by Gabe Thornhill, Austin, Texas
We are all afraid of things we don’t understand and more afraid of new threats than dangers we’ve been exposed to for years. Terrorism is a great example. The U.S. lost over 2,000 people in one attack almost 20 years ago. In the interim, we’ve spent trillions tracking and killing terrorists around the world. At the same time, 40,000 people die every year in auto accidents around the county. In looking at the news on any given day and in analyzing federal spending for two decades, it’s clear that the threat of terrorism–which has in the same time period taken less than one percent of the number of American lives than auto accidents have–is perceived to be the more dangerous. Why? Terrorism is an unknown and outside of our daily experience. Conversely, the dangers from driving are a well understood risk of modern life.
COVID-19 certainly qualifies as new, unknown, and outside our daily experience. According to the CDC, it’s a virus that jumped from bats to humans in November of 2019 in Wuhan, China, a place unknown to much of the world. While the virus initially spread from animals to humans and originated in a market where wild animals are slaughtered for food, evidence of person-to-person spread emerged soon thereafter. The Chinese government seemed to actively hide or downplay the problem in the early days of the outbreak. In fact, several early whistleblowers have even disappeared. Today, information from China seems unreliable. All this to say, the origins of COVID-19 are mysterious and different.
If we step back from our fear of COVID-19, we can think about how we should we feel. We can frame and evaluate. We can ask basic questions: Where does the U.S. and the world stand in terms of in the fight against coronavirus? How many people are infected? What is the death rate? How does it compare to other respiratory diseases? How is the virus trending? AND Where are we getting our data?
Last week, the White House Task Force members estimated the coronavirus could kill 100,000 to 240,000 people in the U.S. this year. For me, saying 100,000 deaths conjures an image of the University of Texas football stadium being nuked on game day. Those are terrifying numbers and more troubling, it’s the same bureaucracy that was telling us not to worry in early March. The message seems to be this is a new danger and we just don’t know how it will play out.
FiveThirtyEight.com, a site known for analyzing statistical information, has reviewed a range of forecasts to ascertain likely death rates and spread of the virus through the population. The forecasts vary widely. The often cited IHME model has been wildly inaccurate in predicting the situation only a week out. FiveThirtyEight.com reviewed the myriad of complexities and assumptions that go into generating and modeling the virus behavior and, in summary, making a reliable model is extraodinarily difficult.
The virus is scary, new and difficult to model. What are we to do?
Given the difficulty of building comprehensive models that can accurately predict COVID-19 behavior, we can use simpler heuristics to understand and evaluate the situation on an absolute and relative basis. Hopefully, it makes the COVID-19 virus a little less exotic and frames its risks more clearly.
How Many U.S. Infections? Johns Hopkins shows 401,166 infections on April 8th. The number is cumulative cases and includes people who have tested positive or are presumed positive since mid-January. But this case total almost certainly understates the actual number by undercounting people who have the virus but are asymptomatic and by the limited nature of the number of tests given. We have data points that we can use to estimate the infected population. The Diamond Princess cruise ship was a closed system where everyone was tested. Roughly ½ of all those who tested positive were asymptomatic. Similarly, Iceland’s deCode testing showed ½ of all cases had no symptoms. According to an article in Live Science, Chinese researchers found their testing probably captured only ½ of the projected cases. U.S. reports have indicated testing is generating almost ½ false negatives. Assuming ½ of people have no symptoms and ½ of tests generate false negatives, we get 4x the number of positive tests or 1,600,000 cases in the U.S. Another way to estimate the total infected population would be to apply Iceland’s testing numbers to the U.S. The country has undertaken a broad testing regime in partnership if deCODE genetics. DeCODE is testing a semi-random portion of the population. Through 4/4, deCODE had tested 18,170 people of the 364,000 Icelandic population. The testing shows 0.78% infection rate. Applying that number to the U.S. generates 2.5 million cumulative cases. We can estimate 1.6-2.5 million U.S. cases without building a complicated model. It’s much larger than the daily reporting, but the larger number of cases also has a dramatic impact on the expected death rate. 
What is the death rate? On April 8th, the Johns Hopkins website shows 12,912 U.S. deaths and 401,166 confirmed cases, implying a 3.2% death rate. That’s a pretty terrifying figure, especially when compared to the flu at 0.1%. However, the number is probably too high by an order of magnitude. We have three distinct ways to estimate a more accurate death rate:
- Using our previous estimate of 1.6–2.5 million cases, we get a death rate of 0.8% to 0.5%.
- In looking that the closed system on the Diamond Princess we know that there were 12 deaths in 712 cases for a rate of 1.7%. However, the average age those that died was roughly 75. Those over 65 are at much higher risk of death from respiratory illness than the regular population. The death rate for flu for those over 65 is 5x higher than the general population. The same is true for influenza hospitalization. The rate for those over 65 is 5x the rate for people between 18 and 49. Applying the 5x multiple to the Diamond Princess death rates generates a death rate of 0.3% for the general population.
- Iceland has had 6 deaths in 1,616 cases for a 0.4% death rate.
These three distinct methods generate a death rate of 0.3–0.8%. That’s more deadly than the flu, but much less deadly as the implied rate of 3.2%. As we learn more about the virus and generate new treatments, the death rate will likely decline. Additionally, with wide scale antibody testing, we’ll get a better idea of actual infection rates.
How is COVID-19 trending and where are we on the epidemic curve? Fear sells, and the consistent media message of danger and death and overwhelmed hospitals systems, subjects us all to framing biases and recency bias. The Johns Hopkins website and its map of red dots implies that the world will be overrun by the red dots and the death toll. We are led to believe the trend is inevitable. There are, however, other ways to frame the virus and the data that provide a distinct view of the trend. Let’s look at three.
The first is to consider COVID-19 a seasonal respiratory virus and likely similar to others in the past. The chart below shows the seasonality of different upper respiratory illnesses. COVID-19 isn’t the first coronavirus. Others viruses in the coronavirus family typically run from December to April. Assuming this novel coronavirus acts like the others in its family, it will tail off this month and into the summer. It may not as it is new. But in the absence of evidence to the contrary, it is distinctly possible that it will. 
A second way to frame the virus is to look at death rates from comparable and derivative diseases like influenza and pneumonia. The CDC generates that data and puts out flu surveillance data on a weekly basis throughout the year. The chart below shows the death rate from pneumonia and influenza (P&I) on a weekly basis over the last 7 years. Two features are striking and not widely reported. First, deaths from respiratory ailments move in a seasonal pattern year after year. The peaks may vary, but the pattern repeats. Second, the death rate has been much worse as recently as 2 years ago. 
The data is lagged by a week and according to the CDC, the current spike is likely the result of COVID-19. However, the death rate is still well below 2 of the last 6 years. It seems likely to move higher in coming weeks, but there’s still room before we hit the death rates from just 2 years ago. The European Union also maintains a similar database that shows excess mortality throughout member states. The chart is below.
This chart doesn’t show respiratory deaths, but deaths overall and the deviation from trend. The pattern is similar to the U.S., deaths are higher in the winter and peak in January or February. Strikingly, this year’s flu season peak was below the last 3 years. There’s a second peak forming that is likely the result of COVID-19. Lower mortality during flu season might have driven higher mortality from COVID-19. As in the U.S., the seasonal death rate is comparable to previous years.
Finally, the New York Times tracks COVID-19 case growth rates throughout the U.S. and the world. It shows the absolute number of cases in countries, states and counties. Most interestingly, it also shows a daily reading for how quickly the cases load is growing. The chart is shaded from faster to slower for every locale. The virus is now growing more slowly in every state in the U.S., even New York and Louisiana. The same is true for the major countries in Europe. If the rate of growth is slowing, the curve is flattening. That means social distancing efforts are working and could mean that the virus is seasonal like other corona viruses and influenza.
These three separate sources all show a similar story: the virus is likely seasonal and seems to be flattening out. That story is completely at odds with the relentless coverage of this “silent killer”.
How can we put 100,000 to 200,000 deaths in perspective? I think it’s worth reviewing the figure of 100,000 to 200,000 deaths put forward at the presidential briefing last week. That is a horrifying number by any measure. But let’s put the projected death rate in context with real mortality numbers for the U.S. According to the CDC, roughly 55,000 people die every week. During the worst weeks of the 2018 that flu season, over 4,000 people were dying every week from pneumonia and influenza (Fluview). That context provides us a point of comparison and helps us imagine a figure of 100,000 to 200,000 deaths as a result of COVID-19. 200,000 is an unimaginable number of deaths, unless you consider we are a country of 327,000,000 with 55,000 people dying every week of the year and up to 5,000 dying each week of peak flu season. Finally, the 100,000-200,000 projection was based on the model from the Institute for Health Metrics and Evaluation (IHME). The model they have built has been over pessimistic. Just since last week, the estimated number of deaths has fallen to 60,000. Based a quick review of the underlying estimates and the known difficulty with modeling, even that number may be way high. However, it’s comparable to the 55,000 deaths attributed to flu in the 2018 season.
Conclusion Although we are all afraid of new things we do not understand, we can step back and try to analyze novel things from a logical perspective. As we ask basic questions, wade through the barrage of media messaging and look at data, we can frame what is new to us in a way that is both understandable and not nearly as frightening. Existing data and historic context help us in evaluating the COVID-19 pandemic. Based on the simplified analysis above, we can draw a few of conclusions. There are many more people infected than are captured by the existing testing regime, the death rate from the virus is significantly lower than the media reports, and social distancing and seasonal factors indicate a slowdown in the numbers of infection. The projections for 100,000 to 200,000 deaths in the U.S. this year are possible but not likely and should be considered in context to other figures. Be positive, continue social-distancing practices and hand washing, and always ask questions.
Sources:
CDC, Deaths and Mortality, https://www.cdc.gov/nchs/fastats/deaths.htm
New York Times, Coronavirus in the U.S.: Lates Map and Case Count, https://www.nytimes.com/interactive/2020/us/coronavirus-us-cases.html#cases
CDC, Pneumonia and Influenza (P&I) Mortality Surveillance, https://www.cdc.gov/flu/weekly/#S2
European Monitoring of Excess Mortality, Pooled Death Data, https://www.euromomo.eu/outputs/number.html
Medscape, Seasonal Patterns in Upper Respiratory Tract Infections, https://www.medscape.com/answers/302460-86798/what-are-the-seasonal-patterns-of-rhinoviral-coronaviral-enteroviral-and-adenoviral-upper-respiratory-tract-infections-uris
Johns Hopkins University, Corona Virus Resource Center, https://coronavirus.jhu.edu/map.html
Medrixiv, Estimating the infection and case fatility ratio for COVID-19 using data from the Diamond Princess, https://www.medrxiv.org/content/10.1101/2020.03.05.20031773v2.full.pdf+html
CDC, National Vital Statistics Report 2017, https://www.cdc.gov/nchs/fastats/flu.htm
Iceland Data, https://www.covid.is/data
IHME Website, https://covid19.healthdata.org/united-states-of-america
Live Science, https://www.livescience.com/undetected-infections-coronavirus-widespread.html