Apr 1, 2020 - Health

What disease modeling could learn from weather forecasting

An illustration of an ambulance trying to following a weathervane

Illustration: Eniola Odetunde/Axios

COVID-19 has brought the arcane work of mathematical disease modelers to the forefront, as politicians search for ways to flatten the curve.

Why it matters: Models are the only way we can plan out effective steps now to prevent more deaths in the future. But modeling a disease in mid-pandemic isn't easy, and important nuance can be lost in the translation between academic modelers and policymakers.

Driving the news: During the White House briefing on Tuesday, Anthony Fauci and Deborah Birx displayed graphs showing government projections that even with mitigation, COVID-19 would kill as many as 240,000 people in the months ahead. It was the first time the White House had officially estimated the potential death toll.

  • What Fauci and his colleagues are trying to do is "flatten the curve" — taking steps now to reduce the speed of COVID-19's spread and therefore prevent hospital systems from being overloaded. The shape of that curve is the product of mathematical disease modeling.
  • To simplify things — which is the point of a model — mathematical modelers take known data like the past number of cases, plug in estimates for things they don't know, and create models for how an outbreak will progress.
  • If those models are accurate, they can help policymakers visualize how an unchecked outbreak will play out and take effective steps to protect their populations.

Yes, but: Any attempt to predict the future will be imperfect, and the less modelers know for certain, the more weight they need to put on their estimates. That's especially true for a new disease, so we've seen models spit out a wide range of potential outcomes for COVID-19 — sometimes by the same modelers.

  • Early COVID-19 models by a team at Imperial College London assumed that demand for intensive care units would be roughly the same as previously modeled flu pandemics. The result was a milder forecast that initially encouraged the U.K. government to adopt a "mitigation" strategy with relatively little social distancing.
  • But when data from China and Italy showed that COVID-19 patients required intensive care at much higher proportions, the same team revised its models, predicting that a laissez-faire strategy could lead to as many as 500,000 deaths in the U.K. alone.
  • The British government eventually abandoned mitigation in favor of a near-total lockdown.

Be smart: As the famous saying (among modelers) attributed to the British statistician George Box goes: "All models are wrong, but some are useful" — meaning that perfect prediction is impossible, but a rigorous model can help politicians see how their actions might bend an outbreak one way or another.

  • The problem is that politicians and mathematical modelers don't speak the same language and don't move at the same pace, especially during an outbreak.
  • "Sometimes you need an immediate answer, not one in five days," says Caitlin Rivers, an epidemiologist and modeler at the Johns Hopkins Center for Health Security who earlier worked with the federal government.
  • It's notable that while other governments publish data-rich updates on the epidemiology of COVID-19, the best dashboard in the U.S. is updated by academics at Johns Hopkins — not the U.S. government.

One possible solution is to try to bridge the gap between the mathematicians who model diseases and the politicians who respond to them.

  • In a new report, Rivers and her colleagues suggest creating the equivalent of a National Weather Service for disease modeling. The proposed federal institution would provide authoritative disease models during outbreaks and embed modelers in the government so that both cultures can learn from each other.
"Right now with modeling the outbreak, it's as if we're building the plane as we're flying it. If it was already there, it would be much smoother and more effective."
— Caitlin Rivers

The bottom line: Because there's so much we still don't know about COVID-19 — and because the field mostly remains an academic endeavor — the models that are being produced are likely flawed. But they remain the best intelligence we have in the war on the pandemic.

Go deeper