Differential effects of intervention timing on COVID-19 spread in the United States

Abstract

Assessing the effects of early nonpharmaceutical interventions on coronavirus disease 2019 (COVID-19) spread is crucial for understanding and planning future control measures to combat the pandemic. We use observations of reported infections and deaths, human mobility data, and a metapopulation transmission model to quantify changes in disease transmission rates in U.S. counties from 15 March to 3 May 2020. We find that marked, asynchronous reductions of the basic reproductive number occurred throughout the United States in association with social distancing and other control measures. Counterfactual simulations indicate that, had these same measures been implemented 1 to 2 weeks earlier, substantial cases and deaths could have been averted and that delayed responses to future increased incidence will facilitate a stronger rebound of infections and death. Our findings underscore the importance of early intervention and aggressive control in combatting the COVID-19 pandemic.

Fig. 1. Model fit and parameter inference.

Posterior fitting to daily cases and deaths in the United States (A and B) and the New York metropolitan area (C and D). Orange dots represent observations. Blue and gray lines are the median estimate and 95% CIs, respectively. The estimated effective reproductive number, R_e, in six metropolitan areas are shown in (E). The black dotted line indicates R_e = 1. (F) The estimated ascertainment rate over time. The blue line and gray dashed lines are the median estimates and 95% CIs, respectively. (G) The estimated cumulative infections (both reported and unreported) in six metropolitan areas. We compare the reported seroprevalence (%) in nine locations on different dates with the inferred percentage cumulative infections on those dates in (H). Whiskers show 95% CIs. Details on the serological survey are provided in Materials and Methods.

Fig. 2. Asynchronous reduction of effective reproductive numbers.

For each county, we show the date when the local effective reproductive number dropped below 1.5 (A), 1.25 (B), 1 (C), and 0.75 (D) and stayed below that threshold until 3 May. Counties in gray are those that either never reached the threshold or failed to remain below the threshold.

Fig. 3. Counterfactual simulations with control interventions beginning in early March, 1 and 2 weeks earlier than implemented.

Daily cases and deaths in the United States (A, B, E, and F) and the New York metropolitan area (C, D, G, and H) under early interventions are compared with the observations (orange crosses). The top and bottom rows present counterfactuals with interventions implemented on 8 and 1 March, respectively. The black lines and surrounding bands show the median estimate, interquartile, and 95% CIs.

Fig. 4. Effects of response time after control measures are relaxed.

We assume a control relaxation (a weekly 5% increase of the transmission rate) starting on 4 May in all U.S. counties. If the local weekly case number in a county increases for 2 or 3 consecutive weeks, a weekly 25% reduction of the transmission rate is imposed for that county. Daily cases and deaths in the United States for a response time of 2 weeks (A and C) and 3 weeks (B and D) are compared. The black lines and bands show the median estimate, interquartile, and 95% CIs.