The solid blue line in the chart above is the number of reported COVID-19 deaths in the U.S. each day, the red line represents an exponential growth curve, and the dotted blue lines are “control limits” that indicate whether growth is on an exponential path. See full primer below. (MDMetrix Chart.)

Reflecting a sentiment being conveyed in some COVID-19 hotspots, Gov. Phil Murphy of New Jersey tweeted this week that the “curve is flattening” in the state’s COVID-19 crisis. But he cautioned that it was too early to celebrate — saying that it was “no time to spike any footballs or to take our foot off the gas.”

However, it is time to sharpen our pencils. And it turns out the math agrees with all of Murphy’s metaphors.

Daily deaths in New York, New Jersey, California, Michigan and Washington state “are still on an exponential growth curve,” according to a new analysis from Seattle health data startup MDMetrix. The company says it’s using artificial intelligence combined with control charts to distinguish genuine trends from less-than-significant changes in data sets that vary widely from day-to-day.

But the analysis also shows the growth trajectory of new COVID-19 cases starting to fall below the exponential range in some of those same states, and growing below an exponential curve for more than a week in the US as a whole.

That could be read as an early signal.

“There are likely a couple of weeks of lag time between the two measures,” MDMetrix CEO Warren Ratliff said via email. “If less people get COVID-19 today, all things being equal, we would expect fewer deaths a couple of weeks later.”

“However, there’s a major caveat,” he said. “The number of reported new cases is tied to the availability of testing. So, a ‘signal’ regarding the number of new cases might represent a flattening of the new-case curve but it could also reflect testing limitations.”

Ratliff added, “While it’s encouraging to see signs of a deceleration in the number of new cases for some states, the key measure that we all want mitigation measures to affect is the number of lives lost each day.”

“Unfortunately, we have not yet seen a data signal that the U.S., or any of the worst affected states, has successfully reduced the accelerating daily death rate. We certainly hope that changes soon. We hope these charts help everyone see the need to continue with mitigation measures and make sure they’re effective.”

Dan Low, MD, associate professor at the University of Washington and MDmetrix chief medical officer, describes control charts as “the best tool for distinguishing signals from noise in real-world medical data.”

The charts are based on an approach developed by noted health statistician Lloyd Provost, an MDMetrix adviser affiliated with Associates in Process Improvement and the Institute for Healthcare Improvement.

The company, which has raised more than $4 million to date, was started in 2016 after Low, an anesthesiologist at Seattle Children’s Hospital, was surprised by the challenges he encountered when trying compare the efficacy of two different drugs in patients, as chronicled in this episode of the GeekWire Health Tech Podcast.

MDMetrix, which offers its “Mission Control” software to hospitals free of charge for COVID-19 cases, says the charts are an example of the type of insights its technology provides to healthcare professionals. The company says its artificial intelligence “automatically identifies ‘signals’ in the data so that leaders and frontline clinicians can understand and adapt their approaches to COVID-19.”

The charts are derived from data complied by The New York Times. MDMetrix says it will update them daily.

From MDMetrix, here is a primer on reading the charts.

• The red line represents an exponential growth curve projection, based on available daily deaths data from the New York Times.

• The solid blue dots, joined by the blue line, represent the actual number of reported COVID-19 deaths per day.

• The dotted lines are “control limits” that are mathematically tied to the projected growth rate. If COVID-19 deaths are on a “stable” path of exponential growth, then the solid blue line should stay within the dotted “control limit” lines.

• Naturally, the blue line moves back and forth across the red line projection, reflecting expected real-world variation (noise).

Decades of quality improvement data science (i.e., industrial engineering) have established several rules for detecting material “signals” using control charts. For example, if the blue line (deaths) crosses a dotted line (a control limit), that would signal that something has occurred that has made the exponential growth “system” unstable. Perhaps a stay-at-home order might decelerate the daily death count, flattening the blue line and crossing the lower control limit. Or, the lifting of a stay-at-home order might accelerate the daily death count, raising the blue line and crossing the upper control limit.

Other data signals that would indicate we’ve turned the corner would be eight sequential points to the right of the red line; or six sequential points, each falling farther to the right (i.e., away from the red line). The converse of any of these rules would mean the situation is getting worse.

See the company’s dashboard for the latest state-by-state analysis.

Like what you're reading? Subscribe to GeekWire's free newsletters to catch every headline

Job Listings on GeekWork

Find more jobs on GeekWork. Employers, post a job here.