Joining us today is Dr. Samir Bhatt to speak about the latest report by the Imperial College COVID-19 response team. This report focused on estimated numbers of infections and impact of non-pharmaceutical interventions in 11 countries in Europe. Welcome. Thanks for having me. What questions did you answer when writing this report? Well, as many know, our department specializes in outbreaks. When we started on this, we wanted to assess what impact and what changes interventions are having on the number of infections. So when I say infections, I'm talking about people who've caught COVID, but not necessarily only the ones that show symptoms. There are a lot of people who potentially are asymptomatic and we don't actually know what this number is. So we wanted to know what were the changes in the number of infections or the number of cases to some degree. So we looked at the data and quickly we realized it was very difficult to disentangle and figure out what the signal would be from looking at the case data. This makes sense. I'm sure that a lot of the audience have been watching Worldometer, particularly during the early stages and watching the ECDC for the changes. Unfortunately, a lot of these cases, they change when a country change in testing policy, and I'm not saying this is a good or bad thing, but the data is profoundly biased on cases. We don't know what's going on just from the case data. It misses people who got ill and might have just stayed home. It misses people who are ill and don't even know it because they aren't showing symptoms. As time goes by, when an epidemic gets more and more in its full swing for a country such as Italy, the case testing data is only showing a vanishingly small amount. So we more sombrely thought, let's look at the death data. Now, the death data is not without its own problems. Of course, quite a lot of you will see a lot in the news that there's a lot of underreporting in Italy, but by and large, it's far, far better than cases. So we thought, well, can we find out the signal of what's going on in the COVID epidemic from the death data? Can we explain what's going on as a result of these interventions? Can we see if they've had no impact? Can we see if they have? All right. How did you answer these questions? So we use a gold standard form of statistical modeling called Bayesian hierarchical modeling. I know it's a fancy term, but basically what it is is a back calculation. We start from deaths, and we say, well, we've observed these deaths reasonably accurately, there is some uncertainty, but not a huge amount. We don't expect deaths to be three orders of magnitude higher than what we see observed, for example. Then, we say, well, these deaths had to occur from some infections, and these infections have a certain infection fatality rate. We then bring in as much information as we can on epidemiological parameters. These, again, sound complicated, but they're not really. It's how long does it take on average from somebody to get from infection to an onset of symptoms, if they get symptoms? Then, how long does that more sombrely result in death, and how long that time goes. We don't treat anything as a single number, we treat everything as a probability representing our uncertainty about all these things. So we back-calculate infections from death and then from infections, we say that infections are driven by a rate of transmission. That's how fast things are happening. If you want an intuition of this, I'm sure many have seen this on the graphs, think of it as a doubling time in its most simplest form, but it's something a little bit more complicated called the reproduction number, which is a rate of transmission. So we wanted to see how this reproduction number was changing over time when various interventions came in. We now have this hierarchical framework that we developed, and we wanted to see, well, what interventions are going to matter. So we had to go out, collect standardized information or intervention data and put those together and try to see what happens when we estimate this latent infection parameter in our model, so to speak. What are the key findings of the report? So the key findings of the report are that it does seem that interventions have had a big impact on the rate of transmission. So we think that interventions have had an effect. For many countries, most of them, in fact, the doubling time or the reproduction number is somewhere close to one, which means it's slowed down quite a bit. Now, that's the good news. The bad news is that there's so much uncertainty in all of this that we don't know exactly how low it is or how high it is. We can just say that it has reduced and it's somewhere around one, but it's most certainly not statistically below one, which means the epidemic would die out and we'd be in a containment phase where we need to prevent repeated epidemics happening. If it's not sufficiently massively above one where we are in an unmitigated epidemic, but if it is above one, we're in this flattening the curve regime and it's a matter of how much we flatten the curve. The epidemiological parameters we have now are based on what we observe in other countries and in unique experiences like the Diamond Princess, which was very unfortunate for the individuals involved in there, but it gave us a lot of epidemiological information, and we take these and we input them in the model. Now, these will get refined as time goes by, but this is the best we know now. We, at Imperial, never just rest on the best we know now, we always update our results to reflect the most current information that we have. So that's caveat number one. Caveat number two is a lot of countries have just implemented interventions today. Now, there's a big lag between deaths and infections. When somebody gets infected, there's some time until they get some symptoms, if they get symptoms, and then there's some time until they end up going through the healthcare system to critical care, to eventual unfortunate death. Because of that lag, just looking at deaths in a country that's implemented interventions a week ago, you might not have enough signal. Yes, some people might have died in that period, but not the bulk that would have died or will die. So we need to leverage information from older epidemics. So a key assumption is that if what we observe as the impact in countries like Spain and Italy is true in other countries, then these are the impacts we'll see. So how did these findings support decision-making in general, and more specifically, how useful are these findings for other countries that were not included in the results of your report? That's excellent question. My personal view is that science should be a multi-faceted endeavor. There are many different lines and strands of evidence that come together that ultimately get reported to those who make decisions, and then they make decisions based on these. This provides one cog suggesting that the reproduction numbers or the rate of transmission has gone down due to these interventions, but we don't know enough decisively to say relax interventions more certainly. There's a huge amount more work looking at Imperial and other really great institutions across the UK, looking at bed occupancy, looking at ICU capacity, conditions on the ground, seeing what's going on, pharmaceutical endeavors trying to search for other interventions, looking at people's behaviors from surveys and seeing how that's working, there's a huge line of other evidence. These things have to be assessed daily based on updated best data and made by those who factor in all the different decisions and evidence out there. Hoping to expand our codebase and start bringing in more different countries. We have people in the team looking at this and being applied in the US setting as well. So I think that we have a lot of interesting follow-ups on this that could be very useful to both the general audience who want to understand what's going on in the country and to researchers and policymakers as one cog of evidence in their decision-making structure. Thank you. What is the next step going forward? What are you focusing your work on now? So currently in the team, we are looking at more European countries: Portugal, the Netherlands, Greece, and a few others. We want to do all of them. It's just we do need sufficient amount of data to actually run this model, but we're watching countries very carefully and working with researchers, if we can, in these countries so that we can make this a very collaborative endeavor, and we're also looking at this in the USA right now. So members of the team are leading this in the USA, and we're trying to work with researchers there as well. Thank you for taking the time to speak with us today. Thanks for having me on.
