April 10, 2020
(On April 3rd, BHV members Kim Bottomly and Wayne Villemez issued a selection of readings on COVID-19, click here.)
A Selection of COVID-19 Readings #2
Based on questions and comments, we have extended our readings to include discussions of vaccines, of projections and estimates, and of future planning for pandemics.
Kim Bottomly and Wayne Villemez
6. Slowly but Surely: The Vaccine Solution to the Pandemic Problem
Comments (HKB): The development of a SARS-CoV-2-specific vaccine is absolutely essential. Life will not go back to normal (working, playing, socializing) until a vaccine is widely available. It is not a quick solution, but it is the ultimate solution.
Yet the most optimistic view of the timetable for vaccine development is 12-18 months. Thus, a vaccine won’t be developed fast enough to protect people from the levels of infection and death seen in the current manifestation of the pandemic, and perhaps won’t be soon enough to protect uninfected people from a second wave of infection.
Even knowing this, vaccine development is key—it is the only solution other than waiting for multiple peaks and troughs (infection and death) until we have a naturally immunized population (herd immunity).
The current recommendations today to actively reduce viral spread (isolate the sick, reduce social contact, close schools, etc.) and to identify promising drugs to prevent serious symptoms and death of infected people simply buy us time—time to develop and test promising vaccines.
Slow as it is, this approach is the only alternative to the development of natural immunity, in which the exposure and spread of the COVID-19 virus goes unregulated, ultimately producing a large percentage of survivors who are immune. The human cost of that approach is unacceptable.
All of this is to say that conquering a new pandemic, where the starting point is no one having neutralizing COVID-19-specific antibodies to protect them when we need 50-60% having protective immunity, requires modifying human behavior to avoid viral transmission, the development of drugs to reduce deaths, and simply holding off the virus until a new vaccine is developed and distributed (it has the added large benefit of better distributing the demand for hospitals and healthcare workers).
There is some good news with this particular pandemic, as seen in the readings below. Fortunately, vaccine development is progressing at blinding speed. That is not sarcasm. The research group I am part of has been developing a vaccine (for something else entirely), and we are in our second decade of research and testing. Our timeline is not unusual for an innovative vaccine to be brought to market. Fortunately, there are starting advantages in the development of a SARS-CoV-2 vaccine. For example, the molecular makeup of the infecting virus was determined early. Using recent innovations in imaging techniques, scientists can now look closely at the structure of the virus, which allows them to gain information about how the virus interacts with the host (us). They can then make predictions about what proteins are important in the virus-host interaction and predict types of molecules that can interfere with the interaction. Also, studies suggest that the coronavirus isn’t mutating quickly, suggesting a vaccine will offer lasting protection. The family of coronaviruses has been studied previously to develop possible vaccines for SARS and MERS. While none of these vaccines were fully developed (past Phase I), this knowledge will be very useful in developing a COVID-19 vaccine.
The articles below describe some of the promising candidates. They also describe the steps and timelines needed to develop a vaccine. As you will see, one of the most time-consuming steps is that of manufacturing a vaccine that can be distributed widely. The manufacturing process is difficult, both technically and organizationally.
Reading: Scientists Rush to Find Coronavirus Cure-but it Still Isn't Fast Enough
Reading: Perspective. Developing Covid-19 Vaccines at Pandemic Speed
Reading: Power Up: The coronavirus vaccine race is on. But we probably won't see one this year.
Reading: With record-setting speed, vaccine makers take their first shots at the new coronavirus
7. Projections of the Epidemic: All Models Are Wrong
Comments (WJV): “All models are wrong, but some models are useful.” This common aphorism from statistics is a 1970s quote from the distinguished statistician George Box; he is expressing well a long-recognized statistical truism. The epidemiological models and projections we are seeing these days are almost certainly wrong, but are nonetheless useful if properly done.
The primary use of projections (and the primary intention of those who make them) is not to nail down a number, but to show what factors increase the number, and what factors decrease it. Epidemiological projections are tools, not hard facts. They are not reports, but prescriptions and guides for action. And they are often specifically designed for a single purpose, and less well suited for other purposes. For example, the two current leading models come from the Imperial College of London, and the IHME group, and the two use different methodologies (Imperial a modified standard SIR model, IHME a more complex curve-fitting model). Christopher Murray, head of the IHME group said, in an interview, “The reason we created our model is to help hospitals plan. How many beds you’ll need, how many ventilators, when the peak is likely coming. The purpose of Imperial’s model is to make people realize government intervention is crucial and what would happen without that.”
It is important to know what is going into the models, as we act on what is coming out of them – some are better than others. But it is also important not to get hung up on the differences among various projections: the best of them are partly guesswork -- intelligent and educated guesswork, but guesswork nonetheless. Much too little is known about the extent of Covid-19 in the population, so a large number of parameters of all the models must be inferred. For an example, deaths from Covid-19 in Italy is a pretty firm number (ignoring the likely undiagnosed deaths), and the number actually testing positive to the disease is as well, so modelers start with those, and “back-estimate” what the initial incidence of disease in Italy probably was. Emphasis on the “probably”, since testing is so lacking (not just from the absence of tests, but from the fact that mild cases aren’t tested, and some victims are asymptomatic). That inferred incidence is then treated as a real number in models projecting incidence and deaths in, say, the U.S. That’s a bit of an oversimplification, but not much of one.
And much is left out of models because of the necessity for simplifying assumptions. For example, some of our early received projections for Boston used a simple SIR model (named after the major population parameters: the Susceptible, the Infected, the Recovered), which assumes, among other things, that every person in the population has an equal chance of encountering every other person in the population. Using Wuhan-derived parameters assumes, e.g., that population density doesn’t alter the pattern of contagion. If it does, then to the extent it does, the projections will be wrong (the density of Wuhan is about 3200/square mile; the density of Boston is almost 14,000/square mile). Using the Wuhan-derived model to estimate all of Massachusetts engenders the opposite problem (the density of our state is only about 900/square mile). The better models adjust as they can for these types of issues, but there is a limit to what they can do, given the data.
Of course, many of the flaws in early models are fixed in later and more complex models, when better data become available. A simple example is the recent projections of needed beds in Massachusetts issued by our governor. The number available and the deficit seems to differ from the IMHE projection for the state, even though the IMHE methodology is used by the state. The state methodology has not been made generally available, but it appears that the state model has updated the number of beds available following emergency steps taken – the state data is a realistic upgrade of the 2018 hospital association survey used by IMHE. IMHE used best-available data; Massachusetts properly updated those data – this is the process by which models are improved over time, using better data as it becomes available.
A better example is the fact that, on April 6, IMHE revised their earlier estimates (of deaths and beds needed) downward, based on new data. Projecting deaths and needs depends on predicting the peak. Last week, the only place that had experienced a peak was Wuhan City, and those data were rightfully employed initially. But this week, seven European regions have peaked in daily deaths – two in Spain and five in Italy – and those data have now been incorporated into the IMHE model.
The early projection models, in general, are the most valuable despite their limitations – the later models, while perhaps more accurate, come too late to guide effective in-time action.
All of these caveats notwithstanding, epidemiological models are extremely useful, when properly interpreted, and if acted upon, can save (and have saved) many lives. The articles below show how and why such necessarily imprecise models are so crucial to successful planning.
Don't Believe the Covid-19 Models: That's not what they're for
Why It's So Freaking Hard to Make a Good COVID-19 Model
How to Model a Pandemic
A Statistician's Guide to Coronavirus Numbers
8. Thinking about the next pandemic
Comments (HKB and WJV): Finally, we recommend a thoughtful and detailed piece about what we might do to better prepare for the next pandemic. It is from the Wall Street Journal, and written by the ex-CEO of the Gates Foundation.
Preparing for the Next Pandemic
[BHV Note: H. Kim Bottomly and Wayne Villemez, members of Beacon Hill Village, are retired academics. Kim (Ph.D, 1975, University of Washington Medical School), taught immunobiology at Yale before becoming the 13th president of Wellesley College. Wayne (Ph.D., 1970, University of Texas), taught sociology at the University of Connecticut and was director of its Center for Population Research.]