Emma Ross
Good morning and thank you for joining us for this week’s Chatham House COVID-19 briefing with Distinguished Fellow David Heymann. Our guest today is Professor Azra Ghani, a Professor of Infectious Disease Epidemiology in Imperial College London. Her research combines the use of mathematical models and statistical methods to understand the transmission dynamics to control of a range of infectious diseases, to explore the impact of interventions, and to help guide policy. Her group in Imperial College has done modelling for the COVID-19 pandemic response in the UK, but also, globally, to understand how the disease is spreading in different contexts, what the best responses might be, and how generalisable those responses might be to other countries. So, we’re really glad to have her with us today.
But before we launch into that, I’ll just cover the housekeeping stuff, again. This briefing is on the record and questions can be submitted during – using the ‘Q&A’ function on Zoom, and upvoted questions are more likely to be selected. So, if you like a question and you want it definitely moved forward, upvote it, and tweeting is absolutely fine.
So, Azra, thank you for being with us today, and welcome.
Professor Azra Ghani
Thank you. Thank you.
Emma Ross
I thought we could start by you taking us through some of the issues around using mathematical models to guide responses to the pandemic, but before we get onto the specifics of modelling, in the context of this pandemic, I was hoping you could start us off by outlining what role modelling plays in responding to epidemics generally, what is modelling for infectious disease control, what does it do for us, what does it not? So, basically, what is it and why do we do it?
Professor Azra Ghani
Sure. So, mathematical modelling has really come to the forefront, I think, in this epidemic, but actually has a very, very long history, going back – right back to some of the early work that was done for malaria, to understand how that particular disease was being transmitted. What it is, is a simplification of our understanding of how a disease is transmitted. So, what we do is we try to look at how it is transmitted onwards, so, for example, from person-to-person, and put that into a very simple set of equations, taking into account, for example, how long somebody might be infectious, how many contacts they might make, and therefore, what that might look like as an infection spreads out in a population.
So, modelling is a simplification, it isn’t reality. We observe reality. The model tries to take our understanding of the epidemiology, formulise that, and put it into a set of mathematics, or increasingly computer code, in order that we can better understand how transmission is occurring, and then also to be able to look at how we best respond, and what the different responses are that one could make and how they might combine together to best reduce transmission.
So, for epidemics, obviously things are highly uncertain early on, so actually, one of the key aspect is not the mathematical modelling per se, but it’s actually trying to get the key parameters that we know determine how fast something spreads and how widely it spreads and then how severe it might be, in terms of its impact on the population. Things like how – the R number, very well-known now, but how many onward infections does one initial infection generate, and that helps us track how fast it’s spreading. Also, the time between each generation of infection is important, so for an epidemic like we’re experiencing now with COVID, it’s very rapid, it’s in a matter of days. Something much longer is – an example would be HIV. So, those aspects are brought in and that’s really the focus early on, and certainly during this pandemic, that was really the focus of our work, how severe is the infection, how rapidly is it transmitting, and trying to understand, at the same time, what the likely interventions could be.
So, once we’ve got some sense of how transmissions occur, we can then use these models to say, well, if we manage to reduce transmission by a certain amount, for example, through isolating cases early on, and reducing their infectious period, what impact will those have? And the types of things we looked at early on for this particular pandemic is things like school closures, or workplace closures, contact tracing of cases and isolation of contacts, and really, the typical epidemiological responses that we might expect to this sorts of pathogen.
Emma Ross
Thank you for that really great introduction. David, I wanted to touch on something around an often cited aphorism in this field, which is that all models are wrong, but some are useful. British Statistician, George Box, who’s credited with that quote, has also said that “Any model is at best a useful fiction and that the practical question is how wrong do they have to be to not be useful? Or is the model good enough for this particular application?” And a bit closer to home, Professor Robert Dingwall, Sociologist and Member of the New and Emerging Respiratory Virus Threats Advisory Group, the NERVTAG, in the UK, which feeds into SAGE, the main government expert advisory group, warns that, to quote him, “Frankly, modelling is not that much more of an upgrade on crystal balls.” So, scientific modelling seems to have a lot of complexity. What would you say is the value and the limitations of using mathematical modelling for shaping epidemic response decision-making, when you have a new virus, where there’s a lack of reliable data, and we hardly know anything about it at the beginning?
Professor David Heymann CBE
I think Azra clearly said that modelling is not the reality, it’s a model. It’s an estimate of what might happen, and this is very useful for public health leaders, for public health people who are trying to plan, because models usually have a best-case scenario and the worst-case scenario. The best-case being what would happen, if you do certain things, and the worst-case, if you don’t do certain things, or if you continue on the current trajectory. So, for that, modelling is very useful and you can sometimes fit in ideas of how you think you could control the outbreak into a model, and that can tell you whether or not that’s a feasible way forward. So, modelling is very important for the public health community.
Where modelling goes bad is when the press gets it, or when the Politicians get it, and treat that maximum worst case scenario as reality and they say, “This is what will happen,” and that causes concern and panic in the general population, it causes Politicians to want to do something to show that they can modify this, and it does a whole series of things that it doesn’t need to do, if it’s kept within the public health community. But it can’t be kept in the public health community because it needs to be peer reviewed, it needs to be published and it needs to be put out so that people can read it. So, interpretation, especially by the press, is what’s very difficult for Modellers and for the public health community in general, because they often take the models as being reality, when actually, they’re only modelling and they only – they depend on, as Azra said, on parameters of what you put into the model, the current understanding, which may change tomorrow and make the model entirely a different way – give the model a different outcome. So, modelling is an important public health tool in the right hands. Modelling is also very dangerous, if it gets into the communication channels in a way that causes undue concern and sometimes panic among the population and the political leaders.
Emma Ross
So, it would seem that a critical element that goes hand-in-hand with modelling is communicating it well. Azra, I wanted to ask you how do you think that has gone in this pandemic and what really should be happening? What is the issue here, why is it so difficult, from a communication standpoint, to interpret the models, when this gets out beyond the public health community?
Professor Azra Ghani
Well, yeah, I think one of the big issues is really trying to communicate uncertainty more generally, it isn’t just an issue of models, but when you’re producing scenarios, and I think that’s essentially what was happening early on, much of the work we were doing and other groups were doing were producing a whole range of scenarios and they all had different outcomes. So, the numbers and the – really that quantitative element can be really fixated on. It’s very easy to fixate on a single number, it makes a good headline and it’s not really reflecting the purpose with which those scenarios are produced. And I think one of the biggest challenges we have in modelling generally, and it’s not just for this epidemic, and more generally, is we often want to produce that worst-case scenario, not because we think there’s any chance of it happening, because we do expect to respond in the public health community, but because we want to be able to measure the need for a response. So, the strength of a response really needs to be balanced by what the threat is. So, that counterfactual example, which is the case of the maybe half a million deaths in the UK, that counterfactual really is just there to give us some sense of how important it is to respond and what those responses should be.
So, I think those comparative numbers early on are misinterpreted. They’re taken as predictions and there’s something very different between scenarios that are there just to generate and trying to understand how different interventions might work, compared to a precise prediction, and that is really very challenging, the latter thing to do during an epidemic, when we have so much uncertainty.
Emma Ross
Do you think, to follow-up on that, to some extent, this pandemic is a huge story that moves beyond the Science Journalists? Normally, Science Journalists can handle communicating modelling studies quite well, but a lot of the Reporters, covering this story, are not experienced Science Journalists. Do you think part of why this has gone the way it has has something to do with a different type of Journalists covering the story?
Professor Azra Ghani
I think there’s two elements to this. As you allude to, the Science Journalists are very well briefed and have a really very good understanding of models and that has been borne out of the many, many years when there’s been a lot of past epidemics, past threats, and I think the communication has improved greatly through that. In this particular situation, of course, there are a wider number of Journalists, who may not be as familiar, but I think it also comes back to the uncertainty that this pandemic is generating in the wider population, and it’s a sort of natural human response to want some certainty and, therefore, a number gives some sort of certainty. So, I think it’s not necessarily just the Journalists, but I think possibly, the wider public really would like to know what the epidemic’s going to do in the next four weeks, businesses would like to know. There’s many people who would like that certainty and unfortunately, at the current time, we still don’t really have that certainty.
Emma Ross
And, on that same topic of interpreting the models, does it feel to you, to any extent, that Modellers – modelling is being blamed for some of the mistakes in the response, and is that valid, if it does feel that way?
Professor Azra Ghani
I think there was sometimes a reflection that a decision’s being made just on the basis of a model, and I think that’s entirely erroneous, that’s not what happens in reality. What is actually happening in the way that science is communicated into government is there are really large committees and many, many Scientists, hundreds of Scientists, contributing information. That information is then synthesised and compared, and then many factors then that go in and are generated up to make a decision on some sort of policy. So, I think there’s a feeling that the modelling has been driving things, and that’s actually just not the reality.
Emma Ross
Well, worldwide, maybe this is one for David, could you give us some examples of some of the decisions made in the response to this pandemic, either in the UK or in other countries, that were informed by models and, as Azra says, to what extent they were informed by them or driven by them, given that modelling is a tool for decision-making?
Professor David Heymann CBE
Well, first of all, when a virus or a bacteria emerges from the animal kingdom, we just don’t know its destiny. We’re beginning to understand the destiny of this virus, it’s transmitting in a different way than flu in many ways, but, at the same time, it spreads through the air in coughs, in droplets generated by coughs or sneezing or even speaking, but it doesn’t actually – it does occur still in discrete outbreaks that can be stopped, and those outbreaks can be stopped and prevent transmission further into the community. But early on, and Azra will correct me if I’m wrong, but early on, the Modellers were still thinking about the way that flu transmits and about the way flu causes mortality and also, were basing their modelling on the fact that this looked like it might spread just like influenza. And so, in many ways, flu pandemic planning that had been done, early on, was used to fit into this disease, whereas, actually, continuing a few of the activities that were started at the beginning of outbreaks in many countries, might have also helped in decreasing or suppressing viral transmission.
So, I guess what I’m saying is that early on, Modellers have to make assumptions and in my view, and Azra may disagree with that, but in my view, they use many of the assumptions for influenza early on and they’re no longer assumptions that they’re using today. But, Azra, you might want to correct me on that. over to you.
Professor Azra Ghani
Sure. I mean, I think that there’s a number of aspects that went into the early modelling, but actually, our experience was also shaped by the 2003 SARS epidemic that we also worked on. So, I think, early on, what we were really trying to get a handle on is, does this look influenza-like in its transmission characteristics versus SARS and its transmission characteristics? And actually, I suspect the reality is somewhere in-between these two. So, obviously, one of the key factors that we really wanted to get a handle on is how much transmission is occurring, prior to the onset of symptoms, and, David, as you’ll know very well, much of that was after the onset of symptoms for the original SARS outbreak in 2003, and that really made it containable, in a sense, with the contact tracing and isolation of patients, once they went into hospital. So, I think, early on, we got a sense that it wasn’t quite SARS 2003, it wasn’t going to be as containable, so some of the characteristics are more like flu, but others aren’t, and as we’re learning and as we know more, as things go on, it’s somewhere in-between those two.
So, I think the models were informed by both. The perception maybe that it looked – the models looked more like flu, but actually, they really derived from having considered both pathogens.
Professor David Heymann CBE
Yeah, thanks, yeah.
Emma Ross
I wanted to zero in a bit on this discussion of assumptions, as David said, you have to make assumptions when there’s so little known. But, Azra, can you give us some examples of some of the assumptions that were made in the modelling of this pandemic that had to be adjusted and how that affected the response? I mean, I guess the similarities with flu, there’s another one we’ve heard about, the timing of the lockdown being acceptable or not, calculations of how fast the virus was spreading, what are some of the assumptions that you had to then tweak, as we went on?
Professor Azra Ghani
Yeah, sure. I mean, there’s a number. So, early – sort of mid-February, we had a first set of parameters, I think, that we were fairly comfortable with, or were using in those early scenarios. The R number was one of those. Actually, we’d estimated that from the outbreak in Wuhan, in China, that – we had our early estimates were 2.4, but actually, as infections started to take off in Europe, we had to revise that upwards, infections spreading more rapidly than we’d estimated from the Chinese data. Severity was a critical one and actually, the early estimates we made of severity around a 1% infection fatality ratio seemed to have been borne out, we haven’t actually changed those particularly.
The other transmission parameter I think that’s really key and actually remains incredibly uncertain, is the degree of asymptomatic infection in the population and also, the degree to which those asymptomatic infections generate onward transmission. And the area which remains a really key uncertainty, and we stressed this early on, was in children. We could see, from the early data coming from China, that there were certainly less cases, or symptomatic cases, being recorded in children. What we didn’t know was whether those children were getting infected and how onward infection might happen from children, and that was actually really – is a fundamental understanding that we need, in order to assess whether school closure, for example, is a policy and that will have a big effect.
So, actually, I think that’s one example of an area where we focused very early on and didn’t take what we knew from flu, because we know from flu that closing the schools is a very, very good strong intervention, in order to reduce transmission. I think that much of the early advice we gave, based on those scenarios, was that we really can’t say, at this current time, whether school closures would be useful.
Emma Ross
David, do you have any thoughts on other assumptions, globally, that have had to be, you know, rejigged a bit?
Professor David Heymann CBE
No, I think Azra’s covered it really well. You know, the Modellers take the data that comes in, and it’s coming in regularly. It’s an unprecedented sharing of data during this outbreak, I think Azra you’d agree with that, it’s quite amazing, and journals are actually publishing peer reviewed articles, in front of their pay walls, so that people can have access to this information, and this helps the Modellers in deciding how to regulate their models, or how to readjust them to meet reality. I think that, you know, modelling, as I say, is extremely important for the public health community and I think it has helped, in many ways, show the way forward. I think if there would be one criticism that I would make of modelling, it’s that – it’s not a criticism, it’s just a – maybe it’s a shortcoming of the field epidemiology people, who can’t get the data incorporated in the models in a way that they feel it’s being used. And this is especially true for such things as how do you get community involvement? How do you get contact tracing and the benefit of that into the models, so that they can show what will happen if you do field epidemiology and better testing strategies?
You know, right now, we only have at our disposition, tools in epidemiology and diagnostic tests. We don’t have a vaccine. We don’t have a drug or a therapeutic that can prolong life or cure infection. So, what we need to do is better use the tools that we have, and I think the Modellers often have difficulty in incorporating that into the models, and Azra will correct me if I’m wrong, but I’m going to go back to the Ebola outbreak in West Africa, because at that time the recommendations of the Modellers, which really influenced policy in the development agencies wise, the money is in isolating patients in hospital beds and in making sure that burials are safe. That – if they’d have been able to incorporate in that such things as here’s what would happen if communities became more engaged and would identify contacts and isolate those contacts or put them under surveillance, it might be more effective. So, we ended up, in many instances, with hospital beds created, which weren’t even opened or available, until after the outbreak was over because the communities had become engaged during that period of time and had really been able to help with the outbreak containment activities. But, Azra, you may want to refute that, or come in with some additional information.
Professor Azra Ghani
No, I think that’s a fair criticism, and I think it’s partly to do with the simple ways some of the models are constructed. So, many of the models that are being used worldwide at the moment are these, what we call a compartmental model, where we just track where the people are susceptible to infection or infected in a very simple, simplistic way. And actually, there’s a real challenge with putting something like contact tracing into those models, it’s just very difficult to actually implement mathematically. So, the simulation models that we also use, I think, do capture that. They didn’t have it in initially, we do have contact tracing in the individual based simulation models, and – but I think that’s part of it. It’s sometimes that the Modellers get so, sort of, focused on their own structure of a model that they can’t always easily put in some other interventions. But actually, the first model I ever published was contact tracing, so I did have that about 25 odd years ago, so I absolutely agree that contact tracing is a critical component of public health and we should have that in all models going forwards.
Emma Ross
Thank you. This one looks like it’s for David first, at least, Scientists, as we know, don’t always agree with each other at the best of times, and in circumstances like this, models can produce wildly differing, seemingly incompatible predictions, and with several modelling groups contributing to the response with apparently competing views, how do policymakers know which models to put their confidence in, and how do you manage that issue?
Professor David Heymann CBE
Well, it’s a real difficult issue for policymakers to decide who they’re going to listen to, and that’s why I think there needs to be a balance in the groups that have people with field experience, people with modelling experience, and people with political experience. You know, if there’s been one difficulty in this current pandemic, it’s been the discourse between the public health community and the Politicians, and it’s always been a push and pull. The Politicians are looking for things that they can do, which will get them the credit they need to move ahead and for the next election coming along, whereas, the public health community, including the Modellers, are looking more at what needs to be done and how can we rationally approach that. And so, the difficulty occurs between the public health community and the Politicians, but when the public health community is not unified, as happens sometimes in the current modelling, one group says one thing, another says another, much of that is because there’s rivalry among the modelling groups, which is important, which is necessary, but then that adds to the confusion as to what the political leaders will take and understand. So, many times, they choose a group that they feel is most credible and they stick with that group, as I believe has occurred in this current pandemic, where SAGE includes people from well-known modelling groups. But, Azra, you have some insights better than I maybe in this.
Professor Azra Ghani
Yes, I mean, I think the UK has actually been very well-prepared, in terms of how it uses the information from modelling, compared to many other countries that don’t have that broader experience. So, there is the modelling subgroup of SAGE and that has six to eight different groups all involved, and I think the strength of that is that the Modellers can get together, look at the projections and come up with a consensus and that is actually what is going forward to SAGE. It isn’t a single modelling output that is going forward, but a consensus, among a number of different groups, all of whom make slightly different assumptions, or parameterisations and take a different perspective on that. I think what would benefit is rather than putting just the Modellers in a room, which has tended to happen in public health, the Modellers go off, you go off and discuss amongst yourselves, it would actually be better to have a mix of specialities in that room when you’re coming up with that consensus, and that no doubt happens later, but I think it would be helpful to happen all – that that would happen all the way through the process.
Emma Ross
So, to what extent should decisions be influenced by models, given, you know, the predicted value? What else is there to rely on, if not this system, then what?
Professor Azra Ghani
There’s many dimensions. I mean, I’m sure David can speak to this better than I can, but there’s many dimensions. The models are giving you one angle, in terms of potential scenarios and, in some cases, short-term predictions. But there’s the understanding of what’s happening from the field epidemiology, there’s the understanding of the public health responses, the clinical input, there’s the Behavioural Scientists. I think all of these aspects of science need to come together and they all form part of that, and that gives you the information really, on what the appropriate public health response options should be, at any given point in time, and then, ultimately, that needs to be passed up to the Politicians, who have to make the decisions and balance that public health response against the wider impacts that it might have.
Emma Ross
Now, David, are there some countries that are using modelling to drive their decisions more than other countries, or is everyone pretty much mixing it in with other…?
Professor David Heymann CBE
I expect all countries are using modelling from somewhere, as they make their plans – as they plan, because models, as Azra said, are extremely important, especially that worst case scenario, in being able to plan for the worst, and all countries want to be prepared for the worst. That’s why, in the UK, they were able to construct the Nightingale excess beds because they felt they might need them, under the worst-case scenario. So, it’s very important in planning that you have worst-case scenarios. It’s very important also, that the press treat those as scenarios and not as reality, as we spoke earlier, and then Politicians, at the same time, have to do the same. But no Politician wants to be caught without having prepared for the absolute worst, and so that’s why they’re looking for the information from the Modellers, in order that they can plan for the worst and hope that the best will come along, which it does, in many instances, as it did in the UK when they didn’t need all the excess beds, but they had those beds ready, in case they would need them. Azra.
Emma Ross
Did you want to add anything on that, Azra?
Professor Azra Ghani
No, no, I think I agree, yeah.
Emma Ross
Okay. Well, I wanted to ask you about some of your modelling work that’s looked into differences and the experience between different countries, ‘cause your work has gone beyond what’s going on in the UK. What have been the key insights on that front, particularly the Imperial College work on what’s going on in the US?
Professor Azra Ghani
Yeah, I think the – well, our more general ones, we’re looking actually globally, and I think there’s a real value to looking globally, because we really want to use the modelling to understand what is working and what might be different in a different context, but also, to understand and learn from other countries responses. And one of the key insights, and it’s not surprising, any public health person might understand this, but those countries that responded earlier have done better so far, that’s a generic, but it’s definitely the case, and I think some of that was from experience, and some – certainly the countries in Southeast Asia had recent experience, both with SARS and, in South Korea’s case, with MERS. So, they had that experience to think we need to respond strongly, and we need to respond rapidly.
Other countries, I think, have responded just partly maybe because of what they saw happening in Europe and those early responses made a big difference. Obviously, our modelling in the US, we’ve done a lot of work supporting New York, which has experienced an epidemic very similar to that in London, and that is now under control. What we are seeing, of course, is cases start to rise in many US States now, and that is concerning, and the surveillance obviously needs to be in place and the appropriate public health responses there.
Emma Ross
Have your models been able to show why that’s going on, or not?
Professor Azra Ghani
We haven’t ben focusing particularly on the US, in terms of those surge in cases. Clearly, they have relaxed some interventions and I think it’s quite difficult to disentangle exactly what’s happening in each of the different States, in this case in the US, but even globally, to try and disentangle the different responses that have happened and try and attribute one response as being more effective than the other. And that’s partly because multiple responses are put in at the same time, so you can’t see them because they’re all happening at exactly the same point in time, but also, because the local context matters and the local context is so important to the quality and effectiveness of the response that it’s very difficult to generalise. And a key example I would put in is that we’re looking in lower middle-income countries that have much younger populations, early on, it was suggested that they would be protected, because of the younger age of the population. But we’re also seeing that something like school transmission is going to be potentially important in that younger population and not understanding exactly how transmission is occurring in children is really, sort of, making it difficult to understand what the appropriate responses would be and whether, for example, schools should stay closed in those settings.
Emma Ross
Okay, before I move onto the questions, ‘cause we’re almost halfway through, is really for Azra, what is modelling telling us about what to expect next, as countries start to ease some of the restrictions? What are you predicting?
Professor Azra Ghani
Oh, well, we’re not making predictions, we should stress that, but we are trying to understand the patterns that are emerging. So, many countries went into either national or localised lockdowns and they have emerged from that and we are seeing different patterns emerging, in terms of the spread. So, as people make more contact and start up their lives again, it is inevitable that transmission will increase, but it’s not inevitable that it should go back to the levels it was prior to the lockdowns, and I think that’s because we’re learning. We’re learning how to act as a society, how to make fewer contacts, how to make less risky contacts, but also, the public health system has been able to develop and respond. So, what we are actually seeing is, as when we went into lockdown, there was a very strong correlation between transmission and what was happening, in terms of movement. So, companies such as Apple and Google track this movement through our mobile phones, and we can see that going downwards. As we’re coming out, what we’re seeing is that that movement is increasing, but transmission is not necessarily increasing at exactly the same rate, and I think this reflects our ability to start to work out how to better live with this virus in the medium-term. So, that will, of course, influence any projections and it’s very, very difficult to say, but if people behave in a way that they try to limit onward transmission, just try to limit the contacts a little bit more than would have been the case prior to the pandemic, and also, the public health response is strengthened, there’s good surveillance and good identification and isolation of cases and contact tracing, then it may well be possible to keep transmission low.
Emma Ross
David.
Professor David Heymann CBE
Yeah, I was just going to say that Azra’s talking about the epidemiological approach to dealing with this pandemic rather than just a reactive lockdown of everything, and the countries in Asia that did that selective locking down and then unlocking, in the end, probably have a more sustainable way forward. But, as Azra said also, if the basis of all this is human behaviour and if people can learn themselves to do their own risk assessment and to be able to protect themselves and protect others by simple means, such as physical distancing, wearing a mask if they can’t physically distance from others, and a few other things, including hand washing, we can see this continue – we can live with this, in a very important way. Remember that early on, nobody realised that it was the elderly that really had to be shielded in nursing homes and if we can already shield those elderly, also make those people with co-morbidities aware that they need to pay special attention, we can deal with already more than half the mortality, just by making sure that these people are shielded or understand how to protect themselves. And so, there are a whole series of things to be done, in addition to just this epidemiological approach of locking down when transmission occurs in one area and then unlocking when it’s safe to do that, as they did in the nightclubs in South Korea or in Hong Kong, when they saw it increased transmission in those areas.
So, we all need to participate in this and, if we do, we can see that we can live with this, I’m fairly sure, and I know Azra said the same thing.
Emma Ross
Well, on that hopeful note, I’ll move onto the questions. This is an upvoted question from Milly Zameta, “At the start of the pandemic, the modelling focus was on infection and survival and on severe cases for ICU beds, but now we are seeing a range of different experiences among survivors over time and across severe, moderate and mild cases. What modelling work is being done around the longer-term health impacts of COVID-19 across the spectrum of cases?” Azra.
Professor Azra Ghani
Yeah, no, I think this is a critically important question actually, and it comes back to how we best try to live with this virus going forwards. So, it’s correct, early on we weren’t aware of the longer-term consequences and particularly in a much broader age range. So the focus has been, and I think it’s very important that it is – remains in protecting the elderly and the vulnerable populations. But I do think we need to start taking into account that spectrum of disease and the long-term morbidity that that is generating. That will shift the balance a little bit, in terms of the appropriateness of interventions, because much of that morbidity is happening in much younger population, working age, therefore, potentially back in the community. So, I think we need to take that into account and we don’t, at the current time, have really good estimates to put into the models. As soon as those data become available, we will, of course, do that, and I think we might want to start generating models that don’t just focus on deaths as a primary outcome, but really focus on wider morbidity. There are methods to do this and it’s been very standard across infectious diseases to do so.
Professor David Heymann CBE
Emma, could I ask Azra a question about modelling the deaths and the excess mortality that’s occurred in other populations that didn’t have COVID, but couldn’t access healthcare, are you working on that modelling as well?
Professor Azra Ghani
We are working on it. We have a wide group working on it, particularly in trying to look at excess deaths and also wider economics, because I think all those things come into fore. A single model can’t do all of that, it would just become an overcomplicated and not particularly useful model, but it is important to keep that balance. I think one of the challenges is what data are available. Early on, the data really focused on COVID-related deaths. As the ONS has started to produce those broader statistics, it’s easier to try and calibrate and compare the magnitude of the response to reduce COVID morbidity and mortality against what the negative consequences of, for example, cancelling elective surgery or routine appointments could be. We won’t know, of course, for quite some time, because the backlog of surgeries is still there. It is challenging to increase that rapidly, because of the social distancing that needs to remain in place in hospital settings, so that will have longer-term consequences and we won’t know that for many years.
Professor David Heymann CBE
Thanks.
Emma Ross
Okay, thank you. Here’s a question from Ian Sample at The Guardian, “What does your modelling say about the value of contact tracing here in the UK and does it suggest how we might better contain flare-ups?”
Professor Azra Ghani
Sure. So, contact tracing is really helpful, in terms of the way in which you can identify earlier on potential cases. So, a big focus, and the modelling generally says the most important thing is that those who are – develop symptoms stay at home. That’s critical because we know those people are symptomatic. The contact tracing can take you an extra level to try and find those cases early on. In terms of broad transmission, it’s really quite difficult to say how relatively important that is, and I think there was a lot of excitement early on about the contact tracing apps that the models showed that that would be very successful, if you do it very well, but, of course, if people don’t use those apps and a proportion drop off and a proportion don’t have phones, then you soon lose efficacy. So, I think it’s – it will always – the models will always, I think, reiterate a sense of what we know from public health, that if you do things well, it will be very effective. If you do them partially or not that well, then you – it probably won’t have much of an impact.
Professor David Heymann CBE
Emma, I think there’s also an example in Europe, in Germany, for example, where they started this contact tracing effort very early on in the first outbreak, which occurred from a Chinese importation, and they have built up a group of what they call contact tracing scouts, who are people trained to just go out and do the contact tracing. They’ve found that, despite the app that their government has developed and wanted them to use, it’s much more effective to do it face-to-face with people, as Azra will certainly, because she’s studied this in the past, but face-to-face and locally, where there’s trust in the community that the information will be treated in the right way. So, Germany has had this dichotomy of a government-imposed app, which is – developed app, which has been recommended for use, and the understanding that their tracing scouts are doing a much better job and have taken it to the local level where it needs to be, and they’ve had relative success in keeping their reproductive number low and dealing with discrete outbreaks when they could, thus interrupting chains of transmission into the community.
So, there are examples, as Azra knows, and in Asia, Azra was talking about those examples as well, where contact tracing, continued throughout, has been – you can’t estimate what percentage of transmission might have been decreased, the Modellers will do that eventually, but at least we do know that it has contributed to that, we believe, or at least we believe it has. So, contact tracing can’t be done without having the trust of the communities and it’s best done when that trust of the communities is developed within the community, from face-to-face contact tracing, where there’s no threat to the people that this information will be used otherwise. And Azra began with, I think, sexually transmitted infection contacting, which is also very sensitive, you might say a word about that, Azra.
Professor Azra Ghani
Sure. I mean, I worked on this 25 odd years ago, but actually, it was an interesting research project because I – we were – I was a Mathematical Modeller and I was working with Anthropologists and I think there’s an – that’s an important insight into how important the behavioural aspect is and actually just that there was a big difference, in terms of the number of contacts that were named in an initial clinical appointment, where the Doctor doesn’t necessarily have much time, and just write down the first one or two that are mentioned, and the subsequent follow-up through an in-depth qualitative interview, where you start to think through where you might have made a contact or something. Even something as, sort of, clear as a sexual contact, the numbers of contacts that were generated through more in-depth interviews was much greater. So, I think, if we start to think about contacts that are made in the community for an airborne disease, that’s really quite difficult, if you try to think through how many contacts did I make yesterday? It’s probably easier now that we’re mostly at home, but in the past, it’d be very difficult for you to identify all of those contacts. So, I think it does require that person-to-person interview skills and techniques to be able to really get a good list of relevant contacts as well, because we may, you know, as individuals, just focus on the person we thought infected us rather than necessarily a representative sample. So, I think it’s challenging to do, but it can be done and it – but, as David said, it needs the local people with the local knowledge and the training in the public health field.
Emma Ross
Thank you. Here’s another upvoted question from Charles Clift on, “If the peak of deaths in the UK was on the 8th of April, and the lockdown proper was on the 23rd of March, was the peak of infections not before the lockdown? Does modelling have anything to say about this? Are there implications for the impact of lockdowns on transmission?”
Professor Azra Ghani
Yes, I mean, we’ve looked at this extensively. It all depends on the time between infection and developing symptoms and the duration between those that are hospitalised and how long they remain in hospital. Our analyses of data that are coming through means that those results still are quite consistent. There has been focus, I think, on some of the early estimates that we obtained from China, where the duration between infection, onset of symptoms and death seems to be a little bit longer, but it was highly uncertain, and I should just say that was made on the basis of around 25 or 26 individuals. We now have far more data in the UK and I think the pattern is quite consistent with a rapid drop in the infection following lockdown, as one would expect from the epidemiology of this disease, if we reduce contacts, that should have a major effect.
Emma Ross
Thank you for that. Here is a question from Dina Mufti, another upvoted question, “What is the ratio of beds and ventilators needed to 100,000 of the population to manage the worst case scenario of COVID-19, so as not to overwhelm the healthcare sector?”
Professor Azra Ghani
That’s a very precise number. I don’t think I can give a number off the top of my head.
Emma Ross
Yes, it’s an interesting question, if you can have a shot at it.
Professor Azra Ghani
I think we are – we had fairly good estimates of this early on, to try and manage healthcare demand, and I think that’s the one place where essentially, very simple models we used to project and that – those were not the complicated models that we’re necessarily talking about here. But just doubling times and exponential growth to try and make early projections of hospital bed demand, and I think that was helpful and probably did inform the building of the extra bed capacity. What is happening, of course, is we’re also getting better at treating this infection. So, we’ve got the new therapeutics coming through, but also, clinical practice has learnt to better manage these cases. So, we are actually seeing now a decline in those that require ICU treatment. So that those numbers, of course, need to be propagated forwards and when we think about planning for any future scenario.
Emma Ross
David, anything to add?
Professor David Heymann CBE
No, that’s absolutely right, the more you deal with a disease, the more you understand it and a better job you can do in dealing with the patients and that’s what’s happening in the UK and everywhere, people are understanding how to deal with this. The tragedy though is that it took so many of the elderly people in these nursing homes at the start, which was really a tragedy in all countries, not – except in the Asian countries where, again, they were right on this from the start. You know, as early as the 22nd of January, they were dealing with outbreaks, they had identified them, and stopping outbreaks in Asian countries, and our countries were just waiting to see what might happen and, when it happened, they then reacted in a different way than Asia, which had the systems and was already looking at those because they knew what could happen, from having SARS or MERS previously, and they were ready and they – when they saw these cases, even before WHO called an emergency, they were out there, identifying new cases, case contact tracing and stopping outbreaks.
Emma Ross
Can I just follow-up on that, when you say that other countries were sitting waiting to see what might happen, what should they have been doing instead, when Dr Tedros was saying, “Get ready, this is your window to get ready,” what did that really mean? What should they have been doing instead of waiting to see, or what preparations should they have been making?
Professor David Heymann CBE
Well, you know, what’s really happened in this outbreak, which is quite difficult for many people to understand, is that the public health agencies, which have really been – usually been charged with doing the preparatory activities and strengthening the surveillance, in doing what was necessary, weren’t really in the forefront. This was, for many reasons I’m sure, some political reasons, other reasons were, as Azra said earlier, testing was very limited at the start, in many countries, and they didn’t take that window of opportunity to increase their ability to do testing and to identify people who were infected. And so, by not taking the warnings that were already there and people were preparing for in Asia, our countries didn’t really do what was necessary. And I think the public health agencies and institutions were doing that, but somehow, it just didn’t get translated into a response in the right way. But, Azra, you’ve been involved in some of the discussions more so than I have in the UK.
Professor Azra Ghani
Yeah, I think that’s right. I think there was an element of thinking it may not happen, and certainly looking at Asia and thinking of Asia as a long way away, I think that inevitably happens. There was a very strong focus on travel associated cases early on and I think, in retrospect, that was probably an error because infection was likely to have been circulating much earlier than we appreciated. So, I think what was really needed was not just that focus on travel histories, but also, a very strong, wider surveillance system and for many reasons those data, particularly on the level of community transmission, didn’t come through early, but also just identifying cases of – in hospital settings earlier would have been helpful. I think the genetics now shows that we were – a lot of the infection came from Europe, in around the middle of February, and that’s due to, of course, circulating transmission, it would have gone both ways, but I think that gives us a better indication of where our focus should have been and maybe our surveillance systems weren’t really in place at that time and to pick up quite where we were.
Emma Ross
So, they weren’t looking for it enough, is that really what you’re saying?
Professor Azra Ghani
Yeah, I think the focus maybe stayed too long on looking at travel associations, particularly with China, rather than – and the switch only happened to Italy really in the middle of February, and by just focusing only on those risk factors, we didn’t recognise the wider problem that was already occurring in – across the population, in people who didn’t have those histories.
Professor David Heymann CBE
But for the UK, on the other hand, has one of the strongest influenza-like illness surveillance systems, which is constantly looking for flu and identifying flu, and they just didn’t have the testing capacity to link up to testing with that excellent surveillance system that would have identified these cases, possibly earlier on. So, there were a whole series of things that happened and, you know, to blame any one of those would be wrong, to blame a country is wrong. Everybody was working to try to do the best they could with what they had, and we’ll see in the end that every country had some positive effects and every country probably had some negative effects. But, in all, I think everybody working together is the way forward, not trying to rip apart what’s going on by saying this is wrong and trying to make changes in mid-course. You continue to do this in a way, working together, as the Modellers have encouraged, I know, all throughout.
Emma Ross
Okay. There’s – we’ve only got time for a few more questions. This is another upvoted question from Michel Reimers, “The New York Times reported that the average age of those testing positive in Florida, over the last few weeks, were 35-years-old, compared to an average of over 60 in March and April. What do you see as the consequences of that?” So, David, do you want to go first on what to make of that?
Professor David Heymann CBE
Well, first of all, it would be nice to know who’s being tested, what the definition of a person is who’s being tested. Did they change their definition since the time before? Have they changed it over time? What are they seeing? And the second thing is that, you know, I can’t answer more than that, I can only say that, interpreted properly, you need to know what the testing strategy is. If it’s remained the same and younger people are now being identified and they weren’t identified previously, then that means the transmission is greater among the younger people than it was in the past, provided they were being tested in the past. If they weren’t, then there’s a void in understanding exactly what the level might have been of transmission in that population. So, I think there’s a lot of information missing and that’s the problem when you read something in a newspaper, you assume it’s reality, when you don’t really stop to think about all the different things that go into that testing strategy and what might have altered the way the results have been coming out recently. Azra.
Professor Azra Ghani
Yeah, no, I absolutely agree, I would say the same thing. I think it’s been a real challenge to try and interpret global statistics, because it depends on the testing strategy and who is being tested, and we saw early on that has moved. So, a lot of the testing early on, in the UK, for example, was happening in hospitals. Now it’s moved out into the community, that’s exactly the sort of pattern you might expect if you move from the more severe disease being tested in hospitals, into the wider community. We know that this infection does affect all age groups, but we also know that the more severe infections are in elderly individuals. So, I think it – we need that interpretation before making any generalisations.
Emma Ross
Thank you. I’m going to move to a Journalist question now, Prashant Rao from The Atlantic, “There are some question as to the credibility of the data provided by China, i.e. underreporting of deaths and infections in Wuhan. Did that have any tangible impact on the modelling and, by extension, the policy response in Europe, the United States and elsewhere?” I think you can both probably address that.
Professor Azra Ghani
I mean, I can start with the modelling. I mean, I think every country has struggled, first of all, to report data in a crisis situation, and I think the UK hasn’t been any different in that sense. So, I think what was coming out of China, we did, of course, all look at it closely, they were experiencing a big epidemic in the same way as the UK has subsequently. What we found from that, and we try to correct for potential underreporting in the analyses we do, and we – the estimates I think we got from there have stayed quite robust. So, a slightly higher reproduction number in Europe than we found from those early data, but that could be anything to do with the way transmission’s occurring in Europe compared to China. But actually, what was interesting is that the estimates of severity that we got from those early data, which, of course, do rely on reporting of deaths and infections, actually have been quite robust. So, there may well be underreporting, there’s underreporting everywhere, simply because the systems are overwhelmed.
Professor David Heymann CBE
Yeah, and I guess I would just add to that that, you know, initially, very early on, it looked like the case fatality ratio, that is the number, the percentage of people who are dying, who are sick, was quite high in China, and that was because the definition that they were using for illness was one of serious illness. As they learn more about the infection in that it did infect younger people and they may even have remained asymptomatic, they broadened that definition of what a disease was and immediately, the case fatality ratio decreased and it was very confusing to the media as to what was going on. But it was very clear to the Epidemiologists, the Modellers, that what they were doing was changing the definition of disease, based on what they were learning. You can’t criticise a country for putting out a definition of what they think is the disease, having a high mortality, and then having it decrease because they’ve broadened their definition. This is the way a new disease is identified and discovered, and so I would, in a certain way, defend China for having shared what they did share and not look for what they didn’t share, because we can look, as Azra said, in every country, and there’s things that aren’t shared, just because they aren’t known. But I think China’s done quite a good job in sharing the information that they have, after they once understood that they needed to be sharing this globally because the disease was spreading around the world.
So, you know, many countries have benefitted from what China has provided. We’ve all benefitted by learning rapidly what was going on with this outbreak.
Emma Ross
Thank you. I think we’re onto the last question now for a quick answer. There’s an upvoted question, Damien Chunilal, “The Imperial model’s prediction of excess deaths was very influential in driving the UK lockdown. Has this model been independently verified and run in no lockdown countries, such as Sweden, Hong Kong and Japan? How have its predictions compared to the empirical reality? If not, how then, can a model be used to influence public policy?”
Professor Azra Ghani
Well, first off, that the Imperial model, we have about five different models in Imperial and I’m sure this is referring to a specific one, and we have many because we want to double check assumptions and everything else, with very simple mathematical calculations through to the more complex modelling, but it – the Imperial model was not the only one. So, as we’ve spoken about earlier, there is a modelling subgroup in the UK and there are multiple models, multiple groups, and multiple Scientists, involved in coming up with the consensus projections for the UK specifically. Yes, the model is being applied to multiple other settings. We are looking at other scenarios, Sweden in particular. It is possible to generate very similar results to what’s being observed, but the important factor is that it – these were never predictions, they were just – they were scenarios that were meant to guide thinking really and a response. So, predictions about lockdown or not lockdown, we didn’t actually – it was more nuanced than that. We were looking at multiple different interventions, how they might combine, and it was really more of a – not a prediction as such, but the basic science that we need to reduce the reproduction number to below one, suppression, as David has already spoken about, in the same way that’s happened in many countries. We don’t think that happened in Sweden, but they have actually reduced transmission because other factors did play in. The population did change, for example, its contact patterns. They didn’t carry on as normal in response. So, things have happened that, if you take those factors into account, you can actually generate, using the same model, epidemics that happened in Sweden, and the UK, and elsewhere.
Emma Ross
Thank you. We’ve run out of time now, so I’m going to wrap up. Thank you both for joining us for a really fascinating and accessible conversation about very complex things. As usual, the full recording of this will be on the Chatham House website this afternoon, and please join David and I the same time next week, when we will be in conversation with a very special guest to mark the Chatham House Centenary, and that will be WHO Director-General, Dr Tedros. So, until then, wishing you all a great week. Thank you very much.