OK. The recording is happening. We're very lucky to have Nick Drew joining us today. He's the professor at London School of Hygiene and Tropical Medicine, as well as a professor at the graduate school at UC Berkeley. So there was some debate over whether or not he'd talk about COVID, but we opted to ask him to speak about something that wasn't covered today so we can have a bit of variety in our lives. So you'll be I'll keep people on mute except for Nick.
And then if you have questions, please put them into chat and I will try and note if something is an urgent question, where if it is, I might politely interrupt Nick to to ask it. Otherwise, I will keep them till the end and try to cover the range of questions as if it's urgent. And I haven't noticed it. Please do the raise hand button and then I will be even more likely to notice you. OK, thank you very much for that, and I will hand over to Nick, who will now share his screen.
Thank you very much, Crystal. Thanks for the invitation. It's a great pleasure to speak about something other than COVID 19 for a while. Let me just share the screen, so I very much appreciate the opportunity to talk about something else that's close and dear to my heart. And as I say to my friends, I only wish I could go back to normal life with Ebola and dengue instead of COVID 19, which always raises a few eyebrows.
But that's the nature of life as a statistician involved with infectious diseases. So what I'm going to talk about today is a trial that is in progress to eliminate or reduce the burden of disease from dengue fever and the statistical aspects. So I'll give a background as to what's going on for those of you who may not be familiar with this approach to controlling dengue. So this is a study I'm going to talk about is sponsored by the World Mosquito Programme.
You can see there in Spanish. It used to be known as the Eliminate Dengue programme and indicate why it's changed its name. It will become obvious in a few minutes. There've been many, many people involved in this trial, including Crystal here and Neil Ferguson at Imperial. And, of course, the local investigators at Yogyakarta. This programme comes out of the lab of Scott O'Neill at Monash University in Australia and several other people.
And the last person I want to acknowledge is my graduate student, Suzanne Duvall, who just graduated with her Ph.D. last week. There's a bunch of references here. I'm not expecting you to look at those. They're just here. If you ever download this or want the slides so you can have the background for some of the references that described some of the work, that's in this talk. So let me say a little bit about dengue fever.
It's one of the most important mosquito borne viral diseases, of course, along with malaria, it's very rapidly increasing, with roughly a 30 fold increase in global incidence in the last 50 years. A significant part of the world is at risk of infection, and that number is growing. I'll show a map in a second. It leads to an enormous number of infections per year and disease burden and dengue.
When you're first infected with one of the four serotypes of the virus that causes dengue, it's likely that you suffer what would be most equivalent to a severe case of influenza with joint pains and the like for an end. But it's rarely fatal. But what can happen with the because of immunity reaction?
If you're subsequently reinfected by another serotype, it can often lead to much more serious health complications, including dengue haemorrhagic fever, and that causes a lot of hospitalisations and a significant number of deaths per year. Dengue is transmitted primarily, but not universally, by the species of mosquito in order to elicit a fight. So a different species from the primary mosquito that causes malaria.
And here is a map of where a decision play lives in the world or a heat map here, with red being very high concentration of very good conditions for breeding over the subject. Slowly declining to the blue and the grey to cold and not appropriate for. For the life cycle, for a dosage of type, and you can see there that it's an area with a covers a significant amount of Latin America, Africa, India and Asia.
And just in north the northeast corner of Australia on here I. You can see up on the United States there, it's just creeping in. And in Florida, a little bit of Texas, but it's creeping up there because with global warming and climate change, more and more of the northern hemisphere is coming, becoming suitable for a subject. And now there are industries by in California not far from where I am right now, beginning to develop.
They're also so that's becoming a more and more risk as we go on, it's not in the United Kingdom yet, but it will become in the United Kingdom if global warming continues at the pace it is in in due course. This is similar a graphic that reflects where the disease burden occurs. In addition to where the mosquito lives and so you can see there on the left and by far the biggest country that suffers dengue is Brazil, followed by Indonesia, which I'll come back to in a minute.
Then Vietnam and so on. And so very significant effects just where you find. Aegypti. And so, of course, humans have battled for aeons against mosquito borne diseases, and there's a number of interesting public health interventions that are designed to try and protect humans from infection from various diseases transmitted by mosquitoes. And I'm not going to go through these in great detail.
Many of them, I'm sure all of you were aware of there can be direct vector control that's trying to kill the mosquito. Stop it breeding by usually spraying larva sites or removing habitats. That's the sort of thing you would have seen during the world before the World Cup and the Olympics in Brazil because of the concern then of Zika,
which is also transmitted by Aedes aegypti. And so you would see these images from the TV of people in white protective suits spraying various streets and areas as much as we're seeing now for COVID 19, and I'll try and resist mentioning COVID 19 as much as possible.
It's largely been ineffective. It's a long term strategy, in part because the efficiency of data subject time breeding is renowned and even a small plastic bottle cap that's left in the street that fills with water from rain can breed an enormous number of subjects in short order. So it's extremely difficult to remove all possible breeding grounds and spray all of them, but it is used. Vector barriers have been very effective for malaria.
That's bed nets, largely our personal insect, which is what I use when I'm travelling in a just aegypti country, and that certainly can work at the individual level effectively. There has been some attempt at vaccine development, and there is now a vaccine for dengue fever called its trade name is Dengvaxia that was developed by Sanofi and was approved for use in many, many countries. It was suspended. Use of Dengvaxia was suspended in the Philippines.
Now I'm going to say almost a year to two years ago because of a cross immune reaction to the vaccine in the sense that individuals, children, particularly who were vaccinated. And had been naive to end any strain of the dengue virus who were subsequently infected, even though they were vaccinated, suffered very severe complications and since the vaccine was only about 60 percent effective, that was not a rare occurrence. In other words, 60 percent effectiveness is not awful.
It's it's useful. It's it's not sufficient to eradicate dengue, but that's very helpful. But when you get the 40 percent of children who were vaccinated but still vulnerable, and then they suffer a severe reaction to a subsequent infection that caused a suspension of the licence of its use, and that has cast a great shadow over the use of Dengvaxia. And now that nor the newer sort of modern high tech strategies for controlling these diseases.
The genetically modified mosquito development from a British company, Oxitec, is still in full operation. It's an attempt, and now there are many different using CRISPR technology. There are many variants of the Oxitec strategy to genetically modified mosquitoes to essentially interrupt the life cycle and not allow them to either breeders effectively or other have other consequences that limit disease transmission.
I'm not going to talk about that. Some of you, I know, know about those experiments to test this as a strategy. What I am going to talk about for the rest of the talk is the second one, which is a more natural version of modification of the mosquito, and that is to deploy mosquitoes that are trans infected with a bacterium known as Wolbachia.
And if those of you are not familiar with Wolbachia, I'll talk about it very briefly to show you why it has an impact, but this is a naturally occurring bacterium. And here's the little graphic, that's you. So you've got the red blob of the virus, and it's now appearing simultaneously with the green dots for you and or vice versa? Doesn't matter. The issue is having both at the same time changes things and let me talk about Wolbachia and what it does to the mosquito and in particular,
what it does to dengue virus living in a mosquito. So Wolbachia is a very natural symbiotic bacteria and you all know about symbiotic bacteria. We carry a huge amount of symbiotic bacteria in our bodies constantly, largely in our gut. We depend on those bacteria. If you remove all of those bacteria from mammals, the mammals die. We know this from mice experiments where they're born and in a complete bacterial free environment, and they don't live for more than a few weeks.
So mammals depend on symbiotic bacteria to digest our food, to protect those largely from other bad things, and so they don't cause us harm. Most bacteria don't cause us harm, and Wolbachia is such a bacterium that exists in the insect population that does not exist in warm blooded animals. So it can't be transmitted from a mosquito to your cat, for example. And so it but it's a very common, in fact, the most common symbiotic bacterium living in insect species.
And you can see it in the graphic there. A lot of different insect species carry Wolbachia naturally. And if there's some doubt about this, but it largely was believed that it was not carried by Aedes aegypti, naturally, it could not be found in Egypt. I know it could be found in other mosquito species. And so it was not existent in the wild.
It is a type population and. Let me just go back a second until it was determined in experiments and Scott O'Neill's lab how to trends in fact, a is a subject a mosquito with while back in the laboratory and then breed the mosquito in the laboratory to produce a number of mosquito that were carrying Wolbachia.
And that took a significant amount of time, it was originally thought that this would be potentially an effective control strategy by shortening the lifespan of the mosquito, perhaps shortening it in some ways similar to genetic modified mosquitoes, so that it would shorten it below the time when they have maximal breeding and therefore reduce mosquito populations.
That turned out to be true, but only slightly, and not really a significant enough effect to have any impact in the wild and reducing mosquito populations so scientifically. This took a great amount of effort that looked like a complete dead end.
Because it didn't do what was hoped, and then in one of the serendipitous discoveries that science is so full of, someone looked at it as suggesting that we're infected with dengue and saw an effect of the presence of Wolbachia, and I'll come to that in a moment. Let me just point out that Wolbachia carries with it cytoplasmic incompatibility. So this is very important in when you introduce it in the wild. It's how quickly Wolbachia infections in mosquitoes can spread.
So an infected male and a uninfected female cannot produce offspring, but in all other combinations, infected females with an uninfected male and with both mosquitoes being infected, Wolbachia hold all of their offspring have. Rollback infections through birth, so through the egg, so that cytoplasmic incompatibility means that there's very rapid transmission.
By the way, this suggests one strategy for using Wolbachia, which is to use to release infected males into a population because they're they're infertile and they compete with uninfected males and therefore reduce mosquito reproduction. And this is actually been used in California. In the Central Valley, which produces a significant amount of the world's food to reduce rollback, I would reduce the subject statutory abundance.
And so that's a different strategy. But that's not the strategy I'm going to use now. I'm going to talk about deployment of both male and female, and there are some reasons for doing that. One of which is it's actually quite expensive to have to separate out when you're breeding mosquitoes separate out the males from the females. And so that isn't done in the experiment I'm going to discuss.
And so that cytoplasmic incompatibility means that when you release or deploy mosquitoes with Wolbachia represented by the green here into an area in the wild where of course there is no Wolbachia. And these are representing the number of weeks. So these are releasing mosquitoes, deploying them usually in larvae farm in a population where there are wild. It is a job type.
You can see over about three months the infection spreads very rapidly because of that cytoplasmic incompatibility and within three or four months, essentially, almost all the mosquitoes carry Wolbachia infection, so they're carrying the bacterium. So this means it's a relatively straightforward way of introducing Wolbachia into the population that doesn't have ecological consequences per se and reducing mosquitoes and therefore upsetting the animals that depend on mosquitoes.
So it actually doesn't reduce that abundance at all, but it does ultimately rather quickly lead to a sustained infection with Wolbachia and in experiments that are now going on for many years. There isn't really a significant reduction. Once it's there, it's there and it doesn't disappear. So unlike genetically modified mosquitoes, you don't have to continuously deploy Wolbachia mosquitoes to try to keep a high level of infection.
And so this is the experiment I was talking about that led to that evidence or some of the evidence, there's more evidence now, but this was where they were originally. Released in the wild in a pilot in the northeast of Australia, in Cairns, and you can see this was done almost 10 years ago, over a couple of month period where Wolbachia was deployed. By the way, Cairns is one of those areas in Australia that does suffer from dengue, usually not endemically, but introduced through travel.
With travellers going to Asia getting infected with dengue returning, they get bitten by a local, a subject which sustains an outbreak for a while, and then it disappears until a new introduction. So there's interest there in reducing that burden. It was introduced in 2011 over a couple of months. So there you can see that the percent of wild mosquitoes in that area of Cairns, two areas of care and so we're releases were done in a pilot.
Fashion has sustained over 100 percent now for actually much longer than five years. OK, so what's the big deal about getting Wolbachia into a mosquito? What does it do to dengue? So the key discovery was that made it reinstate interest after noticing it didn't do a great deal to the lifespan was that it blocked the ability of dengue virus to replicate inside the guts of the mosquito. So the mosquito bites and infected human it, it draws in dengue virus through the blood.
It goes into the gut of the mosquito where it normally will replicate. It will then move through the body of the mosquito up to the saliva of the mosquito bites. Another naive human, it transmits the virus from the saliva of the mosquito into the blood stream of the human on which it's feeding. And that's how dengue is transmitted. There are these four serotypes, as I've mentioned, a very brilliantly named dengue one, dengue two, dengue three and dengue before.
And it was noticed in the lab that the ability to of the mosquito to find dengue virus took from an infected mosquito after they've been infected with Wolbachia was significantly reduced somewhere in the order of 70 to 90 percent reduction in being able to detect dengue in the saliva of infected dengue infected mosquitoes after they were infected with with Wolbachia. And this is pretty consistent across all serotypes and most areas of the world. So there's various prevalence.
Differences carry all four serotypes, and that's part of the issue around cross immunity that occurs in these countries. Well, that was interesting.
But really, what's more interesting, the stakes went up considerably higher because dengue is a member of a family of viruses like the coronaviruses, a family of viruses and the dangly is a member of another family of versus known as flaviviruses and many of the familiar and horrible viruses that humans suffer from, including yellow fever, which was a great scourge of the 19th and early 20th century West Nile virus, which is a more recent worldwide concern chicken junior Zika,
which became a huge concern several years ago, and Japanese encephalitis and so on, and the ones I've checked on the left. And actually, I think now this slide is slightly outdated. One or two on the right have also been checked that they have exactly the same response to the presence of Wolbachia in these viruses can't be replicated within the body of the virus within the body of the mosquito. So this suddenly raises the stakes because it means that it is aegypti infected with the bucket.
Not only might be enable unable to transmit dengue, but unable to transmit yellow fever, chikungunya, Zika and many countries suffer from all of these viruses. Brazil suffers from yellow fever. Still, even though there's a very effective vaccine for yellow fever.
They certainly suffer, as we all know, from Zika and chicken. And so suddenly, this robot here has the appearance of maybe being a magic cure all, it's like taking one pill or having one integrated pharmaceutical intervention that protects you from multiple diseases simultaneously, which is a very rare outcome pharmaceutically. And suddenly, this was a tantalising possibility.
And so this really raised the interest of whether this could be an effective strategy for protecting not only against dengue and against all serotypes of dengue, but against these other flaviviruses. And so now interest around the world in places that suffer from the burden of these diseases showed great interest in this. And here are some of the current sites in the world where Wolbachia has been released in various levels.
And I'm going to focus entirely on here on Indonesia, which is currently in the midst of a rather severe dengue outbreak. At the same time as COVID 19. But there's significant effort going on elsewhere in the world that I'd be happy to answer questions about. So the real question now is you're there, so this looks like a great possibility, but this is all in the lab. Everything I've said has been larger than the lab other than make showing that you can logistically deploy will be accurate.
Mosquitoes and in fact, a wild population. But does it really protect people in in in the in the field? Does this really work at scale? Can we do this out there in in in urban and city environments? And so now you've got to try it. So the question when Crystal and I will remember being at a meeting where basically the question on the table to a bunch of infectious disease statisticians and biologists were?
How do we design a study that will actually produce convincing evidence or not that this is it could be an effective intervention strategy? And that's what I'm going to talk about for the last twenty five minutes ago is that design of that study. And basically, we sat down and over a few days in a workshop came up with what at that point I thought it was a case control design.
The natural experimental design was to use a cohort study because there's a natural aversion to case control studies, because of concern of confounding and the ability to control confounding.
So a cohort study, I won't describe it in detail that was envisaged in this was enrolling a large cohort of susceptible to dengue that's largely minors and children because most adults in this country have developed have been infected already and are immune to one or more of the one or all of the serotypes of dengue.
So you're you're establishing a large cohort of children and then following them in and maybe a cohort of children who are in an area where you've deployed Wolbachia and comparing them to a cohort of children in an area where Wolbachia has not been deployed, following them for several years and intermittently bleeding them to determine whether they've been infected or not. Because since dengue often produces symptoms that are similar to many viral infections, fever, high fever and so on.
It's not distinguishable without a definitive test diagnostic test. So this raises the prospect of trying to. Recruit and retain a cohort of many, many thousand of children and bleeding them constantly. Many of those bleeding episodes, of course, would be uninformative to the extent they would be negative. We wouldn't be expecting everyone, every child to be infected with dengue over a even a several two or three year period.
And so that was conceived of being a very difficult study to actually do effectively to recruit and retain and ethically draw blood samples so frequently. And so we moved to a more case control kind of issue. And here the idea was to enrol dengue cases and non dengue controls in hospitals and clinics in an area, and this is schematically represented here.
Here are the clinics. And the idea is people present if if it right, if the infection within it rises to the level of symptoms again, largely in children and minors, the parents would bring them to a clinic for treatment. In fact, it wouldn't be clear as the arrived that they had dengue fever or some other viral infection. And so blood samples would be drawn and treatment would be given while test results were weighted. The test results would either be positive or negative for dengue.
The positives would form the basis of your case control, design and the negatives of the words. Children infected with some other viral infection would be the controls. The controls would actually have to remove other saliva virus infections like Zika and chikungunya and so on. But so the controls would have to be injected with some other non-Florida virus and they would form then your ability. You would then check with the parents where they lived and their location.
And then if you had an area where you'd partially deployed Wolbachia in the green areas and not in the way you'd look at the place of residence of the child or the infected individual, and look at the rates at where the cases and controls arose from and do a comparison there to determine if that distribution of locations differed depending on.
From cases to controls. And that's the basic structure of the design, and I'm going to follow on as it happened, I quickly discovered I went back and I thought, Well, well, why is this interesting? So we're doing case control. I mentioned what you worry about confounding other factors because people don't deploy across a geographic area uniformly.
The socioeconomic status, population density all differ substantially geographically, and many of those are known risk factors for infection of dengue. So wouldn't you worry a lot about confounding? Well, the beauty here is that the that we then moved forward is if we randomly allocate the regions to Wolbachia deployment, then in fact the exposure is randomly allocated.
And therefore, there can't be any confounding in principle, because we have this ability to randomly export to do a random allocation of exposure. It's very rare and actually started when I went away from Australia. After thinking about this design, try to think of other experiments in epidemiology where there was a natural random allocation of exposure.
And then you superimposed on top of that case control ascertainment, which is efficient because you can find cases more easily and yet not worry about confounding. And I look for examples, and to be honest, I didn't really find a great example. I'd still be curious if there are those examples in the history of epidemiology.
What I did find was a essentially a design which mimicked what we had come up with independently, which is called the test negative design, and it's suddenly become very popular. And in fact, people are not going to talk about this. People are trying to promote test negative designs to study COVID 19. Currently, a test negative design had been developed initially, largely to test the seasonal influenza vaccine after original clinical trials to test influenza vaccine.
We don't test them at clinical trials year after year, as you all know. And yet, year after year, you will see on the BBC some and the ITV you'll see some mention of the effectiveness of the seasonal flu vaccine this year. Is it 60 percent effective with last year's 30 percent effective? And you might ask yourself, Well, where do they how do they estimate the effectiveness? Well, those numbers are long largely coming from test negative designs, and it's largely come from taking sentinel clinics.
Getting data on everyone showing up at the Sentinel clinic with symptoms of a viral infection. Drawing a sample to test for the presence of influenza or some other form of viral or some other cause of the symptoms. Separating the two into two groups to test positive for the influenza cases of the test negative on influenza. And then at the same time as you recruit the participant and take the sample asking them whether they've received the flu vaccine.
And then you compare the frequency of vaccination amongst the flu test. Positives to the vaccination rate in the flu or test negatives, and that comparison leads an estimate of the efficacy of the seasonal vaccine. And again, now you immediately should think, wait a minute, vaccine status is not randomly allocated. It's clearly not, and it's clearly selectively allocated in a way that also due to factors that may also influence the risk of influenza.
But the big advantage of why it's used in influenza is it this test negative design controls or eliminates the confounding due to health care seeking behaviour? So people who get vaccines are more likely to seek care for influenza, which is not usually a very severe disease or fatal disease. And so you might worry that people who get vaccines are more connected to the health, health or health care system and therefore seek health care when their child or they get influenza.
And therefore, that will bias the results. But that's not possible with the tests negative design since the individual does not know on arrival at the clinic whether they're infected by influenza or some other virus that is only determined post recruitment. So this is why it's been used, and there's a significant literature about its use for in the context of influenza vaccination assessment, but also in other respiratory diseases and other diseases that I'm not going to talk about today.
So now let me move on to the actual experiments in the last 10 or 15 minutes or so the site should send from here. And there's a lot of care going into choosing the sites because once you've moved in and you can't use this site ever, ever again in history to test things because you're going to contaminate the site by releasing Wolbachia. But the city chosen with junk Jakarta in Java in Indonesia, it's a very large city.
I usually say it's a city the size of Oakland, but I realise that doesn't tell you much if you're a native Californian, but it's a fairly big city urban environment. There's a picture of it and the right not far from Jakarta, but it's much, much smaller than Jakarta, which is huge. And this was piloted in external areas around the city just for logistics. And then we implemented this design in starting in around 2017.
Late 2017 and what was done was the city was by the statisticians, the city was divided into regions and the deployment was randomly determined by the region. And there you can see on the left here the map of drug Toccata broken up into these patches or geographic regions contiguous and then a random allocation of deployment. Now I should have said when when I pointed out this test negative design and the advantage of randomisation.
And randomising exposure, unlike vaccination, this is not an intervention you can implement at the individual level because you can't give each individual their own personal Wolbachia mosquitoes to carry around with them. So you have to do it geographically. So now we've introduced a statistical complication because this is a cluster randomisation that we're super imposing in addition to sort of a case cohort sampling design of of outcomes.
But there are the regions they were allocated and there were 24 clusters. Obviously, you can't have 2000 clusters because then the clusters become very small geographically and there would be spread of infected mosquitoes into neighbouring regions. So the spread was considered in designing how many regions that could be restart. The regions had to be a roughly one to two square kilometres. So there's about 24 of them are exactly 24 of them. At City, 12 were allocated at random. I do, too.
Wolbachia deployment and 12 to nothing and. In just two, I won't go into all of the technical details because time doesn't allow it, but in choosing in these regions, there was an attempt to use natural boundaries like major highways or rivers to limit the contamination or the drift of mosquitoes from one region to another, so that the Wolbachia infections would be contained within the 12 clusters that had been chosen.
And that was checked throughout the experiment, using mosquito traps to measure the prevalence of Wolbachia in all 24 regions and. Yeah, let me keep going here. And I should say the other thing, I was just pausing there, it says to the sensors a very small number of cluster randomised units here 24 and you worry about balance.
Actually, a constrained randomisation beer was used to balance the green, the Wolbachia areas from the great on a variety of factors, including things like population density and and past history of dengue infection. And a bunch of other factors. So that was there was a constrained randomisation that I'm not going to talk about here.
There's an interesting issue here, also in the pilot areas, there were two pilot areas that had been done several years before where systematic surveillance thing, decent population density, surveillance of severe cases haemorrhagic fever were detected and we could follow then the incidence of those before and after in a pilot region where Wolbachia was employed in an area where it wasn't.
And that leads to interrupted time series. But I'm not going to talk about that that much, but here you can just sort of see, it's quite a difficult set of data to analyse with interrupted time series because of the great volatility and seasonal appearance of dengue fever.
Dengue causes a certain amount of herd immunity if there's a huge outbreak and then it goes away for a year or two and then it comes back again, you can see this phenomenon here going on before and after you might introduce an interruption or an intervention. And so it was quite tricky to do that, and that's described in some of the results that were described in some of the references I gave at the beginning. But I want to go back in the last five minutes or so to looking at now.
The results are the outcomes data from a test negative design at this clustered level. So here's the the raw data that you will get that's being collected. So in the two intervention clusters, there will be a cumulative count of the number of tests positive, the number of people testing positive for dengue in a negative in the in the area, in the whole population. The rules here represent the intervention. Areas in the control is.
In the whole population, of course, there's a lot of people who get infected by one virus or another and never show up at a clinic or never get infected. And so what's observable to you from the way I described the data collection are only over here, ABG and age, and you don't observe this. Ideally, what you'd like to measure efficacy is you'd like to know the rate of dengue in the exposed area, which is over nine and the rate in the unexposed, which is over and see.
But now in and steer unobserved and they're not really useful in. It's not really the total number of people in the population who can't usually use that as a proxy because the selection isn't appearing at a clinic. So I can see here the numbers of people in the population who would have shown up at the clinic had they been affected by a virus that produced these kind of symptoms. And that's really just impossible to measure. So you can't really measure and I don't see.
But the idea of the test negative is is to exploit the fact that other viruses are not affected by, well, Baqir. And therefore the rate of occurrence of PNH should be an exact proportion to the sizes of the population that seek health care so that be over any is approximately the same as h over NC because the distribution of test negative should be completely independent of the intervention because Wolbachia does nothing to the other viruses.
And that's of course, the key aspect to the test negative design. I shouldn't have said that a flu vaccine you've got to assume the flu vaccine does nothing with compared to other respiratory viruses that might cause similar symptoms. But if you assume that, then you can substitute the ratio of P over H for the the population sizes that you would like and you get an estimate then of this relative risk are one minus the efficacy.
And the idea here is that you have removed or certainly severely reduced confounding. The problem here statistically was is that you've got this at the cluster level. So actually what you have is you have one of these tables for every cluster and then you collect the cumulative odds ratio across all clusters. But the statistics has to account for the clustering while exploiting the randomisation.
And I was a great fan and still am of using permutation tests here, particularly given the small number of randomisation units. And what I want you to understand for power calculations and design was the properties of that permutation distribution, and that can be done relatively straightforward. I'm not going to talk about this measure. I'm going to talk about the the odds ratio estimates. I'm going to jump over those sites.
The odds ratios say here's the cumulative odds ratio estimate here that I described before comparing the intervention areas and the control accumulating over the clusters. And by using finite sampling ideas and techniques, you can actually figure out from the property on the permutation distribution, an approximation to the variance of that under the null. The mean of the log odds ratio will be zero just by symmetry, but you need to know the variance.
And here's the variance that looks complicated. It's not that complicated. And each of these variances within within each of these terms can be easily estimated from the observed data. So that allows you to simulate the permutation distribution for power calculations, which is important because the permutation distribution to compute is hard to do repeatedly.
And then you can do these power calculations, and I didn't talk about this alternative, you're just there the odds ratio test, but it shows you that the the the the with the number of clusters we were using, we had reasonably good power for detecting a 50 percent reduction in dengue, comparing the deploy to the UN deployed areas, and I'm drawing to a close here.
There are lots of interesting statistical questions that I've just touched on, and I hope the main point of talks like this are not to overwhelm you with technical details, but just to stimulate your interest and go away and think about things. You can actually just use the case only day to year. That's just using the test positives because of randomisation, because you could just compare the frequency of tests positive in the intervention area to those in the non-intervention area.
That comparison, we've studied it, but it depends critically on the randomisation working and balancing everything else effectively. Interestingly, you can use it in the test negative to see and check the assumption that the test negative distribution should be independent of the intervention areas. There's interesting questions here. Well, what else can you estimate beyond estimating the overall efficacy of the intervention?
Can you estimate and distinguish the efficacy that's due to the direct objective Wolbachia blocking, but also the herd immunity effect? Because if you're in an area with Wolbachia, fewer people are being infected is the idea, and that will mean that your risk is reduced because you're living around people in your household and next door who themselves are protected. So like vaccination, there will be some herd immunity. And so whether you can estimate from that data is an interesting question.
There's also the problem of even though we've tried to protect the contamination of the infected areas by essentially taking advantage, the mosquitoes don't travel a great deal during their lifetime. Humans are mobile and do move across the city. They may go to school in a different area. They may visit grandparents in a different area. And so that hasn't all been measured in this study.
We're humans, we're the test positive test negative moved in the few weeks before they showed up at the clinic, and that allows one to measure a sort of continuous exposure measurement. And how that should be used in the analysis is interesting. There's a considerable interest in the transferability of these efficacy estimates.
If I estimate an odds ratio for drug Jakarta, will it be useful in Rio de Janeiro because where there's a completely different population distribution of mosquito distribution? And that's interesting because the herd immunity aspects will be different in different locations, and that's still ongoing work.
And then in some places in Colombia in particular, there were politically opposed to doing randomisation randomisation design, but they were not opposed to a deployment that happened sequentially over time with a random choice of the deployment schedule, which leads to extending the results to stepped wedge design. And let me just finish this and then take some questions if there aren't any by giving an update.
And unfortunately, that's where COVID 19 appears. This study was designed to end at the end of 2020, probably around November. There's still, as I said, a significant amount of dengue transmission going on in Jakarta at the moment. However, unfortunately, this study had to be stopped eight months early on March 18 because the triage of patients in the clinic with fever was modified by the appearance of COVID 19.
There wasn't sufficient protective equipment to protect our research staff that were working to recruit patients with fever and to service the mosquito traps, and the diagnostic labs had to be seconded to support the COVID 19 testing effort going on in Indonesia. So the data have now been locked, and I think I'm actually expecting the data to arrive in the next week or so.
So I can't tell you the results just yet, but I'll stop there, and I hope I find something to at least attract your attention not only as a great public health. The. Prevention, but also some very interesting statistical questions surrounding cluster randomised trials in which trials and I'll stop there. Thank you so much for your attention. Thank you very much, Nick.
That was really interesting. It covers the lessons of, I mean, I've heard a lot of things before, but having not thought about it for a living, oh, we're getting lots of raised hands. OK, so since we don't have it in chat, I will just pick people, unmute them. So Emmanuel, I'll unmute you. And you can ask your question. OK, Manuel, you want to ask? Yeah, that's a cloud. Chris, though not OK. Oh, sorry. This is my first time, my first time hosting, so I didn't realise that.
All right. Sorry, I actually didn't see any raised tens. OK, so one of the things I was thinking about was about blinding, because if you is not blinded, right, still that's. The obviously, the participants in the experiment were not blinded people in principle, you could determine which region you lived in.
In fact, when I visited Yogyakarta during the trial to visit the clinics and observe, I think it's important for statisticians to actually go to the place where data is being collected to understand the data. I deliberately found out where my hotel was and whether I was in a black region or not. I wasn't there as it happened, but so the participants were not.
In fact, it was a great attempt to bring the community in to this experiment and engage them, and there was a big public drawing of ping pong balls like the World Cup does. A draw was made on which region, which there. So it was certainly I'm yes, so that they knew it was fair. It was not. I'm blinded. However, all the clinic staff and the testing staff were blinded to where these samples were coming from in the region,
so there was some attempt to do blinding. But clearly this was not a situation where blinding could be inferred, and that is interesting and using the test negative to see if there's any any impact of that, that will be a check to see that somehow the unblinding of participants didn't change health care seeking behaviour. So in different regions, we should be able to check that as I mentioned, the and will you have demographics about those people so that you can look to see?
I mean, there might be a certain level of. Education or awareness that would affect whether or not it had even occurred to them to affect their decision to go into health care, you would. There's a fairly significant amount of demographic data available on the participant, not a huge amount, but their socioeconomic status, their exact location, their movements and a bunch of their sex, of course, and age. And I know a lot of standard demographic information.
OK, so one of the other questions was how confident can you be that it doesn't affect other viruses? Well, you certainly have to check that and that has been done in the lab to the largest extent possible that I mentioned very briefly that with the appearance of Zika, there isn't a great deal of Zika in Indonesia, but there is a little historically.
And so the samples are all being tested for the presence of other flaviviruses, and they would not be used as controls, by the way, in separately with providers simultaneously an estimate of the efficacy for other viruses distinct from dengue. But I don't think there will be in sufficient numbers from preliminary look at the data. But yeah, you have to make sure that influenza, for example, is unaffected by Wolbachia.
And as far as we know in the lab, it has no impact on the the ability of influenza to replicate. But that's key. Absolutely key, correct, and that's a big issue on trying to use this design for COVID 19. That I am concerned about the people are promoting it because you're testing negative. There are usually other respiratory infections and any intervention that we might have that is of interest for COVID 19 is likely to have similar effects on other respiratory infections.
So that night. Makes it much less useful. OK. Are there any other questions? But is it just one of the other things that I wondered if you're going to do is the. A new wave already planned for this. The as you get near the edges of the areas. Is there going to be any sort of sort of excluding people within a certain distance of the boundary with those as being?
Likely to you're in effect, yes. So what we have observed, as I said, throughout every region, mosquitoes were trapped, so we know exactly the Wolbachia prevalence by the man in every region and much more frightened, granular detail than just which region. So there were many traps for each region. And actually so some drift finally was observed in the late part of last year. Some of the control areas started on the edges to show 20 percent of prevalence, so they started to be partially.
Treated, if you will. And that data is available, so that will go into a measure of Wolbachia exposure. That's not just a binary, yes or no, where was an original or not, but a continuous measure and that can be done at the time level. So we know when you know the prevalence of Wolbachia in the region at the time you presented at the clinic. So that will be used.
I mean, an extreme version, as you say, would be just to eliminate those positives and negatives that are obtained after after contamination. Not not many of the, I would say three or four of the 12 control regions started to show contamination in the last few months. So that will be an issue that will be looked at in the analysis, for sure. Well, the intention to treat will probably dominate because that's what we said.
But we did pre disclose that we would look at a specifically constructed Wolbachia exposure index that would account for that contamination. And is there a threshold beyond which could you sort of show that as well back is sort of put into the area. It grit its prevalence, gradually increase and then you expect it to go to 100 in the absence of everything else. There's some threshold isn't there below which that won't happen.
There would be a threat as it happens, we know all 12 treated areas, well above 90 percent, OK, 95, 98 percent prevalence because this was being measured throughout the entire study. So the and in fact the the level in the controls was close to zero percent for at least two plus years. So that was monitored throughout. That worked really well. So we know the experiment worked really well that way. It is true that let me see what was your question then.
Well, I think that the threshold for the threshold? Yeah, yeah. Yeah, there is a threshold. And and that's why clinic ascertainment wasn't used for a month or so after deployment until until the thresholds went above, I think it was above 80 percent. You weren't allowed to recruit patients, but once above that, we started recruitment in the in the intervention areas. And so that was you use what's interesting from what your previous question is about it.
In the areas now where there's been some contamination, it may allow a sense of which what threshold is needed for protection against and you would if you were living in an area where only 50 percent of the mosquitoes had Wolbachia, would that be sufficient to confer protection through a sort of herd immunity kind of thing? Or would you need it to be higher? No one knows that at the moment, we've clearly designed this to be a 100 percent deployment or zero percent.
But we do now have this in between for the last few months for a few of the control areas. So it it would be interesting. It's unlikely we'll get really detailed data on that. It's only happened in the very last few months of data collection, but it's possible we may see if we see some drift in a controlled area of increasing or decreasing dengue as as contamination increased, that would suggest we could detect where that threshold might be. Would you have been it?
I can understand why they didn't include the first cases, but you can imagine if you didn't put an 80 percent threshold to start accumulating data, you could have looked at the cases early on to get some idea of where it was going to you.
Yes, and there may be data like that I can't recall if we if they were piloting the clinic recruitment and so that there was pilot data before the definitive light was flicked on for actual data collection, which clearly have that kind of information in the pilot studies of the before and after, which is a very powerful. We don't have that here, of course, because we did the deployment all at once.
That's a little bit of the interest in step wedge design, where you not only have cross region comparison, but then meet the comparisons of before and after deployment, so that twice that wedge is sort of interesting in its own way. But it raises additional statistical complications because of temporal trends in dengue infections over time that you have to worry about with step, which decides that we don't have to worry about here because they were affecting all areas simultaneously.
OK, great. Are there any further questions? I don't see anything else in the chat. You're OK with it. I think we've gotten just over an hour, so thank you very much, Nick, for making time to do this and doing it online since we couldn't have you in all sorts of. A sometimes when you're visiting, we have a dinner that you were promised initially. Yes, I look forward to getting back to London and getting back to Oxford.
So thank you very much, everybody. Thank you. Thanks for your attention, right? And I will see you next week, Crystal. Yes, indeed. Virtually. All right. Bye bye.