I have fished for trout for many years. I have learned that there are some predictors that seem to be associated with success. There are certain months of the year when I am more successful, as well as selected times of day. I believe that selecting the right fly, coupled with a decent cast and a natural drift into a prime pool is predictive for prompting fish to strike. But sometimes I get skunked, only to see a fisherman come behind me and catch the fish that moments before did not exist.
Fishing is a metaphor for mathematical models of the epidemiology of infectious diseases; COVID-19 infections are especially relevant. Let me explain.
Modeling COVID-19 infections has dominated news for the past couple of months. We hear that the predicted deaths in the US may be as many as 2 million, or as many as 240 thousand, or perhaps 100 thousand, or maybe as few as 60,000. We have heard that the peak incidence will occur in April, about the time you are reading this, or as late as July. We are told we need to cluster in groups of no more than 100, then 50, then 10, and then “get used to being alone”. What the hell is going on?
What is going on is that some very smart and dedicated professionals are trying to understand a new infection and predict its future behavior. They are using data about as “squishy” as my predictors for successful fishing. The accompanying figure illustrates what I mean.
This figure comes from the Institute for Health Metrics and Evaluation (IHME) at the University of Washington. IHME predictions are widely cited at daily briefings that include President Trump and Drs. Fauci and Birx. The figure shows predicted COVID-19 deaths per day in the US (the Y-Axis) versus time (the X-axis). The solid red line represents the observed (true) deaths and the dotted red line represents predicted deaths. The shaded area above and below the dotted line represents the expected uncertainty or “error” around the dotted line.
A news reporter who has to make sense of this might write that on April 15, the date of today’s Chieftain, (1) nearly 2,000 Americans are predicted to die from COVID-19 infection, or (2) as many as 4,600 persons are predicted to die, or (3) as few as 800 persons are predicted to die. All three are correct.
Similarly, the figure predicts that deaths will drop to zero as early as the middle of May or as late as the first of June. In other words, the first wave of the COVID-19 pandemic in the US is predicted to be over by the first of June. We should know in 6 weeks if these predictions are accurate.
What are the reasons for the uncertainty in these predictions? Like trout fishing, the assumptions that determine model predictions are multiple and usually imperfect.
For example, a key concept for most epidemiology models is the average number of newly infected persons following contact with an infected person. This is called the basic reproductive number [Ro] (pronounced R-naught). The higher the number, the greater the infectivity. For measles and chickenpox, Ro values are around 12; these infections rapidly spread through a vulnerable population.
The median Ro for COVID-19 was 2.28 when measured from passengers on the Diamond Princess cruise ship. But like all numerical values determined from a small sample of individuals, there is uncertainty in the true value in the larger population; the expected true value lies between 2.06 and 2.52. Because of this uncertainty, predictions of future death rates for the US population in the figure include the lowest expected Ro (2.06) and the highest (2.52) resulting in a range of expected deaths. An excellent reference to the many obstacles for predicting the epidemiology of COVID-19 is: https://fivethirtyeight.com/features/why-its-so-freaking-hard-to-make-a-good-covid-19-model/
For the short term, the primary role of “stay at home” policies is to decrease Ro--to decrease the number of new infections. For the longer term, the great majority of people infected with COVID-19 will survive and develop active immunity. Once an effective vaccine is available the remainder of the population should be protected; Ro will have declined to a value less than 1 and the infection is not sustainable.
Like the guy who catches fish when I do not, some models will prove to be better than others and predictions will improve. Despite these limitations, modelers of the COVID-19 pandemic should be immensely proud knowing their predictions will have been instrumental in guiding us through these tough times.
Ron Polk lives in rural Lostine and is a Fellow in the Society of HealthCare Epidemiology of America. [Google: Ron Polk SHEA]. He is not an epidemiologist.