Projections of COVID-19 in the US at county level generated based on latest data.

Generated by Shaman group.

Projection visualizations, descriptions, and associated publications can be found here: https://blogs.cuit.columbia.edu/jls106/publications/covid-19-findings-simulations/

County-level case and death data are compiled from Johns Hopkins University and USAFACTS. https://coronavirus.jhu.edu/us-map https://usafacts.org/visualizations/coronavirus-covid-19-spread-map/

Projections are generated for daily new confirmed case, daily new infection (both reported and unreported), cumulative demand of hospital beds, ICU and ventilators as well as daily mortality (2.5, 25, 50, 75 and 97.5 percentiles).

Projections for daily new confirmed case and daily new infection are reported in Projection_*.csv with column headings 'report' and 'total', respectively.

Projections for cumulative demand of hospital beds, ICU, ventilators and daily mortality are reported in bed_*.csv Note that hospitalizations and ICU bed numbers indicate the current demand on a given day (NOT just the new persons needing beds).

The most recent projection files and estimates of R are also found in LatestProjections. Projection files in this folder are assigned the names high, low, and mid. The metadata files in the folder explain the corresponding scenarios.

We generate projections based on a number of hypothetical scenarios describing changes in disease transmission rates due to individual behavior and government policies.

• nochange - Assumes that current contact rates will remain unchanged in the future.

• 5_1xbeta (starting 2020-11-12) - This scenario assumes a one-time 5% increase in contact rates at the start of the projection. The following week, the effective reproduction number R is set to 1. As the simulation progresses, R_eff decreases as the fraction of susceptible individuals in the population decreases.

• 5_2xbeta (starting 2020-11-12) - This scenario assumes a weekly 5% increase in contact rates for two weeks. The following week, the effective reproduction number R is set to 1. As the simulation progresses, R_eff decreases as the fraction of susceptible individuals in the population decreases.

• 5_1xhold (discontinued 2020-11-11) - This scenario assumes a one-time 5% increase in contact rates at the start of the projection. The following week, the reproduction number R is set to 1 for the remainder of the projection.

• 5_2xhold (discontinued 2020-11-11) - This scenario assumes a weekly 5% increase in contact rates for two weeks. The following week, the reproduction number R is set to 1 for the remainder of the projection.

• season4 (starting 2020-10-18) - This scenario assumes seasonal changes to R(t). In the summer months, it assumed a seasonal decrease in disease transmission leading to a weekly 4% decrease in reproductive number R(t). In the winter, it assumes a weekly 2% increase in reproductive number R(t).

• reduce10 - We project a weekly 10% decrease in contact rates.

We have initiated the following scenarios:

• nochange - Assumes that current contact rates will remain unchanged in the future.

• 5_1x (2020-06-14 through 2020-07-17; 2020-08-23 through 2020-09-27) - We project a one-time 5% increase in contact rates due to loosening restrictions as states continue to reopen economically, applied at the beginning of the projection.

• 5_2xhold - This scenario assumes a weekly 5% increase in contact rates for two weeks. The following week, the reproduction number R is set to 1 for the remainder of the projection.

• season4 (2020-06-14 through 2020-08-30) - This scenario assumes seasonal changes to R(t). In the summer months, it assumed a seasonal decrease in disease transmission leading to a weekly 4% decrease in reproductive number R(t).

• reduce10 (starting 2020-07-18) - We project a weekly 10% decrease in contact rates.

We modified the Adaptive Interventions control scenarios to account for increases in contact rates due to loosening restrictions in states that have begun to reopen economically. We project four scenarios. In states maintaining or increasing current social distancing restrictions, we continue to apply the 20% weekly reductions of contact rates, as in the 20% contact reduction Adaptive intervention scenario described below. However, in states that reopen, we apply an increase to the contact rate of counties in those states. In the first scenario, we apply a one-time 10% increase to the contact rate during the week that the state is scheduled to reopen and maintain this new increased rate for the remainder of the projection. In the second scenario, the contact rate is increased by an additional 10% each week to represent progressive loosening of restrictions and increased public confidence and frequenting of businesses. The third and fourth scenarios are the same as the first and second scenarios, but with a 5% rather than 10% increase in contact rate in reopening states.

The scenario names used in the Projection and Bed output files for these interventions are:

• 80contact1x10p - Weekly 20% decrease in places with growing weekly cases and a one-time 10% increase in places with return to work

• 80contactw10p - Weekly 20% decrease in places with growing weekly cases and a weekly 10% increase in places with return to work

• 80contact1x5p - Weekly 20% decrease in places with growing weekly cases and a one-time 5% increase in places with return to work

• 80contactw5p - Weekly 20% decrease in places with growing weekly cases and a weekly 5% increase in places with return to work

Four different adaptive scenarios of contact reduction were projected, 0% (no contact reduction via social distancing controls and behavior change), 20%, 30%, and 40% contact reduction. These scenarios are meant to mimic adaptive adjustments in contact imposed both by government regulations (e.g. school closures, restrictions on mass gatherings) and population self-regulation (e.g. isolation, mask wearing, social distancing). Upon initiation of projections on a given day, all counties with 10 or more confirmed cases impose a 0%, 20%, 30% or 40% contact rate reduction, depending on the projection scenario; all other counties implement no contact reduction. Each week following, if a county newly exceeds 10 confirmed cases, a 0%, 20%, 30% or 40% contact rate reduction is imposed, depending on the projection scenario. Counties that had previously exceeded 10 confirmed cases and also experience an increase in the number of weekly reported confirmed cases impose a further, multiplicative 0%, 20%, 30% or 40% contact rate reduction, depending on the projection scenario. Counties with fewer than 10 confirmed cases continue not to implement control. This multiplicative ratcheting of contact reduction levels is continued until the end of simulations and is meant to represent increasing reactive social distancing imposed within counties as long as confirmed weekly cases of COVID-19 continue to rise.

Scenario names used in output files are Projection_nointerv.csv, Projection_80contact.csv, Projection_70contact.csv and Projection_60contact.csv for 0%, 20%, 30% or 40% contact rate reduction, respectively.

Several qualifications with respect to these projections must be noted and considered during interpretation. Firstly, the model is optimized using observations through the projection date; however, those observations, i.e. confirmed cases by county, represent infections that were acquired by individuals 1-2 weeks earlier during a time prior to the implementation of many of the social distancing and isolation measures in place on the projection date. Because of this long delay between infection acquisition and case confirmation, any flattening of the curve due to these effects is not yet apparent in observations nor communicated to the model during optimization. Consequently, the no intervention scenario roughly represents the transmission potential of the virus around March 12th projected forward. As many new control policies were effected after March 12, 2020, the contact reduction projections may depict paths that some counties are already following and thus may be more representative of future activity. In this fashion, the 3 scenarios provide references against which the effectiveness of control measures already enacted can be assessed. We also ask readers to note the cones of uncertainty associated with each scenario projection, which broaden into the future.

Secondly, the landscape to which this model has been optimized is highly variable in space and time, due to differences in contact behavior, population density, control measures and testing practices. These differences in space and time make the fitting of any model of this scale challenging. In this shifting landscape, it is very important to reiterate that it will take 10-14 days before the effects of real-world interventions—any flattening of the curve weeks—become apparent.

Please cite: Sen Pei, Jeffrey Shaman, Initial Simulation of SARS-CoV2 Spread and Intervention Effects in the Continental US. medRxiv.doi: https://doi.org/10.1101/2020.03.21.20040303