key: cord-0295796-zmiiq2d3 authors: Runge, M.; Richardson, R. A. K.; Clay, P.; Eagan, A.; Holden, T. M.; Singam, M.; Tsuboyama, N.; Arevalo, P.; Fornoff, J.; Patrick, S.; Ezike, N. O.; Gerardin, J. title: Modeling robust COVID-19 intensive care unit occupancy thresholds for imposing mitigation to prevent exceeding capacities date: 2021-07-03 journal: nan DOI: 10.1101/2021.06.27.21259530 sha: 298a7a78fa37af22e75fc6b09284f3f423803807 doc_id: 295796 cord_uid: zmiiq2d3 In managing COVID-19 with non-pharmaceutical interventions, occupancy of intensive care units (ICU) is often used as an indicator to inform when to intensify mitigation and thus reduce SARS-CoV-2 transmission, strain on ICUs, and deaths. However, ICU occupancy thresholds at which action should be taken are often selected arbitrarily. We propose a quantitative approach using mathematical modeling to identify ICU occupancy thresholds at which mitigation should be triggered to avoid exceeding the ICU capacity available for COVID-19 patients. We used a stochastic compartmental model to simulate SARS-CoV-2 transmission and disease progression, including critical cases that would require intensive care. We calibrated the model for the United States city of Chicago using daily COVID-19 ICU and hospital census data between March and August 2020. We projected ICU occupancies from September to May 2021 under two possible levels of transmission increase. The effect of combined mitigation measures was modeled as a decrease in the transmission rate that took effect when projected ICU occupancy reached a specified threshold. We found that mitigation did not immediately eliminate the risk of exceeding ICU capacity. Delaying action by 7 days increased the probability of exceeding ICU capacity by 10-60% and this increase could not be counteracted by stronger mitigation. Even under modest transmission increase, a threshold occupancy no higher than 60% was required when mitigation reduced the reproductive number Rt to just below 1. At higher transmission increase, a threshold of at most 40% was required with mitigation that reduced Rt below 0.75 within the first two weeks after mitigation. Our analysis demonstrates a quantitative approach for the selection of ICU occupancy thresholds that considers parameter uncertainty and compares relevant mitigation and transmission scenarios. An appropriate threshold will depend on the location, number of ICU beds available for COVID-19, available mitigation options, feasible mitigation strengths, and tolerated durations of intensified mitigation. 2 transmission, strain on ICUs, and deaths. However, ICU occupancy thresholds at which action 23 should be taken are often selected arbitrarily. We propose a quantitative approach using 24 mathematical modeling to identify ICU occupancy thresholds at which mitigation should be triggered 25 to avoid exceeding the ICU capacity available for COVID-19 patients. We used a stochastic 26 compartmental model to simulate SARS-CoV-2 transmission and disease progression, including 27 critical cases that would require intensive care. We calibrated the model for the United States city of 28 Chicago using daily COVID-19 ICU and hospital census data between March and August 2020. We 29 projected ICU occupancies from September to May 2021 under two possible levels of transmission 30 increase. The effect of combined mitigation measures was modeled as a decrease in the 31 transmission rate that took effect when projected ICU occupancy reached a specified threshold. We 32 found that mitigation did not immediately eliminate the risk of exceeding ICU capacity. Delaying 33 action by 7 days increased the probability of exceeding ICU capacity by 10-60% and this increase 34 could not be counteracted by stronger mitigation. Even under modest transmission increase, a 35 threshold occupancy no higher than 60% was required when mitigation reduced the reproductive 36 number Rt to just below 1. At higher transmission increase, a threshold of at most 40% was required 37 with mitigation that reduced Rt below 0.75 within the first two weeks after mitigation. Our analysis 38 demonstrates a quantitative approach for the selection of ICU occupancy thresholds that considers 39 parameter uncertainty and compares relevant mitigation and transmission scenarios. An appropriate 40 threshold will depend on the location, number of ICU beds available for COVID-19, available 41 mitigation options, feasible mitigation strengths, and tolerated durations of intensified mitigation. In the first half of 2020, the global spread of SARS-CoV-2 left many countries with no option other 48 than to shut down their economies and encourage people to isolate by staying home. In the United 49 States (US), stay-at-home policies implemented in late March and April of 2020 reduced the number 50 of new infections and deaths [1] . In mid-2020, US states began to relax their stay-at-home policies 51 [1, 2] despite a lack of effective treatments or a vaccine. In late 2020, many states experienced 52 epidemic waves as large as, or larger than, their initial epidemics, putting renewed strain on hospital 53 resources and requiring new mitigation measures [1] [2] [3] . 54 55 Intensive care resources, particularly staffed beds and ventilators, are limited [4, 5] especially in rural 56 areas [6, 7] . In early 2020, many intensive care units (ICUs) in the US and other countries operated 57 near and above capacity limits [8] [9] [10] [11] [12] . To ensure continued life-saving care and a functioning health 58 system, ICU occupancies must stay below capacity, and multiple guidelines for managing ICU 59 capacities during COVID-19 surges have been formulated [5, [13] [14] [15] . 60 61 In response to fluctuations in SARS-CoV-2 transmission, states formulated COVID-19 response 62 strategies to guide transitions between mitigation and relaxation policies [1] . These mitigation and 63 relaxation policies defined setting-specific COVID-19 prevention measures such as occupancy limits 64 for businesses, constraints on indoor activities, work from home recommendations, or population-65 wide stay-at-home orders ('lockdowns'). For instance, in the US state of Illinois, thresholds used to 66 spur increasing mitigation measures included test positivity rate (if surpassing 8%), increasing or 67 decreasing trends in occupied hospital beds, and total ICU bed availability (if below 20%) [16] . The 68 selection of robust yet sensitive thresholds to trigger a strategic mitigation response is challenging 69 but critical, as health departments require time to appropriately prepare for and respond to a 70 potential increase in transmission and hospital bed demand but prefer not to impose unnecessary 71 mitigation (Fig 1) . Thresholds that are too low could lead to premature restrictions or harmful effects 72 on the economy and the community due to unnecessarily remaining under mitigation for too long. 73 Thresholds that are too high could lead to late action, strained hospital resources, and elevated rates 74 of severe COVID-19 cases and deaths. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 3, 2021. The first SARS-CoV-2 infection was reported in Chicago in mid-January 2020 [21]. On March 21, 2020, 105 a statewide stay-at-home order was announced to contain the spread of the virus. The stay-at-home 106 order was gradually relaxed at the end of May 2020, and restaurants and recreational locations were 107 allowed to reopen at the end of June [30] . Although the number of reported cases stayed relatively 108 low during the summer, transmission increased during fall and on November 20, 2020, a second stay 109 at home order was issued [31] . Time-varying detection rates for severe and mild cases were taken from an analysis of Illinois case 130 and death data [35] (S1 Fig 2- from literature [37, 38] . Other parameters were based on research studies outside Illinois (S1 Table 136 1). Triggered mitigation measures after October 1, 2020, were applied as a decrease in transmission 139 rate, non-specific to the mitigation measures that would cause this decrease (such as the closure of 140 retail business, stricter mask-wearing protocols, or shelter-in-place). We set a feedback loop 141 between the population in the critical detected (Cd) compartment (COVID-19 ICU occupancy) and the 142 transmission rate parameter, such that Cd triggers mitigation at specified occupancy thresholds (Fig 143 2 ). 144 145 146 Time-varying transmission rate prior to September 1, 2020, was fit to confirmed daily COVID-19 ICU 147 census and COVID-19 med/surg hospital census in Chicago between February and August 2020 (Fig 148 3A , S1 Fig 6-8) . The census data included all confirmed COVID-19 patients currently occupying ICU or 149 med/surg beds in Chicago hospitals, and no data was available on location of residence for individual 150 patients. Each of the two data series was smoothed with a 7-day centered moving average prior to 151 comparison with simulation outputs. The time-varying reproductive number Rt was calculated from 152 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 3, 2021. smoothed using a rolling average for 3 days to reduce high variation in these estimates (Fig 3) . In the fitting process, we first estimated the infection importation date (date with 10 infections), 169 initial transmission rate, and transmission rate under mitigation for March 2020. We then fitted 170 twice-monthly adjustments to the transmission rate between April and August. Other parameters 171 were set to their mean value according to local data or epidemiological studies (S1 Table 2 ). Best fit 172 parameter combinations were those that minimized the negative log likelihood of the simulated 173 trajectories, based on a Poisson distribution. In the fitting, ICU census and med/surg census were 174 weighted equally. The model fit was validated against COVID-19-like illness (CLI) hospital admissions 175 data for Chicago hospitals and against COVID-19 deaths from I-NEDSS with Chicago listed as county 176 or ZIP code of residence. To account for parameter uncertainty, we ran simulations with 400 unique 177 parameter combinations using fitted parameter ranges and, for the data-informed parameters, 178 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.27.21259530 doi: medRxiv preprint values sampled from uniform distributions. We then chose 100 trajectories (unique set of 179 parameters) that best fit the ICU census data and used these parameter sets in the later analysis. 180 To describe the fitting accuracy we calculated the mean absolute error (MAE) [41], using the metrics 181 R package [42] , for the median prediction compared against the weekly moving average of the data 182 (S1 Fig 7) . Simulated scenarios 185 Based on daily ICU occupancies and total bed availability in Chicago, we calculated that in 2020, on 186 average 44% of all ICU beds were occupied by non-COVID patients, theoretically leaving 56% of beds 187 available for COVID-19 patients. The average number of ICU beds available for COVID-19 patients 188 during the week immediately preceding September 15, 2020, (516 ICU beds) will be referred to as 189 ICU capacity in this work. We assumed the capacity of 516 ICU beds to stay constant capacity during 190 the simulation period, whereas in practice the capacity ranged between 407 to 744 beds on a seven-191 day rolling average (Fig 3A, S1 Fig 14) . We imposed a gradual increase in the transmission rate beginning on September 1, 2020, and Between October 2020 and May 2021, mitigation (immediate reduction in transmission rate) was 206 triggered either one or seven days after the COVID-19 ICU occupancy threshold was reached. 207 Mitigation was simulated to reduce the transmission rate by 20, 40, 60, or 80% ('weak', 'moderate', 208 'strong', or 'very strong'). Once applied, changes in transmission rate due to mitigation were never 209 reversed. A table comparing the assumed transmission increase and reduction values to other 210 studies is included in the Supplement (S1 Table 5 ). We explored scenarios in which we varied the increase in transmission (two levels), mitigation 213 effectiveness (four levels), and mitigation delay (two levels), and ICU occupancy threshold that 214 triggered mitigation (eleven levels), resulting in a total of 176 unique scenarios (Table 1) . Each 215 scenario was simulated with 400 sets of sampled parameters, drawn from uniform distributions (S1 216 Table 6 ). The top 100 trajectories that best fit to ICU census data up to September 1, 2020, were 217 retained for each of the 176 scenarios. Trajectories in which the ICU occupancy threshold to trigger 218 mitigation was not reached by May 2021 were excluded (5-6%, S1 Table 6 ). The sampled parameters 219 were summarized using the mean and 90% prediction interval (PI). Daily COVID-19 ICU occupancy was the primary outcome in the analysis. The probability of exceeding 222 ICU capacity (for COVID-19 patients) was calculated by dividing the number of trajectories that 223 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. Simulating a September 2020 epidemic wave in Chicago 234 We fit a compartmental model of SARS-CoV-2 transmission (Fig 2) to hospitalization and intensive 235 care unit census data from Chicago between March and August 2020 (Fig 3A, S1 Fig 6-7) . The fitted 236 infection importation date was February 28, 2020, with an initial transmission rate of 1.14 and 237 reproductive number R0 of 5.00 (90% prediction interval (PI) 4.57-5.41). After the stay-at-home 238 order starting on March 22, 2020, we estimated a 92.5% reduction in the transmission rate (Fig 3B) , 239 reducing the time-varying reproductive number (Rt) to 0.74 (90% PI 0.72-0.77) (Fig 3C) . At the end of is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. Preventing ICU overflow strongly depends on ICU occupancy threshold for action 258 Mitigation was triggered when the simulated ICU occupancy reached a pre-defined threshold 259 relative to the ICU capacity (Fig 4) . Stronger mitigation (>60% reduction in transmission rate) led to 260 lower peak ICU occupancy and peak occurred sooner (Fig 4A, 4B) . At the higher level of transmission 261 increase, new infections dropped after mitigation was triggered (S1 Fig 13) and estimated Rt reached 262 a minimum after around two weeks, before increasing again and leveling off below 1. The estimated 263 Rt varied slightly across the simulated scenarios, and at the high transmission increase scenario, the 264 Rt two weeks prior mitigation was estimated at of 1.23 (90% PI: 1.18-1.29) and was reduced to 1.03 265 (90% PI: 0.99-1.07) at weak, to 0.92 (90% PI: 0.88-0.96) at moderate, to 0.75 (90% PI: 0.71-0.79) at 266 strong, and to 0.47 (90% PI: 0.40-0.54) at very strong mitigations (Fig 4D, S1 Fig 18) . Compared to no mitigation, immediate mitigation decreased peak ICU occupancy by 34.5% (90% PI: is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. ICU occupancy continued to grow for a short time after mitigation was imposed (Fig 5A) . At the same 285 mitigation strength, peak ICU occupancy was reached at a similar length of time (12 days) after 286 mitigation regardless of the threshold ICU occupancy. Lower occupancy thresholds for triggering 287 mitigation led to a lower probability of ICU overflow and lower peak ICU occupancy (Fig 5B) . We calculated the probability of exceeding ICU capacity under different possible ICU occupancy 300 thresholds at which mitigation was dynamically triggered (Fig 5B) . We compared the probabilities by 301 transmission level, mitigation strengths, and delay between trigger and reduction in transmission 302 due to mitigation. The probability of overflow increased with a higher ICU occupancy threshold. The 303 probability was, on average across the ICU occupancy thresholds, 33% (range across mitigation 304 strengths: 17%-63%) higher at the higher level of transmission increase compared to the lower level. 305 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.27.21259530 doi: medRxiv preprint Weak mitigation (20% decrease in transmission rate), where Rt was not reduced below 1, had a 306 substantially higher probability of overflow than the other mitigation levels, and differences were 307 greater for the higher level of transmission increase. At the lower level of transmission increase, the probability of ICU overflow was almost identical for 310 moderate, strong, and very strong mitigation, whereas at the higher transmission increase the 311 difference between the mitigation levels was more pronounced with consistently high probability of 312 overflow for weak mitigation. At the lower level of transmission increase, the probability of ICU 313 overflow increased at thresholds above 30% occupancy when mitigation was weak, whereas when 314 mitigation was moderate or stronger, the probability remained near zero until an ICU occupancy 315 threshold of 60-70% after which the probability increased sharply. The incremental difference in the 316 overflow probability between weak and moderate mitigation was 10% and less than 3% for the 317 other mitigation strengths at the low increase level. In comparison, at the high increase level, the 318 probability of exceeding capacity increased at thresholds above 40-50% occupancy and reached 319 100% for occupancy thresholds above 60-70% at strong and very strong mitigation. The difference in 320 the overflow probability was 42% between weak and moderate mitigation, 16% between moderate 321 and very strong mitigation, and negligible between strong to very strong mitigation (<1%) (Fig 5, Fig 322 S1 Fig 14) . A delay of seven days shifted the probability curves to the left, with higher probability of overflow at 325 each of the ICU occupancy thresholds. For instance, at an 80% ICU occupancy threshold and very 326 strong mitigation, the probability of overflow increased from 41.8% to 89.6% under the lower level 327 of transmission increase when mitigation was delayed by seven days. At the higher level of 328 transmission increase, a 60% occupancy threshold and very strong mitigation had 37% probability of 329 overflow if mitigation was immediate but 97.4% probability of overflow if mitigation was delayed. 330 When assessing mitigation strengths against delay, the probability of overflow was higher for strong 331 mitigation that was delayed by seven days compared to moderate mitigation with immediate action. 332 For a hypothetical risk tolerance of 25% probability of ICU overflow, the required ICU occupancy 333 thresholds for action were 40 to 60% across the tested scenarios. A policy of 80% ICU occupancy to trigger mitigation did not prevent exceeding capacity, as mean 336 peak ICU occupancy was at or above ICU capacity for all mitigation strengths and both levels of 337 transmission increase. A 60% ICU occupancy threshold for mitigation was barely sufficient for 338 preventing ICU overflow: mean peak ICU occupancy remained below capacity for the lower 339 transmission increase at all mitigation strengths but remained below capacity for the higher 340 For simulation trajectories where ICU capacity was exceeded, we measured the number of days in 349 each trajectory where ICU occupancy exceeded capacity (Fig 6, S1 Fig 14) . Without mitigation, the 350 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. duration above ICU capacity was on average 81 days at the higher level of transmission increase and 351 68 days at lower transmission increase. Under immediate mitigation, the average duration above 352 ICU capacity for the higher transmission increase was reduced to 78 days with weak mitigation and 353 further decreased under stronger mitigation (to 43, 21, 14 days for moderate, strong, and very 354 strong), averaged across ICU occupancy thresholds. The corresponding average duration above 355 capacity for the lower transmission increase were 26, 14, 10 and 8 days for weak, moderate, strong 356 and very strong mitigation respectively. Higher ICU thresholds, and hence later action, resulted in 357 longer duration above capacity. A delay of seven days in mitigation did not substantially extend the 358 time above capacity beyond the seven days among those trajectories that exceeded the capacity (Fig 359 6B) . In an initially expanding epidemic, ICU occupancy of COVID-19 patients will continue to increase for 375 around two weeks after imposing mitigations. Higher ICU occupancy thresholds for action thus 376 increase the probability of overshooting ICU capacity during those two weeks. Furthermore, 377 mitigation measures could be delayed to give individuals and businesses warning in advance of 378 changing policies. Scaling up of hospital beds and staff might require even longer notice times of 379 three to four weeks [44, 45] . These delays would result in additional hospitalizations and increase the 380 probability of ICU overflow. We found that mitigation strength could not compensate for a delay in 381 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.27.21259530 doi: medRxiv preprint action. Other researchers have also noted a critical window during which policies need to be 382 implemented and have shown that even short delays can result in substantial increase in infections 383 [46]. If immediate action (i.e. less than 3 days) is not feasible, an alternative would be to reduce the 384 threshold for triggering mitigation to allow more time for planning and implementation. Anticipating 385 a delay is crucial when selecting a threshold if ICU capacity is not to be exceeded. This study models mitigation as an abstracted decrease in transmission rate. In current practice, 388 mitigation is achieved through a mix of social distancing, masking, diagnostic testing, isolation, and 389 contact tracing [47] . The strongest mitigation considered in this study reduced transmission to levels 390 below what was observed in Chicago during the stay-at-home order in March 2020, when mitigation 391 relied heavily on social-behavioral changes as access to diagnostic testing was limited and contact 392 tracing had yet to be implemented. Practical implementation of the "very strong" mitigation 393 modeled in this study would therefore require both interventions that reduce contact rates and 394 interventions to promote early diagnosis and isolation. Studies in the US and Canada estimated that 395 social distancing alone would be not enough to prevent ICU overflow during the first epidemic wave 396 of 2020 [19, 20] . Higher mitigation strength might be easier to achieve in higher populated areas 397 than in more sparsely populated areas with less mobility and already relatively low contact rates. 398 However, determining an expected reduction in transmission given specific mitigation plans is 399 challenging since transmission is influenced by many behavioral factors that vary geographically, 400 demographically, and over time. Reductions in between 20% and 95% have been estimated across a 401 range of studies (S1 Table 5 ). In Chicago and other regions in Illinois, the ICU occupancy threshold to spur transition to the next 404 COVID-19 mitigation phase was at 80% of total occupancy (20% total availability), corresponding to 405 around 40% occupancy of beds available for COVID-19 patients [16] . In theory, different 406 geographical areas could have different ICU occupancy thresholds for action tailored to their specific 407 context, determined by factors such as current Rt, anticipated population behavior, ICU flexing 408 capacity, and overall risk tolerance for exceeding ICU capacities. In practice, using region-specific 409 thresholds risks an uncoordinated response [48] , and mitigation might not be as effective due to 410 spillover effects across neighboring regions. During Chicago's October 2020 epidemic wave, 411 mitigation was implemented on November 20 [31] when the COVID-19 ICU occupancy was 53% (S1 412 Fig 16) . However, Rt had already begin to decrease prior to implementation of official mitigation 413 measures as individual action preceded government policy (S1 Fig 17) . Our suggested threshold of 60% for Chicago aligns with thresholds used in a modeling study that 416 simulated multiple on-off cycles based on a fixed 50% ICU occupancy threshold [ Illinois Department of Public Health defined an 80% threshold on total ICU occupancy, translating to 422 a occupancy threshold of around 50% for COVID-19 patients when assuming a maximum 60% 423 occupancy by non-COVID-19 patients. In our analysis, a 40% occupancy threshold is associated with 424 relative low probabilities of overflow. In practice, local health departments monitor multiple 425 indicators and the decision to act depends on the combination of all indicators or the most 426 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted July 3, 2021. Hospital, one of the largest hospitals in Chicago, 25-30% of COVID-19 admissions were not Chicago 450 residents (S1 Fig 18) . Hence, while capacities are higher in urban areas, accounting for potential 451 patient transfers may require lowering the threshold for action. As occupancy nears capacity, and 452 patients are admitted to nontraditional ICU areas or transferred to hospitals outside the region [49], 453 ICU data becomes less reliable as an epidemic indicator for local transmission. 454 455 The model does not include vaccination, and we assume 34% of hospitalized COVID-19 patients will 456 require ICU care [50] . As vaccine programs are scaled up, hospital and ICU admissions will 457 substantially decrease and the demographics of admitted patients may also shift. Younger patients 458 may reside longer in the ICU than the elderly if elderly patients are more likely to move to hospice 459 care or are less likely to survive COVID-19. The model also did not include shifting virulence due to 460 spread of new SARS-CoV-2 variants. Vaccination, changing demographics, and changing virulence 461 may mean that ICU occupancy is no longer a good indicator of transmission and hence should not be 462 used to make mitigation decisions. Whether and when to implement mitigation will need to 463 increasingly depend on more direct measures of transmission. Nevertheless, it remains crucial to 464 monitor COVID-19 hospitalizations and ICU occupancies and to have fail-safe thresholds in place to 465 allow timely action to prevent severe illness and deaths. 466 467 We used a SARS-CoV-2 and COVID-19 disease transmission model to evaluate how ICU occupancy 469 can be used as an indicator for triggering new mitigations in response to increasing transmission in 470 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 3, 2021. COVID-19 pandemic in the United States. Health Policy Technol Times TNY. See Reopening Plans and Mask Mandates for All 50 States. The New York Times Coronavirus cases: These states face biggest potential shortfalls in hospital 484 ICU beds What is an intensive care 488 unit? A report of the task force of the World Federation of Societies of Intensive and Critical 489 Care Medicine Care Resources in COVID-19 An Implementation Guide for Regional Allocation: An Expert Panel 492 Report of the Task Force for Mass Critical Care and the American College of Chest Physicians. 493 CHEST Rural America's Hospitals are Not Prepared to Protect 495 Older Adults From a Surge in COVID-19 Cases Urban and rural differences in 498 coronavirus pandemic preparedness Critical Care Utilization for the COVID-19 Outbreak in 502 Lombardy, Italy: Early Experience and Forecast During an Emergency Response ICU outcomes and 505 survival in patients with severe COVID-19 in the largest health care system in central Florida Geographic Differences in COVID-19 Cases, Deaths, and Incidence -United States Characteristics of Hospitalized Adults With COVID-19 in 511 an Integrated Health Care System in California Covid-19 in Critically Ill 514 Patients in the Seattle Region -Case Series Critical Care Surge Response 517 Strategies for the 2020 COVID-19 Outbreak in the United States Managing ICU surge during 519 the COVID-19 crisis: rapid guidelines Rapid response to 522 COVID-19, escalation and de-escalation strategies to match surge capacity of Intensive Care 523 beds to a large scale epidemic Actions to Combat a Resurgence of COVID-19 Timing social distancing to avert 529 unmanageable COVID-19 hospital surges Design of COVID-19 Staged Alert 532 Systems to Ensure Healthcare Capacity with Minimal Closures. medRxiv Projecting hospital 535 utilization during the COVID-19 outbreaks in the United States Projecting demand for 538 critical care beds during COVID-19 outbreaks in Canada Real-time forecasting 541 of COVID-19 bed occupancy in wards and Intensive Care Units. Health Care Manag Sci Impact of vaccination and 547 non-pharmaceutical interventions on SARS-CoV-2 dynamics in Switzerland COVID-19: a simple statistical model for 550 predicting intensive care unit load in exponential phases of the disease COVID-19: Short-term forecast of ICU beds in times 553 of crisis A modelling 555 study highlights the power of detecting and isolating asymptomatic or very mildly affected 556 individuals for COVID-19 epidemic management American Community Survey Single-Year Estimates First known person-564 to-person transmission of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in the 565 USA. The Lancet State of Illinois. State of Illinois: Coronavirus (COVID-19) Response. New. 2020 Compartmental Modeling Software: A Fast Discrete Stochastic Framework for Biochemical and Epidemiological Simulation Computational Methods in Systems Biology Python 3 Reference Manual Core Team. R: A language and environment for statistical computing Geographic and 581 demographic heterogeneity of SARS-CoV-2 diagnostic testing in Illinois, USA Illinois National Electronic Disease Surveillance 584 System Clinical Characteristics of 138 Hospitalized 588 Patients With 2019 Novel Coronavirus-Infected Pneumonia in Wuhan, China Analysis of 391 cases and 1,286 of their close contacts Python package to estimate the time-varying effective reproduction 594 number of an epidemic from reported case numbers EpiEstim: 597 Estimate Time Varying Reproduction Numbers from Epidemic Curves Mean Absolute Error. Encyclopedia of Machine Learning Evaluation Metrics for Machine Learning Incubation period of 2019 novel coronavirus (2019-nCoV) 604 infections among travellers from Wuhan, China Transforming a PICU Into an 607 Adult ICU During the Coronavirus Disease 2019 Pandemic: Meeting Multiple Needs Navigating 610 hospitals safely through the COVID-19 epidemic tide: predicting case load for adjusting bed 611 capacity A novel COVID-19 epidemiological model 613 with explicit susceptible and asymptomatic isolation compartments reveals unexpected 614 consequences of timing social distancing Effective public health 617 measures to mitigate the spread of COVID-19: a systematic review Interdependence and the cost 620 of uncoordinated responses to COVID-19 Locally Informed Modeling to Predict Hospital and 623 Intensive Care Unit Capacity During the COVID-19 Epidemic We thank Ariel Chandler Eickelberg for technical and management support during the 656 pandemic early in 2020. We thank Sebastian Rodriguez and Ben Toh for helpful comments on the 657 draft manuscript This research was supported in part through the computational resources and staff contributions 660 provided for the Quest high performance computing facility at Northwestern University, which is 661 jointly supported by the Office of the Provost, the Office for Research, and Northwestern University 662 Information Technology MR was 666 supported by a COVID-19 rapid response grant via NUCATS (UL1TR001422). MR and JG were 667 supported by a BMGF grant (INV-002092) The funders had no role 669 in the design of the study and collection, analysis, and interpretation of data or in writing the The COVID-19 transmission 676 model is maintained under Illinois hospital census 678 data and I-NEDSS data can be received from IDPH upon reasonable request All other data are publicly available as mentioned in the text S1 Appendix: Technical supplement and additional results figures We thank Chinyere Alu, Stacey Hoferka Jensen, Megan Patel, and Dejan Jovanov from IDPH and 643Michele Atkinson and Sara Rogers from Civis Analytics for data extraction and data management.