As I argued a month ago, there’s no fundamental reason that putting out a vaccine for the coronavirus should take the “more than a year” or “at least 18 months” that is commonly quoted; those numbers come from the assumption that we’d follow the usual rules. At the same time, though, there’s considerable experience and wisdom incorporated into those rules; the idea shouldn’t be to discard that wisdom, but rather to understand it and adapt it to the emergency.

The idea of a randomized controlled trial, for instance, is fundamentally sound. Having a control group (which only gets placebo) puts an end to an awful lot of quarreling about how exactly people were selected for the trial and what exactly their infection rate would have been without the vaccine. (Anyone who doubts how ugly such quarreling can get is invited to do a web search for Didier Raoult, who has been the subject of much personal vilification after his authorship of uncontrolled studies on hydroxychloroquine and azithromycin.) Having a control group isn’t something that has to slow trials down: it’s more work, since the control group has to be recruited and given a mock vaccine, but that work can be done in parallel.

The idea of ramping things up, likewise, is sound: there’s not much point in suddenly trying a new vaccine candidate on a million people. What doesn’t have to be adhered to are the stylized distinctions between Phase I, Phase II, and Phase III trials, where Phase I purely tests safety, Phase II is a first stab at efficacy, and Phase III is the big trial that nails everything down. Safety and efficacy can be tested at the same time. Indeed, in the case of a vaccine, the two are intertwined: one of the concerns that has been voiced about a coronavirus vaccine is a phenomenon known as “disease enhancement”, where people who are vaccinated actually get a worse form of the disease than the unvaccinated. It doesn’t seem like a huge concern with this particular virus, as opposed to, say, dengue fever, where it happens even with the natural infection itself: the first time you get dengue you might not even notice it, but the second time it’s a nightmare of pain. But although not a huge concern here, disease enhancement is still worth testing for. It could be considered as a safety issue or as an efficacy issue (efficacy less than zero); but either way, it can’t be tested for without trying the vaccine on people who are at high risk. It is thus dismaying to read of Phase I coronavirus vaccine trials being started in which they specifically try to find subjects who are at very low risk of getting the disease. That’s what the rules for Phase I dictate, but the rules obviously weren’t optimized for speed.

On the other hand, recent suggestions for “challenge trials”, where subjects are deliberately exposed to the virus to test the vaccine’s efficacy, seem unnecessary. There are quite enough people being challenged in the wild these days to get enough statistical power. To get quick results in the wild you might have to vaccinate twenty times as many subjects as you would in a challenge trial (since even picking volunteers to be vaccinated from a high-risk population, only one in twenty of them might be exposed to the disease during a short trial period). But the precautions that are recommended by advocates of a challenge trial seem like they’d make each subject much more than twenty times as expensive: not only would the trial’s organizers pay for quarantining the subjects for two weeks to prevent spreading the challenge virus, but also for their medical treatment in the event that the vaccine failed. And a lot of vaccine candidates fail; while there’s much talk of vaccine side effects, the most common way of failing seems to be simple lack of effectiveness. Even successful vaccines often give immunity only to 90% of the people who are vaccinated; the remaining 10% are protected because the herd immunity of the 90% keeps the disease from spreading widely. (That’s the proper use of the concept of “herd immunity”, before it became a miserable excuse for failure.)

Although much of the ritual of the three phases could be dispensed with, the basic pattern inherent in the three phases is still worth using. That pattern is exponential ramp-up; each phase is much larger than the one that came before. As the virus itself has illustrated, exponential growth can get you very far, very fast, even starting from very modest origins. If you started by testing on ten people and doubled that number every week, it would only take ten weeks for you to get to ten thousand people tested. The exponential ramp-up wouldn’t have to be divided into discrete stages: instead you could just keep one set of statistics covering everyone who had been enrolled in the trial, continuously updated as information came in. When continuously updating and assessing results, the criteria for statistical significance would have to be toughened a bit: if you keep looking at every random up and down movement, there’s more chance of falsely concluding that the trial is a success or a failure than there is if you wait until the end and look only at the final numbers. (To compensate for that, you’d use something resembling the Bonferroni correction, though not exactly that since its assumption of statistical independence doesn’t apply. (Update, Nov 2021: the name for such techniques is “sequential analysis”.))

Though that doubling-every-week schedule would be fast, during any one week we’d be testing the vaccine on only about the same number of people as it had been tested on in all the previous weeks combined. If, say, there were a risk from the vaccine that only became evident after two weeks of incubation, only four times as many people would get exposed to the risk as had been necessary to show the risk’s existence in the first place. For example, say it was a risk that only hit one in a hundred people; then you might need to test a thousand people to show its existence. (The first bad outcome might just be random chance, but as it repeated ten times the odds of it being random chance would dwindle.) With two weeks of incubation and on the doubling-every-week schedule, by the time you realized the risk four thousand people would have been exposed to it and about forty injured by it. This is not so different from the sorts of numbers of injured that conventional trials might yield.

One could of course imagine there being some secret horror that lurked undetected for months after vaccination before emerging; that would indeed only be detected by a long trial. Disease enhancement could be delayed like this, if the trial were done on people who were at little risk of catching the virus. But that’d be a crazy way to do the trial in the first place: to judge effectiveness you need to run the trial on people at high risk. Other than that, I’ve never heard of any such secret horror; even the anti-vaccine nuttery I’ve encountered has spoken of harm starting right after vaccination. Nor does it seem like there’s a plausible mechanism for a secret horror: the immune system can be dangerous, but immune responses such as are invoked by a vaccine tend to be strongest at first, then decay. I’m not a subject expert on this particular matter, so I may be missing something. Still, if there is some real prospect of a secret horror, defenders of the current rules should say so (and name it), not just put on the deer-in-the-headlights look of people who have never seriously thought about which parts of the rules are essential and which parts are accidental.

Of course, to change the rules would require political backing. Now, at a time like this, one might consider breaking rules and asking for forgiveness later. That seems like the one real advantage of challenge trials: a small group of people could do them in secret defiance of the rules. They would, however, have to expect some very ugly behavior from all the people they’d upstaged: failure would probably mean prison time and success would probably mean other people stealing all the credit. (The one exception is if the people who made the vaccine tested it on themselves; that always commands respect.)

Also, it would be best to make these sorts of rule changes permanent, because getting vaccines out quickly isn’t just a coronavirus issue. With the flu vaccine, effectiveness numbers below 50% have been usual in recent years; and the reason usually given is that they failed to accurately predict the strains of flu that would circulate. If we could redo the flu vaccine quickly to fit the circulating strains rather than having to live with guesstimates from nine months before, we might actually be able to do a good job of it.