Using test positivity and reported case rates to estimate state-level COVID-19 prevalence in the United States

Accurate estimates of the prevalence of infection are essential for evaluating and informing public health responses to the ongoing COVID-19 pandemic in the United States (US), but reliable, timely prevalence data based on representative population sampling are unavailable, and reported case and test positivity rates may provide strongly biased estimates. A single parameter semi-empirical model was developed, calibrated, and validated with prevalence predictions from two independent data-driven mathematical epidemiological models, each of which was separately validated using available cumulative infection estimates from recent state-wide serological testing in 6 states. The analysis shows that individually, reported case rates and test positivity rates may provide substantially biased estimates of COVID-19 prevalence and transmission trends in the U.S. However, the COVID-19 prevalence for U.S. states from March-July, 2020 is well approximated, with a 7-day lag, by the geometric mean of reported case and test positivity rates averaged over the previous 14 days. Predictions of this semi-empirical model are at least 10-fold more accurate than either test positivity or reported case rates alone, with accuracy that remains relatively constant across different US states and varying testing rates. The use of this simple and readily-available surrogate metric for COVID-19 prevalence, that is more accurate than test positivity and reported case rates and does not rely on mathematical modeling, may provide more reliable information for decision makers to make effective state-level decisions as to public health responses to the ongoing COVID-19 pandemic in the US.