13)
# Assume symmetric errors for purpose of linear regression.
# This is an approximation, probably with a small effect that
# reduces apparent effectiveness of treatment.
# Weights are inverse-squared confidence intervals.
w = (0.5/(hrh - hrl))^2
lm_fit <- lm(y~x, data=data.frame(x=days, y=hr), weights=w)
summary(lm_fit)
Please keep comments on topic.
Another point: Why have weak observational studies resulted in strident pro-mask guidance, while comparatively stronger randomized trials result in vehement anti-ivermectin guidance? What happens when people with runny noses wear masks, as would be expected with respiratory infections? Mechanistically, one could expect mucus to wet the fabric, followed by widespread constituent (including virus) dispersal through exhalation.
Your footnote touches on something very important. The RCTs on masks fail to show a benefit, yet on the basis of a handful of observational studies masks are recommended - despite the authors of some of those studies stating explicitly that they're not causal studies. Yet Ivermectin, which is safe and has stronger evidence in favor, is ignored. Clearly medicine now follows the money - no one has a patent on Ivermectin so no manufacturer is championing the drug, and the CDC and FDA work for the pharmaceutical industry.
On the subject of runny noses and masks, some of the mask RCTs actually show more illness among the masked group vs. unmasked. But the studies are all lab testing for influenza specifically and since lab tests come back negative for influenza virus the study authors ignore the issue.
But as I keep pointing out, those people had fevers and coughs - as diagnosed by medical doctors. Clearly something was made much worse by the masks. Here's a list of the 18 RCTs and a very short summary of each, noting the negative results in some:
https://johnhowardroark.substack.com/p/enough-with-the-masks-already