Abstract
During the 2009 H1N1 influenza pandemic nearly every decision associated with new vaccine development and dissemination occurred from the Spring of 2009, when the novel virus first emerged, to the Fall of 2009, when the new vaccines started reaching the thighs, arms and noses of vaccinees. In many ways, 2009 served as a crash course on how mathematical and computational modeling can assist all aspects of vaccine decision-making. Modeling influenced pandemic vaccine decision-making, but not to its fullest potential. The 2009 H1N1 pandemic demonstrated that modeling can help answer questions about new vaccine development, distribution, and administration such as (1) is a vaccine needed, (2) what characteristics should the vaccine have, (3) how should the vaccine be distributed, (4) who should receive the vaccine and in what order and (5) when should vaccination be discontinued? There is no need to wait for another pandemic to enhance the role of modeling, as new vaccine candidates for a variety of infectious diseases are emerging every year. Greater communication between decision makers and modelers can expand the use of modeling in vaccine decision-making to the benefit of all vaccine stakeholders and health around the globe.
Key words: influenza, H1N1, vaccine, modeling, pandemic, vaccine development, vaccine distribution, vaccine administration
Introduction
The 2009 H1N1 influenza pandemic fast-forwarded the new vaccine development, production, distribution and administration time-line from multiple years to less than half a year. Nearly every decision associated with new vaccine development and dissemination occurred from the Spring of 2009, when the novel virus first emerged, to the Fall of 2009, when the new vaccines started reaching the population (Table 1). Prior influenza pandemics (i.e., 1918, 1957 and 1968) over the past century did not have available the mathematical and computational modeling expertise and techniques that we have today. In many ways 2009 served as a crash course on how modeling can assist all aspects of vaccine decision-making.
Table 1.
2009–2010 H1N1 influenza pandemic timeline
| H1N1 influenza timeline | H1N1 Vaccine timeline | |
| 2009 | ||
|
April | |
|
May |
|
|
June |
|
| July |
|
|
| August |
|
|
| September |
|
|
|
October |
|
|
November |
|
|
December |
|
| 2010 | ||
| January | ||
| February |
|
|
| March | ||
|
April |
|
| May | ||
|
June | |
| July | ||
|
August |
The circumstances were ripe for the use of modeling: complex decisions with far-reaching implications, multiple stakeholders, time and practical constraints precluding traditional epidemiologic and clinical studies, and the world's eyes watching every step. Modeling has been a mainstay of decision-making in other industries (e.g., weather forecasting, stock and options pricing, transportation planning, manufacturing, natural resource exploration and aeronautical engineering) for years. In fact, vaccine decision-making may be ideally suited for modeling: retrospective data are often limited, prospective studies are difficult and in some cases impossible to perform, and decisions are complex with wide reverberations. Vaccine decisions cross a wide variety of disciplines and involve an array of people and resources.
The 2009 H1N1 pandemic also introduced much of the public to influenza modeling, as articles in newspapers and magazines described the ongoing influenza modeling activities. Modeling is the use of mathematical or computational equations to represent decisions, phenomena, and processes. Models can range from decision trees portraying the steps and options comprising a decision to large-scale agent-based models that simulate the people, locations and activities in a geographic region such as Allegheny County, the Washington, DC metropolitan region, the state of Pennsylvania, or the United States.1–5
Modeling did influence 2009 pandemic vaccine decision-making. During the pandemic, our modeling team from the University of Pittsburgh Models of Infectious Disease Agent Study (MIDAS) National Center of Excellence worked closely with the Office of the Assistant Secretary for Preparedness and Response (ASPR) at the US Department of Health and Human Services. This included being “embedded” in Biomedical Advanced Research and Development Authority (BARDA) offices for over two months. We also worked with the Department of Homeland Security (DHS) and Centers for Disease Control and Prevention (CDC) as well as public health officials in the State of Pennsylvania and the Allegheny County Health Department. Much of our work involved exploring different vaccination scenarios to assist decision-making.
Did modeling contribute to its fullest potential to vaccine decision-making during the H1N1 pandemic? The answer is a resounding no. Previous work on H5N1 avian influenza may have influenced opinions; some stakeholders, such as manufacturers, may have relied on internal models; and as mentioned above, some modelers worked directly with public health officials.6–9 However, the majority of decisions that could have or even should have benefited from modeling did not. While decision makers in many industries would not think of proceeding without reviewing model results (imagine launching the space shuttle, tracking a hurricane, or making a major investment decision without a model), the public health and biomedical arenas have not yet embraced modeling with the same fervor. What follows is a chronicle of the types of vaccine decisions encountered during the 2009 H1N1 pandemic and how modeling helped or could have helped.
Questions Facing New Vaccines
Is a vaccine needed?
When the novel H1N1 influenza virus emerged, one of the first decisions was whether developing a vaccine would even be needed or useful. For nearly any infectious disease, modeling can help forecast the value of a potential vaccine by estimating the morbidity, mortality, and economic burden of the disease and determining whether a vaccine could better mitigate the disease than other existing measures.8,10 Outbreaks that are too small or too rapid may not benefit from a vaccine. Also, other pharmaceutical (e.g., antivirals) or non-pharmaceutical interventions (e.g., social distancing measures) may be alternatives.
It is unclear to what degree decision makers relied on previous H5N1 and concurrent H1N1 model results when giving the green-light to develop H1N1 vaccines. Regardless, model explorations supported development by demonstrating the value of an H1N1 vaccine over that of other potential strategies. For example, school closures alone may not have a noticeable impact unless maintained strictly for a long period of time (at least eight weeks). In fact, short-term (one or two week) school closures have the potential to worsen an epidemic by re-releasing susceptible schoolchildren back to schools in the middle of the epidemic.3 Similarly, while antiviral medications could be helpful, models suggest that they would have done little to quell the epidemic as an isolated intervention.11,12
Lesson for future vaccine decision-making. Models can help investigators, policy makers, investors, and manufacturers decide whether to pursue developing a vaccine.
What characteristics should the vaccine have?
Following the go-ahead to develop and manufacture a vaccine, questions about its target characteristics emerged. How efficacious does the vaccine need to be to quell the epidemic and to be worthwhile? How many doses should each individual receive? What price points would be reasonable? What side effect probabilities would be acceptable? Modeling can help forecast the impact of varying different vaccine candidate characteristics, set targets and thresholds for these characteristics, and in turn, guide development and prepare a vaccine candidate for the market.1,13,14 Some stakeholders (e.g., manufacturers) may have used modeling to help establish target characteristics and facilitated price negotiations for the H1N1 vaccine, but most of this modeling was internal and did not enter public discussions.
Lesson for future vaccine decision-making. Constructing models early in development, when a vaccine's characteristics can still be altered, can enhance its chances of success. Such modeling could have helped some past vaccine candidates (e.g., FluMist vaccine against influenza and LYMErix vaccine against Lyme Disease) avoid obstacles that they encountered once they reached the market.15
How should the vaccine be distributed?
Since a vaccine has to reach vaccinees to work, setting up an effective vaccine supply chain (the series of steps required to get a vaccine from the manufacturers to patients) is essential. The design and operation of a supply chain determines when the population is actually immunized. The timing of immunization can greatly impact an individual's and a population's risk of disease, especially during an epidemic.16–18 Even a perfect vaccine can do little if it does not reach people.15
Because vaccine distribution can be quite complex, modeling may be one of the few methods available to predict the costs and effects of different strategies. Prospective studies can be costly and time consuming. There are too many variables to simply think through the problem or rely on gut instincts. Models essentially make decision-making transparent and place it “on the table” for others to see, comment, and adjust. Distribution companies such as McKesson, which helped handle the H1N1 vaccines, already utilize models extensively to plan their operations. Logistics experts for vaccine manufacturers often utilize models as well. However, other vaccine decision makers (e.g., public health officials and scientists) may not be using vaccine distribution models as much as they should. The public health and biomedical literature certainly could use more vaccine distribution modeling studies.
Lesson for future vaccine decision-making. Not all vaccine decision makers are fully utilizing models to inform vaccine distribution decisions.
Who should receive the vaccine and in what order?
The vaccine became available in limited quantities in October 2009, necessitating initial rationing. Public health officials had to select the initial target populations and the order in which people would receive the vaccine. Should immunization be “first come first served”? Alternatively, should children, older adults, health care workers, pregnant women, or other higher-risk individuals receive vaccine first?4,19 Once an immunization order is established, how strictly should this be followed? In late August 2009, the Advisory Committee on Immunization Practices (ACIP) released its recommendations for the use of the 2009 H1N1 vaccine, based on reviewing the literature and expert opinion.20 Subsequently, our work with ASPR included evaluating different vaccine prioritization strategies and determining the effects of varying compliance to each of these strategies.5 Ultimately, model results favored early allocation to ACIP priority groups (versus other strategies). In the future, perhaps modeling could more actively assist expert panels, such as the ACIP, when formulating initial recommendations.
Lesson for future vaccine decision-making. Because vaccinating an entire population may not be possible or indicated, models can help identify and prioritize the vaccine's target populations.
When should vaccination be discontinued?
As the pandemic approached its peak in October 2009 when vaccines first became available, the next question was whether continuing vaccination was worthwhile. Some questioned the utility of mass vaccinating a population that had already been widely exposed to natural infection. Would immunization have much effect when the pandemic seemed to have already run its course? Could there be a subsequent upsurge in the pandemic without vaccination?21 Our explorations with ASPR supported continuing the immunization program by demonstrating how it could prevent a third pandemic wave (with the first two waves occurring in the Spring and Fall of 2009) from emerging in early 2010.18
Lesson for future vaccine decision-making. Once a vaccine is available, the epidemiology of an infectious disease may change. Models can help determine whether a vaccine and vaccination strategy need updating.
Expanding the Role of Modeling
As the 2009 H1N1 pandemic demonstrated, there are many opportunities for modeling to facilitate vaccine decision-making. Decision makers and modelers capitalized on some of these opportunities, but also missed many of them. So, how can the role of modeling in vaccine decision-making be expanded in the future, both in epidemic and non-epidemic settings? Decision makers and modelers could do following:
Decision makers: understand what models can and cannot do. Modelers: be better at communicating what models can and cannot do. The underuse or misuse of models often arises from a misunderstanding of the purpose of models. Models are simplifications, not replicas of real life; models can assist but should not make decisions. Modeling helps people better understand their own and others' decision-making processes by bringing them out into the open for everyone to view. Even the most brilliant and experienced minds cannot account for every factor in a complex decision the way that models can. Models are best at identifying important relationships, key factors in a decision, important questions and information that needs to be gathered. Anyone expecting models to give exact predictions, completely mimic real life, or represent every possible eventuality will be sorely disappointed.
Modelers: be as transparent as possible about models. Decision makers: give modelers the opportunity to fully explain their models. Models should not be “magic boxes” that spew out results. Without understanding a model's structure and parts, decision makers should neither implicitly trust nor disregard a model. Therefore, clear written and oral communication between decision makers and modelers is essential. Modelers must clearly state the advantages, disadvantages, assumptions and limitations of their models.
Decision makers: understand that all models are not the same and the value of multiple models addressing the same question. Modelers: encourage questions and competing models. Would you ever rely on a single retrospective or prospective study to make decisions? Should encountering a poorly constructed clinical study mean you should disregard all clinical studies? Similarly, no model is perfect—every model has its strengths and weaknesses. There is a wide range in model quality and comprehensiveness. Therefore, having different modelers and models tackle the same question can provide valuable insights. Comparing and revealing the differences among various models can be enlightening, similar to bringing multiple experts to the decision-making table.
Modelers: fully understand the important questions and the accompanying circumstances. Decision makers: tell modelers what questions are relevant. Models should reflect relevant decisions and incorporate important factors. Therefore, modelers should either be or work closely with subject matter experts. Otherwise, the model may be too conceptual or unrealistic and therefore be of limited value to decision makers.
Modelers: keep models as simple as possible. Decision makers: communicate which details matter and why. Models should only be as complex as needed. The purpose of modeling is to distill a decision down to its most important components and relationships. Adding unnecessary detail to a model only clouds the picture.
Decision makers: understand that all decisions can be modeled. Modelers: show how each decision can be modeled. When each of us makes a decision, we consciously or subconsciously model the decision in our heads. Even instinctual decisions are the result of rapidly operating internal mental models built from years of experience. If you can think it, you can model it.
Decision makers: help provide or find data for models. Modelers: clearly identify data needs. The 2009 pandemic modeling efforts revealed some important data gaps. No one knew the current status of the pandemic. At a given point in time, was the pandemic waxing or waning, near or far from its peak? How many people had been infected? What percentage of the infected individuals exhibited symptoms? Models were highly sensitive to this information. A vaccination program early in a pandemic would be much more effective than one late in the pandemic. This underscored the need for a national close-to-real time serologic surveillance program.
Decision makers and modelers: communicate and understand each other. Open communication is key. Modelers need to present their models and results in formats that are easily understandable and digestible by decision makers. During our H1N1 work with public health officials, we found that more traditional scientific graphs and charts were not always effective in communicating results. We had to speak the “language” of decision makers. Therefore, a substantial part of our efforts was devising better ways to convey our work (e.g., visualizations). At the same time, it is helpful for modelers to know what decision makers are thinking.
Conclusions
As the 2009 H1N1 pandemic demonstrated, modeling is a potentially powerful methodology that is currently underutilized (and in some cases mis-utilized) in vaccine decision-making. If used appropriately, modeling could benefit nearly every decision in new vaccine development, distribution and administration. There is certainly no need to wait for another pandemic to enhance the role of modeling, as new vaccine candidates for a variety of infectious diseases are emerging every year. Greater communication between decision makers and modelers can expand the use of modeling in vaccine decision-making to the benefit of all vaccine stakeholders and global health.
Acknowledgements
The University of Pittsburgh MIDAS National Center of Excellence H1N1 influenza modeling team, led by Donald S. Burke, MD, Dean of the Graduate School of Public Health, consisted of Kristina M. Bacon, MPH, Rachel R. Bailey, MPH, Shawn T. Brown, Ph.D., John J. Grefenstette, Ph.D., Bruce Y. Lee, MD, MBA, Margaret A. Potter, JD, MS, Roni Rosenfeld, Ph.D., Ronald E. Voorhees, MD, MPH, Ann E. Wiringa, MPH, Shanta M. Zimmer, MD and Richard K. Zimmerman, MD, MPH. From September 2009 to October 2009, Drs. Lee and Brown were “embedded” in ASPR. This work was supported by the National Institute of General Medical Sciences Models of Infectious Disease Agent Study (MIDAS) grant 1U54GM088491-0109, and the Vaccine Modeling Initiative (VMI) through support from the Bill and Melinda Gates Foundation. The funders had no role in the preparation, review or approval of the manuscript.
Abbreviations
- MIDAS
Models of Infectious Disease Agent Study
- ASPR
Assistant Secretary for Preparedness and Response
- BARDA
Biomedical Advanced Research and Development Authority
- DHS
Department of Homeland Security
- CDC
US Centers for Disease Control and Prevention
- ACIP
Advisory Committee on Immunization Practices
- FDA
US Food and Drug Administration
- WHO
World Health Organization
- NIH
National Institute of Health
References
- 1.Cooley P, Lee BY, Brown S, Cajka J, Chasteen B, Ganapathi L, et al. Protecting health care workers: a pandemic simulation based on Allegheny County. Influenza Other Respi Viruses. 2010;4:61–72. doi: 10.1111/j.1750-2659.2009.00122.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Lee BY, Bailey RR, Wiringa AE, Afriyie A, Wateska AR, Smith KJ, et al. Economics of employer-sponsored workplace vaccination to prevent pandemic and seasonal influenza. Vaccine. 2010 doi: 10.1016/j.vaccine.2010.07.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Lee BY, Brown ST, Cooley P, Potter MA, Wheaton WD, Voorhees RE, et al. Simulating school closure strategies to mitigate an influenza epidemic. J Public Health Manag Pract. 2010;16:252–261. doi: 10.1097/PHH.0b013e3181ce594e. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Lee BY, Brown ST, Cooley PC, Zimmerman RK, Wheaton WD, Zimmer SM, et al. A computer simulation of employee vaccination to mitigate an influenza epidemic. Am J Prev Med. 2010;38:247–257. doi: 10.1016/j.amepre.2009.11.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Lee BY, Brown ST, Korch GW, Cooley PC, Zimmerman RK, Wheaton WD, et al. A computer simulation of vaccine prioritization, allocation and rationing during the 2009 H1N1 influenza pandemic. Vaccine. 2010;28:4875–4879. doi: 10.1016/j.vaccine.2010.05.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Longini IM, Jr, Nizam A, Xu S, Ungchusak K, Hanshaoworakul W, Cummings DA, et al. Containing pandemic influenza at the source. Science. 2005;309:1083–1087. doi: 10.1126/science.1115717. [DOI] [PubMed] [Google Scholar]
- 7.Ferguson NM, Cummings DA, Cauchemez S, Fraser C, Riley S, Meeyai A, et al. Strategies for containing an emerging influenza pandemic in Southeast Asia. Nature. 2005;437:209–214. doi: 10.1038/nature04017. [DOI] [PubMed] [Google Scholar]
- 8.Ferguson NM, Cummings DA, Fraser C, Cajka JC, Cooley PC, Burke DS. Strategies for mitigating an influenza pandemic. Nature. 2006;442:448–452. doi: 10.1038/nature04795. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Halloran ME, Ferguson NM, Eubank S, Longini IM, Jr, Cummings DA, Lewis B, et al. Modeling targeted layered containment of an influenza pandemic in the United States. Proc Natl Acad Sci USA. 2008;105:4639–4644. doi: 10.1073/pnas.0706849105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Lee BY, Bedford VL, Roberts MS, Carley KM. Virtual epidemic in a virtual city: simulating the spread of influenza in a US metropolitan area. Transl Res. 2008;151:275–287. doi: 10.1016/j.trsl.2008.02.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Lee BY, Bailey RR, Wiringa AE, Assi TM, Beigi RH. Antiviral medications for pregnant women for pandemic and seasonal influenza: an economic computer model. Obstet Gynecol. 2009;114:971–980. doi: 10.1097/AOG.0b013e3181bdbfed. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Lee BY, McGlone SM, Bailey RR, Wiringa AE, Zimmer SM, Smith KJ, et al. To test or to treat? An analysis of influenza testing and antiviral treatment strategies using economic computer modeling. PLoS One. 5:11284. doi: 10.1371/journal.pone.0011284. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Lee BY, Burke DS. Constructing target product profiles (TPPs) to help vaccines overcome post-approval obstacles. Vaccine. 2010;28:2806–2809. doi: 10.1016/j.vaccine.2009.09.047. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Lee BY, Wiringa AE, Bailey RR, Lewis GJ, Feura J, Muder RR. Staphylococcus aureus vaccine for orthopedic patients: an economic model and analysis. Vaccine. 2010;28:2465–2471. doi: 10.1016/j.vaccine.2009.12.075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lee BY, Burke DS. Constructing target product profiles (TPPs) to help vaccines overcome post-approval obstacles. Vaccine. 28:2806–2809. doi: 10.1016/j.vaccine.2009.09.047. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Lee BY, Tai JH, Bailey RR, Smith KJ. The timing of influenza vaccination for older adults (65 years and older) Vaccine. 2009;27:7110–7115. doi: 10.1016/j.vaccine.2009.09.056. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lee BY, Tai JH, Bailey RR, Smith KJ, Nowalk AJ. Economics of influenza vaccine administration timing for children. Am J Manag Care. 16:75–85. [PMC free article] [PubMed] [Google Scholar]
- 18.Lee BY, Brown ST, Cooley P, Grefenstette JJ, Zimmerman RK, Zimmer SM, et al. Continuation of vaccination deep into a pandemic wave: Potential mechanisms for a “Third Wave” and the impact of vaccination. Am J Prev Med. 2010 doi: 10.1016/j.amepre.2010.07.014. In Press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Beigi RH, Wiringa AE, Bailey RR, Assi TM, Lee BY. Economic value of seasonal and pandemic influenza vaccination during pregnancy. Clin Infect Dis. 2009;49:1784–1792. doi: 10.1086/649013. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Use of influenza A (H1N1) 2009 monovalent vaccine: recommendations of the Advisory Committee on Immunization Practices (ACIP) 2009. MMWR Recomm Rep. 2009;58:1–8. [PubMed] [Google Scholar]
- 21.Shi P, Keskinocak P, Swann JL, Lee BY. Modelling seasonality and viral mutation to predict the course of an influenza pandemic. Epidemiol Infect. 2010:1–10. doi: 10.1017/S0950268810000300. [DOI] [PMC free article] [PubMed] [Google Scholar]