Why So Many COVID Models are Broken

When I was in college, a friend noticed how often I squinted to see the whiteboard.  She told me I needed glasses.  I didn’t believe her.  I went to the eye doctor anyway.  Two weeks later, I learned it was possible to see leaves on trees from more than a few feet away.  With glasses on, I could even make out a family of cardinals sitting in a sugar maple tree aflame with red and orange leaves.   The cardinals had always been there – I’d just never before been able to see them.

Through the fog of the COVID pandemic, legions of brilliant, well-intended people have worked tirelessly to create models to educate and inform policymakers.  These models are powerful, elegant, and – through no fault of their makers – unfortunately largely wrong.  The vast majority of COVID models I’ve seen to date fail to find the cardinals in the sugar maples – and Wayne County, North Carolina is a prime example of why.

Wayne County sits about an hour east of North Carolina’s Research Triangle Park.  The largely rural, typically beautiful Carolina county has a population of about 120,000 and is home to the city of Goldsboro, Seymour Johnson Air Force Base, and well over $300 million in livestock, poultry, and field crop production.

As of April 27th, it was also home to 630 cases of COVID according to the North Carolina Department of Health and Human Services COVID Dashboard.  That’s 525 cases per 100,000 people, almost half the intensity of hard-hit Manhattan and more than nine times the case rate of nearby Wake County, North Carolina’s most populous county with nearly 1.2 million residents.

How does this make any sense?  Simple.

Wayne County is also home to the Neuse Correctional Institution, a prison with nearly 800 inmates of whom an astonishing 470 have tested positive for COVID-19.  Add in the positive staff members and the prison-attributable COVID population accounts for 80% of the entire county’s case count.

Knowledge of these kinds of localized outbreaks or “epicenters” isn’t new.  Across the nation, we’ve seen them pop up in prisons, skilled nursing facilities, homeless shelters, and food processing plants.  While we’ve known about them since the earliest days of the pandemic, their impact on modeling is asymmetric, significant and as yet unaccounted for in any model I’ve seen.

Try plugging Wayne County into your favorite county-level COVID model.  I used the Columbia model as an example, and found that Wayne County was projected to experience 15-18 new cases per day from April 27 through today.  In reality, as of May 3 Wayne County had reported only 47 new cases in the past week, or just 7 new cases per day.

Most COVID models fail to capture the reality that America is living through two pandemics – the first, measured by true community spread, has had its trajectory dramatically impacted by massive societal interventions like stay-at-home orders.  These incredibly successful yet incredibly painful efforts have unquestionably flattened the curve in places like North Carolina.  They have worked, and in doing so have kept the majority of America from living through the horrors of New York City, Spain, or Italy.

But these measures cannot last forever, and their impact on America’s second pandemic is more variable.  In North Carolina and elsewhere, epicenters keep popping up in meatpacking plants and long-term care centers, driving the state’s biggest one-day increase on May 2 despite nearly simultaneous indications of good news statewide.

That’s because spread within an epicenter is linked less to community-wide interventions and more to 1., localized efforts to keep the virus out of the epicenter and 2., limiting its spread once it gets inside.  Said another way, while community-wide measures reduce the risk that someone shows up to an epicenter with COVID-19, we know they can’t eliminate it.

Once someone with COVID-19 gets to an epicenter – which has repeatedly happened despite stay-at-home – no amount of school, restaurant, or business closures will protect the epicenter population from rapid spread.  Instead, powerful local measures like banning visitation and screening people for symptoms prior to entry are needed to keep it out.  And once its inside, the CDC recommends even stronger local interventions like within-center social distancing, limiting the use of shared spaces, universal masking, cohorting, and rigorous cleaning to keep the community safe.

Why does any of this matter?  It matters because policymakers are facing the greatest public policy challenge in generations – how to prevent death and destruction from both COVID-19 itself and from the socioeconomic impact of everything we’ve had to do to keep the virus at bay.  There are no great options, and harm will come to good people whether policymakers err to one side or the other.  Failing to identify an uptick in real community-spread as we relax social distancing may send a state back on the path towards exponential, New-York-style viral growth and thousands of preventable deaths.  Failing to separately understand, analyze, and project epicenter outbreaks may keep community-level interventions in place longer than necessary and cause undue socioeconomic harm.

Right now, current publicly-available models make it impossible for policymakers to isolate the impact of their decisions on either community-spread or epicenter-spread. Instead, they’re forced to stare into a sea of September sugar maples, left to wonder whether the flickering scarlet mass is a family of cardinals or just a bunch of leaves.

Dr. Chris DeRienzo is a physician leader from North Carolina and author of the book Tiny Medicine – One Doctor’s Biggest Lessons from His Smallest Patients.  All views expressed here are his alone and are not attributable to any entity with which he may be affiliated.  Follow him on Twitter at @ChrisDeRienzoMD and on LinkedIn.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.