Each day we are deluged with new predictions of the levels of disease and death that lie ahead of us with the COVID-19 pandemic. Many are traumatized and confused by the changing predictions, leading some to simply dismiss such predictions out of hand.
So where do these predictions come from? And, how should we everyday citizens think about them?
Where these COVID models come from
When it comes to predicting epidemics and pandemics, researchers and planners use algorithms based on the three basic components of the spread and consequences of a disease:
- The transmission rate of infection
- The level of contact that people have in response
- The consequences of the disease
In the case of a disease that has been around for a while (e.g., measles), the values of these parameters are well known. In these cases, public health planners can map with considerable accuracy the likely spread and consequences of a contagion. Most importantly, they can estimate how much the infections will burden a country’s health system, such as how many people will need to be hospitalized.
But, in the midst of a pandemic caused by a previously unknown virus, such models are very imperfect tools. When it comes to COVID-19, we knew relatively little in the beginning about the transmission of the virus, or about what it could do to the body. For instance, the finding that asymptomatic carriers could spread the disease introduced great uncertainty in modeling the transmission of the disease. As for disease effects and consequences, we are continually discovering new and distressing impacts it has on all parts of the body.
In the face of such uncertainties, epidemiologists use data from known diseases to try to estimate these basic parameters. For instance, transmission rates from past influenza outbreaks were used in some of the initial models. They then run many simulations and develop multiple models. Worst-case models are particularly important because they help health planners anticipate when a health system might become swamped and develop contingency reasonable contingency plans.
What about the human contact factor?
Estimating human behavior and response to an infectious outbreak is the most fraught element of epidemiological forecasting. The worst-case scenarios typically project that ordinary rates of social interaction continue and that there are no active efforts to reduce or mitigate the transmission of the disease.
However, when the rates of transmission are high and the consequences of a disease are quite negative, humans naturally react to those findings and try to reduce exposure and disease transmission. The current social-distancing measures that societies around the world have undertaken are an obvious example of this. Estimating the effects of these policies, if fully complied with, is manageable. But, estimating compliance with such measures over time is an uncertain enterprise, because policy compliance decays over time.
The more-negative recent models
The public was shocked this week by a dramatic increase in the projections of COVID-19 infections and deaths from one modeler: the University of Washington’s widely followed model from the Institute of Health Metrics and Evaluation (i.e., IMHE).
The IMHE changed their estimates of likely deaths in the U.S. by August 4 to 134,000 – a dramatic increase from their earlier estimate of 75,000 deaths. Data from cell phone tracking, as well as daily case totals, suggest that compliance with social distancing is diminishing as states develop plans to reopen business and social life.
But remember this …
Remember, an epidemiological model says something like the following, “If the infection has this level of transmission and morbidity, and we practice this specified level of mitigation, we can anticipate outcomes something like this.” If we keep in mind that such models are decision-making tools and not destiny, some of the public controversy and confusion regarding them might diminish.
As we gain more knowledge about COVID-19, the models are improving. Scientific research and new data clarify things, but this takes time. One implication of this is that public health communicators and politicians should probably take some time to explain that epidemiological models are useful but fallible tools. By acknowledging their limitations, but emphasizing their usefulness, the public controversies about their use might diminish a bit.
Governor Beshear has done an admirable job of communicating with the public during this pandemic. I would humbly suggest one additional message someday soon: “These models that we consult are tools; they are fallible, but they do help us make better decisions in these very difficult and uncertain times. But remember, they are only tools. They do not have to be our destiny as long as we do our part to change the numbers.”
–30–