Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Medical costs of keeping the US economy open during COVID-19

Abstract

We use an individual based model and national level epidemic simulations to estimate the medical costs of keeping the US economy open during COVID-19 pandemic under different counterfactual scenarios. We model an unmitigated scenario and 12 mitigation scenarios which differ in compliance behavior to social distancing strategies and in the duration of the stay-home order. Under each scenario we estimate the number of people who are likely to get infected and require medical attention, hospitalization, and ventilators. Given the per capita medical cost for each of these health states, we compute the total medical costs for each scenario and show the tradeoffs between deaths, costs, infections, compliance and the duration of stay-home order. We also consider the hospital bed capacity of each Hospital Referral Region (HRR) in the US to estimate the deficit in beds each HRR will likely encounter given the demand for hospital beds. We consider a case where HRRs share hospital beds among the neighboring HRRs during a surge in demand beyond the available beds and the impact it has in controlling additional deaths.

Introduction

As states push to end social distancing and reopen businesses, it is important to understand the cost of opening in terms of lives lost and medical costs incurred. A premature opening will likely cause more deaths and infections as the healthcare system will likely get overwhelmed, and may wipe out all the gains made in the initial shutdown. We use an agent-based model and simulation framework to estimate the immediate medical cost of COVID-19 under different mitigation scenarios. The scenarios consider social distancing with different durations and varying compliance levels. The simulation framework uses a detailed representation of the US population and their social interactions to study the spread of COVID-19. An SEIR (susceptible-exposed-infected-recovered) model captures the time varying health states of the individuals. The infected individuals arrive at one of the three health states i.e. medically attended, hospitalized, or ventilated before getting to the final health state i.e. recovered or dead, as shown in Fig. 1.

Medical costs are applied based on the three health states i.e. medically-attended, hospitalized and ventilated. In addition, if an infected individual dies, then the “value of statistical life” is used to estimate the cost of death. We also estimate the shortage of hospital beds that is likely to occur in each Hospital Referral Region (HRR) given the demand for hospital beds and the number of available beds in each HRR in the US. Data on the number of beds in each HRR is obtained from the American Hospital Association (AHA) and a fraction of them are assumed to be available for COVID-19 patients. We consider cases where neighboring HRRs share or do not share hospital beds during a surge in demand.

This information is then used to calculate additional deaths and medical costs for each of the mitigation scenarios. Policy makers can apply this kind of analysis to decide where the temporary hospitals may need to be built to offset the deficit in demand for beds. Our goal is to use this knowledge to provide guidance to public health officials and policy makers on the trade-offs between the length of lockdown, compliance to social distancing, infections, deaths and the medical costs. Our scenario-based analysis estimates the burden of the disease in monetary terms, and helps rank-order mitigation strategies.

In related work, authors in1 consider potential health care costs and resource usage under different attack rates which vary from 20 to 80%. However, it does not consider any interventions or mitigation strategies. Our research focuses on counterfactual mitigation scenarios and their respective costs. We use recent cost estimates for COVID-19 available from the Kaiser Family Foundation (KFF)2 which uses cost of pneumonia cases as a proxy. Our detailed network based model, that captures heterogeneous social interactions and contact times among the individuals in the population, is one of the unique features of the analysis. Additionally, no other research has provided an estimation of medical costs for such detailed mitigation scenarios for the entire US.

Our results show that (1) Without mitigation, the total medical costs would be a significant fraction (5%) of the US GDP; (2) a lockdown of just 2 months, if done early in the epidemic, and with sufficient compliance, could have reduced the medical costs by more than 90%; (3) if 90% compliance could be achieved, then even a 45 day lockdown period would have been enough to contain the epidemic and the medical costs; (4) if HRRs do not share hospital beds with other HRRs, a significant deficit of beds will cause medical costs to skyrocket, through increase in deaths. However, if HRRs shared beds with their neighboring HRRs, the bed-deficits and additional deaths could be reduced to almost zero; and (5) a sensitivity analysis of the parameters shows the costs are most sensitive to the duration of the stay-home order.

Data and methodology

We build on our modeling and simulation framework for epidemic spread3,4,5,6,7,8,9 using an individual level synthetic social contact network5,10—which represents each individual in the population along with their demographic attributes (e.g., age, gender, income), and their social interactions. The main steps in the first-principles based construction of synthetic populations and social contact networks are: (1) construct a synthetic population by using US Census and other commercial databases; (2) assign daily activities to individuals within each household using activity and time-use surveys (American Time Use Survey data and National Household Travel Survey Data); (3) assign a geo-location to each activity of each person based on data from Dun and BradStreet, land-use, Open Street Maps etc.; and (4) construct a dynamic social co-location based social contact network that is induced when people simultaneously visit locations. These networks have been validated and used for numerous public health analyses before such as3,5,11,12,13,14,15,16. For details on the construction of social networks, see11,12,17.

The SEIR disease model and parameters used here for COVID-19 have been defined in the best guess 2020-04-14 version of “COVID-19 Pandemic Planning Scenarios” document prepared by the Centers for Disease Control and Prevention (CDC) SARS-CoV-2 Modeling Team18. The sequence of health-state transitions and possible paths are shown in Fig. 1. There are many possible health states and paths an individual can move through as it transitions from susceptible to its final health state. This model is age stratified for the following categories i.e. preschool (0–4 years), students (5–17) adults (18–49), older adults (50–64) and seniors (65+) and calibrated for each of the age groups separately. Details on the transition probabilities between health states for each age group and the length of the stay in each health state are shown in the table in the Appendix. Our models and simulation framework have been used in some of our ongoing SARS-CoV-2 response work for the Virginia Department of Health and the US Department of Defense19.

Medical costs

To estimate the medical cost of treating COVID-19 patients, we use the average cost of treatment for pneumonia, paid among “large employer health insurance” plans, as a proxy2. The costs for each health state are shown in Table 1. Note that each infected individual’s medical cost is counted only once. For example if a person is in ventilated state, after having gone through “medAttend” and “Hosp” state, costs are cumulative to the “vent” state.

Value of life

The total medical costs can be measured through multiple criteria in terms of number of deaths and treatment costs, or a single criterion in dollar terms by converting deaths into dollars using “value of a statistical life”. There are various ways in which the value of a statistical life has been measured. US federal government uses $10 million dollar per life lost regardless of a person’s age while others have used estimates in the range of$160k to $2.4 million based on metrics like average lifetime earnings of a college graduate, 9/11 victim compensation, wrongful death claims, insurance policy value etc.20,21,22. We use an estimate of$2 million dollar as the value of life to convert deaths into costs23.

Interventions

We consider a number of mitigation scenarios that comprise of various social distancing strategies, compliance levels and durations of stay-home order. The following social distancing strategies are used:

1. 1.

Voluntary home isolation (VHI): Symptomatic people choose to stay at home (non-home type contacts are disabled) for 14 days.

2. 2.

School closure (SC): Schools and colleges are closed (school type contacts are disabled).

3. 3.

Stay home (SH): People follow public health “stay-home” directive (non-home type contacts are disabled).

School closure and stay-home interventions start on different days in different states as stated in24,25. Once closed, schools are assumed to remain closed until end of August after which they reopen. Other social distancing interventions stop at 30, 45 or 60 days from the start date of the intervention, depending on the SH duration. Note that this implies all mitigation efforts end by the end of summer 2020.

Durations of stay-home order vary from 0, 30, 45 to 60 days. Compliance to stay-home and voluntary home isolation vary from 60%, 70%, 80% and 90%. Table 2 shows a factorial design with 12 mitigation scenarios and an unmitigated case, resulting in a total of 13 cells experiment. For each cell, 25 replicates are run and their averages reported. Table 3 shows a complete list of variables and their parameter values.

Hospital bed capacity

We use hospital bed capacity data available for each HRR in the US from AHA, to calculate the deficits that are likely to be encountered by each region. We assume three scenarios regarding the use of hospital beds: (1) All COVID-19 patients who need a bed, will have one available i.e. there is no shortage of hospital beds, (2) bed capacity in each HRR is limited and beds cannot be shared among hospitals in other HRRs to accomodate the surge in demand for beds and (3) bed capacity in each HRR is limited but beds can be shared among hospitals in the neighboring HRR regions to accomodate the surge in demand.

Other factors considered in these scenarios are the average length of hospital stay of patients who are hospitalized, and the percentage of beds dedicated to COVID-19 patients in each HRR. Length of stay is 7 days or 14 days, and the percentage of beds available to COVID-19 patients is 70%, 80%, 90% or 120%. More than 100% bed capacity (120%) has been considered since many hospitals are able to temporarily increase their capacity beyond normal levels26.

Each day the demand for the number of beds in each HRR is determined by the simulations. The simulation results provide the counts of individuals who are in “hospitalized” state and depending upon the assumed duration of the hospital stay (7 days or 14 days), the demand for beds per day is calculated. Note that this includes individuals who are in “ventilated” state as well since everyone in “ventilated” state has to be in the “hospitalized” state first according to our disease model, see Fig. 1. The difference between the bed capacity and the counts of patients who need it, determines the deficit in hospital beds.

We assume that patients who need a hospital bed and cannot get it, will die. This is a strong assumption and hence four different values of dedicated bed capacities have been considered to show its sensitivity. Note that only the death counts and overall costs will change in scenarios (2) and (3) when limitations to bed capacity are considered. This assumption of “all bed deficits result in deaths” provides an upper bound on the medical costs resulting from the shortage of beds. If only a fraction of them die then the additional deaths and the costs can be appropriately scaled down.

Results

This section reports the simulation results and the medical costs that are likely to incur under various mitigation scenarios. For each of these scenarios we also consider four different possibilities for the availability of dedicated hospital beds for COVID-19 patients. Note that the hospital beds are considered as a post-processing step, after the simulations have been run and infected individuals have arrived at “hospitalized” state.

Figure 2 shows the epidemic curves with daily new infections in the US for all the mitigation scenarios. The four subplots refer to VHI and SH compliance levels of 60%, 70%, 80% and 90%. Each solid line in each subplot corresponds to a different stay-home duration. The shaded area around the solid line shows the stochasticity in the simulation results and is marked by one standard error band. The following observations can be made from Fig. 2: (1) a higher SH duration lowers the peak of the epidemic curve; (2) in all cases, a longer SH duration either delays the peak and/or flattens it; (3) a large second wave hits in Fall 2020 unless VHI and SH compliance rates are at least 80% and SH duration is at least 45 days. Note that by Fall all interventions end including schools closures; and (4) if VHI and SH compliance rates are 90% and SH duration is 45 days or longer, the epidemic ends by the end of the year.

Medical cost under mitigation: unlimited supply of hospital beds

Figure 3 shows medical costs for the 12 mitigation scenarios as well as the unmitigated one. Counts for health states, for each scenario, are estimated by our model and the simulations. Costs for categories ventilated, hospitalized and medically-attended are calculated by multiplying the per capita costs for each of these health states given in Table 1 by the counts of individuals who reach that health state before recovering or dying. If a person dies, then an additional cost of death is incurred.

Conclusions

This study estimates the medical costs of COVID-19 in the US under different mitigation scenarios and helps understand the tradeoffs between deaths, costs, infections, compliance to social-distancing and the duration of stay-home order. Our scenario-based analysis estimates the medical burden of the disease in monetary terms, and helps rank-order mitigation strategies.

It shows that costs are most sensitive to the length of the stay home order and then to the level of compliance. A stay home order of 60 days with 80% compliance, can drop the medical costs from $1 trillion to a meager$35 billion. Stay home duration also mitigates the effect of a high $$R_0$$, as shown in the sensitivity analysis. As stay home duration increases, sensitivity of costs to $$R_0$$ drops because the lockdown reduces the effective $$R_0$$ by cutting off contacts. However, to get these results, the social distancing strategies must be applied consistently nationwide. Even though policy makers cannot control the level of compliance among people, they can control the duration of the lockdowns and can make adjustments to the public health directives based on the level of compliance observed on the ground.

We also show that sharing of hospital beds among neighboring “Hospital referral region” during a demand surge can reduce the additional number of deaths to almost zero. Cooperation in sharing medical resources between neighboring regions can save lives and money. This kind of analysis can also help decide where additional bed capacity and temporary hospitals may need to be built to offset the surge in demand.

Data availability

All the output data reported in the paper is available upon request, but restrictions apply on the commercially available data used in the construction of the social contact network and hence the availability of the social network data itself.

Code availability

Code developed to analyze the results and support the findings in this paper is available upon request, from the corresponding author.

References

1. 1.

Bartsch, S. M. et al. The potential health care costs and resource use associated with covid-19 in the united states. Health Aff. 39(6), 927–935 (2020).

2. 2.

Rae, M., Claxton, G., Kurani, N., McDermott, D. & Cox, C. Potential costs of coronavirus treatment for people with employer coverage. In Peterson Center on Healthcare and Kaiser Family Foundation, Vol 13 (2020). https://www.healthsystemtracker.org/brief/potential-costs-of-coronavirus-treatment-for-people-with-employer-coverage/. Accessed 15 July 2020.

3. 3.

Chen, J., Marathe, A. & Marathe, M. Feedback between behavioral adaptations and disease dynamics. Sci. Rep. 8, 1–15 (2018).

4. 4.

Dorratoltaj, N. et al. Epidemiological and economic impact of pandemic influenza in Chicago: Priorities for vaccine interventions. PLoS Comput. Biol. 13, e1005521 (2017).

5. 5.

Eubank, S. et al. Modelling disease outbreaks in realistic urban social networks. Nature 429, 180–184 (2004).

6. 6.

Chen, J., Levin, S., Eubank, S., Mortveit, H., Venkatramanan, S., Vullikanti, A. & Marathe, M. Networked epidemiology for COVID-19. SIAM News 53(5), 3, 7 (2020).

7. 7.

Adiga, A., Venkatramanan, S., Schlitt, J. & et al. Evaluating the impact of international airline suspensions on the early global spread of covid-19. Technical Report, medRxiv (2020). https://www.medrxiv.org/content/10.1101/2020.02.20.20025882v2.full.pdf. Accessed 1 July 2020.

8. 8.

Machi, D. et al. Scalable epidemiological workflows to support covid-19 planning and response. Technical Report SC-TR-2020, Network Systems Science and Advanced Computing Division, BII, University of Virginia (2020). https://www.dropbox.com/s/2uk2wnbte95mk5y/main.v87.7020f2c.pdf?dl=0. Accessed 10 July 2020.

9. 9.

Singh, M. et al. Impact of demographic disparities in social distancing and vaccination on influenza epidemics in urban and rural regions of the united states. BMC Infect. Dis. 19, 221 (2019).

10. 10.

Beckman, R. J., Baggerly, K. A. & McKay, M. D. Creating synthetic baseline populations. Transp. Res. Part A Policy Pract. 30, 415–429 (1996).

11. 11.

Barrett, C. et al. Generation and analysis of large synthetic social contact networks. In Winter Simulation Conference, 1003–1014 (Winter Simulation Conference, 2009).

12. 12.

Halloran, M. E. et al. Modeling targeted layered containment of an influenza pandemic in the United States. In Proceedings of the National Academy of Sciences (PNAS), 4639–4644 (2008). PMCID:PMC2290797.

13. 13.

Barrett, C., Eubank, S. & Marathe, M. An interaction-based approach to computational epidemiology. AAAI 1590–1593 (2008).

14. 14.

Eubank, S. et al. Detail in network models of epidemiology: Are we there yet?. J. Biol. Dyn. 4, 446–455 (2010) (PubMed PMID: 20953340; PMCID: PMC2953274).

15. 15.

Marathe, A. et al. Comparing effectiveness of top-down and bottom-up strategies in containing influenza. PLoS ONE 6, e25149 (2011) (PMCID: PMC3178616).

16. 16.

Rivers, C., Lofgren, E., Marathe, M., Eubank, S. & Lewis, B. Modeling the impact of interventions on an epidemic of Ebola in Sierra Leone and Liberia. PLoS Curr. 6 (2014). PMCID: PMC4399521.

17. 17.

Bisset, K. & Marathe, M. A cyber-environment to support pandemic planning and response. DOE SciDAC Mag. 13, 36–47 (2009).

18. 18.

Centers for Disease Control and Prevention. Covid-19 pandemic planning scenarios. https://www.cdc.gov/coronavirus/2019-ncov/hcp/planning-scenarios-h.pdf (2020). Accessed 8 July 2020.

19. 19.

Virginia Department of Health. Uva covid-19 model: Weekly update. https://www.vdh.virginia.gov/coronavirus/2020/05/19/uva-covid-19-model-weekly-update/ (2020). Accessed 8 July 8 2020.

20. 20.

Merrill, D. No one values your life more than the federal government. Bloomberg (2017). https://www.bloomberg.com/graphics/2017-value-of-life. Accessed 15 June 2020.

21. 21.

Alvarez, F. E., Argente, D. & Lippi, F. A simple planning problem for covid-19 lockdown Technical Report, National Bureau of Economic Research (2020).

22. 22.

Liu, J.-T., Hammitt, J. K., Wang, J.-D. & Tsou, M.-W. Valuation of the risk of sars in taiwan. Health Econ. 14, 83–91 (2005).

23. 23.

Mrozek, J. R. & Taylor, L. O. What determines the value of life? A meta-analysis. J. Policy Anal. Manag. 21, 253–270 (2002).

24. 24.

Education Week. Map: Coronavirus and school closures. https://www.edweek.org/ew/section/multimedia/map-coronavirus-and-school-closures.html (2020). Accessed 30 June 2020.

25. 25.

Wikipedia contributors. U.S. state and local government response to the covid-19 pandemic. https://en.wikipedia.org/wiki/U.S._state_and_local_government_response_to_the_COVID-19_pandemic (2020). Accessed 30 June 2020.

26. 26.

US News. Gov. andrew cuomo orders hospitals to increase capacity by 50% (2020). https://www.usnews.com/news/health-news/articles/2020-03-23/new-york-gov-andrew-cuomo-orders-hospitals-to-increase-capacity-by-50. Accessed 8 July 2020.

27. 27.

Moghadas, S. M. et al. Projecting hospital utilization during the covid-19 outbreaks in the united states. Proc. Natl. Acad. Sci. 117, 9122–9126 (2020).

28. 28.

McKay, M. D., Beckman, R. J. & Conover, W. J. Comparison of three methods for selecting values of input variables in the analysis of output from a computer code. Technometrics 21, 239–245 (1979).

29. 29.

Iman, R. L., Helton, J. C. & Campbell, J. E. An approach to sensitivity analysis of computer models: Part I–introduction, input variable selection and preliminary variable assessment. J. Qual. Technol.. 13, 174–183 (1981).

Acknowledgements

The authors would like to thank members of the Biocomplexity Institute at the University of Virginia (UVA) for useful discussion and suggestions. We also thank the staff members at the UVA’s high performance computing center and at the Pittsburgh Supercomputing center for providing the much needed high-performance computing resources. This work was partially supported by National Institutes of Health (NIH) Grant R01GM109718, NSF BIG DATA Grant IIS-1633028, NSF DIBBS Grant ACI-1443054, NSF Grant No.: OAC-1916805, NSF Expeditions in Computing Grant CCF-1918656, CCF-1917819, US Centers for Disease Control and Prevention 75D30119C05935, DTRA subcontract/ARA S-D00189-15-TO-01-UVA, and a collaborative seed grant from the UVA Global Infectious Disease Institute. The content is solely the responsibility of the authors and does not necessarily represent the official views of the sponsoring agencies.

Author information

Authors

Contributions

JC, SH, HM, SE, SV and BL built the model and the software. AV, JC, WY and AM processed and analyzed the data. AV, JC, AM, MM and CB conceived the project. All authors helped write, edit and review the paper.

Corresponding author

Correspondence to Achla Marathe.

Ethics declarations

Competing interests

The authors declare no competing interests.

Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and Permissions

Chen, J., Vullikanti, A., Hoops, S. et al. Medical costs of keeping the US economy open during COVID-19. Sci Rep 10, 18422 (2020). https://doi.org/10.1038/s41598-020-75280-6

• Accepted:

• Published:

• Data-driven optimized control of the COVID-19 epidemics

• Afroza Shirin
• , Yen Ting Lin
•  & Francesco Sorrentino

Scientific Reports (2021)