TrumpLyftAlles | 17 points
Unitaid, Part of the World Health Organization Funding Ivermectin Research Targeting COVID-19 led by British Expert (England 2020-12-29) Dr. Andrew Hill does meta-analysis of 11 ivermectin RCTs (TrialSiteNews)[-] Haitchpeasauce | 3 points
There was a good comment on the article:
I, and surely every statistician or epidemiologist worth their salt knows p is far more important than n.
... Even with a (so far) smaller n, the p for ivermectin is far far smaller (more impressive) than that for remdesivir.
As one who isn't a statistician, I don't fully appreciate this. Demanding high n is arbitrary, a moving goalpost? TLA I believe you showed there is a diminishing return of error once a certain n is achieved?
What I know so far is that n must be higher if the statistical significance is low, and n must be very high for vaccines because not all subjects will be exposed to the disease.
This comes right back to ivermectin, where all subjects are either exposed to the virus or are already infected, and the statistical significance is shown repeatedly.
I hope Dr Hill gets his data soon so we can have a recommendation to the WHO.
[-] TrumpLyftAlles | 3 points
Sorry if this is too basic and it seems like I'm insulting your intelligence. There are people reading the sub who never took a stat class and I'm writing to them too.
The p-value is an estimate of the probability that the observed effect is due to random error. If you do 100 trials, all treating covid19 patients with (picking something) red M&Ms, then just due to random variation in data, 5% of the 100 trials will have a p-value < .05. Statistical tests are about making sure the observed effect is real, NOT do to that random variation.
The power of a trial, its ability to distinguish real effects from random noise, is a function of (1) the size of the effect and (2) the trial's N. If ivermectin has a big effect, then a smaller N will be sufficient to produce a statistically significant (statsig) result. As some people anticipated, Dr. Chaccour's trial with N=24 failed to produce any statsig results. IIRC, there is a vitamin D RCT that shows statsig results with N=26: in the vitamin D group 2 of 13 patients ended up in the ICU vs 7 of 13 of the untreated patients ending up there. Small N + BIG effect = statsig.
A study is called underpowered if the effect is in the expected direction (the ivermectin-treated patients got better faster, for example) but given its N, the difference between the treated and untreated groups wasn't big enough to get p < .05. Dr. Chaccour's trial was underpowered.
Trusting that https://ivmmeta.com/ has done the math correctly, the way that underpowered studies can be combined to get statsig overall results is illustrated by that site's analysis of Early trials. On the chart, a trial isn't statsig if the line representing it's 95% confidence interval (corresponding to p < .05) does NOT cross the vertical line at 1. Only the Espitia-Hernandez trial meets that standard; the other 5 trials have lines that cross the vertical and are NOT statsig.
Notice, however, that all 6 have very good effects. The leftmost column of numbers is the Risk Ratio. The Espitia-Hernandez RR of .03 indicates that ivermectin reduces the risk from covid19 (whichever outcome measure was used in the trial) by 97%. Even Chaccour's trial showed an RR of .44. Ivermectin shows benefit in all 6 trials.
5 of the 6 trials were not statsig, but look at the summary of the chart's data at the bottom with the bold label Early. It has an excellent RR of .13 and its diamond-shaped line combining the data from the 6 studies is very statsig: it ends a good distance from the vertical 1 line and is shorter than every other line except Espitia-Hernandez's line. Shorter = more statsig. Not crossing 1 = p < .05.
The effect size of the summary Early data is the average of the 6 studies, so in some sense the same. The difference is the N: the Early summary combines all the Ns so it has more power and is more statsig.
Dr. Hill may have done math on the 11 RCTs that (making up the number) produced p < .07. He sees the consistency in the results. He can do the math, anticipating the next 2-3 studies, and predict that if they show results like the 11 trials, then adding the 2-3 additional RCTs will bring p < .05.
So he's waiting to get p < .05 before announcing results! If he wanted, he could release a report right now saying "The ivermectin RCTs do not show a statistically significant impact on covid19." Thankfully, he sees that ivermectin is working and it's very close to that p < .05 -- so he's holding off with the expectation that he will be able to report that when 2-3 more trials report.
He's a good guy!
Ugh: this stuff is difficult to express. Does that help?
Thank you for the tutorial on statistical significance. You did explain it in a very clear way.
Dr. Hill may have done math on the 11 RCTs that (making up the number) produced p < .07. He sees the consistency in the results. He can do the math, anticipating the next 2-3 studies, and predict that if they show results like the 11 trials, then adding the 2-3 additional RCTs will bring p < .05.
Where have you seen this information that the 11 RCTs produced p<0.07?
[-] TrumpLyftAlles | 2 points
(making up the number)
LOL. Sorry, I didn't see the number. I made it up.
I think it was the Gorial trial that had a result with p = .07. I repeatedly entered numbers into an online statsig calculator and eventually figured out how many more subjects had to have the positive outcome to get that p-value under .05. The p-value was surprisingly slow to change. That led to my selecting p < .07 because if the p-value is much larger than that, then it's going to take more than 2-3 studies to get p < .05. Unless a couple studies have nice big Ns so they have more weight than the known 11 RCTs.
Here is what Dr. Hill presented regarding p-values:
Clinical recovery: 43% higher rates of clinical recovery (95% CI: 21%-67%) p<0.0001
Reduction in death rate: 83% (95% CI: 65%-92%) p<0.0001
I couldn't find a single piece in his presentation where he expressed a concern that any of the IVM benefits showed statistical insignificance. Quite the contrary, all are highly statistically significant. So, his insisting on more trial results with at least 3000 patients is having another basis. I think, the following transcript of what he said clears the dilemma:
It's not long until the meta-analysis that we've just done on 1400 reaches 3000, which is about the evidence base that was used for the original approval of remdesivir. The approval of dexamethasone was based on over 5000 patients.
The idea is simple and more of a politically rather than scientifically sound criterion. Remdesivir and dexamethasone approvals were based on 3000 and 5000 patients, respectively, so IVM should not be an exemption. This is an irrational criterion in times when people die in large numbers every day, but obviously health authorities don't care.
[-] TrumpLyftAlles | 2 points
Excellent excellent post, Mass -- as usual. Thanks a lot for correcting my error. I should watch the dang video, it's only 10 minutes long.
[-] Haitchpeasauce | 2 points
You should watch the video because it is independent meta-analysis outside the pro-ivermectin crowd, from a group who will be presenting to the WHO.
[-] TrumpLyftAlles | 2 points
Will do, next time my tired brain is fully functional. I spent all day banging my head against C# code. Mysterious magic happened, which I don't understand -- and I gave up on that until tomorrow. {brain with something crossing the blood brain barrier emoji}
[-] Haitchpeasauce | 2 points
No insult felt at all, this sub is for the education of everyone who reads it, glad I could prompt an informative discussion. Something often not discussed when explaining statistical analysis is what is an acceptable n, which varies depending on situation.
what large trials have results coming in the next few weeks?
[-] TrumpLyftAlles | 1 points
I don't know, they are always surprises to me. There are 14 trials marked as completed on ClinicalTrials, only 6 of which have results. Maybe 1 of the 8 which don't yet have results? I browsed through several, didn't see any with large Ns. Sometimes trials are reported without any notice in ClinicalTrials. Some trials aren't registered in CT. Some aren't registered anywhere.
So we get to wait.
[-] TrumpLyftAlles | 1 points
Nice job, catching the good comment!
n must be very high for vaccines because not all subjects will be exposed to the disease.
That's the important point: even now as the virus rages, it's a rare event for someone to catch covid. The trials were set up to report when certain Ns were infected. From This Week in Virology, IIRC the Pfizer trial was supposed to report at a couple lower Ns, but it blew right past those and ended up reporting at N=192. That's a nice big N for finding statistical results if the vaccine has an effect -- which thankfully it does.
It's more like the vaccine trials had to enroll 40,000 subjects in order to get N's of 192, etc. The relevant N is the number of enrollees infected with the virus.
[-] Ok-Film-9049 | 6 points | Dec 31 2020 10:55:56
As someone whose job involves stats, but I am not an expert... I have the feeling that the stats are already overwhelming to say IVM works unless: 1. All the research teams have collaborated to commit fraud 2. There has been extraordinarily selection bias I. E. There are lots of studies that show IVM doesn't work or harms. Given that Gilead are making so much money from Remdisavir and IVM would be a threat to this. They would be highly motivated to promote any existing trials that show harm. I don't see any!! I'm convinced!
permalink
[-] TrumpLyftAlles | 4 points | Dec 31 2020 18:37:57
They would be highly motivated to promote any existing trials that show harm.
This is a really important point, IMO. I agree, that would be in the news if it ever happened, because of Gilead or a lobbying organization that represents Big Pharma, who knows or cares? It's not in the news because it hasn't happened, and that's important.
Ivermectin has an unassailable safety record. It is ridiculous that ivermectin has to prove its safety AGAIN with covid19. All those hecking articles and government authority warnings about toxic effects, brain damage. So much BSS. As the Drs. Varon, Kory and Marik said over and over at the FLCCC press conference: there is no downside to treating patients with ivermectin.
Even if critics find the ivermectin trials flawed and statisticians conclude that they are insufficient to prove that ivermectin works, by their RCT Golden Calf standards, the trials have consistently shown:
1) If ivermectin has an effect (which it almost always does) it is in the positive direction. In NO CASE do patients get worse because of ivermectin.
2) Adverse effects are ALWAYS MILD. No subjects died or went into coma or got Bell's hecking palsy!
WE DO NOT NEED ANY MORE TRIALS TO ESTABLISH THOSE FACTS.
First do no harm, doctor? You may set aside that concern with ivermectin; we have the proof.
A 3rd relevant fact is ivermectin is relatively cheap. Even at the US's inflated prices for ivermectin, a heavy treatment (36mg/day for 5 consecutive days) is only 7.5% as expensive as a treatment with Remdesivir. A more typical ivermectin treatment (18mg on days 1 and 7) would cost 1% of Remdesivir expense.
IMO, these 3 facts should lower the bar for US trials to the hecking floor.
I'm not sure the rest is interesting or meaningful enough to deserve your time and effort -- but I'll leave it for those with long attention spans.
A while back I wrote this (adapted it from something, not my original idea):
A man is floundering in the ocean, a fair distance from shore, on the verge of drowning.
Lifeguard: “I’m going to throw you this Ivermectin Floating Life-Saver, sir!"
FDA: “You are not allowed to use that specific flotation device with people in that man's condition. We have not authorized its use for people drowning in the ocean.”
LG: “But it’s been used 4.2 BILLION times to save drowning people in lakes and rivers, and no one was ever harmed. MANY MANY PEOPLE WERE HELPED”.
FDA: “We are well aware of Ivermectin Floating Life-Saver excellent safety record with other drownings but that's irrelevant: we need a trial to establish that the IFLS is safe to use in the ocean.”
LG: “But he's drowning!"
Drowning man drowns.
FDA: Ignores the Ivermectin Floating Life-Saver for the next 8 months while 100s of 1000s of people drown, even as trial after trial shows that IFLS could save a lot of people who are drowning in the ocean -- because the trials have flaws and because none of them were done on American shores. Whose to say that the proof that IFLS saves people drowning in the Indian Ocean would carry over to the Atlantic and Pacific? Those oceans could have some crucial difference that makes the IFLS fail!. Only an RCT done off American shores will do. It doesn't matter, that so many people are drowning in the Atlantic and Pacific, and lifeguards on US shores only have one very expensive way to try to save people that doesn't work, the Remdesivir MaybeItFloats(tm) device that costs $3000 each.
Hmm: Maybe that allegory is too long.
permalink
[-] lovethebigones | 2 points | Jan 01 2021 08:00:41
Wouldn’t be the first time Gilead tried to sink the opposition!
permalink
[-] Ok-Film-9049 | 3 points | Jan 01 2021 10:24:35
It is natural for a business to protect its own interest. But something is seriously wrong when the government(s) isn't seeking the most effective treatment in an emergency.
permalink
[-] lovethebigones | 3 points | Jan 01 2021 10:38:07
I couldn’t agree more. The fact that IVM is so safe, cheap and effective is absolutely beyond me the governments aren’t giving it the time of day, going as far as mocking the doctors who have the solutions.
permalink