fbpx

Kinsa’s COVID-19 Predictions, One Month Out

On June 29, the New York Times published an op-ed exploring the few options available to detect viral outbreaks early, and highlighted Kinsa’s new illness measure—Atypical Transmission (Rt)—which we used to make some predictions about what might happen in a number of specific states. Today, in the interest of transparency, we revisit our month-old predictions from that article to assess their accuracy.

Atypical Rt measures the rate at which influenza-like illnesses, including COVID-19, are spreading in different parts of the country in real-time; higher than seasonally-normal illness transmission is considered to be ‘Atypical Rt’. Our early evidence suggests that Atypical Rt is a leading indicator of COVID-19 outbreaks—essentially, when people are transmitting illness at seasonally elevated rates, those illnesses are likely to include COVID-19. While we validate the exact nature of this relationship, we’re sharing these results because they illustrate the potential of using Atypical Rt to indicate what’s coming as we prepare to enter the flu season.

Texas, Arizona, and Florida

By late June, daily COVID-19 cases were rapidly increasing in Texas, Florida, and Arizona, but it was unclear whether these trends would continue or level off. Reporter Charlie Warzel wrote of Kinsa’s prediction (on June 29): “the company told me it was seeing high atypical transmission in Texas, Florida and Arizona, suggesting the states have yet to peak.” 

Did our predictions come to pass? – Yes. Daily new cases in Texas cases did not peak until 7/20, in Florida until 7/17 and in Arizona until 7/6. 

For each of these states, Kinsa observed elevated Atypical Rt every day for more than a month up to when the article was posted, suggesting that transmission of influenza-like illness, likely including COVID-19, was significantly higher than normally seen during this time of year. Transmission steadily increased until Rt peaked on approximately 6/17, 6/20, and 6/18 for Texas, Florida, and Arizona, respectively (see figures 1, 2 and 3 below).

Kinsa’s Atypical Rt peaks preceded COVID-19 daily case peaks by ~3-4 weeks in each state, which is consistent with Atypical Rt being a leading indicator for COVID-19 outbreaks. While the relationship between Atypical Rt and COVID-19 may not always be this beautifully clear, Atypical Rt can and should be used as an early warning indicator—a signal that people are spreading infectious illness before it becomes apparent in the confirmed case numbers

Figure 1: Kinsa Rt compared with Texas daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 2: Kinsa Rt compared with Arizona daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 3: Kinsa Rt compared with Florida daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Missouri and Utah

Kinsa cited Missouri and Utah as “states where daily case counts will climb, though absolute case numbers may not look out of control.” 

Did our predictions come to pass? – Yes. Both states have continued to see steady increases in COVID-19 confirmed cases, though case numbers per capita are still lower than some of the harder hit states like Texas.

At the time of our predictions, the states showed both elevated Rt levels and recent case growth that caused us some concern — not enough to predict an out of control outbreak, but enough that we felt they should be monitored closely.

As we predicted, cases in Missouri have steadily increased since 6/29 (see figure 4). Rt has only recently dropped below expected transmission levels, suggesting that there may be some case growth in front of them, but that a peak is likely coming.

Though daily COVID-19 cases in Utah initially increased in line with our prediction, they soon plateaued, unlike Missouri (see figure 5). We observed increasing transmission levels until about 7/8, and while levels have declined, they are still above 1. At the same time, COVID case transmission has remained steady. We have no clear indication of what will happen next here.

Figure 4: Kinsa Rt compared with Missouri daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 5: Kinsa Rt compared with Utah daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Indiana, Illinois, Wisconsin, Michigan, and Pennsylvania

Kinsa’s prediction called the “Midwest an area to watch closely… seeing ‘both increasing atypical Rt and increasing fever counts’ in Indiana, Illinois, Wisconsin and Michigan, suggesting a forthcoming rise in confirmed cases. Pennsylvania, too, ‘might see an uptick’.”

Did our predictions come to pass? – Yes. All states saw steadily increasing confirmed COVID-19 cases.

Each of these states has seen daily case growth continue on a steady trajectory since 6/29 (see figures 6 – 10). While none have extremely high daily new case counts currently, we expect they will continue to experience case increases in the near term, with a few states likely peaking within the next two weeks.

Figure 6: Kinsa Rt compared with Indiana daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 7: Kinsa Rt compared with Illinois daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 8: Kinsa Rt compared with Wisconsin daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 9: Kinsa Rt compared with Michigan daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Figure 10: Kinsa Rt compared with Pennsylvania daily COVID-19 cases from 6/1/20 through 7/28/20.

 

Conclusion

Kinsa’s predictions have held up well, particularly in states with high per-capita cases of COVID-19, in states where Kinsa’s Rt trends were clear, and in states with large population centers where our signal strength is strongest. While we continue to learn more about how various Kinsa data measures relate to COVID-19 case growth, we have strong evidence to suggest that Kinsa data can be used to accurately predict the directionality and relative timing of COVID-19 case growth across states.

With real-time knowledge of increasing transmission, decision-makers can enact interventions to slow downstream effects of contagious illness. As we continue to explore our data’s relationship to COVID-19, we look forward to publishing more state-level data, including COVID-19-specific metrics, a signal strength indicator, and more.

 

Notes: