Sense-making in the fire service during COVID-19
What we know
The Coronavirus is “serve acute respiratory syndrome coronavirus 2” (SARS-CoV-2) typically shorted to COVID-19. Corona is Latin for “crown.” Corona is the name given to a group of viruses that look like a crown under an electron microscope; they include SARS of 2003 and MERS of 2012. SARS of 2003 had a “case fatality rate” (CFR) of 10%, while MERS had a CFR of 34%. The case fatality rate (CFR) is the number of deaths among the confirmed cases. CFR is not the “crude mortality rate” or the “infection fatality rate.” The media frequently interchange the terms, resulting in wrong conclusions. We will expand on this later.
What we want to know
We would like to know the chance of death from COVID-19 expressed in the metric “infection fatality rate (IFR).” IFR is the number of fatalities among everyone who contracted the virus. But to know IFR, we would have to test everyone. Currently, we only test those with symptoms, missing asymptomatic carriers, false negatives, and fatalities that don’t present with COVID-19. So instead of knowing IFR, which includes everyone who has the virus, we use the number of deaths per the number of confirmed cases to calculate “Case Fatality Rate” (CFR).
There is a big difference between IFR and CFR. IFR tends to be much lower than CFR. Even worse, when making comparisons between COVID-19, the seasonal flu, and the pandemic of 1918 (H1N1), the media frequently confuses the two. Comparing one virus’s CFR to another’s IFR, either downplay or overestimate the severity of COVID-19. As a result, the typical person has a difficult time making sense of the situation. Is COVID-19 just like the flu, as deadly as the Spanish Flu of 1918, or no worse than riding in a car? The reality is somewhere in between.
Although we don’t know the IFR, we estimate the CFR by dividing the number of confirmed cases by the number of fatalities. As of 04/08/2020, the estimates of COVID-19’s CFR are between 12.47% in Italy and 1.8% in Germany. The current CFR in the U.S. is 3.23%.
What CFR means to the fire service
“We are all going to get it” is not a good mindset. If we use CFR’s from counties around the world, the results are scary. If we allowed 500 firefighters to contract COVID-19, the results would be as follows:
CFR of 12.47% (Italy on 4/8/20) = 63 dead firefighters
CFR of 1.8% (Germany on 4/8/20) = 9 dead firefighters
CFR of 3.23% (US on 4/8/20) = 17 dead firefighters
Estimated case fatality rate (CFR) tends to rise overtime due to the increase in data. The CFR estimate of SARS-1 in 2012 started at 2% but concluded at 10%. As a fair comparison, the CFR of the seasonal flu is 0.1%. Now that we are comparing apples to apples, we see why COVID-19 sparked such a response once the estimated CFR of COVID-19 went above 1%, or ten times deadlier than the flu. At 12%, COVID-19 is 120 times deadlier than the flu.
Age Matters
But the average CFR in each country can misrepresent the number of deaths if you consider age group. For instance, if you are 40 years old in Italy, your CFR is 0.4% opposed to the average 12.47%. This is good news for young firefighters, but not great news for firefighters in their 50's.
Unfortunately, not all countries provide enough demographic data to make estimates base on age. Again, it’s important to stress that the CFR simply represents the number of deaths divided by the number of confirmed cases. It does not give us a true risk of death, which would be IFR. Differences between countries do not reflect real differences in the risk of dying from COVID-19; but rather, they reflect differences in testing, demographics, and the stage of the outbreak.
The Data Sucks
CFR is estimated by data pooled into files shared on GitHub. Unfortunately, the data is always behind real conditions on the ground. First, it takes four to six weeks for a patient to go from symptoms to death. Second, deaths go through a chain of reporting before they get analyzed.
The typical steps involve:
- Doctor or laboratory diagnosis COVID-19 based on testing or symptoms.
- The doctor or laboratory submits reports to the local health department.
- The health department receives and records each case in its reporting system.
- The ministry or governmental organization brings the data together and publishes its latest figures.
- International data bodies such as the WHO and ECDC then collate statistics from hundreds of nations.
The reporting chain can take several days and produces errors. Additionally, countries such as China are accused of purposely altering data by the US Intelligence Community. As a result, the data is continuously late, full of errors, and frequently misinterpreted by the media.
Modeling — “All models are wrong; some are useful.”
“Exponential” is the buzz word for 2020. “Exponential curve” and “exponential growth” are descriptions of why pandemics are different than vehicle deaths or smoking. And although no curves are truly “exponential,” COVID-19 is entirely different than vehicle accidents in the fact that it doubles in short time frames. While annual deaths due to car accidents fit on a bell curve, a pandemic forms an “S” shaped curve that starts slowly, rises rapidly, and finally levels off after it has consumed its network. Over a year, vehicle accidents will march along steadily day-by-day because they are unconnected to one another. They are random events tabulated at the end of the year. They may fall or rise by small amounts each year, but rarely will they double, and they will never grow by tenfold.
Opposed to vehicle accidents, pandemics are a series of connected events that build on each other. Currently COVID-19 is doubling in size every few days. Comparing unconnected events such as vehicle accidents to self-replicating events like COVID-19 creates a false impression. For example, if vehicle accident deaths behaved like COVID-19 and doubled every five days, we would have about 23 million deaths by the end of March each year. Instead, we only have about 1,600 deaths from vehicle accidents by March each year. Pandemics such as COVID-19 have the potential to produce extremely high numbers, even if they start off small. COVID-19 is a self-replicating contagion and should never be compared to vehicle accidents, smoking, cancer, or heart disease deaths.
Modeling self-replicating contagions is a complicated science. Unlike predicting annual car deaths under a bell curve, pandemics like COVID-19 use models that are ultra-sensitive to new data. It is near impossible to draw accurate conclusions when modeling self-replicating contagions. A predicted 100,000 deaths can quickly become 2 million deaths with new data. All we really know is that the curve will climb quickly in the beginning, but will eventually slow down as the network is consumed.
Due to the difficulty of forecasting growth in complex environments, most institutions don’t even try. The brave that do attempt to forecast self-replicating systems swing wildly in their predictions. Consider the “Institute for health and medical evaluation” (IHME) model. The IHME model is currently the leading forecaster on the web, making predictions of what days individual states will peak. Their model is impressive, and their science is reliable, but their predictions have swung wildly from day-to-day. IHME used the curve from China to get its initial forecast. China’s curve may not reflect reality due to deliberate data tampering. As new data comes in from more reliable sources, their curve shifts.
Consider the Bass diffusion model (BDM), which assumes a standard “S” shape to all such curves. The BDM is fantastic at predicting power laws such as COVID-19, but only after a certain time frame. Which means the BDM needs to get more than halfway through the curve before its predictions work.
Both of these models are useful at making sense of the pandemic to some extent, but they deliver less certainty than we are accustomed to. We are used to academics telling us that “the evidence suggests” an outcome with a “high level of confidence.” Instead, we currently get models that give predictions ranging from 100,000 to 2 million deaths. And to make it worse, the same sources change their predictions each day, creating either exuberant paranoia or passionate disbelief over COVID-19.
But The Show Must Go On
Regardless of the poor data, misunderstanding of its meaning, and purposely downplaying of cases, we still have to make decisions.
COVID-19 will consume its available network. We may not have control over the data or the modeling, but we can control the size of the network. Decisions and actions must focus on breaking the network at every opportunity.
Future discussions about COVID-19 should revolve around terms such as links, nodes, and density.
- Nodes are the items and people COVID-19 jumps to,
- Links are the way it travels
- Density is the number of additional nodes it has access to.
We break links by wearing PPE, we eliminate nodes by hygiene, and we reduce density by distancing. This is the language of Network science, the foundation of modeling contagion spread. All of the other answers we seek, such as CFR, IFR, and total morbidity, will come later, but we can kill the network now.
Why the new terminology? Old habits have inertia. The habits of shaking hands, eating together, training together, and moving in tight groups are hard to break. Breaking ingrained behavior requires a deliberate shift in thinking. We think with the words we know. New words create novel thoughts. Novel thoughts are the path to overcome old habits. If we aggressively break old habits and create new ones, we can avoid adding new names to the Firefighter memorial.
