Relevant and even prescient commentary on news, politics and the economy.

Accelerating Vaccine Development

How can deployment of a Sars Cov2 vaccine be speeded up ?

One important step is to make a lot of candidate vaccines while testing for effectiveness. Usually, there is testing, approval than producing. Since producing a lot then testing makes no business sense, it is a project for states or charities. Unsurprisingly the Gates foundation is on it committing to mass produce 6 candidate vaccines.

A remaining problem is that proof of effectiveness takes a long time (and a huge sample). Fortunately, even without a vaccine, most people, even in the highest risk occupations, don’t get Covid 19. Therefore it is hard to prove that vaccinated people do even better. An effectiveness trial will take time.

The most advanced candidate is the Oxford made adenovirus based vaccine. Because a very similar vaccine against (I forget) MERS or Sars Cov1 is known to be safe, they can go straight to testing effectiveness. A phase 1/2 trial is recruiting volunteers. A phase III (yhat means very big) trial is being organized.

The phase 1/2 trial has estimated primary completion date May 2021. One step in the trial is follow the subjects for 6 months to see if they catch Covid 19. I am citing the entry at

One basically clearly unethical shortcut is to vaccinate then challenge with the virus. The hope is that the experimental subjects won’t be killed. Acting on such hope is no longer allowed (nor should it be). I am willing to volunteer to participate (a safe offer because it won’t happen and they would use young volunteers if it did).

Of course most people have few qualms about doing this with animals and there is already a vaccine (good old killed virus vaccine) known to protect Rhesus monkeys.

I think another approach would be to vaccinate then test serum. One thing would be to test if the serum blocks infection in vitro, that is contains neutralizing antibodies. It is know that the serum of (most) people who have had Covid 19 contains such neutralizing antibodies.

Another would be a trial of serum of vaccinated people as therapy for people with Covid 19. Serum from people who have recovered is being used. It is in short supply. A theraputic trial of serum from unexposed people, from vaccinated people, and from people who have recovered from Covid 19 as therapy for people with Covid 19 makes sense. As we have seen, theraputic trials don’t have to take very long (the course of the disease is within 28 days the course of not catching the disease has been all of time until now).

Serum of recovered patients has not been proven to be effective, so this trial would be legal.

Comments (1) | |

Remdesivir 7

The NIAID Trial of Remdesivir has closed early, because they concluded it was not ethical to treat people with placebo given what they consider proof that Remdesivir is effective. This is huge news (I am surprised that the Dow Jones only went up 2% not that I care about the Dow Jones).

This is a large double blind randomized controlled trial. The null of no effectiveness was rejected using the principal outcome measure. This is the sort of outcome which causes the FDA to approve drugs.

The data shows that remdesivir has a clear-cut, significant, positive effect in diminishing the time to recovery,” said the institute’s director, Dr. Anthony Fauci.Results from the preliminary trial show remdesivir improved recovery time for coronavirus patients from 15 to 11 days.

Also preliminary results from the Gilead 5 days vs 10 days study were announced. This is an odd study as there was no control group. The result is that the null that 5 days of treatment are as good as 10 days of treatment was not rejected. This is very useful information, since Remdesivir is likely to be a lot of excess demand for Remdesivir very soon. Many wondered why do a study without a control group. I think the aim was to get Remdesivir in as many people as possible as soon as possible. The study has 6,000 participants. This is in addition to the controlled trials, the expanded access program and individuals who have obtained compassionate use.

The scientific result is not critical. If Remdesivir doesn’t work, the 5 days are as good as 10 days and no one cares as 0 days are also just as good at 0 cost and with 0 side effects. However managers at Gilead believed (with it appears good reason) that the cost of the trial is huge and negative. Certainly people beg to participate. I think that in extreme cases it can be a good idea to use drugs based on preliminary (even in vitro) evidence while waiting for the results of the phase III controlled trial. The Gildead 5 day v 10 day trial is one example of this, and I applaud their clever approach to dealing with regulations.

Also, last and least, the disappointing Chinese study has been published in The Lancet. This is the study which caused widespread dismay and headlines including “Remdesivir fails”. Given the results of the NIAID trial, it appears that some people misunderstood the brief note accidentally published by WHO. More people correctly asserted that the question was still open. However, I think misunderstanding of the note (maybe also by the person who wrote it) is a good example of what happens when people try to use mathematical statistics but do not understand the Neyman Pearson framework, that is don’t know what a null hypothesis means or what failure to reject a null hypothesis implies. This is a very common elementary error (actually more universal than common).

I am not going to provide links, but many articles and especially many headlines contained the completely incorrect claim that the study showed that Remdesivir failed to perform better than a placebo. This is simply and obviously false (this is obvious without looking at the data just based on an understanding of what data can and can not imply). The correct statements are that the Chinese trial failed to show that Remdesivir performed better than a placebo or (equivalently) that Remdesivir failed to perform statistically significantly better than a placebo.
Removing the words “statistically significantly” makes a true statement absolutely false. It is not acceptable even in a headline.

In fact, in the study, on average patients treated with Remdesivir recovered more quickly than patients treated with placebo, however, the difference was not large enough to reject the null of no benefit at the 5% level. the ratio was not huge with a ratio of hazards of improvement of 1.2. Notably the ratio 1.31 in the NIAID trial is not huge either. The difference between headline success and headline failure is almost entirely due to the sample size. This is a failure to understand what it means to test a null hypothesis against and alternative hypothesis. The statement that the Chinese study was underpowered does not even begin to approach a demonstration of an understanding of elementary mathematical statistics. I will try to explain after the jump.

An even more alarming headline result of the Chinese study was that Remdesivir did not have a statistically significantly greater effect on the viral load. The failure to help patients statistically significantly more could occur even if Remdesivir blocked viral replication in people as it does in vitro. The explanation is that late in the disease the often fatal trouble is due to the patients’ immune response not the virus directly. Cytokine storm can kill in the absence of a virus.

However, a failure to reduce the viral load would be terrible news. Of course the study didn’t show that. It showed that the reduction was not statistically singnificant not that it was zero. Now there are two ways to compare treaments — the viral load after days of treatment or the reduction of the viral load from when treatment started. In principle and with large enough samples, this makes little difference. The two treatments are randomized and the average viral load at the beginning of the trial for the two treatments will go to the same number by the law of large numbers, which is true asymptotically (recall my personal slogan “asymptotically we’ll all be dead”). It did not apply in this case.

The average lower respiratory viral load (from now on just viral load) at the beginning of the trial was not similar for those treated with Remdesivir and those treated with placebo. It was roughly 10 time higher for those treated with Remdesivir. Notice the recovered more quickly in spite of having an initial viral load on average 10 times higher.

In this case it seems reasonable to look at each patients viral load after n days of treatment divided by his or her viral load aftr 0 days of treatment. The FDA will not allow doing what seems reasonable after looking at the data (nor should they). To avoid cherry picking the test must be described *before* the data are collected. I think that, reasonably assuming the distributions of initial loads would be similar for the two groups, the researchers said they would look a viral load on day n not that ratio, so the placebo started out with (on average) a huge head start.

Also Remdesivir caught up. After 2 days of treatment, the average viral load was lower in remdesivir treated patients. The ratio changed (roughly eyeballing) 100 fold. This is the raw data which was reported as Remdesivir fails to reduce the viral load.

The authors tested if this (apparently huge difference but I am cherry picking after 2 days) was statistically significant and got a p level of 0.0672. This means that, even if allowed to divide by initial levels, they would not have rejected the null of no benefit at the 5% level. It would have been reported as “remdesivir failed to affect the viral load”. This is would have been crazy.

I think the particular issue of divide by initial load or not is less important than the point that a p level of 0.0672 is not the same as a p level of 0.5, yet it is treated as the same.

5% is not a scientific concept. Calling 5% statistically signficant is an arbitrary choice due to the fact that the (smallest) 95% interval of a normal distribution is about 4 standard deviations. The idea that science requires one to look only at whether an number is greater or less than 0.05 is crazy and extremely influential.

The correct brief description of the results of the Chinese study is that, in the study, patients on Remdesivir did better than patients on placebo, but the difference was not significant at the 5% level. The direction the results points depends on the point estimate. Technicalities like p values are also important as are technicalities like the power function. But considering p values the main result and power a technicality is an error. It lead people to conclude that a ratio which was 100 after 2 days of treatment had been shown to be constant at 1.

I should point out that everything I am writing about hypothesis testing is elementary statistics which statisticians try again and again and again to explain.

After the jump I wonder why and type more about statistics.

Comments (5) | |

Does recovering from Covid 19 cause immunity to new infection by Sars Cov2 ?

The WHO warns that it is not proven that we acquire immunity to Sars Cov2. If we don’t we are pretty much doomed. However, I don’t see how people could recover without developing immunity or develope immunity without memory with human immune systems. Here, as often, the burden of proof is placed on the optimistic hypothesis. Here again (third time) I dare to be optimistic (third time’s a charm).

At Daily Kos Mark Sumner has an excellent post on the topic. I cut and pasted my comment below. I agree with everything he wrote except the part about the common cold (which is a syndrome caused by one of dozens or hundreds of diffferent known viruses). In particular a key open question is whether all people who recover are immune but some are more immune than others (as he notes it is known that some have low anti Sars Cov2 antibody levels). Also it isn’t known if and when there will be new strains which evade the currently existing neutralizing antibodies.

His post is brief and too good to excerpt. Just click this link (then come back to read my post or not — it’s not at the same level)

First there is strong evidence that exposure to Covid causes people to make neutralizing antibodies, you mention this.  Here is my first google hit  .  In that article, the cell making such an antibody was immortalized by fusion with a leukemic cell and so there is a neutralizing monoclonal which can, in principal, be mass produced and used as a treatment.   The problem (as usual) is that it isn’t proven that it works and such proof takes a lot of time and money.

I don’t think there are viruses which infect us again and again.  We are infected by new viruses (here I include mutants of the old virus as in seasonal flu). The common cold is a syndrome not a virus. There are dozens to hundreds of known viruses which cause colds (with similar symptoms).  By your definition we get many different viral diseases with similar symptoms.

There is certainly a risk that Sars Cov2 will mutate in a way such that it is no longer neutralized by currently neutralizing antibodies,   and so there will be a new epidemic even if enough people are vaccinated to provide herd immunity to Sars Cov2 1.0 (influenza does this which is why there is seasonal flu).  It is also possible that this won’t happen (there wasn’t a smallpox 2.0 which infected people who were vaccinated).

Importantly, antisera against Sars Cov2 can be tested (in cell culture) against isolates from different continents.  So far, I think, the evidence is that sera (polyclonal) antibodies which block one strain block all existing strains (which sure doesn’t mean all strains which will ever exist).  There is a vaccine (good old fashioned killed virus vaccine, which has protected Rhesus macaques from infection by Sars Cov2 squirted in their lungs b
If our immune system can’t stop Sars Cov2, then how to people recover ? Why has anyone gone from positive to negative ?

Again, we don’t get the same common cold virus again and again.  There are many different cold viruses.  Most are picornaviruses (double stranded circular RNA genome, small, like polio) some are coronaviruses, and some are adenoviruses (which cause fierce colds).

I think the WHO is putting the burden of proof on the hypothesis that, whatever happens in us to beat one Sars Cov2 infection lasts for a while.  If this were not true, it would be unique (I think).

update: there is now more direct evidence in a non peer reviewed preprint here

Neutralizing antibody responses to SARS-CoV-2 in a COVID-19 recovered patient cohort and their implications
Fan Wu, Aojie Wang, Mei Liu, Qimin Wang, Jun Chen, Shuai Xia, Yun Ling, Yuling Zhang, Jingna Xun, Lu Lu, Shibo Jiang, Hongzhou Lu, Yumei Wen, Jinghe Huang

“neutralizing antibodies (NAbs)”

Plasma collected from 175 COVID-19 recovered patients with mild symptoms were screened using a safe and sensitive pseudotyped-lentiviral-vector-based neutralization assay. [skip]
SARS-CoV-2-specific NAbs were detected in patients from day 10-15 after the onset of the disease and remained thereafter. [skip]
Notably, among these patients, there were ten patients whose NAb titers were under the detectable level of our assay (ID50: < 40)

So according to We et al 165 out of 175 people had produced antibodies which prevent infection of cells. So this is evidence that people recover because of acquired immunity. On the other hand, not all subjects had measurable quantities of the antibodies (and the measure is what dilution blocks virus entry). So an exposed and immune passport would be unreliable — those 10 people are assessed as still vulnerable. This isn’t the end of humanity (165/175 is enough for herd immunity to get R under 1). But it matters for policy (as WHO insisted). Also “remained thereafter” doesn’t mean forever. It can only mean up until now which is, at most, months after the recovery (even if the blood was collected as soon as some people had recovered).

Good news.

Comments (10) | |

Remdesivir 6

Bad news from China. There was not a statistically significant benefit for patients with severe Covid 19 in the large Chinese controlled trial of Remdesivir for patients with severe Covid 19. The death rate was actually a tiny bit higher in the treated group.

New data on Gilead’s remdesivir, released by accident, show no benefit for coronavirus patients. Company still sees reason for hope

As a long time Remdesivirenthusast, I feel I have to note this result here. Now I will go off and cry.

OK also be stubborn. One argument is that the treatment was given too late when autoimmunity not viral replication was the main problem. A counterargument is that the trial of treatemnt for patients with moderate Covid was suspended and one suspects not with promising data.

Another argument is that the better end of the 95% confidence interval of the estimate of the relative hazard of improvement (mostly needed less extreme oxygen therapy) is 1.75 which is high. Part of the dismay is due to the error of considering failure to reject evidence in facor of the null. Priors should be updated with the likelihood and roughly in the direction of the point estimate. Here, as usual, there is a problem with treating current regulations as the scientific method. This is not as bad as looking only at the most pleasing part of a 95% interval, as I just did. I couldn’t resist.

On the other hand note the (slightly statistically insignificantly) worse death rate for the treated group.

But all in all a horrible disappointment (there were warning signs including the premature end of the trial)

Gilead, of course, contests the Statnews interpretation. They note that due to trouble recruiting patients the power of the test was low (my point above). They note there was (statistically insignificant) evidence that it helped. Needless to sau, they are even more biased than I am.

update: Again someone who might be biasd pro remdesivir, but definitely interesting.

Frederick Hayden, an infectious disease expert at the University of Virginia School of Medicine who helped the Chinese doctors conduct the study, disputed the characterisation that the study had failed.

“That is not correct,” Hayden said in an interview, when asked whether the results showed remdesivir had flopped. “My interpretation of them is not consistent with that headline.”


Hayden said there was a mistake in the original manuscript and that it had since been revised. He said that the study was under review at a journal, which he would not identify. and that the authors were still waiting for page proofs. He declined further comment or elaborate until the study was published.

I want to believe that the mistake was saying that Remdesivir didn’t show ntiviral activity based in RT PCR

Comments (6) | |

Covid 19 Shutdown Politics

Advocates of quickly ending the shutdowns are in the news. Mostly, because one of them is President. Like most Americans, I think reopening soon would be a mistake (and remember I am in Italy where the shut down is severe compared to any State in the USA). I’m just going to assume that reopening by May 1 is a bad idea and try to understand who advocates it and why they do.

First the vast majority of Americans do not support reopening soon. There was the poll with 81% suppoert for a national lockdown. Also here is a poll which shows 10% believe gatherings of 10 or more will be safe by May 1. Interestingly the article reporting the poll is illustrated with a photo of an anti-lockdown demonstration. The silent majority is not photogenic.

So who and why

1. Wishful thinkers. The shut down is painful. It would be nice for it not to be necessary, so some people decide it is not necessary.

2. One of them is named Trump: I don’t pretend to understand him, but I think this is mostly infantile wishful thinking. He wishes there were no lockdown and no epidemic, so he believes it is possible. I am fairly sure this is also true of a majority of the small minority who thinks gatherings of 10 or more will be safe soon or in another 40 days.

3. Trump followers. I am sure that there would be even fewer advocates of reopening soon and no or very few demonstrations if Trump were’t against lockdowns. There are clearly many Americans who believe what he says and follow him blindly. I count the Fox News editorial staff among them. Fox follows Trump as he zigs and zags, and tens of millions of Americans follow Fox.

I am sure this is most of it, but I would like to go on.

Comments (23) | |

What has Science Established ?

I want to try to punch above my weight (and be a troll) and contest this

Dr. Kathryn Hibbert, director of the medical intensive care unit at Massachusetts General Hospital … said it will take weeks or months to get results and doctors and patients will need to wait for the results of the clinical trials before knowing which Covid-19 patients — if any — should be getting remdesivir.

“I think we should be extremely cautious about responding too enthusiastically to early data that has not been thoroughly vetted,” Hibbert said. “In the scientific community, over decades, we’ve established that critical review of clinical trials is the best way to determine if the data is strong enough to inform clinical decisions.”

I think the verb “establishd” is doing a lot of work here. Scientists have established two sorts of things.

One is the set of empirical results, styized facts and well supported theories. It is reasonable to accept that science has established that information can not be transmitted faster than the speed of light and that DNA (and topically for some viruses RNA are genentic materials.

The other is that scientists have established rules for what can be published peer reviewed journals (and rules for peer review) and Congress has established a Pure Food and Drug Act which authorizes the FDA to establish rules for which pharmaceuticals can be sold. There is a difference between natural laws and, you know, ordinary laws. Both are binding and both have been established, but one set is discovered and the other set is made.

Comments (26) | |

Two Cheers for the FDA

Only two because the FDA did play an important role in the US Covid 19 testing fiasco.

the first cheer is for extreme speed in approving phase 1 trials of candidate vaccines (vaccine in a person within 2 months of publication of the Sars Cov2 sequence) and phase III trials of Remdesivir (results leaking already) and hydroxychloroquine.

The second and even louder cheer is for the expanded access to Remdesivir policy. I actually predicted that this wouldn’t happen. I was wrong. I guessed there would be consideration of compassionate use in individual cases. I think the reason there is a systematic program is that the FDA and Gilead (the manufacturer) were overwhelmed with requests.

In particular, I note that one onlusion criterion is eligibility for a clinical trial. If one is eligible then one must risk getting the placebo. The logic is that people have the right to standard of care but do not have the right to an experimental drug which is more likely to help than to harm them. I argue for this logic here.

On the other hand, an exclusion criterion is “Requiring invasive mechanical ventilation (e.g., via endotracheal intubation or tracheostomy)”. This doesn’t make much sense. There is good reason to think that antivirals are useful especially if given early before the cytokine storm breaks. I think part of the logic here is first do no harm so an experimental drug can be used only on people who have very poor chances without it (the death rate of the intubated is roughly 66%). Also I guess Remdesivir will soon be in short supply (I think they have enough for 140,000 patients now and basically everyone in the world will want some).

So the expanded access is radical, but I don’t think it is radical enough.

Comments (6) | |

Premature Hydroxychloroquine After Action Report

The action sure isn’t over. The armies are still on the field, but I fear the outcome is no longer in doubt and hydroxychloroquine will lose this battle. There are two good studies which show almost exactly no benefit of hydroxychloroquine. They are reveiewed here. Basically the only benefit detected is as a nonspecific anti-inflamatory. There is essentially no evidence that hydroxychloroquine blocks replication of Sars Cov2 in people as it does in vitro.

I was an enthusiast for hydroxychloroquine so I have a lot of explaining to do. I will do it after the jump.

Comments (20) | |

hydroxychloroquine and Remdesivir Update

I have repeatedly argued that, based on in vitro evidence, hydroxychloroquine should be prescribed while we wait for the results of clinical trials. Such results were just published. They are very disappointing.

The link is to an LA Times Article: 2020-04-17 malaria-drugs-fails-to-help-coronavirus-patients-in-controlled-studies

On the other hand
Early peek at data on Gilead coronavirus drug suggests patients are responding to treatment

The University of Chicago Medicine recruited 125 people with Covid-19 into Gilead’s two Phase 3 clinical trials. Of those people, 113 had severe disease. All the patients have been treated with daily infusions of remdesivir.

“The best news is that most of our patients have already been discharged, which is great. We’ve only had two patients perish,” said Kathleen Mullane, the University of Chicago infectious disease specialist overseeing the remdesivir studies for the hospital.


“It’s always hard,” she said, because the severe trial doesn’t include a placebo group for comparison. “But certainly when we start [the] drug, we see fever curves falling,” she said. “Fever is now not a requirement for people to go on trial, we do see when patients do come in with high fevers, they do [reduce] quite quickly. We have seen people come off ventilators a day after starting therapy. So, in that realm, overall our patients have done very well.”

She added: “Most of our patients are severe and most of them are leaving at six days, so that tells us duration of therapy doesn’t have to be 10 days. We have very few that went out to 10 days, maybe three,” she said.

back to hydroxychloroquine and the very disappointing results

Quick summary of Tang et al pdf warning “The overall 28-day negative conversion rate was not [statisically significantly argg] different between SOC plus HCQ and SOC group (Kaplan-Meier estimates 85.4% versus 81.3%, P=0.341). ”

Conclusions The administration of HCQ did not result in a higher negative conversion rate but more alleviation of clinical symptoms than SOC alone in patients hospitalized with COVID-19 without receiving antiviral treatment, possibly through anti-inflammatory effects. Adverse events were significantly increased in HCQ recipients but no apparently increase of serious adverse events.

Mahevas et al was not a randomized controlled trial as explained by the title “No evidence of clinical efficacy of hydroxychloroquine in patients hospitalised for COVID-19 infection and requiring oxygen: results of a study using routinely collected data to emulate a target trial”

This study included 181 patients with SARS-CoV-2 pneumonia; 84 received HCQ within 48 hours of admission (HCQ group) and 97 did not (no-HCQ group). Initial severity was well balanced between the groups. In the weighted analysis, 20.2% patients in the HCQ group were transferred to the ICU or died within 7 days vs 22.1% in the no-HCQ group (16 vs 21) events, relative risk [RR] 0.91, 95% CI 0.47–1.80). In the HCQ group, 2.8% of the patients died within 7 days vs 4.6% in the no-HCQ group (3 vs 4 events, RR 0.61, 95% CI 0.13–2.89),and 27.4% and 24.1%, respectively, developed acute respiratory distress syndrome within 7 days (24 vs 23 events, RR 1.14, 95% CI 0.65–2.00). Eight patients receiving HCQ (9.5%) experienced electrocardiogram modifications requiring HCQ discontinuation.

The comparison doesn’t just fail to reject the null that hydroxychloroquine doesn’t help. the outcomes are very similar with and without. It sure suggests any possible benefits are very small (recall the warning that this wasn’t a randomized trial).

Comments Off on hydroxychloroquine and Remdesivir Update | |

Economic Policy for the Pandemic III What About Aggregate Demand ?

Unemployment has shot up in many countries especially including the USA. Is it time for aggregate demand stimulus ? One interesting thing (noted by Krugman in his newsletter which should still be a blog) is that Keynesians (including Krugman) are not treating everything as a nail, because all they have is a hammer. He (and others) haven’t just reflexively said stimulate until the unemployment rate is normal. This might be a bit of boasting about not making a stupid mistake, because the straw Keynesian demanding full employment now would be very dumb. Much (but not all) of the high unemployment is socially desirable. It is the result of the technology shock that, right now, the provision of services to customers carries the unusual cost of sharing Covid 19. So baristi and waiters should stop working. Their sitting at home is not a waste of productive capacity — it is an efficient response to the pandemic.

But note the parenthetical “(but not all)”. I will argue that it is also true that, in addition to the efficient unemployment caused to reduce contagion, there is also inefficient unemployment due to low aggregate demand. This means that stimulus makes sense. This includes cutting interest rates back to zero again and also broad fiscal stimulus including the $1200 which many of you may have received today. The reason is that the loss of income from currently socially undesirable work (tending bar and waiting tables and such) causes a larger decline in total production through the good old Keynesian multiplier. That additional decline is inefficient and can be prevented with monetary and fiscal aggregate demand stimulus.

This is so obvious that it is even obvious to Donald Trump. Or rather this is a point where Trump, Pelosi and Krugman agree so it must be obvious. I am going to play with some counter-intuition explaining how it can be non obvious to someone who knows a little general equilibrium theory but missed developments in the field since 1950 (I am thinking of fresh water macroeconomists who have been very quiet lately).

I realize that this post makes an obvious point in a boring technicl way so I put all the rest of it after the jump.

Comments (1) | |