10x worse than the Flu, 1/10th as Bad as 1918 Spanish Flu Pandemic

For A Video Explanation of The Drivers Of the Forecast, see The ARF Town Hall

1 March 2020—I started coverage of the novel coronavirus a month ago, when I saw the model coming out of China to be far too low (based on my analysis). One month later, we know it was 10 times worse than the initial model out of China projected, with over 85,000 cases globally. Yet, the data trends show it could have been 100x worse had China not taken actions to reduce the spread through extensive quarantine measures.

There are many questions. This forecast addresses the following:

  1. Will countries succeed in containing the virus?
  2. What happens if COVID-19 isn’t contained -- how bad will it get?
  3. What are the implications to business and society?

A COVID-19 Forecast And Scenarios

For the first question, it is possible that countries will contain the virus. Singapore seems to be having the most success. At the same time, there are countries containment seems extremely challenging. In the US, I am reminded that H1N1 spread to nearly 20% of the US in 2009. Will COVID-19, which appears to have similar spread dynamics, be any different?

For the second question, I provide key data points and a model to estimate how bad it could get. What follows is an analysis and a forecast. It is not a guaranteed outcome. This is my safe-harbor statement: The model is a calculation based on data and assumptions. I am publishing my data and assumptions so you can make your own adjustments and draw your own conclusions. There are many things that can change that can make the situation better or worse. There are many factors that can influence the outcome. With the caveat emptor stated, here is my Forecast:

US by end of 2020:

For context, the Spanish Flu of 1918 infected an estimated 27% of the population and killed about 2% of the total population. My forecast is that COVID-19 in the US is that the total deaths, on a percentage basis, will be about one tenth as bad as Spanish Flu of 1918. COVID-19 could be bad. Not society ending bad, but it will touch the lives of people we know and love.

COVID-19 vs. Spanish Flu and Seasonal Flu - A Forecast of How Bad COVID-19 Could Get in 2020.
COVID-19 vs. Spanish Flu and Seasonal Flu - A Forecast of How Bad COVID-19 Could Get in 2020

On the other end of the spectrum is annual influenza, what we colloquially call the flu. Many are comparing COVID-19 to the flu. What we call the flu kills about 80,000 in a bad year. My forecast projects COVID-19 will be about ten times more deadly than a bad year of annual influenza.

H1N1 is another useful comparison point. COVID-19, I forecast, will be about 25 percent less than the spread of H1N1 in terms of infections in the US, but more deadly. H1N1 infected approximately 20% of the US population, however it wasn't very deadly. As people come to terms with the deadliness of COVID-19, I am forecasting behaviors will change and we will land at least 25% lower than the spread of H1N1.

COVID-19 will disproportionately infect and kill those 65 and older. About 80 percent of all deaths will be among those 70 and older. Based on the forecast, this would mean over half a million deaths in our senior population.

If this forecast is correct, COVID-19 will lead to significant business of policy alignment.

I hope my forecast is too high. This article encourages you to get inside the models and understand what drives the forecast. In the weeks ahead, you will find more forecasts from me and from others. I’ll refer to my forecast as “Briggs COVID-19 Forecast 1.0” so we can compare and contrast as more data becomes available. I would encourage you to look inside the models to understand the assumptions that drive the forecast and make your own adjustments to come to your own forecast.

For Briggs COVID-19 Forecast 1.0, it is driven by three key factors:

  1. How fast it spreads and how high it gets in terms of population penetration.
  2. The mortality rate.
  3. The ratio of serious and critical to the mortality rate.

I share a table of forecast drivers so you can adjust the assumptions to create higher and lower forecasts. As more data comes in, I will save the original model as a reference, and provide updated models. I will thoroughly explain the basis for the forecast.

(UPDATE 3 March 2020: Yesterday, Mark Lipsitch of Harvard released a range of 40-70% on CBS Evening News -6min video- so I am including that range in the table.)

COVID-19 Table of Forecast Ranges And Assumption Drivers.
Table of Forecast Ranges And Assumption Drivers.

But, first, let's review my Forecast through the end of February compared to actuals.

COVID-19 February Forecast vs. Actual.
Graph of February 12 Forecast vs. Actual

Two weeks ago, I estimated we’d end the month with about 100,000 cases globally. WHO reported 85,403 cases (100,000 was crossed one day later than I foretasted). I modeled a reduction in the spread in China while at the same time calculating spread into other parts of the world.

Coming in below forecast is good news. Hopefully, we see this trend continue, though I worry the there may be under-counting / poor surveillance in China, Iran, and even the US. Therefore, the 85,403 reported by WHO may be an under-representation of actual infections.

Basis of Briggs Forecast 1.0: I have extended this analysis by using the Diamond Princess Cruise (DPC) to consider the likely penetration of the virus. The fact the DPC didn't get worse than 20% suggest to me that there are limits to the spread of the virus. These limits may be biologic, or may be entirely human behavior as we adapt our actions to address a rapidly spreading virus.

COVID-19 Confirmed Cases By Date on Diamond Princess Cruise.
COVID-19 Confirmed Cases By Date on Diamond Princess Cruise.

The 15% penetration is based on stratifying this data between the passengers and crew due to the bi-modal age distribution. I estimate the passengers to have an average age of 70, and project their rates to the cohort of 65 and older in the US. I estimate the crew to have an average age in their 30s and use the population to project to those under 64. I then apply a weighted average of the percent infected in each age cohort to calculate the percent of the population likely to become infected in the US in 2020. This produces a 15% overall population infection rate. Next, I evaluate the weekly rate of spread (more on this point later) and consider the timing of US taking actions to limit the spread. Finally, I compared the 15% penetration to H1N1 (20%) and the 1918 Influenza Pandemic (27%) as reference points.

This 15% penetration in the forecast is a data informed assumption. Make your own assumption. The table shows the difference in forecast if we assume different rates from 0.01% to 30%.

The DPC spread to about 15% of the population may be a worst case in terms of population density and the spread of the virus. Some argue it could have gotten worse had quarantine measures (as imperfect as they were) not been implemented. Still, when extrapolating to the US, where our rural and suburban population will see a slower spread simply because of a lower rate of human to human interaction, it may be that the 1% penetration or lower is a more realistic assumption.

COVID-19 Table of Forecast Ranges And Assumption Drivers.
Table of Forecast Ranges And Assumption Drivers.

A Rosier Scenario:

0.1% penetration is about the rate in Hubei province right now. It is a population of 59 million with 66,337 cases as of the end of February. It is possible to come in below the Hubei 0.1% level. The rest of China has reported a total of 79,394 for a population of 1.4 billion, so they are below 0.0001% level. If the US performs as well as China in limiting the spread, cases in the US will be well below the 321,000 infected rate and 5,265 deaths listed in the 0.1% scenario (leftmost column).

However, there is risk we don’t contain the virus.

Weekly Rate of Spread:

Why do I use the 15% penetration rate in the US for my forecast? Consider how a weekly reproductive rate of 3 grows by the end of the year. As the chart illustrates, one person infects three.

These three each infect three more for a total of nine. These nine infect three each, adding another 27. The 27 grows to 81 a week later -- and that is just the first month of the cycle. COVID-19 Table of Forecast Ranges And Assumption Drivers. Consider we have 71 cases as of February 29th. By the end of March, if the weekly R0 is 3, we’ll have about 6,000 cases. By the end of April, we’d be at over 500,000. By the end of May, we’d be north of 40 million.

If the weekly rate is lowered to 2, we would still get north of 40 million before the end of the summer. Use this simplified table below to track which weekly spread rate we are observing over the course of the next eight weeks.

I’ve included the Flu as a reference point. If COVID-19 is no worse than the flu, then we should be at about 200 cases in the US at the end of March. (I will bet we will see more than 200 cases at the end of March.)

Weekly Growth Rate In COVID Reproduction.
Weekly Growth Rate Scenario From March 1 Confirmed Cases of 71.

You will note the math of a weekly R0 cycle shows the number of weeks until “ALL” in the population is infected. The math of an unmitigated spread is daunting. However, my forecast assumes that there is a natural ceiling of 15% of the US population. This, as noted previously, is an assumption based on the DPC data set and comparing to the 2009 spread of H1N1. Unfortunately, the forecast of 15% by end of 2020 is a plausible scenario. I consider it like a weather forecast for rain. It is better for people to seriously consider the risk of rain, and pack an umbrella and not need it than to underestimate the risk of rain and be all wet.

Some may wonder, "What if it is worse than 15%?"

I’ve included a 30% scenario so you can consider if the spread is as bad as the 1918 Spanish Flu, what would be the forecast for serious/critical cases and mortality. I am not forecasting penetration to exceed 15% and I seriously hope we will come in well below the 1% level. (Note, Table Updated 3 March with 40% using Harvard's lower end projection)

COVID-19 Table of Forecast Ranges And Assumption Drivers.
Table of Forecast Ranges And Assumption Drivers.

Mortality Rate: My Age Adjusted Mortality Rate Lower Than Other Estimates

The next part of the forecast is for the number of deaths and the number of serious and critical cases. Here I combined the data from the DPC with my analysis of deaths from Hubei. My analysis calculated the risk of death based on sex, age and chronic conditions. Subsequently, a study was released from China using a similar and larger data set that corroborated my findings. To read a full explanation of the data, and method, here is my pre-publication working paper.

My 1.6% mortality rate is significantly lower than the 3.4% case fatality rate from WHO. As of February 29th (2,924 deaths / 85,403 cases = 3.4%). As I have reported previously, I believe the denominator, that is the total number of people infected, is very difficult to know in China, Iran, the US, etc. The numerator, Death is not asymptomatic. The denominator, COVID-19 infection is often asymptomatic. I suspect the count of deaths is more accurate than the count of infections. Therefore, the case fatality rate may overstate the severity of COVID-19.

The 1.6% mortality rate may be better news than we've had, but keep in mind it means COVID-19 is more than ten times worse than influenza's mortality rate.

The DPC is a tragic case where do in fact know the total number of people to test positive (705), we know the denominator (3700 passengers and crew) and we know the number of deaths. Over the next month, I will update the number of DPC passengers and crew that recover or succumb to COVID-19. The mortality rate may change. The best case scenario is that no-one else dies from the DPC. In such a case, the age adjusted rate would be 0.13, which is pretty close to the influenza we deal with on an annual basis. I’ve included this scenario, labeled "No Additional Deaths scenario."

I sincerely hope that all recover and I will be adjusting the mortality rate downward to something similar to influenza. However, if you see additional deaths from the DPC reported in the news, you will know COVID-19 is more deadly than the flu.

It can certainly be argued that the DPC is the best case, medically, because professionals caught the infection early, and treated it immediately at a time when hospitals are not overwhelmed. I am including the 3.4% current case fatality rate, but I believe we will do better in the US that that rate would suggest. You can evaluate for yourself the forecast implications.

Forecasting Serious and Critical Cases:

The final element of the forecast is the serious and critical cases listed in the middle of the table. Based on the current trends, the number of serious and critical cases are a ratio of approximately 5:1 to each death. This figure was used in my forecast.

To date, about 80% of deaths are among those 65 and older. At the same time, those who are older are more likely to need care from a hospital. When making a forecast for the entire population, it is important to factor this into the calculation. It is also important to consider the implications to society of a disease that is most threatening to our older population.

With this sobering forecast of 15% penetration and 1.6% mortality rate, the risk is about 1 in 425. With a lower penetration rate of 0.1% (Hubei's level), the risk is 1 in 75,000.

I plan to provide a calculator for estimating the risk of contracting COVID-19. Based on the Hubei data I’ve analyzed and posted previously, if you are under 50, without underlying risk factors such as Hypertension, Diabetes, Heart Disease or Cancer, and there are no reported cases in your area, the risk is very low. If you are 65 and over, with these underlying conditions, be thoughtful about the risks and methods to enhance your safety.

Business and Societal Implications:

My wife and I have volunteered to help the older and wiser members of our neighborhood with their shopping needs, should the need arise. We can each be prepared to respect any quarantine if it is asked of us. We can wash our hands and stay away from others if we get the slightest inkling we might be coming down with something. We have actions we can take to change the trajectory of COVID-19. It is up to us to decide on the risk and take appropriate actions.

More broadly than our own personal actions, we should consider how COVID-19 might reshape business and society. Will we find it acceptable that the cost of screening falls to individuals with poor health care as reported for some of the early cases?

Will our approach to business forecasting change to be more in line with fast moving changes? Will we finally ditch three year lagged mix models for weekly or daily analytic models?

Will quarantines at home change the way people shop and be the death knell for many retailers?

How will our views on the safety net change if over 75 percent of service workers can't perform their services because most everyone is staying at home for a month?

Will we experience a quick economic bounce back because we contain the virus quickly, or a prolonged recession because we don't contain it quickly?

It is too soon to answer these questions, but it not too early to start to think about them.

A Forecast Is Not Destiny:

It is possible that good hand washing hygiene, social distancing, and quarantines could result in a significantly lower penetration rate than the 15% assumption.

It is also possible that, like the seasonal flu, COVID-19 does not spread as well in warmer months such as April through October, thus lowering the penetration significantly below 15%.

There is the potential of a viable vaccine identified in three months. I've counted a dozen companies that have announced they are working on a vaccine and they are close. At the same time, it appears Singapore has succeeded in breaking the spread of the virus, and perhaps China has as well, demonstrating we can stop the virus even before a vaccine becomes available.

In sum, there are lots of ways we can come in well below the 15% penetration assumption I’ve used for this forecast. I certainly hope we do keep the penetration below that of Hubei’s 0.1% level.

I hope this forecast and the assumptions that drive are helpful in dimensionalizing the risk of COVID-19. It is a serious threat. You can make your own judgment about penetration and mortality rate. A forecast is not our destiny. We can make changes to come in well below this forecast.

I would very much like to find that my forecast was far too high. It would be an accomplishment of human cooperation to get the R0 below that of the flu. There are many actions we can take as everyday citizens of the world to protect those most at risk of the virus.

A Final Thought:

The two most important tools in my forecasting toolbox are 1) data and 2) A concept I learned at WIRED: “The future is already here, it is just unevenly distributed.”

The DPC is a tragic time machine that gives us data on what our future could look like if we don’t take the right actions to practice excellent hand washing hygiene, social distancing, etc. Italy is a time machine that lets us see how fast a country can go from a handful of cases to thousands. It lets us see what happens under quarantine conditions. Singapore gives us another glimpse into a future where the actions of the government seem to have stopped the spread of the virus. The Singapore government reports each case by number, and shares much more details than we are seeing from the US Government to date. Perhaps the US will find a way to release similarly helpful case data.

A forecast is not our destiny, but it may become our reality if we don’t change some of our behaviors. Collectively, it is OUR choice as to whether COVID-19 continues to spread and kill or whether we work together to prevent COVID-19 from taking more lives.

Recent Analysis

  • 1 March 2020:
    10x worse than the Flu, 1/10th as Bad As 1918 Spanish Flu Pandemic (A Forecast for how bad it could get in 2020)
  • Forecasts (Jan to April) Vs. Actual :
    Summary