Modeling between-population variation in COVID-19 dynamics in Hubei, Lombardy, and New York City

Significance We present an individual-level model of severe acute respiratory syndrome coronavirus 2 transmission that accounts for population-specific factors such as age distributions, comorbidities, household structures, and contact patterns. The model reveals substantial variation across Hubei, Lombardy, and New York City in the dynamics and progression of the epidemic, including the consequences of transmission by particular age groups. Across locations, though, policies combining “salutary sheltering” by part of a particular age group with physical distancing by the rest of the population can mitigate the number of infections and subsequent deaths.

suggest that encounter rate data from mobile phones may be a reasonable proxy for the reduction in physical contacts; our 170 posterior mean estimate for δc in New York City was 0.97, while Unacast encounter data showed a peak reduction in the 171 encounter rate of 90-99% across the various boroughs of New York City. 172 We set the prior range for p inf to be [0.03, 0.07], again set to include all values with non-negligible likelihood. d mult was 173 given a uniform prior over the range [1,4], allowing for but not mandating a higher IFR than Hubei. We handled the starting 174 conditions of the epidemic differently in New York City than in Lombardy due to reports of multiple distinct importation sheltering by a single age group. For each age group, we simulate the impact of 25%, 50%, or 75% of the members of that age 191 group sheltering (in addition to physical distancing at each level δ second by the rest of the population). We run the simulation 192 until the end of 2021 to ensure that the epidemic has had time to run its course in all scenarios and report the final median 193 number of new infections and deaths for each scenario. 194 We also simulate a corresponding set of scenarios where the population begins in a completely susceptible state. For these 195 simulations, population-level parameters are sampled from the posterior distribution as in the second-wave scenarios, but the 196 population is initialized to be completely susceptible (apart from the randomly-sampled initially infectious individuals, as in not sheltering) by a factor of two. Here, we consider specific ways that this kind of physical distancing could be achieved by

Sports, exercise, and recreation
Gyms could stagger the times that individuals use the facility and take steps to limit close 240 contacts, e.g., ensuring appropriate spacing between people in a group exercise class. Improving ventilation in gyms may also 241 be important to avoid increased transmission rates due to physical exertion in an enclosed space.

Attending or hosting social events
Restaurants and bars could be required to maintain appropriate spacing between parties.

243
Events involving more than a specific number of people could still be restricted.

Religious and spiritual activities
In order to reduce the density of contacts at religious services, possible measures include  -depending on the age category, occupation type, and industry -were already more likely to work from home than others 255 before this pandemics. Taking these factors into consideration helps rate the feasibility of our proposed policies and provides 256 guidance for immediate intervention.

Age
Half of telecommuters in the US are 45 years of age or older, as compared to just 41% of the overall workforce (61).

258
Specifically, the greatest disproportional participation in home-based work is observed for the following age groups: 65+ years 259 old (odds ratio, OR=1.7) and 55-64 years old (OR=1.2). personal care and service (OR=2.0), and business and financial operations (OR=1.9).

Industry
The largest share of telecommuters is due to three industries (61): professional, scientific, and technical services 266 (17%); healthcare and social assistance (11.6%); and finance and insurance (9.7%). The prevalence in telecommuting vs.

Access to telecommuting options
In the US, having the possibility to telecommute is mostly a function of both the worker's 270 company size and employment status. Telecommuting is more prevalent among large vs. small companies. 12% of firms with 271 more than 500 employees offer such a possibility, whereas only 5% of those with less than 100 employees do so. Moreover, 272 employers are more likely to offer the option to work from home to their full-time rather than part-time employees (8% vs. 2% 273 respectively as of 2017). It is expected that these tendencies would be confirmed in other industrialized countries as well.  The 90% credible interval of the predictive posterior includes contains the held-out data at almost all points, including when the model is fit using only data from the earliest portion of the epidemic. The model over-predicts deaths early in the epidemic, though the timing of the peak is correctly captured early on. Much of the over-prediction is corrected with additional training data even before the peak is observed. . We find that for all populations, 25% or less contact is sufficient to suppress the epidemic. At 50% contact, a significant portion of each population becomes infected (approximately 10-40% depending on the population, which group shelters, and what fraction of that group shelters). Across populations, sheltering by the 20-40 and 40-60 age groups reduces infections by the largest amount; sheltering by the 60+ group has only a minor impact. given level of physical distancing by the entire population (specified as the percentage of normal contact levels). The x-axis within each figure gives the fraction of a single age group which adopts salutary sheltering. Each bar represents a scenario where the given fraction of a single age group adopts salutary sheltering, with the color of the bar representing the identity of the group (see legend). In scenarios with 25% or less contact, the outbreak is effectively surpressed (see Figure S5) resulting in correspondingly few deaths. At 50% contact, the larger number of infections results in a larger number of deaths. For Hubei and New York City at 50% contact, deaths are reduced more effectively via sheltering by the 20-40 or 40-60 groups than by the 60+ group. In Lombardy, sheltering by the 60+ group is always the most effective at reducing deaths but the margin between the number of deaths under sheltering by the 60+ group compared to other groups is smaller under 50% contact than under higher contact levels. At 75% or higher contact, this pattern is replicated in Hubei and New York City, where sheltering by the 60+ group has the greatest marginal impact on deaths and the gap between the 60+ and other groups is larger at 100% contact than at 75%.  Figure S5, 25% contact is not always sufficient to suppress a widespread outbreak. This reflects two factors. First, the importance of acquired immunity accumulated during the first outbreak in reducing the effective reproduction number. Second, the potential for the timing of when interventions are applied to influence the total extent of infections. E.g., it is possible for the total number of eventual infections to be lower when more people are infected in the first wave than when more stringent control measures are imposed from the start (66). At 50% contact and above, the dynamics become more similar to the second-wave scenarios, with substantial fractions of each population infected. As in Figure S5, sheltering by the 20-40 and 40-60 age groups reduces infections by the largest amount; sheltering by the 60+ group has only a minor impact. the fraction of a single age group which adopts salutary sheltering. Each bar represents a scenario where the given fraction of a single age group adopts salutary sheltering, with the color of the bar representing the identity of the group (see legend). Deaths are limited by contact levels at 25% or lower. As in the second-wave scenarios for Hubei and New York City, at low levels of contact, sheltering by the 20-40 and 40-60 age groups is more effective at reducing deaths than sheltering by the 60+ group. However, due to the larger number of infections at a given contact level in the completely-susceptible scenario as compared to the second-wave scenario, a lower level of overall contact is sometimes needed to realize this effect (25% contact in New York City instead of 50%). Once contact levels rise to 50%, only Hubei shows greater effectiveness for sheltering by the 20-40 and 40-60 age groups, and at 75% contact it is more effective for the 60+ age group to shelter for all populations.  Log-standard deviation time to progress from exposed to mild 0.418 (12) λm→s Mean time to progress from mild to severe 7 days (67) λs→c Mean time to progress from severe to critical 7.5 days (using 14.5 days from onset to mechanical ventilation in (4) First date with at least 5 infected individuals Free parameter † This setting for α is likely pessimistic in that Li et al.'s estimate for reduction in transmissibility is for undocumented cases, including asymptomatic cases, presymptomatic cases, and those with limited symptoms (8). Future work should examine the impact of a potentially lower α as better information on transmissibility in the asymptomatic or presymptomatic state becomes available.

Location
Poisson AIC Negative binomial AIC       Table S8. Infections (in thousands) for a fully susceptible population in New York City. Each major row heading denotes the age group which adopts salutary sheltering, and the sub-headings denote the fraction of the group which shelters. The major column headings give the level of contact amongst individuals who do not shelter. The entry "Total" gives the median number of total infections (in thousands) in each scenario, while "0-59" and "60+" give the median number of total infections in each segment of the population (under or over 60 years of age).  Table S9. Deaths (in thousands) for a second-wave scenario in Hubei. Each major row heading denotes the age group which adopts salutary sheltering, and the sub-headings denote the fraction of the group which shelters. The major column headings give the level of contact amongst individuals who do not shelter. The entry "Total" gives the median number of total deaths (in thousands) in each scenario, while "0-59" and "60+" give the median number of total deaths in each segment of the population (under or over 60 years of age       Table S13. Deaths (in thousands) for a fully susceptible population in Lombardy. Each major row heading denotes the age group which adopts salutary sheltering, and the sub-headings denote the fraction of the group which shelters. The major column headings give the level of contact amongst individuals who do not shelter. The entry "Total" gives the median number of total deaths (in thousands) in each scenario, while "0-59" and "60+" give the median number of total deaths in each segment of the population (under or over 60 years of age  Table S14. Deaths (in thousands) for a fully susceptible population in New York City. Each major row heading denotes the age group which adopts salutary sheltering, and the sub-headings denote the fraction of the group which shelters. The major column headings give the level of contact amongst individuals who do not shelter. The entry "Total" gives the median number of total deaths (in thousands) in each scenario, while "0-59" and "60+" give the median number of total deaths in each segment of the population (under or over 60 years of age      55. NBC News, Walmart will limit customers and create one-way traffic inside its stores (2020) https://www.nbcnews.com/news/ 378 us-news/walmart-will-limit-customers-create-one-way-traffic-inside-its-n1176461.