Responding to Simulated Pandemic Influenza in San Antonio, Texas
Objective. To describe the results of a simulation study of the spread of pandemic influenza, the effects of public health measures on the simulated pandemic, and the resultant adequacy of the surge capacity of the hospital infrastructure and to investigate the adequacy of key elements of the national pandemic influenza plan to reduce the overall attack rate so that surge capacity would not be overwhelmed.
Design. We used 2 discrete‐event simulation models: the first model simulates the contact and disease transmission process, as affected by public health interventions, to produce a stream of arriving patients, and the second model simulates the diagnosis and treatment process and determines patient outcomes.
Setting. Hypothetical scenarios were based on the response plans, infrastructure, and demographic data of the population of San Antonio, Texas.
Results. Use of a mix of strategies, including social distancing, antiviral medications, and targeted vaccination, may limit the overall attack rate so that demand for care would not exceed the capacity of the infrastructure. Additional simulations to assess social distancing as a sole mitigation strategy suggest that a reduction of infectious community contacts to half of normal levels would have to occur within approximately 7 days.
Conclusions. Under ideal conditions, the mix of strategies may limit demand, which can then be met by community surge capacity. Given inadequate supplies of vaccines and antiviral medications, aggressive social distancing alone might allow for the control of a local epidemic without reliance on outside support.
Received May 10, 2007; accepted December 5, 2007; electronically published February 28, 2008.
The threat of biological weapons use has become a national security preoccupation during the past 15 years, although pandemic influenza probably represents a greater threat to the US population.1 The US General Accounting Office has noted that “[t]here is little or no excess capacity in the health care system for accepting and treating mass casualty patients.”2 The surge in demand for healthcare services as a result of a bioterrorist attack or naturally occurring epidemic would significantly affect the ability of a community healthcare infrastructure to respond. Previously, we used discrete‐event simulation to assist communities in planning their response to a bioterrorist attack with pneumonic plague and smallpox.3,4 In this article, we describe an analysis that uses the same tools to address pandemic influenza.
Previous influenza pandemics occurred in 1918, 1957, and 1968. If the next pandemic influenza is similar to its predecessors, it will likely be highly infectious in close contact and have a 1.9‐day incubation period, a 4.1‐day infectious period, a 1‐week to 2‐week recovery period, and a mortality rate of approximately 2.5% of those infected.5‐7
The concern that the world faces a pandemic threat has spurred the publication of the US National Strategy for Pandemic Influenza, of which the strategic intents are to prevent or limit the spread of the disease to the United States and within the United States; to reduce morbidity and associated mortality; and to minimize its economic and social impact. There are 3 pillars to the strategy: (1) preparedness and communication, including stockpiling of vaccine and antiviral medications; (2) surveillance and detection; and (3) response and containment, which include aggressive social distancing by means such as limiting public gatherings and quarantine.8 The national plan to implement the pandemic influenza strategy predicts that susceptibility will be universal: up to 35% of the population could develop the disease, 50% of those infected will seek medical care, approximately 20% of those seeking care will require hospitalization, and up to 2.1% of those infected will die. The plan calls for domestic vaccine production sufficient to vaccinate every American, a prepandemic vaccine stockpile of 20 million doses, an antiviral stockpile sufficient to treat 25% of the US population, and protective health measures to reduce person‐to‐person disease transmission.9
A vaccine specific for the pandemic strain, the best countermeasure, will not likely be available for the first wave of the disease; however, other measures to slow disease spread are available. Close examination of historical records of the 1918 pandemic has shown the tremendous efficacy of social distancing and other community containment measures if begun early in a pandemic.10 Mathematical modeling of the geographic spread of a future pandemic demonstrates that these community interventions, when applied in combination with antiviral medications, may have a synergistic impact in reducing disease spread, limiting attack rates, delaying the peak of the wave, and reducing the burden on hospitals and infrastructure.11‐14 Our methods and results extend the analysis beyond the spread of the disease to include the treatment of individuals who need medical care. Our analysis was designed to assess the feasibility and effectiveness of the national plan in the context of the ability of a major US city to respond to an influenza pandemic.
Our simulation was set in the urban environment of San Antonio, Texas. With a population of 1.4 million, the city is served by a university medical center, 2 large military hospitals, several major hospital systems, and several smaller hospitals.15 (Applying the worst‐case scenario from the national plan to this population results in 490,000 people infected, 245,000 new patients seeking care, 49,000 hospitalizations, and 10,290 deaths. Our simulations do not use these numbers directly but dynamically generate similar results.) The community‐wide acute care capacity is considered adequate to accommodate steady‐state demand for healthcare services. Area hospitals have dedicated 990 medical or surgical beds to the National Disaster Medical System, which we assume represents the total medical or surgical bed surge capacity for the community. We further assume that 50% of the total intensive care unit beds present at the hospitals (238 appropriately staffed beds) could become available 14 days after the initial infection in San Antonio, after elective surgical procedures and elective admissions are discontinued at the advent of the pandemic. We received these and other significant inputs from local officials, infection prevention and control experts, and local emergency management and healthcare organizations.
Methods
We developed an end‐to‐end simulation method to represent the consequences of an epidemic from initial infection to final patient outcome (Figure 1). We developed and modified 2 discrete‐event simulations for this purpose: the casualty prediction model and the healthcare complex model. Both models are generic in that they are structured to represent any community and disease, depending on the appropriate input data.
Figure 1. Summary of the end‐to‐end simulation method representing the consequences of an epidemic from initial infection to final patient outcome. Surveillance and alert systems notify that the event has occurred; subsequently the public health and acute medical care delivery systems respond. The casualty prediction model simulated the impact of the public health response on the spread of the disease and the resultant time‐phased stream of patients who present to the acute care system (see Figure 2), and the healthcare complex model (HCM) simulated the subsequent diagnosis and treatment of these patients (see Figure 3).
Casualty Prediction Model
The casualty prediction model simulates contacts and disease transmission between infected and previously unexposed individuals, both within the infected individual’s household and in the broader community (Figure 2). The model also simulates the effects of various possible public health responses that might be taken to halt an epidemic by reducing the contact and transmission rates between infected and susceptible individuals. By varying the assumptions about the specific actions taken by the public health community, we can use the casualty prediction model to improve our understanding of the most effective mix of responses.
Figure 2. Generic‐disease timeline represented in the casualty prediction model, which simulates the activities of a patient from the time of transmission of the disease agent to the end of the infectious period, during which the patient randomly comes into contact with others in the community; household contacts are assessed at the onset of the infectious period. A new timeline is created for each contact to whom the disease agent is transmitted.
We developed and applied the casualty prediction model to represent pneumonic plague, smallpox, and influenza. For influenza, we validated the model16 by comparing its representation of the historical spread of seasonal influenza A virus (H3N2) within a community with published accounts of that history.17,18 For validation purposes, the model was designed to distinguish between individuals with and without prior immunity to H3N2; our subsequent representation of pandemic influenza assumed no prior immunity in the population.
Like other epidemiology models, the casualty prediction model can be used as a stand‐alone tool to investigate the effect of alternative public health interventions or to conduct sensitivity analyses of the impact of variations in parameter values, such as the initial number of patients, contact rates, the start dates of various interventions, and vaccination effectiveness. However, analysts can also use the model in conjunction with the healthcare complex model to obtain outputs describing morbidity, mortality, and resource consumption associated with treatment of these patients.
Healthcare Complex Model
The healthcare complex model simulates healthcare delivery to the patient throughout a network of medical facilities. Patient information is input into the model by using output from the casualty prediction model. Each patient appears for diagnosis and treatment at 1 of several medical facilities. The facilities are characterized by quantities of available resources, including multiple provider types, multiple bed types, and various ancillary resources. The model uses treatment protocols linked to patient characteristics to identify the resources needed for each service within the diagnosis and treatment processes (Figure 3). For this application, we used 10 protocols to capture the effects of patient age (using 5 age groups: younger than 2 years of age, 2‐18 years of age, 19‐40 years of age, 41‐64 years of age, and 65 years of age or older) and immunization status (distinguishing whether or not a patient seeking care has been vaccinated and/or treated with antiviral drugs). The model records resource consumption, the time a patient spent waiting, patient disposition, and other information for subsequent analysis, including identification of bottlenecks in healthcare delivery.
Figure 3. Example of the healthcare complex model treatment protocol for patients with influenza who are 19‐40 years of age. ICU, intensive care unit; LOS, length of stay.
During the past 10 years, we developed the healthcare complex model, validated it against the performance of a complex of US Department of Defense medical treatment facilities, and applied it in numerous studies to help guide the reengineering of the healthcare delivery process within the US Department of Defense, US Department of Veterans Affairs, and community settings. Recent initiatives have enhanced the model, so that it now simulates a community’s ability to meet the surge in demand associated with a natural disaster or terrorist attack.
Application and Analysis
We simulated pandemic influenza and the community’s response to it using the 2 models in combination, to study the spread of the disease through the community, the effects of potential public health responses, and the effects of the pandemic on resource requirements in the acute healthcare delivery system. We adapted the 2 models to simulate the specifics of the disease and the environment by collecting data on the natural history of the disease from the published literature, on the parameter estimates developed from existing plans or expert opinion, on the capabilities and capacities provided by public sources, and on existing emergency response plans. We assumed that the disease would have the same infectiousness that H3N2 has in individuals with no prior immunity.18
Our analysis addressed the effectiveness of increased social distancing, the prescription of antiviral medications for treatment or prophylaxis, and the targeted vaccination of susceptible populations, including first responders, healthcare workers, and elderly individuals. Our analysis of social distancing included an assessment of its effectiveness as the only intervention to mitigate the risk of current planning assumptions that rely on the supply and efficacy of vaccine and antiviral medications. We use the term “excursion” to refer to a particular model variation that involves changing the input variables that alter the interventions. Results presented for each excursion are mean values from 30 replications of the casualty prediction model and 10 replications of the healthcare complex model.
Results
We investigated a progression of model excursions that involved changing the following conditions: (1) exclusion of the public health intervention, (2) addition of increased social distancing (resulting from measures to decrease transmission risk by reducing interactions among members of the community to 80% of their normal levels), (3) addition of the use of antiviral medication, and (4) addition of targeted vaccination of susceptible populations. The Table gives the comparative results of these excursions.
With no public health interventions to halt the growth of the epidemic, the overall attack rate (ie, the cumulative incidence rate during the epidemic) was 38.8%, which is slightly higher than the estimated 35% attack rate experienced during the 1918 influenza pandemic and the 35% upper limit estimated in the US Department of Health and Human Services Pandemic Influenza Plan.7,9 The major portion of the epidemic occurred within approximately 60 days, which is similar to the 6‐week to 8‐week duration anticipated in the national plan.
When antiviral medications were added to the model excursions, their availability was assumed to be unconstrained, but prescription by a licensed clinician was assumed necessary. Antiviral medications were prescribed as prophylaxis to all members of infected households in which at least 1 infected member sought medical care, and to all first responders and healthcare workers. We assumed that 2 5‐day courses of antiviral medication would be prescribed and would provide protection for a total of 12 days, and we explicitly modeled 90% compliance with the prescription. A total of 1,590,920 courses of antiviral medication were consumed, which is the equivalent of 2 5‐day courses for 54% of the population. This quantity far exceeds the recommended stockpile size that would provide 1 course for 25% of the population.9,19 To investigate the impact of using the recommended stockpile size, we conducted an excursion with the number of courses of antiviral medication constrained to 360,000 (ie, 1 course for nearly 25% of the population) and compared it with our unconstrained results. The 2 curves of daily infections diverged when the supply of antiviral medication was exhausted in the constrained excursion. From that point onward, the slope of the curve in the excursion with constrained antiviral medication increased, because the protective effects of the antiviral medication wore off. The epidemic continued to grow until the density of remaining susceptible individuals in the population decreased to the point at which the epidemic began to die out.
The final excursion, which represented the essential elements of the national plan, added the vaccination of first responders (70%), healthcare workers (80%), and at‐risk people (40%). The amount of vaccine required was equal to 1 dose each for approximately 7% of the population (and was consistent with 20 million doses of vaccine nationally). The vaccine was assumed to take effect 2 weeks after administration. When vaccination was added to the model excursion, the required number of courses of antiviral medication consumed declined to 508,000, which is much closer to the 25% recommended stockpile.
Figure 4 displays the capacity of and demand for medical and surgical beds in each of the 4 main excursions listed in the Table. Even with unconstrained availability of antiviral medication, the anticipated supply of medical and surgical beds was sufficient only in the excursion in which targeted vaccination was added to the mix of interventions; in this excursion, the peak demand for 517 beds was on days 22 and 23. We generated similar results for the demand for and supply of intensive care unit beds and acute care nurses. For each of these resource types, only the excursion for which all of the interventions were used produced a demand that could be met by the anticipated supply.
Figure 4. Comparison of capacity and demand for medical and surgical beds in the initial 4 model excursions (for definitions, see Methods).
Several assumptions contributed to the success of the excursion with all interventions used, and the uncertainty embedded in these assumptions increases risk. For example, we assumed that vaccine would be available for administration to approximately 7% of the population and that the vaccine would have an efficacy of 70%‐80%.20,21 A vaccine with that efficacy may not exist in the requisite quantities when needed. Additionally, it is not known how effective antiviral drugs such as oseltamivir would be or how many of them would be available. If a variant strain of influenza developed pandemic characteristics today, the performance of the public health interventions would be significantly worse than indicated in the simulations described herein.
Of the interventions that we have modeled, only the increase in social distancing would be possible today. To explore this option further, we created 2 model excursions that reduced the number of community contacts to 50% of normal by day 21 and, alternatively, by day 7 (recognizing that such aggressive social distancing might not be feasible) after the first infection in San Antonio. This magnitude of social distancing was intended to represent closing schools and churches, banning public gatherings, and encouraging people to work from home if possible. As in the severe acute respiratory syndrome epidemic, “work quarantine” could be imposed, meaning that only essential workers, such as healthcare workers, would go to work and then return home; they could not go to the grocery store, pharmacies, social gatherings, and so on.22 No other interventions were assumed to be used.
The Table also summarizes the results of these 2 excursions, and Figure 5 displays the demand for and supply of medical and surgical beds for each of them. Figure 5 also shows that a significant increase in social distancing can halt the epidemic. However, delaying the decision to significantly restrict contacts outside the household would result in surges of demand that would exceed the supply of beds. In the excursion that reduced the number of community contacts to 50% of normal by day 21, peak demand for medical and surgical beds was 11% greater than capacity and occurred on day 28. This excess demand could possibly be absorbed by the creative allocation and use of resources or by temporary changes in treatment patterns.
Figure 5. Comparison of capacity and demand for medical and surgical beds in the 2 social‐distancing model excursions (for definitions, see Methods).
Of greater concern are the results for intensive care unit beds and nursing staff. The demand for nurses that is associated with a delay in aggressive social distancing exceeded capacity by 64% on day 26 and would limit any initiative to supply more medical and surgical or other beds. Demand for intensive care unit beds also peaked on day 26 and exceeded capacity by 71%. Close management, reallocation of patients, or the creative use of resources probably could not rectify these shortages.
However, for each resource type, the supply was adequate to meet the peak demand when social distancing was doubled by day 7 rather than by day 21. These results illustrate the importance of quick action to reduce community contacts in the face of a pandemic for which vaccine and antiviral medications are not available.
Discussion
As in the national planning assumptions, the conditions set in our models caused 50% of the patients with influenza to seek medical care and approximately 20% of those to be hospitalized. Under these conditions, our simulations suggest that the surge capacity of the San Antonio healthcare infrastructure becomes inadequate to accommodate a surge in demand for services at attack rates in excess of approximately 4%. (This finding should not be confused with the 15% attack rate that is commonly factored into the planning for hospitals that deal with seasonal influenza, which has a much lower morbidity that results in lower hospitalization rates and mortality rates of approximately 0.1%.) It is evident that a successful outcome for the community must be achieved by reducing demand through early and effective public health intervention to drastically reduce the overall attack rate.
Application of the national plan within our simulations shows that the recommended mix of strategies could possibly result in a limited, local epidemic that would not exceed the surge capacity of the healthcare infrastructure. This finding suggests that the plan would probably be successful under the conditions simulated.
However, the national plan relies on optimistic assumptions, such as the availability and efficacy of antiviral medications and vaccines, that carry associated risks. As we have simulated it, the plan would probably fail to adequately limit demand for healthcare services without the requisite quantities of antiviral medications and vaccine with efficacies of 70%‐80%—quantities that currently do not exist. In addition, if the pandemic influenza strain is H5N1, the dose of antiviral medications and the vaccine dosage may need to be doubled,23,24 making the deficiencies of antiviral medications and vaccines even more significant. To investigate alleviating this risk, we conducted excursions to identify alternatives that might be used to overcome such deficiencies. Because aggressive social distancing is an intervention that is currently available and over which local officials have control, it was the focus of our excursions.
Previous studies that compared the effects of public health measures among cities in the 1918 influenza epidemic found that cities that implemented social distancing measures showed decreased mortality, compared with cities that did not.10,25 Communities that practiced “protective sequestration” (shielding a healthy group from risk of infections by allowing no outsiders into the community during the epidemic) achieved the best protection if protective sequestration was implemented early enough and continued long enough.26
Aggressive social distancing was a mitigating or dominant factor in our scenarios. Our simulations that doubled social distancing outside the home (ie, reducing casual contacts by half) demonstrate that social distancing, if executed in a timely manner, can limit a local epidemic to the degree that demand for care will not exceed the community’s surge capacity. This finding is significant because it indicates that communities can actively participate in controlling a local epidemic without reliance on outside support, which may be inadequate and/or delayed, and despite the inadequate supplies of antiviral medications and vaccination.
Acknowledgments
We thank our partners within the San Antonio Metropolitan Health District, the Texas Department of State Health Services, the Greater San Antonio Hospital Council, the Southwest Texas Regional Advisory Council, the University of Texas Health Science Center at San Antonio, and the Texas A&M University Health Science Center, for their assistance in data acquisition and for their assistance with assumptions used in this analysis.
Financial support. This research program was funded by grant 1U18HS013683‐01 from the Agency for Healthcare Research and Quality of the Department of Health and Human Services as part of the Partnerships for Quality Program.
Potential conflicts of interest. All authors report no conflicts of interest relevant to this article.
References
- 1. Leitenberg M. Assessing the biological weapons and bioterrorism threat. Strategic Studies Institute, US Army War College, 2005. Available at: http://www.cissm.umd.edu/papers/files/assessing_bw_threat.pdf. Accessed December 28, 2006.
- 2. US General Accounting Office. Bioterrorism, Federal Research and Preparedness Activities. Washington, DC: US General Accounting Office; 2001:22. GAO‐01‐915.
- 3. Miller G, Randolph S, Gower D. Simulating the response of a rural acute health‐care delivery system to a bioterrorist attack. Intl J Disaster Med 2004; 2:24‐32.
- 4. Miller G, Randolph S, Patterson JE. Responding to bioterrorist smallpox in San Antonio. Interfaces 2006; 36:580‐590.
- 5. Elveback LR, Fox JP, Ackerman E, Langworthy A, Boyd M, Gatewood L. An influenza simulation model for immunization studies. Am J Epidemiol 1976; 103:152‐165.
- 6. Cunha B. Influenza: historical aspects of epidemics and pandemics.Infect Dis Clin North Am 2004; 18:141‐155.
- 7. Billings M. The Influenza Pandemic of 1918. June 1997. Available at: http://virus.stanford.edu/uda/. Accessed December 15, 2006.
- 8. Homeland Security Council. National Strategy for Pandemic Influenza. Washington, DC: Homeland Security Council; November 2005.
- 9. US Department of Health and Human Services. HHS Pandemic Influenza Plan. Washington, DC: US Dept of Health and Human Services; November 2005.
- 10. Bootsma CJ, Ferguson NM. The effect of public health measures on the 1918 influenza pandemic in US cities. 2007. Available at: http://www.pnas.org/cgi/doi/10.1073/pnas.0611071104. Accessed May 1, 2007.
- 11. Longini IM Jr, Nizam A, Xu S, et al. Containing pandemic influenza at the source. Science 2005; 309:1083‐1087.
- 12. Ferguson NM, Cummings DAT, Cauchemez S, et al. Strategies for containing an emerging influenza pandemic in Southeast Asia. Nature 2005; 437:209‐214.
- 13. Germann TC, Kadau K, Longini IM Jr, Macken CA. Mitigation measures for pandemic influenza in the United States. Proc Natl Acad Sci U S A 2006; 103:5935‐5940.
- 14. US Department of Health and Human Services, Centers for Disease Control and Prevention. Interim pre‐pandemic planning guidance: community strategy for pandemic influenza mitigation in the United States. February 2007. Available at: http://www.pandemicflu.gov/plan/community/mitigation.html. Accessed February 26, 2007.
- 15. 2000 US Census. Available at: http://www.census.gov/population/www/projections/st_yr01to05.html. Accessed July 14, 2006.
- 16. Randolph S, Miller G. Validation of the Casualty Prediction Model. Altarum Institute. July 2007. Available at: http://www.altarum.org/FCKeditor/UserFiles/File/CPMValidation_ib.pdf. Accessed November 19, 2007.
- 17. Longini IM, Koopman JS, Monto AS, Fox JP. Estimating household and community transmission parameters for influenza. Am J Epidemiol 1982; 115:736‐751.
- 18. Longini IM, Koopman JS, Haber M, Cotsonis GA. Statistical inference for infectious diseases, risk‐specific household and community transmission parameters. Am J Epidemiol 1988; 128:845‐859.
- 19. Barnitz L. The Health Care Response to Pandemic Influenza: A Position Paper. Washington, DC: American College of Physicians; 2006.
- 20. Gani R, Hughes H, Fleming D, Griffin T, Medlock J, Leach S. Potential impact of antiviral drug use during influenza pandemic. Emerg Infect Dis 2005; 11:1355‐1362.
- 21. Longini IM, Halloran ME, Nizam A, Yang Y. Containing pandemic influenza with antiviral agents. Am J Epidemiol 2004; 159:623‐633.
- 22. Patterson JE. SARS in Toronto: what does it mean for us? [commentary] Contagion 2005; 2:1‐4.
- 23. Hayden FG, Pavia AT. Antiviral management of seasonal and pandemic influenza. J Infect Dis 2006; 194:S119‐S126.
- 24. Treanor JJ, Campbell JD, Zangwill KM, Rowe T, Wolff M. Safety and immunogenicity of an inactivated subvirion influenza A (H5N1) vaccine. N Engl J Med 2006; 354:1343‐1351.
- 25. Markel H, Lipman HB, Navarro JA, et al. Nonpharmaceutical interventions implemented by US cities during the 1918‐1919 influenza pandemic. JAMA 2007; 298:644‐654.
- 26. Markel H, Stern AM, Navarro JA, Michalsen JR. A historical assessment of nonpharmaceutical disease containment strategies employed by selected US communities during the wave of the 1918‐1920 influenza pandemic. Defense Threat Reduction Agency, January 31, 2006. Available at: http://www.med.umich.edu/medschool/chm/influenza/. Accessed August 13, 2007.