It’s not the mean, it’s the variance.

Many of you know that I work in Connecticut, which is currently in the midst of our COVID-19 surge. It’s going OK – pretty intense social distancing up here has made the situation manageable. We haven’t run out of vents. We’re low, but not out of critical medications.

In other words, what we’re doing is working, which makes a lot of nervous when we hear talk about rolling back some of the restrictions that have been so successful so far.

The truth is, most people don’t have the same experience with COVID-19 as health care providers do. Many don’t know anyone who’s had the disease.

But if you talk to a healthcare worker about COVID-19, they’ll tell you they are scared. But I’m realizing that we may not be doing a great job communicating what makes COVID-19 so scary to healthcare workers.

And because in my day job I’m a clinical researcher and a data science guy, the best way I can explain it is with some math terms.

It’s not the mean. It’s the variance.

Let me explain with a less grim example.

If you’ve ever played a casino game, like blackjack, you know the house has an edge. With perfect strategy, that edge is only about 0.5% - meaning that, on average, you lose about 50 cents of every 100 dollars you bet. Of course, if you’ve played blackjack you know the average hardly applies – it’s the variance – sometimes you win a lot, sometimes you lose a lot. Here’s a simulated distribution of winnings (or losses) after playing 60 hands of $10-a-hand blackjack.

On average, you’d lose 3 dollars but look at that variance. That’s what makes blackjack so exciting – the chance to win or lose big. The mean matters to the house. The variance matters to you.

What’s exciting in a casino when all you can lose is money becomes terrifying in a hospital when your life is on the line.

We still don’t know what the true death rate is in COVID-19. Estimates range from 0.1 percent to as high as 3 percent, but that isn’t what scares hospital workers. That’s the mean. That matters on a population level.

On an individual level, what matters is the variance.

For many infectious diseases, the overall death rate is somewhat misleading as the deaths tend to cluster among the very sick. The best way to think about it is on a spectrum of health:

A given disease may push everyone further away from health and towards sickness – but only those that were already on the end of the spectrum can get pushed all the way to death.

There is very little variance, in other words. And in a weird way, it’s comforting, because you can tell yourself (right or wrong) that those who died would probably have died soon anyway. This happens to be a talking point among some people trying to downplay the severity of the pandemic.

But it’s not true, as any healthcare worker who has cared for these patients can tell you. Yes, mortality rates are higher in those who are older, and with more comorbidities, but all of us who have taken care of these patients have seen horrible outcomes, including death, in young, otherwise healthy people. That’s high variance. That’s what’s so scary. Some people get mild disease, some people get ventilated, and some people die, despite having very similar baseline characteristics.

That’s why COVID-19 feels like gambling. Like Russian roulette.

We can actually quantify this. A recent JAMA Network Open letter reports on 168 patients who died from COVID-19 in China. 30% of those who died had no comorbidities at all.

I asked healthcare workers on twitter to tell me what scared them about COVID-19. One IM resident captured it perfectly: “really scared of how people my age just die despite all the ARDS tricks we try, and that it can easily be me next”.

That’s why we’re scared. And we owe it to the rest of the country to talk about that as we start to make decisions about opening back up. The law of averages tells us that chances are, it won’t be you next. But that doesn’t matter too much when you’re letting it all ride on one spin of the roulette wheel. 

This commentary first appeared on medscape.com.