Lecture 8: “Epidemiology”

RICHARD YOUNG: Hello, I’m Richard Young, your host, with Facundo Batista and Lena Afeyan for MIT’s course on COVID-19, SARS-CoV-2, and the Pandemic We’ve now heard experts discuss the pandemic, the coronavirus, and our immune response to it And today we have the pleasure of hearing from one of the world’s experts in epidemiology Michael Mina is a physician scientist and assistant professor at the Department of Epidemiology and Immunology and infectious Diseases at the Harvard School of Public Health He’s also a pathologist at Brigham and Women’s Hospital and Harvard Medical School, where he helps to oversee molecular virology He’s a recent recipient of the National Institutes of Health Director’s Early Independence Award And his public health efforts have been highlighted by The Economist magazine, where he was named one of the eight global progress makers He’s received numerous national and international recognitions for his research on childhood infectious diseases and vaccines Michael has been on the front lines of the COVID pandemic since it began, helping to advise state, national, and international institutions and governments And he has been instrumental in setting up high throughput COVID-19 testing programs at the Broad Institute at MIT and Harvard, where he is also an Associate Member His research focuses on monitoring and tracking outbreaks of novel and seasonal viruses through the global population His laboratory at the Harvard School of Public Health develops new technologies and statistical algorithms that measure the immune system response in fine detail, and leverages immunological data to detect and forecast pathogen movement His clinical work at the Brigham and Women’s Hospital is as Medical Director overseeing molecular virology diagnostics Most recently, Michael is working with colleagues at the Ragon Institute, the MIT Sloan School, and Princeton University, to develop a global immunological observatory– a not for profit network of high throughput laboratories to monitor emerging and existing pathogens in the US and globally The objective is to create a virus monitoring system that matches the weather forecasting systems in use globally today Michael, thank you for joining us MICHAEL MINA: Thank you for the invitation And I’m very happy to be here, and to get to speak virtually to this class and to other people who are watching So I’ll just get my slides up So today, I want to talk about testing strategies and the epidemiology of this virus And to suggest new approaches to testing, and how we can potentially use testing not just to let people know if they’re sick, not just to let testing serve to allow us to contact trace individuals, but actually to change the paradigm of what a test can do, and to hopefully use tests almost as a stopgap as we await vaccines for this virus, and hopefully to discuss new approaches that we can employ in the future So I’m going to not just talk about viral testing I want to start with discussing some surveillance efforts and clarify what we mean by surveillance, and what are different approaches, give a little bit of an introduction to some of the research we’ve been doing more recently on using the PCR cycle threshold values– that’s the quantified values that come in the PCR tests for COVID and other uses– and then talk about some of these other screening modalities that exist for this virus But first, I want to talk about this slide And I think that this slide is one of the most important that we should all be paying attention to This is discussing the seasonality of this virus And it was very easy during the summer, for example, to say that cases were persisting in the summer in North America and in the northern hemisphere And that ultimately led to, I think, confusion where people started thinking– A, that our ability to control this virus was readily in our hands That we understood how to control it And we generally did a decent job in New England, for example, in March and April to get it under control by mid-May But we also had the help, I believe, of seasonality It was on our side We then persisted to see transmission continue throughout the summer in much of this country

And my fear is that that was taken too as an example of this not being a particularly seasonal virus To make that assumption is very dangerous– a dangerous idea And I think what we’re seeing now is that despite the fact that the virus continued to transmit throughout the summer, that does not mean, and should not mean, that it is not a seasonal virus I’ve been concerned for quite a while, and I think others of my colleagues also have been, that this is more likely than not to spike during the winter, and I am very worried that that’s essentially what we’re seeing now And what you can see here is the average of 10 years on the right side of the seasonality of each of these different strains of coronaviruses There are the nearest neighbors to this particular coronavirus And what we see is that they all go up Whether it’s in October and November, or January and February– they are seasonal And I think we have to be prepared for this to potentially get much worse before it gets better, and we’re unfortunately now seeing massive increases in much of this country, which I think go beyond just behavioral But it’s really a perfect storm of more lackadaisical behavior, and the seasonality of this virus So there’s a couple of different modalities I’m sure most people, at this point, have heard about serology or antibody testing as well as viral testing The swab that you get shoved up your nose if you’ve gotten a viral test That is looking for the virus genome for the most part It could also be looking for the antigen– the proteins of the virus And that’s to really understand who’s infected now and how to act right now That signal, the positive status of the virus, is usually transient It will stay for one or two or maybe three weeks while somebody is positive potentially And in some people it will continue persisting for longer, but likely without live virus And these types of tests can be very, very sensitive for infected individuals On the other hand, antibody tests– there’s been a lot of confusion on how we could potentially use antibody testing Should it be used as a diagnostic clinical tool, for example All these different questions have abounded about it But in general, an antibody test turns positive on an individual two weeks or so after they’ve been infected, and so it’s not the best diagnostic tool, but what it is, it’s a very good tool for public health monitoring, because it can stay positive for weeks, months, or potentially years, especially after repeat exposures And so it becomes very sensitive as a useful tool for population transmission, and understanding where the virus is in a community So I look at these two as having different uses I think that there’s been a lot of confusion about when to use an antibody test, when to use a viral test, and things have migrated to almost exclusively virology testing But that was unfortunately, I think, because of this confusion And so serology can be very, very powerful for surveillance And in fact, we’ve been working on different efforts to try to understand better how we can actually use serology in high throughput to start to develop new efforts to monitor viruses Not just for coronavirus, though we do have a pilot project going on to do that, but really with the idea of looking long term What can the world do today to start preparing us to make sure that this doesn’t happen again? Our most up to date estimates suggest that before we ever detected a single case in New York City that this virus had already infected tens of thousands of people Had there been a way for the governor of New York and the mayor of New York City to be able to have the data to confidently shut things down earlier, it could have saved tens of thousands of lives And so we should be all aiming for prevention and public health, and part of that is surveillance in the long term And so our effort is really to start thinking through how to build something just like the weather system, but for viruses And instead of satellites and other aspects, we could have drops of blood All of us, every human, for the most part, who walks around Earth has an extraordinarily well evolved sensor for pathogens If we can tap into that sensor to know what people are getting exposed to and when, all of a sudden we have an extraordinary bank of information that we can start tying into to start understanding the rules of transmission of this virus, understanding how it transmits throughout populations, and knowing more about its seasonality effects, so that we can actually not go into every flu season, for example, blindfolded and with our hands tied behind our back, but we can actually act on it in a way that is potentially more

appropriate than our current approaches, for example, with flu and other viruses And we can use it to detect earlier new infectious diseases that are arising So we have a pilot study going on with the company Octapharma at the moment This is a plasma donation company And every week we get tens of thousands of samples into our laboratory And we’re processing them as fast as we can to look for COVID antibodies, to reconstruct– from January, we’re getting the samples all the way back from January, up through this full year– to reconstruct the introduction of this virus into the United States, and then monitor prospectively for what this virus is doing in the United States So I think that we can use this as a stepping point, but really this is the type of approach that, especially at MIT, that we can start creating I would like to create a whole new field of engineering that I call public health engineering, for example And I think that these are the types of things we need to be focusing on in this world to make sure that what we’re seeing today doesn’t happen again Back to being on the ground Serology and virology can both be used for surveillance But I think that what has been lost is just how powerful serology can be And this is a depiction of why If you take these two graphs– these are both showing the same individuals Just bear with me Pretend like the little boxes on the left are individual people Each column being a week in time, and each row being a person And what you can see is– what I’m showing is during an outbreak, if you’re doing sparse surveillance or testing of individuals, you can look for people’s viral proteins or the nucleic acids But they might only stay positive for a short amount of time So if your testing is infrequent or sparse, then you run the risk of just missing that an outbreak is emerging You could take a cross section of people and sample them, which I’m showing with the vertical line, and you run the risk of just totally missing altogether that there are some infected people On the other hand, because antibodies persist for so much longer than the virus signal, if you take that same cross section of people– yes, the signal comes up a week or two after the virus signal does, but it persists And that means that it can become a very powerful tool for surveillance in the long run, during what I call peace time surveillance If you need a tool that’s going to allow you to most effectively identify when new cases are beginning to emerge– for example, in a community– antibody testing can be very powerful And this is an example of that These are agent-based models This is with some work with Nolan Tombe and Bryan Wilder– at the engineering school at Harvard and [INAUDIBLE] And what we’ve been building out agent-based models to use to look at how quickly would an outbreak that starts to emerge become detectable if you’re testing different fractions of a population at different frequencies, and you’re using virus testing or antibody testing Of course, most people say, well antibody testing will always be slower because the antibodies come up later But if you’re missing whole infections, then all of a sudden there are situations when antibody testing can become more powerful And so for example, if we’re testing 10% of the community, or even 20% of a community, on average every month, then if we start to use serology, we actually have a better chance at detecting an outbreak in fewer days than virology necessarily would It’s not much faster but essentially this suggests that we could actually do much more with serology And in fact, because antibody testing is so much cheaper, say $0.50, you can actually scale it up to much larger numbers as well, which would add extra benefits As you become more frequent in your testing approach, then virology will ultimately be the faster approach just because biologically the signal turns up quicker But for the most part, we’re not doing, outside of the wealthy institutions like MIT and Harvard, most of the world is not testing 100% of their community, or close to that, on a near daily basis And most of it actually looks more like the middle column here, or the left column– monthly of testing of a small fraction of people So serology can be very, very powerful But these two can also work well together They don’t have to be one versus the other If we look at– this came from some work we were doing back in April to monitor and test the nursing homes And our governor put in place a plan in Massachusetts to test all of the nursing homes What we found, though, was when we go into a nursing home– if you find that 5%, for example, of individuals were positive for virus, it left us scratching our heads a bit

and saying, well, we’re not quite sure what to make of it in that particular institution Are the cases about to explode upwards? Or are they on the downward trajectory? Has the outbreak already passed? And so without having any additional information besides the virus, we’re kind of stuck But if we add antibodies into the mix– because the antibodies signal reflects a previous infection– if for example we had 55% of that same community already having had antibody formation against this virus and we found a 5% viral PCR positive rate, we would know that the epidemic is on its way down, or the outbreak is on its way down in that community If on the other hand, we found a very limited number of people who are serologically or antibody positive, we would expect instead that the outbreak is on its way up And we’d probably want to pool the resources into that particular community to prevent the outbreak from continuing upwards So these two can really be used powerfully in sync So there’s this other aspect of PCR testing The PCR test is not just a binary result Most people only ever see a binary result But it’s called a quantitative PCR test for the most part And that’s what’s generally being used across the globe What that implies is there is a quantitative value associated As humans we’re very quick to want to binarize everything We want to make everything binary It’s easier to think about It’s easier to understand It’s easier to make decisions about binary results But what if the value isn’t considered binary? Is there more we can do about it? And I’ve talked a lot about this with how we can potentially use the quantitative values for clinical diagnostics and decision making in the hospital But this is a slide to suggest how we can use this very powerfully as a public health tool The real question that we wanted to answer was, how can we know the epidemic trajectory of a outbreak or an epidemic? Is it going up, or is it going down? And we’ve seen, for example in the media, a firestorm of discussions surrounding testing And our President famously has said multiple times that cases aren’t going up It’s just that we’re doing more testing That is a concern How do we cut through the bias that’s associated with increased testing to really understand what the underlying dynamics of the epidemic are? Well it turns out we can actually take the quantitative values And take our understanding of exponential growth and decay at a population level, and start to infer which direction the epidemic is going The reason is, during the upswing of an epidemic when growth of cases is occurring exponentially, the average duration that somebody has been infected on the upswing is short Most people on day five were infected on day four, three or two And that’s just a feature of exponential growth But on the other hand, when the epidemic is decaying, most people who remain positive have been infected potentially for two or three weeks And we can bank on the idea that there is an asymmetry in the viral RNA detection What I’m showing in this middle column here– which I’ll show again, but it’s so small here– is that the virus grows to very high titers very early in infection RNA loads will get to– sometimes to a trillion viral RNA copies per milliliter And then they will wane But they’ll wane and get cleared as the virus will get cleared The live virus will disappear after about a week or so But then the RNAs persist for a very long time Potentially for weeks or months in some people And that’s very similar to DNA sitting around at a crime scene The crime is no longer being committed, but the DNA is still there The virus has done the same thing here So we can bank on this asymmetry of viral loads to actually say, if we know that on– when the epidemic is increasing exponentially, most people have been infected earlier, more recently So that means, on average we should see more viral loads from people’s CT values that are very high viral RNA copies On the other hand, if the virus is decaying, we should be capturing people more in this yellow area When they have largely been exposed maybe two or three weeks ago, and just have lingering RNA And that’s actually what we see These are simulation models that I’m showing right here On the left, I’m showing an epidemic curve in the black line here And then simulated viral titers with all the variants associated with actual viral growth in individuals, and swabbing inconsistencies, and things like that We plug all of that in

And what we can see is that the days since infection as the epidemic does wear on, the median days since infection goes from about not eight or nine to about 15 or 20 over here at the end of an epidemic And that can be reflected then in the distribution of CT values, or the distribution in viral loads And what you can see here on the very bottom of this graph is just how this distribution changes It really gets skewed towards the left there as the epidemic wears on And you can even see that here when we then take the– if we calculate the reproductive rate at a given time, the RT of this virus, really the growth rate– if we plot that against the skewness, or the median value of the CT distributions, we see a very clear relationship down here on the bottom right And if we plot those against each other, we see in the red that the higher growth rate is when you have a skew that’s more towards zero, and a median lower in terms of absolute numbers And then as the virus starts to slow down and turnover and decay, at the population level we start to see that become more skewed negatively And the median starts to increase So this suggests that we really can use the quantitative values as a whole new metric to know what’s happening using just a cross section in time We only need one time point, and we can actually uncover what the epidemic is doing And so this is data from Massachusetts We’ve added a lot since I put this particular slide together This is now in a preprint But what we’re seeing on the top is data from Massachusetts in terms of cases The green line is the reproductive rate at each time point of the virus And then when you can see in B are the CT values And you can see just how skewed they’ve become over time as the epidemic starts to wear down and die out Back in June in Massachusetts the median CT values get into the high 30s Which means the viral loads that we’re detecting are extraordinarily low And you can see again with this real data that the RT value of– that the reproductive rate of the virus was highest when the skew was towards zero and lowest median, and then really increases as the skew gets more and more negative That implies that the epidemic had turned over We could see that here as well Using this, we’ve been building more mathematical maximum likelihood models and Bayesian inference to understand growth rate But we can see that these distributions of CT values really do indicate the growth rate of the virus Above zero is positive, it’s exponentially increasing, and below zero is negative and it’s starting to decay So this is really all to say that the CT values– despite the fact that we are making all of our results binary At the very least, we should be sharing and utilizing the cycle threshold values, the quantitative values, that are in these tests for our benefit, to help us control this virus And this is just for public health observations of the virus, but we could be using it for all sorts of clinical and other uses, which I won’t go into detail right now So now I want to shift gears a little bit and talk about new approaches to clinical virus testing This is an ABI 7500 quantitative PCR instrument I like to show this because this is what the CDC wanted the public health laboratories, and really the country, to use when this virus started to come onto our shores The initial CDC protocol demanded an instrument like this and really a manual qPCR assay The problem there is that what wasn’t really reflected or wasn’t considered very well is that most modern day clinical microbiology laboratories don’t use these types of instruments The education of the technicians is not to run open well format quantitative PCR assays This is not a high throughput PCR instrument by any stretch And so that led to a lot of problems early on where we just weren’t able to scale the testing for this virus To give some idea of what actually the inside of a modern clinical microbiology molecular virology lab is, these two instruments on the bottom left, that Roche– it’s a 6800, that’s huge That’s the size of a car almost And then the Hologic is a little bit smaller But these are high throughput instruments that are very simple to use You take a sample from somebody– essentially you put it into a tube and throw it onto the instrument These can process thousands and thousands of samples a day by one person So they are much higher throughput instruments that require much less molecular know-how in terms of setting up quantitative PCR assays And then up on the top left are also some of the earlier rapid, very cartridge-based tests

that had become available We’ve seen the Cepheid and Abbott ID NOW and then Quidel And we got a whole slew of additional ones now But this is what modern molecular virology is used to And the idea that the CDC was banking on these types of instruments to bail us out of this pandemic and scale testing enough was really, I think, was just the start of all the problems that we’ve had in this country with testing And now what we’re seeing are rapid paper strip tests I call them paper strip tests They’re generally antigen-based tests, though some of them are synthetic biology or CRISPR based tests And what these are– these are essentially just like a pregnancy test You put a sample onto it You can have somebody swab the inside of their nose on their own Put it into a little bit of solution Put that solution onto a card or onto a piece of paper The fluid will flow laterally It’s called the Lateral Flow Assay– LFA And if there is virus there, little antibodies embedded on the paper– essentially it’s paper It’s nitrocellulose and a few different materials can be used But there are little antibodies that capture the virus They’re specific, for, in this case, coronavirus nuclear capsid or spike protein And when they capture the virus, there’s some other secondary reagents that will cause it to turn into a dark line So these are new tests that we will see more and more frequently This Abbott BinaxNOW card, which you see on the right, was just recently approved by the FDA and the White House And the federal government purchased all of their supply through December really– 150 million tests that they’re now distributing throughout the country One of the problems though is that these antigen based tests, these paper strip tests, despite being cheap– you can make them for pennies Sell them for $0.50 or $1 That particular card there is $5 But it’s a whole lot better than $50 PCR or the $25 PCR They can be made much, much more widely And the volume of them can be much higher However, they do have a trade-off Their sensitivity is not quite that of a PCR test And this has been considered problematic But over the next few slides I want to describe how, if we take off our clinical hats– where we think about tests as medical devices to treat people and to diagnose people– and we think about how testing can actually be used to curb epidemics– and we put on our epidemiology hat– we can start to understand these trade-offs, they’re trade-offs that can be made really without any real problem if what it means is more testing becomes available So normally the priority, when we’re thinking about test characteristics, is sensitivity This has been on the front– I never thought test sensitivity would be on the front page of the New York Times, but it has been This is what the FDA has largely put forward as the primary metric to evaluate new tests And everyone– as soon as this epidemic got going, there was essentially a race to get the most sensitive test Unfortunately what was not prioritized was cost And we’ve seen costs exceeding hundreds of dollars for a single test which largely limits the availability of tests for many people in this country and the world We’ve also seen that the speed to get results has been completely ignored Of course it’s been on NPR, and the New York Times, and all these different media channels But it has not been prioritized by the FDA We’ve heard Bill Gates and other people, not to liken myself to Bill Gates, but I’ve said it very broadly, that a test that gives you results after three days should just be garbage Nobody should pay for it It shouldn’t be allowed But we’ve seen test results get get given back after 10 or 15 days during this pandemic So the speed to getting results has been so deprioritized, all in order to prioritize sensitivity in a way that required reprocessing the tests And so what about a test that gives us very fast results– can give you results in minutes, can be done very inexpensively, and can be produced at very high volume, so you could potentially use them daily, but is 1,000 times less sensitive This has been the question A lot of medical clinical microbiologists and others would say, I don’t want, I don’t want a 1,000-fold less sensitive test, that’s not a good test I wouldn’t want to use that on a patient And largely, as a physician, I would say that that’s correct I also wouldn’t necessarily want to use a 1,000 times less sensitive test for my patient I’d rather the PCR that’s done on a $1 million instrument But when we’re dealing with a pandemic,

we can’t always have everything we want And sometimes getting fast results daily can be much better than focusing on the sensitivity And that’s because– we have to come back to this graph here And this is now been borne out in a number of different studies– this type of graph And this is the viral load And what’s really important here is that if you have a low sensitivity and a high sensitivity test, the difference might be 100 times or 1,000 times different, even 10,000 times different, the important piece is that once the virus turns positive on a PCR test, the speed at which the virus is screaming north, in terms of its titers, it’s growing exponentially And it has a pretty short generation time So it’s growing exponentially inside of a person And because of that, it will pass through many orders of magnitude within hours Within a day you’ll see the viral titers go from near the limit of detection of a PCR test, which might be 10 or 100 viral copies, for example, in the assay, to 1,000 or 10,000 within hours And then usually within a day or two, maybe three, you’ll end up seeing peak titers The virus will achieve peak titers of a million or a billion or sometimes a trillion And so it is a very, very fast upswing But then it’s a slower decay And this is very important What I’m showing here is, in the red is when you might have an antigen test be negative, but with a positive PCR test And this is very important A lot of people look at that window and say, we wouldn’t want to miss those people because they’re about to become infectious Well, the important thing is to recognize that the chances of actually finding somebody in that window, through surveillance modes, are pretty minimal Because the time frame is very, very short It’s maybe an hour– I mean, maybe a day, that you’ll be in that window of time But it turns out that most of the false negatives that we’re finding on these lower sensitivity tests, when we find PCR is positive but these antigen paper strip tests are negative, those are happening out here in the yellow section When the RNA is just remaining to get cleared Like the DNA getting cleared from a forensic scene– slowly But the virus is no longer replicating in any dangerous way there, and might not have any truly viable virus left The RNA gets captured in these little double walled vesicles inside of the nasopharynx and the oropharynx, and can just persist for very long periods of time without getting fully cleared But RNA is so easily detectable with PCR that it can continue getting picked up So it turns out that we might not have to worry so much about the true limit of detection of a PCR test if our goal is to find transmitting people Not clinical tests, but public health tools to isolate people most efficiently Identify them, pull them out of the community, and stop spread And so we modeled this also with SIR type compartment models and agent based models And without going into detail about what those models contained, they were essentially communities of people that we can put into mathematical frameworks And we looked at different types of tests Here I’m just showing two tests that are 100-fold different in terms of their sensitivity, the bright pink is reflective a PCR test in the darker columns are reflective of lower sensitivity antigen test if we do this with 1,000-fold or even 10,000-fold difference, we really don’t see too much of a difference from these results What turns out to be the most important– what we’re showing here is how much infectiousness or infectious days are removed from the community given different types of testing relative to not having any testing And on the right is how many total infections happen during the course of an outbreak Turns out that, more than anything else, the frequency of testing is the most important And that’s because if we’re not testing frequently, we’re not going to find people before they become infectious This is a virus that is largely– has some asymptomatic spread or presymptomatic spread initially And then a lot of people will continue to spread even completely asymptomatically So we can’t bank on people going to the doctor every time they get sick That means the only way to really catch people is through frequency If you’re not testing frequently, you’re unlikely to capture them in the first few days So the moment we get out to testing say every two weeks of the population, we start to really have a hard time controlling outbreaks And it really starts to look more like this situation if we had no testing But if we’re doing daily or every three day testing or even weekly testing, we can do a very good job at removing people early before they have a chance to go and infect others And this is really regardless of just how sensitive the assay is– where the limit of detection is

Because again, the limit of detection is sufficiently low still in any of these tests relative to just how quickly the virus grows beyond those thresholds So it will still be detected very early in the course of infection And then it’s not just the frequency of testing that matters It’s also the turnaround time Like I said, we’ve almost placed no priority frankly on how quickly results can get back Some institutions certainly have The Broad Institute has done a tremendous job, for example, at getting results back within 24 hours to most people throughout– really since flipping the switch to turn that system on back in April And they’ve just continued doing tremendous work But many assays and many labs haven’t We’ve seen many, many people, probably some people on this call, have waited five, six, seven, eight days to get a result. And what you can see here is you have the frequency bins, you have the different tests, and the different colors But then zero, one, and two are the different turnaround times in the test– how long does it take from the time you get a swab to get a result The moment you get to a 48-hour turnaround time, you’re really losing your ability to prevent infections from spreading We can’t be doing weekly testing and asking everyone to isolate during those two days while you’re waiting for a test to come back So people are going to keep living their lives But if you’re allowing them to just go and live their lives despite technically having a positive swab waiting to get processed, then that’s two days lost in terms of stopping them from transmitting to others So it really, really becomes crucial to get it within minutes or a 24-hour period at most And so the real power in these types of tests are not necessarily to use them as entry screening devices, meaning using them to get into dinner or using them to go home for Thanksgiving, or using them to get into a class That’s generally how they’re being used to a certain extent Or at least that’s the use that seems most appealing, because it’s very easy to think about And it’s an extraordinarily important use But actually, the real power in this type of testing comes from taking a slightly different more epidemiological approach to the question And that’s, can we actually suppress whole outbreaks through testing? Not through using the test to identify who needs to be contact traced, but actually to use the test in such frequency and in such abundance that it can actually create what we call herd effects And similar to how herd immunity can be elicited with just a fraction of the community vaccinated– so say with herd immunity, if we want to vaccinate 50% of a community we might actually be able to stop overall transmission of an outbreak, for example These tests can do the same thing But instead of using our immune system to stop onward transmission, we use knowledge And this worked with for example, HIV There are huge campaigns to know your status And by knowing your HIV status, people were able to change and modify their behavior appropriately, so that they would not spread to other people with high frequency Same thing here If we allow people and a sufficient number of people to know their status and know if they’re infected, then they can modify their behavior And we’ve seen this for example, in our models here And what I’m showing is even if we had 50% of a community totally flat out refused to use tests, but the other 50% did choose to use them and agreed to use them every three days or every seven days, and let’s say 10% of all those tests failed, we would still be able to take an outbreak that’s raging upwards, and get it under control within four or six or eight weeks And so this is the power of herd effects We don’t need everyone to be participating, and we don’t need the most sensitive test in the world If more people are participating, you can really relax your sensitivity metrics But these results come from using a test that’s 100 times less sensitive or 1,000 times less sensitive– like an antigen test And that’s because frequent testing will define who’s infected early on, and stop enough people so that the effective R value– the average number of new infections that arise from each infection– gets below one And once that value gets below one, the outbreak starts to turn over and die If it’s above one, it increases exponentially Below one it decreases So that’s our goal here And so I think although our priorities have been sensitivity over all else, what we’re seeing is frequency maybe all else should be prioritized And then speed to get results And sensitivity can take a huge backseat in this fight At least when using a test like this for public health use Not for clinical use, but for public health use And I won’t go into too much detail, but I want to show some data that came from Massachusetts in June and July And if you look on the bottom left here– this is real data

Each of those violin plots is a week And so it’s hundreds of samples And these are positive people in June and July And what you can see here is that the average, the average viral load had a CT value about 35 Which, if you’re not used to CT values, that’s a minuscule amount of virus What that suggests to us is that the median person who is detected, probably no longer had transmissible virus left in them The PCR is so sensitive that it’s continuing to detect people late in their course of infection After you’ve already lost your opportunity to necessarily isolate them, and treat them appropriately So unfortunately by focusing so much on low frequency but very high sensitivity testing, we have lost the forest for the trees We’re doing a great job at detecting virus in the samples that come across our desk, but what we’re missing is that the vast majority of people who A, don’t even get detected at all, and those who we do detect we’re detecting out here– like what I’m showing, which has passed the period of transmissibility Once people have very low viral loads, that’s when we’re detecting them And so it suggests that PCR testing while not actually giving false positives, but if your goal is to detect transmissible people, what we find is that during, especially during this summer in Massachusetts, most of the positives we found were probably past the period of transmissibility, and maybe didn’t even have to be isolated, for example And it all points at the direction that we need to be doing higher frequency testing if we want any chance at capturing this virus And so I like to think of rapid tests as two different types I think that there’s the Nespresso machine test, and the instant coffee type test And so far we’ve seen a lot of antigen tests that fit the Nespresso machine model, but these are generally not going to be sufficiently scalable because they have an instrument associated with them What I’d like to see is, for example, for the federal government to just start producing millions of the paper strip tests without any bells and whistles, without reporting necessarily– if we can get these into enough hands in this country, we can actually stop outbreaks We can work with Google and Verizon and AT&T to make reporting possible from the home, for example But we should really be prioritizing just the production, mass production, of these At least until we have vaccines, these can create herd effects that can stop outbreaks We just have to get them to enough people And I want to finish with this slide here, and make it very clear that while we have focused so much on getting very high sensitivity PCR tests, we’ve really missed the boat on this when it comes to public health We’ve been trying to squish and cram a public health tool through a medical lens And what that’s meant is we’ve essentially limited our ability to do a lot of testing by focusing so much on lab based PCRs, and in slowing the regulatory process down by trying to achieve PCR type metrics on a piece of paper And so it’s slowing down the companies that are trying to develop paper strip type tests And I think that this bottom piece is probably the most important thing I’ll say about this And that’s that right now our surveillance systems– we might have tests that can detect extraordinarily low numbers of virus, but at the end of the day, our surveillance system in this country probably is detecting fewer than 5% of infected people in time to actually isolate them appropriately and stop onward spread So that should be the sensitivity metric we’re focusing on We should really start focusing on sensitivity not to detect molecules, but sensitivity to detect infectious people should be our basic metric in public health testing And if we don’t focus on that soon, we’re going to keep just allowing this virus to spread and barely scratching the surface when we do capture people infrequently So I’ll end there, and I’m happy to take any questions RICHARD YOUNG: Michael, thank you very much There are a number of questions that students have for you They want to know, what is it about winter that makes coronaviruses more prevalent? MICHAEL MINA: Yeah it’s a great question, and it’s a question that’s under really intense investigation, and frankly has been for quite a while There’s some nice data that came out that shows, for example– it’s just one piece– in lower absolute humidity, for example, we see mucociliary clearance starts to go down So it could be something– one question is, is it

true transmission that’s changing out in the environment? Is it people’s symptoms and then they’re getting tested more frequently during normal times? Probably not just that Is it that the severity of the disease is actually getting worse in the winter, and so it’s driving people to get tested for flu and things along those lines But here we’re seeing that cases are truly going up And what we’ve also seen with flu and other viruses is that the viruses can change how long they’re viable out in the environment based on some different seasonal features And temperature and absolute humidity have been top of mind, and some of the most well researched But in general, I would say that the jury is still out in terms of really understanding what exactly it that is driving greater transmissibility in the winter But what we can say is that it’s probably some combination of biology, of temperature, of humidity, and of human behavior We can’t discount human behavior as a part of the biological process We can’t all– well I guess we could if we had enough clothes But we’re not going to go out and just live outside in 20 degree weather all day and sleep there We congregate indoors as humans And that’s part of it for sure And so I think all of these essentially make it very difficult to parse out which ones are the true drivers But there’s a number of features RICHARD YOUNG: And students wanted to know how powerful is the viral RNA detection in sewage systems for the types of public monitoring? MICHAEL MINA: Yeah, I think that sewage surveillance, I think, is one of the most exciting areas to discuss And some of it’s really been driven by MIT faculty and labs And this can be very powerful Now the question is, how many cases need to occur before you can detect a case And that really depends on how far down stream you’re looking If you’re out at Deer Island and you’re looking at massive numbers of sewage pipes all centering into the same area, then maybe you need actually more cases But again, this virus grows up to trillions of copies in a given person One person could potentially shed huge amounts, so it could potentially be very, very sensitive even to small numbers of infections The tricky part is if you’re testing downstream in order to really conserve resources and not be testing a lot of individual households for example, you then have to backtrack and figure out where the cases are coming from But I think from a peacetime surveillance mode, for example, they can be very powerful And what I would recommend as one example is institutions that don’t want to just test 365 days a year, they can potentially, if there is no cases hanging about in their community– maybe you can turn off your frequent testing, turn on your sewage testing daily And then if you start to see cases, then turn on your individual level frequent testing with antigen tests for example That would be a dynamic testing approach to use surveillance powerfully to inform when you need to start acting And I think all of these can be very powerfully used together RICHARD YOUNG: Can you speak a little bit about the concerns over false positive rates of antigen tests that– are these a big concern on a population level, or on an individual level? MICHAEL MINA: They’re a big concern on both really So in particular, if we start rolling out antigen test to millions and millions of people, then 1% false positives is a huge number We can’t have one in 100 people in this country turn in a false positive every single day Technically, we probably have the PCR capacity to deal with that at this point But we do not want that to happen And so what I’ve been advocating for widely is that we don’t just start rolling out antigen tests like we saw in Nevada The federal government has been good to purchase all these BinaxNOW now tests, and the Quidels and Sofias and BD Veritors But then they just willy-nilly gave them out to the country and without any rigorous algorithms on how to use them, nor necessarily really discussing in any powerful way how they should best be used And what we’ve seen is confusion has abounded And part of that is because of false positives And in particular, in a nursing home, you don’t want to take somebody who’s negative who gets a false positive and then put them with a bunch of positives That could be a lethal false positive for that individual

And so we have to be very careful, and there’s ways around this We can really drop the number of false positives in the same way that we deal with other infectious diseases HIV is a great example In our hospital, for example, half of all of our positive results on our HIV screening tests are false positives But we don’t say that HIV testing is not useful– it’s because the prevalence is so low, so we just reflex everything to a confirmatory test And so what I think we should be doing is every time we ship out antigen tests, they should come with confirmatory antigen tests that meet the timescale of the initial test You can’t give somebody a five minute test and then tell them that if they are positive they need to wait five days to get confirmed on a PCR That’s just not appropriate, and it’s not going to work And it’s going to cause people to lose confidence So what we need is for the CDC and the FDA to work together, and develop algorithms that are very clear, they’re very concise, and they say, if you’re using this test and you get a positive, and you and your patient is not– the individual doesn’t have any known recent exposures and low predictive value, then you immediately confirm on a sufficiently orthogonal rapid test that is unlikely to turn false positive for the same reason So we need to be matching these up in the same way that we’ve done for many years with other laboratory tests So I think that this is something that has been not taken sufficiently seriously enough And I can say personally, I’ve spoken with the highest up that need to be spoken to about this at the FDA and the CDC, largely and unfortunately to deaf ears, about this particular issue, but where we’re now seeing the consequences of it We’re seeing that Nevada is trying to curb their use of antigen tests, a very powerful tool, because it wasn’t explained well enough about how exactly to use this And right now the only suggestion is if you get a positive, to reflex to some PCR test That isn’t good enough And I think we can do better But I would say that even these antigen tests, some recent data from them, is showing that actually the false positives on a single test are now getting to one in 500, one in 800, so they’re really starting to improve more and more And if we couple them together, we can get that down to one in a few thousand And that starts to get into a very reasonable number of false positives to then have to confirm on a more rapid PCR test RICHARD YOUNG: Can you talk about how increased testing and an understanding of infectiousness — how that’s changed the way you and others have modeled the trajectory of the pandemic? MICHAEL MINA: Yeah I would say that there’s been an interesting phenomenon with this virus And I won’t say that we had a crystal ball going into it, but certainly– there is this prevailing idea that because this is a novel virus, that we had to rewrite every textbook and relearn everything immunologically and transmissibility about this virus That I think has really hindered our approach to do good work, and has led to a lot of confusion So I would say that going into this, we didn’t actually have to learn that much to start building the models about when people would be infectious and transmissible I would say a lot of people did rebuild those models, and try to collect the data And probably made poor inferences based on imperfect knowledge As we’ve learned more, frankly the empirical data is matching more and more to the textbooks, and to what we would have expected going into this with very minimal data at hand And so that’s allowing us to more officially hone in our models, I would say And to understand more and more about how to deal with viruses like this, or this particular virus For example, we’re learning more about super spreading in this virus There was a while there where it was thought that there wasn’t super spreading as a major role of transmission with this, unlike SARS and MERS, and that maybe it was truly a pretty smooth R naught of 1.5 to three or four or something But it turns out that we actually have– that at an individual level, that the R naught is still the same But at an individual level you might have some people go their whole infection not transmit, and then other people will go and transmit to four, 10, or 40 people And that has been– it’s been instrumental to understand just how quickly viral loads increase very, very rapidly, and then get driven down And I think the more we continue to learn about it– we’re probably going to continue honing in on finding that most people are transmissible for an extremely short amount of time– maybe two or three days is when 90% or 95% of the transmission happens And so that will help us continue to learn how to tackle this virus the best

For example, if people aren’t necessarily being cautious when they are positive, maybe we can start to make different policies that say, instead of just going and isolating for 10 days, really isolate for three days And then just do your best to do all the isolation you can for the other seven You know, just simple policy changes that might allow public health to work more with society in a more holistic way, I think, we’ll continue to be able to frame things in more efficient ways RICHARD YOUNG: Well with that note, I want to thank you very much for your lecture today, Michael We really appreciate it MICHAEL MINA: Absolutely Well, very, very happy to be here