Managing emerging vaccine safety signals during an influenza pandemic is challenging. Federal regulators must balance vaccine risks against benefits while maintaining public confidence in the public health system.
We developed a multi-criteria decision analysis model to explore regulatory decision-making in the context of emerging vaccine safety signals during a pandemic. We simulated vaccine safety surveillance system capabilities and used an age-structured compartmental model to develop potential pandemic scenarios. We used an expert-derived multi-attribute utility function to evaluate potential regulatory responses by combining four outcome measures into a single measure of interest: 1) expected vaccination benefit from averted influenza; 2) expected vaccination risk from vaccine-associated febrile seizures; 3) expected vaccination risk from vaccine-associated Guillain-Barre Syndrome; and 4) expected change in vaccine-seeking behavior in future influenza seasons.
Over multiple scenarios, risk communication, with or without suspension of vaccination of high-risk persons, were the consistently preferred regulatory responses over no action or general suspension when safety signals were detected during a pandemic influenza. On average, the expert panel valued near-term vaccine-related outcomes relative to long-term projected outcomes by 3∶1. However, when decision-makers had minimal ability to influence near-term outcomes, the response was selected primarily by projected impacts on future vaccine-seeking behavior.
The selected regulatory response depends on how quickly a vaccine safety signal is identified relative to the peak of the pandemic and the initiation of vaccination. Our analysis suggested two areas for future investment: efforts to improve the size and timeliness of the surveillance system and behavioral research to understand changes in vaccine-seeking behavior.
Citation: Maro JC, Fryback DG, Lieu TA, Lee GM, Martin DB (2014) Responding to Vaccine Safety Signals during Pandemic Influenza: A Modeling Study. PLoS ONE 9(12): e115553. https://doi.org/10.1371/journal.pone.0115553
Editor: Jodie McVernon, Melbourne School of Population Health, Australia
Received: August 12, 2014; Accepted: November 25, 2014; Published: December 23, 2014
This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. All relevant data are within the paper.
Funding: This project is supported by the Mini-Sentinel Contract, which is funded by the U.S. Food and Drug Administration (USFDA) through the Department of Health and Human Services Contract Number HHSF223200910006I. The funders had no role in data collection and analysis. The USFDA reviewed and approved this report before submission.
Competing interests: The authors have declared that no competing interests exist.
Responding to influenza vaccine safety signals experienced during a pandemic is a scientific and public policy challenge. Not only must federal decision-makers balance the immediate consequences of pandemic disease against uncertain vaccine risks, they also must weigh how federal actions might affect future vaccine-seeking behavior. For instance, in 1976, after initiating a National Influenza Immunization Program in response to a localized swine flu outbreak, federal authorities suspended vaccination after ten weeks because preliminary surveillance suggested that the incidence of Guillain-Barre Syndrome was approximately seven-fold greater among vaccinees .
Given that this particular swine flu virus was never isolated outside of Fort Dix , the benefit-risk calculus appears simple in hindsight. However, the decision to initiate and then withdraw a mass vaccination campaign was regarded by some as a public health failure , resulting in sustained and unforeseen consequences on vaccine-seeking behavior, and loss of public confidence in decision-making. Firsthand accounts – and historical assessments ,  have emphasized the difficulty of compressed decision-making under conditions of uncertainty. While improvements in near real-time vaccine safety surveillance now allow earlier detection of vaccine safety signals , , the need to act in the context of scientific uncertainty has not changed.
These circumstances are ripe for simulation and decision models. Recent pandemic threats and the pandemic potential of H5N1 and H7N9 viruses have stimulated multiple preparedness efforts – including scenario-based mathematical modeling , . Prior models have focused on influenza transmission , , optimal vaccine allocation –, social distancing –, antivirals , and layered interventions –. However, none have considered regulatory responses to vaccine safety signals emerging during the course of a mass vaccination program.
We addressed this gap using a multi-criteria decision analysis (MCDA) that explored regulatory decision-making in the context of emerging vaccine safety signals experienced during pandemic influenza. Specifically, the MCDA we developed evaluates the effect of several alternative regulatory responses on the transmissibility and severity of the pandemic, the burden of adverse events, and the potential for sustained changes in vaccine-seeking behavior.
Materials and Methods
The MCDA included several linked models. First, the vaccine safety signal was simulated in a model of the surveillance system. Next, this signal and four potential regulatory responses were the triggering input in a pandemic influenza transmission model. Each response affected short-term vaccine-associated benefits and risks (i.e., within the pandemic period), and future vaccine-seeking behavior. The outputs of the influenza transmission model were inputs to an expert-derived multi-attribute utility function. The multi-attribute utility function is used to weight and combine multiple outcomes into a single figure of merit whose expected value was maximized to select the preferred regulatory decision.
Model of Influenza Vaccination Surveillance System
We simulated surveillance in the Post-licensure Rapid Immunization Safety Monitoring (PRISM) system, which is currently being tested for influenza vaccine safety surveillance . We projected 4.3 million adopters of influenza vaccination in the PRISM system based on prior data. The methodology for the simulation model is described elsewhere  and a detailed description of the model is in S1 File.
Strength of Vaccine Safety Signals.
In each scenario, we evaluated three vaccine safety signals based on historical precedent: 1) vaccine-associated febrile seizures –, 2) vaccine-associated Guillain-Barre Syndrome , , and 3) both febrile seizures and Guillain-Barre Syndrome. We simulated a vaccine-associated febrile seizures effect size as an incidence risk difference of ∼150 excess febrile seizures per 100,000 doses in a cohort of 0–5 year olds when compared with a historical cohort of seasonal influenza vaccinees. We simulated a vaccine-associated Guillain-Barre Syndrome effect size as an incidence risk difference of 40 excess cases of Guillain-Barre Syndrome per one million doses when compared with a historical cohort of seasonal influenza vaccinees. We chose these levels of risk because they might plausibly have escaped detection in clinical trials and thus pose a particular challenge for decision-makers. We also presumed that vaccine safety signals were not unique to a single manufacturer or to a specific lot or batch number.
Influenza Transmission Model with Bass Diffusion of Influenza Vaccine
We adapted an age-structured disease transmission model  and added an influenza vaccination adoption function modeled as a Bass diffusion process while assuming a universal vaccination policy . Bass diffusion models are commonly used in the marketing literature to describe the diffusion of innovations , . A detailed description of the deterministic, compartmental Susceptible-Exposed-Infectious-Recovered (SEIR) influenza transmission model and Bass diffusion process are included in S1 File.
Pandemic Influenza Parameters.
We characterized influenza epidemics by their transmissibility, severity, and timing. We used R0, the basic reproduction number, to characterize transmissibility. Severity was measured by influenza morbidity and mortality. Timing refers to the amount of circulating virus present at the time vaccination began in the modeled U.S. population. Modeled circulating virus determined whether the peak of vaccination coverage was likely to precede, run concurrent with, or follow the peak of influenza transmission.
We implemented two scenarios: mild and severe influenza. The “mild” scenario was an influenza that had low transmissibility (i.e., R0 = 1.4), low severity, and the peak of vaccination preceded the peak of influenza transmission. The “severe” scenario involved high transmissibility (i.e., R0 = 2.0), moderate-to-high severity, and influenza transmission and vaccination peaked concurrently.
Influenza Vaccine Parameters.
We assumed vaccination began September 1. We chose vaccination effectiveness parameters and expected vaccination coverage based on data observed during the H1N1 pandemic. Parameter details are in S1 File.
Multi-Criteria Decision Analysis Model
Simulated Regulatory Responses.
When a vaccine safety signal was detected, the MCDA was evaluated using four simulated regulatory responses: 1) No Action – no communication from the regulatory agency to the public; 2) Risk Communication Alone – risk communication issued (e.g., a “Dear Healthcare Provider” letter or website announcement) that described the vaccine safety signal and identified “at-risk” individuals for a vaccine-associated adverse event but did not recommend changes in vaccine use; 3) Selective Suspension – risk communication issued and vaccination suspended in “at-risk” individuals; and 4) General Suspension – risk communication issued and vaccination suspended for all individuals. These four responses were an informative sample as the complete range of responses was beyond the scope of this activity.
The No Action response did not alter vaccine-seeking behavior and universal vaccination coverage continued as it had before. The Risk Communication Alone response reduced vaccine-seeking behavior among the “at risk” individuals targeted in the risk communication. “At risk” individuals were defined as children age 5 and below for febrile seizures and adults age 50 and above for Guillain-Barre Syndrome based on historical data associated with these vaccine safety signals –. The individuals who were not at highest risk for the adverse event also reduced their vaccine-seeking behavior by a minor amount during the pandemic influenza. The Selected Suspension response reduced vaccine-seeking behavior to zero in “at risk” individuals, and all others reduced their vaccine-seeking behavior by a minor amount during the pandemic influenza. The General Suspension response reduced vaccine-seeking behavior to zero because the vaccine became unavailable.
Expert-Derived Multi-Attribute Utility Function.
In MCDAs, decisions are characterized by multiple competing criteria, and decision-makers must consider all criteria when evaluating possible decision options. A multi-attribute utility function is a mathematical equation used to characterize the overall value (or “utility”) of each decision option relative to the others based on the specified criteria . We developed an expert-derived additive multi-attribute utility function to evaluate the MCDA. The four criteria of interest were: 1) expected vaccination benefit from averted influenza as measured by a composite index of influenza cases, hospitalizations, and deaths averted; 2) expected vaccination risk from vaccine-associated febrile seizures as measured by attributable cases; 3) expected vaccination risk from vaccine-associated Guillain-Barre Syndrome as measured by attributable cases; and 4) expected change in vaccine-seeking behavior in future seasons as a consequence of public reaction to changes in federal vaccination policy during the pandemic. The first three criteria were directly calculated from the influenza transmission model and were limited to the pandemic time period. The fourth criterion describes vaccine-seeking behavior in future seasons. For this criterion, we used a qualitative variable with three levels: a) no change in future vaccine-seeking behavior, b) minor change: anticipated 10% reduction in future vaccine-seeking behavior, and c) major change: anticipated 25% reduction in future vaccine-seeking behavior. These levels were based on anecdotes related to the 1976 swine flu experience – and small studies showing that perceptions about vaccine-associated adverse events can meaningfully reduce vaccine-seeking behavior –, but they were not validated by empirical research.
We convened an expert panel of six physicians who were currently serving, or had previously served, on vaccine-related federal advisory committees to elicit expert preferences , ,  on prioritization of the four criteria of interest as shown in Table 1. We performed separate elicitations for the mild and severe scenarios, thereby creating separate multi-attribute utility functions for the two situations. We combined each expert panelist's multi-attribute utility function to derive the “average” decision-maker. A detailed description of the multi-attribute utility function elicitation process and result is in S1 File.
Conditional Probability Elicitation.
In the MCDA, we were interested in modeling the general public's reaction to federal vaccination policy changes and how that reaction translated into vaccine-seeking behavior in future seasons. We lacked any public preference surveys similar to those described in ,  and therefore, we asked the expert panel to hypothesize about changes to future vaccine-seeking behavior as a result of reaction to federal vaccination policy. For example, we asked panelists to describe the probability that the public would a) not change their future vaccine-seeking behavior, b) reduce it by 10%, or c) reduce it by 25%, if the government received a vaccine safety signal in the mild scenario and responded with No Action. We repeated this procedure for both scenarios and all four regulatory responses. Table 2 lists their averaged probabilities.
All modeling and subsequent analyses were completed using MATLAB and R.
Vaccine Safety Signal Detection
Based on our assumptions of the surveillance system, the febrile seizures signal was detected most often two months after the start of vaccination whereas the Guillain-Barre Syndrome signal was detected most often six months after the start of vaccination (Fig. 1). Statistical power to detect a vaccine safety signal at the effect size we tested was nearly 100% for febrile seizures and 90% for Guillain-Barre Syndrome. That is, in 10% of the simulations, the increased risk of vaccine-associated Guillain-Barre Syndrome was missed.
Left panel is febrile seizures; right panel is Guillain-Barre Syndrome. For Guillain-Barre Syndrome, the safety signal remains undetected (is missed) 10% of the time. “o” is the median, “x” is the mean, “*” is the 80th percentile. Other details in S1 File.
Multi-Criteria Decision Analysis Results
The MCDA selected the regulatory responses shown in Table 3 primarily based on the prioritization, or relative weight, assigned to the four competing criteria by the expert panel in Table 1. In both scenarios, the average decision-maker valued near-term vaccine-related outcomes relative to long-term, projected outcomes by 3∶1. Table 2 shows how the expert panel linked regulatory responses to those long-term projections. In particular, General Suspension maximized negative long-term impacts on vaccine-seeking behavior in the mild scenario whereas No Action did in the severe scenario. Figs. 2 and 3 are representative instantiations of the mild and severe scenario respectively and show the near-term vaccine-related outcomes associated with each regulatory response. Interpreted together, outcomes in Figs. 2 and 3 are modified by the weight assigned to them in Table 1 and the probability of undesirable long-term effects in Table 2. Fig. 1 represents the chance of vaccine safety signal detection in the months following the start of vaccination.
The Multi-Criteria Decision Analysis model selected Risk Communication Alone or Selective Suspension as the preferred regulatory response.
The outcomes that occur with each of the four simulated regulatory responses are shown in each panel. The Multi-Criteria Decision Analysis model selected Risk Communication Alone as the preferred regulatory response.
In the mild scenario, by the time intervention opportunities occurred with a probability described in Fig. 1, the peak of vaccination had passed and the indirect benefits of vaccination had been achieved. Therefore, vaccine associated-benefits for all regulatory responses except for General Suspension were roughly similar. Over the multiple intervention opportunities and types of vaccine safety signals in Table 3, General Suspension was not preferred because of its lower vaccine-related benefit (see Fig. 2). (Note: the MCDA used a composite index for benefits but we show saved hospitalizations here.) Also, No Action was not preferred because of its projected impact on future vaccine-seeking behavior as described in Table 2. With those regulatory responses eliminated, the MCDA was indifferent between the remaining two options.
Generally, Risk Communication Alone had marginally higher vaccine-associated benefit than Selective Suspension (e.g., a difference of ∼2000 saved hospitalizations in Fig. 2), but also had higher vaccine-associated risk (e.g., a difference of 24 Guillain-Barre Syndrome cases in Fig. 2).
In the severe scenario, intervention opportunities occurred with the same probability as in the mild scenario, but the timing of peak vaccination coincided with peak influenza transmission, which meant that decision-makers had almost no influence on the pandemic's near-term outcomes (see Fig. 3). With near-term vaccine-related benefits converging for the four regulatory responses, the MCDA was driven by projected long-term impacts on future vaccine-seeking behavior. In other words, with little leverage in the present pandemic, decision-makers focused on the future. Under these circumstances, the MCDA favored Risk Communication Alone most consistently over multiple intervention points in Table 3 because it minimized negative impacts on future vaccine-seeking behavior as shown in Table 2.
Depending on the month of detection of the vaccine safety signal, the MCDA was sometimes indifferent between Risk Communication Alone and other options when a Guillain-Barre Syndrome safety signal was detected. This occurred because avoided vaccine-associated Guillain-Barre Syndrome cases compensated for greater negative impacts on future vaccine-seeking behavior. However, the number of vaccine-associated Guillain-Barre Syndrome cases that offset the projected negative impact was quite low (e.g. 5 avoided cases). Compensation sufficient to alter the MCDA's calculated preferences did not occur with a febrile seizures signal because of the low weight placed on avoided vaccine-associated febrile seizures.
Timing of Vaccination Availability.
We varied the availability of vaccination from August 1 (day 1) to November 1 (day 90). In the mild scenario, the decision was unchanged from the results shown in Table 3, but the logic contributing to the decision did change. In cases when the timing of vaccine safety signal detection preceded the peak of the influenza epidemic (February 1), the MCDA was indifferent between Risk Communication Alone and Selective Suspension because the former was associated with greater vaccine-associated benefits and greater vaccine-associated risk. However, when vaccine safety signal detection aligned with the peak of the influenza pandemic, then vaccine-associated benefits and risks converged among decision options, and the projected impact on future vaccine-seeking behavior determined the decision. Unlike in the severe scenario when Risk Communication Alone has the best projection on this attribute, in the mild scenario, Risk Communication Alone and Selective Suspension are nearly tied on this attribute and the MCDA is indifferent among them for this reason.
In the severe scenario, if vaccine is available August 1 at the start of the pandemic scenario and thus precedes the peak of influenza, Risk Communication Alone is the preferred decision option for any receipt of a vaccine safety signal. It performs the best on vaccination-related benefits and projected impacts on future vaccine-seeking behavior.
We recalibrated the multi-attribute utility function with equal prioritization for the four criteria in Table 1 (i.e., all set to 0.25) and re-ran the mild scenario. Here, vaccine-associated risk was more important to the decision-maker since its weight increased from 0.01 to 0.25 for vaccine-associated febrile seizures and 0.16 to 0.25 for vaccine-associated Guillain-Barre Syndrome. However, these adjustments did not change the projected impact on future vaccine-seeking behavior (i.e., the probabilities in Table 2), and No Action and General Suspension remained undesirable. The MCDA consistently preferred Selective Suspension to Risk Communication Alone regardless of when the safety signal was received because more weight was assigned to avoiding excess vaccine-related risk.
We also evaluated the MCDA while re-parameterizing the vaccine to be half as effective. While absolute levels of vaccine associated-risks and benefits changed, these changes did not affect the relative standing among regulatory responses. Therefore, the decision was insensitive to the changed parameters.
We developed an MCDA to evaluate regulatory decision-making following the emergence of vaccine safety signals and evaluated potential regulatory responses. The MCDA selected the best relative regulatory response among available choices according to the averaged, expert-elicited preferences by maximizing the expected value of the utility of these options. Numerically, the expected values were often close among options, even though these regulatory responses carry different logistical, social, and risk communication implications. Over multiple safety signal timing situations, Risk Communication Alone was consistently the preferred option in both scenarios, closely followed by Selective Suspension in the mild pandemic scenario cases only. The preferred decision might change if the multi-attribute utility function were developed with other stakeholders' preferences.
Why a Multi-Criteria Decision Analysis Is Important
The MCDA prompts decision-makers to transparently and explicitly describe the determinants of their decision-making by weighting multiple competing criteria. While we present two scenarios herein, the model is flexible enough to be re-run quickly with multiple sets of differing assumptions to understand which regulatory responses are robust under numerous circumstances.
Other regulatory agencies have vetted the use of MCDA to aid decision-making . The general structure of our pandemic influenza MCDA enables regulatory decision-makers outside of the U.S. to utilize our approach. However, sub-model components would benefit from enhanced specificity for extension beyond the U.S. For instance, we chose simulated regulatory responses that are globally generalizable, but other users might want to precisely tailor the responses to their country or region. Also, we modeled detection of vaccine safety signals in the U.S. PRISM medical product safety surveillance system, but other regional medical product safety surveillance systems would need to be explicitly modeled to determine their performance in a pandemic context. The multi-attribute utility function could be adopted, but it could also be re-parameterized using input from experts in other countries and regions.
What This Multi-Criteria Decision Analysis Tells Us
When the near-term benefits and risks of regulatory responses converge, the MCDA highlighted the weight decision-makers gave to the long-term stability of vaccination programs. That is, Risk Communication Alone and Selective Suspension were more desirable because of perceptions that No Action or General Suspension would have created undesirable effects on future vaccine-seeking behavior. Timing was critical to the MCDA, particularly the temporal relationships between the influenza epidemic peak, initiation of vaccination, and detection of the vaccine safety signal. These temporal relationships were most directly affected by the availability and adoption of vaccine. For example, early cycle vaccinations (i.e., September or October) had a greater per-vaccination ability to retard influenza transmission than late-cycle vaccinations (i.e., February or March) because there was higher potential for indirect vaccine-related benefits. Also important was the vaccine-attributable risk of adverse events. If the vaccine-attributable risk were higher, then signal detection would likely occur earlier and decision-makers would have more opportunity to affect near-term (i.e., within the present pandemic) vaccine-related benefits and risks.
Limitations of the Model
First, the weakest element of the MCDA was the mapping of simulated regulatory responses to future vaccine-seeking behavior, for which we relied on the expert panel's assessment. This element drove decision-making whenever intervention opportunities occurred too late in the pandemic to influence its near-term outcomes. Recent work by Blyth et al.  suggests that a major reduction in vaccine-seeking behavior following a selective suspension in a mild scenario might be more likely than the probability assigned during this exercise. Additionally, Blyth et al. 's paper suggests that a “major reduction” in vaccine-seeking behavior could be modeled as high as 35%. If the probability of a major reduction in future-vaccine seeking behavior following Selective Suspension in the mild scenario were higher, then we would expect that Selective Suspension would be a less desirable option. Consequently, we would expect Risk Communication Alone to be the preferred decision option in most mild scenarios.
While the present probabilistic assumptions about long-term vaccine-seeking behavior may be imperfect, we know decision-makers are focused on preserving the integrity of such behavior. Future studies should ascertain how vaccine safety guidance or warnings in one influenza season/pandemic affect the public's vaccine-seeking behavior during subsequent seasons.
Second, as with any modeling effort, the influenza transmission sub-model and the surveillance system sub-model required multiple assumptions during parameterization. The outputs of these models – vaccine-related outcomes and the timing of signal detection – reflected these assumptions.
Third, within the multi-attribute utility function, we created a composite index of vaccine-related benefits. Within that index, we weighted all influenza-associated deaths equally, regardless of age. In the future, it may be preferred to weight morbidity by age, or to adopt a structure analogous to cumulating quality-adjusted life years .
Fourth, we examined risks and benefits with respect to the entire population. Future modeling efforts could focus on risks and benefits for subpopulations such as the elderly, infants or pregnant women.
Fifth, we chose not to model brand-specific risks although vaccine safety problems in recent years have been attributed to particular products . Future extensions of the model could make assumptions about the age-specific market share of various products and model a safety problem that was isolated to a particular manufacturer.
Lessons Learned for Future Pandemic Influenza Preparedness Efforts in PRISM
Low attributable risks for rare adverse events pose a significant challenge to any medical product safety surveillance system. If PRISM can detect safety signals earlier in a pandemic, then decision-makers will have a greater impact on near-term vaccine benefits and risks. Opportunities for earlier detection necessitate increasing sample size early in a pandemic. This can be accomplished by encouraging earliest possible vaccination within a pandemic period or season, adding additional electronic healthcare databases to the PRISM system to increase the number of vaccinees under surveillance, and accessing the existing PRISM databases more frequently (e.g., using a weekly update instead of a monthly update) .
Disclaimer: The findings and conclusions in this paper are those of the authors and do not necessarily represent the views of the U.S. Food and Drug Administration.
The authors are grateful to Laura Hou for programming support, Charlene Gay, MPH, and Linda Pointon, MPH, for project management, Lisa A. Prosser, PhD (University of Michigan, Ann Arbor, MI), Mark L. Messonnier, PhD (Centers for Disease Control and Prevention, Atlanta, GA), and Michael D. Nguyen, MD (U.S. Food and Drug Administration, Silver Spring, MD) for their comments and suggestions, and Robert S. Daum, MD (University of Chicago, Chicago, IL); Kathryn M. Edwards, MD (Vanderbilt University School of Medicine, Nashville, TN); Ruth A. Karron, MD (Johns Hopkins Bloomberg School of Public Health, Baltimore, MD), John F. Modlin, MD (Children's Hospital at Dartmouth, Lebanon, NH); Geoffrey L. Rosenthal, MD, PhD (University of Maryland School of Medicine, Baltimore, MD), and Cindy Weinbaum, MD (Centers for Disease Control and Prevention, Atlanta, GA), for serving on the expert panel.
Conceived and designed the experiments: JCM TAL DBM. Performed the experiments: JCM DGF TAL DBM. Analyzed the data: JCM DGF. Wrote the paper: JCM. Critical revision of the manuscript for important intellectual content: JCM DGF TAL GML DBM.
- 1. Schonberger LB, Bregman DJ, Sullivan-Bolyai JZ, Keenlyside RA, Ziegler DW, et al. (1979) Guillain-Barre syndrome following vaccination in the National Influenza Immunization Program, United States, 1976–1977. Am J Epidemiol 110:105–123.
- 2. Gaydos JC, Top FH Jr, Hodder RA, Russell PK (2006) Swine influenza a outbreak, Fort Dix, New Jersey, 1976. Emerg Infect Dis 12:23–28
- 3. Wecht CH (1977) The swine flu immunization program: scientific venture or political folly? Am J Law Med 3:425–445.
- 4. Sencer DJ (2011) Perspective: Swine-Origin Influenza: 1976 and 2009. Clin Infect Dis 52:S4–S7
- 5. Sencer DJ, Millar JD (2006) Reflections on the 1976 swine flu vaccination program. Emerg Infect Dis 12:29–33
- 6. Krause R (2006) The Swine Flu Episode and the Fog of Epidemics. Emerg Infect Dis 12:40–43
- 7. Fielding JE (1978) Managing public health risks: the swine flu immunization program revisited. Am J Law Med 4:35–43.
- 8. National Research Council (1978) The Swine Flu Affair: Decision-Making on a Slippery Disease. Washington, D.C.: National Academies Press. Available: http://www.nap.edu/catalog.php?record_id=12660. Accessed 9 May 2014.
- 9. Stoto MA (2002) The precautionary principle and emerging biological risks: lessons from swine flu and HIV in blood products. Public Health Rep Wash DC 1974 117:546–552.
- 10. Hilleman MR (1996) Cooperation between government and industry in combating a perceived emerging pandemic. The 1976 swine influenza vaccination program. JAMA J Am Med Assoc 275:241–243.
- 11. Yih WK, Lee GM, Lieu TA, Ball R, Kulldorff M, et al. (2012) Surveillance for adverse events following receipt of pandemic 2009 H1N1 vaccine in the Post-Licensure Rapid Immunization Safety Monitoring (PRISM) System, 2009-2010. Am J Epidemiol 175:1120–1128
- 12. Salmon DA, Akhtar A, Mergler MJ, Vannice KS, Izurieta H, et al. (2011) Immunization-safety monitoring systems for the 2009 H1N1 monovalent influenza vaccination program. Pediatrics 127 Suppl 1: S78–S86
- 13. Iskander J, Strikas RA, Gensheimer KF, Cox NJ, Redd SC (2013) Pandemic influenza planning, United States, 1978–2008. Emerg Infect Dis 19:879–885
- 14. Pandemic Influenza Preparedness and Response: A WHO Guidance Document (2009) Geneva: World Health Organization. Available: http://www.ncbi.nlm.nih.gov/books/NBK143062/. Accessed 26 August 2013.
- 15. Institute of Medicine (US) Forum on Microbial Threats (2005) The Threat of Pandemic Influenza: Are We Ready? Workshop Summary. Knobler SL, Mack A, Mahmoud A, Lemon SM, editors Washington (DC): National Academies Press (US). Available: http://www.ncbi.nlm.nih.gov/books/NBK22156/. Accessed 26 August 2013.
- 16. Lee BY, Wiringa AE (2011) The 2009 H1N1 influenza pandemic: a case study of how modeling can assist all stages of vaccine decision-making. Hum Vaccin 7:115–119.
- 17. Wagner M, Tsui F, Cooper G, Espino JU, Harkema H, et al.. (2011) Probabilistic, Decision-theoretic Disease Surveillance and Control. Online J Public Health Inform 3. doi:10.5210/ojphi.v3i3.3798.
- 18. Chao DL, Matrajt L, Basta NE, Sugimoto JD, Dean B, et al. (2011) Planning for the control of pandemic influenza A (H1N1) in Los Angeles County and the United States. Am J Epidemiol 173:1121–1130
- 19. Kenah E, Chao DL, Matrajt L, Halloran ME, Longini IM Jr (2011) The global transmission and control of influenza. PloS One 6:e19515
- 20. Knipl DH, Röst G (2011) Modelling the strategies for age specific vaccination scheduling during influenza pandemic outbreaks. Math Biosci Eng MBE 8:123–139.
- 21. Conway JM, Tuite AR, Fisman DN, Hupert N, Meza R, et al. (2011) Vaccination against 2009 pandemic H1N1 in a population dynamical model of Vancouver, Canada: timing is everything. BMC Public Health 11:932
- 22. Lee BY, Brown ST, Korch GW, Cooley PC, Zimmerman RK, et al. (2010) A computer simulation of vaccine prioritization, allocation, and rationing during the 2009 H1N1 influenza pandemic. Vaccine 28:4875–4879
- 23. Yang Y, Sugimoto JD, Halloran ME, Basta NE, Chao DL, et al. (2009) The transmissibility and control of pandemic influenza A (H1N1) virus. Science 326:729–733
- 24. Medlock J, Galvani AP (2009) Optimizing influenza vaccine distribution. Science 325:1705–1708
- 25. Bansal S, Pourbohloul B, Meyers LA (2006) A comparative analysis of influenza vaccination programs. PLoS Med 3:e387
- 26. Matrajt L, Halloran ME, Longini IM Jr (2013) Optimal vaccine allocation for the early mitigation of pandemic influenza. PLoS Comput Biol 9:e1002964
- 27. Potter MA, Brown ST, Cooley PC, Sweeney PM, Hershey TB, et al. (2012) School closure as an influenza mitigation strategy: how variations in legal authority and plan criteria can alter the impact. BMC Public Health 12:977
- 28. Cooley P, Brown S, Cajka J, Chasteen B, Ganapathi L, et al. (2011) The role of subway travel in an influenza epidemic: a New York City simulation. J Urban Health Bull N Y Acad Med 88:982–995
- 29. Brown ST, Tai JHY, Bailey RR, Cooley PC, Wheaton WD, et al. (2011) Would school closure for the 2009 H1N1 influenza epidemic have been worth the cost?: a computational simulation of Pennsylvania. BMC Public Health 11:353
- 30. Longini IM Jr, Halloran ME, Nizam A, Yang Y (2004) Containing pandemic influenza with antiviral agents. Am J Epidemiol 159:623–633.
- 31. Davey VJ, Glass RJ, Min HJ, Beyeler WE, Glass LM (2008) Effective, robust design of community mitigation for pandemic influenza: a systematic examination of proposed US guidance. PloS One 3:e2606
- 32. Halloran ME, Ferguson NM, Eubank S, Longini IM Jr, Cummings DAT, et al. (2008) Modeling targeted layered containment of an influenza pandemic in the United States. Proc Natl Acad Sci U S A 105:4639–4644
- 33. Longini IM Jr, Nizam A, Xu S, Ungchusak K, Hanshaoworakul W, et al. (2005) Containing pandemic influenza at the source. Science 309:1083–1087
- 34. Nguyen M, Ball R, Midthun K, Lieu TA (2012) The Food and Drug Administration's Post-Licensure Rapid Immunization Safety Monitoring program: strengthening the federal vaccine safety enterprise. Pharmacoepidemiol Drug Saf 21 Suppl 1: 291–297
- 35. Maro JC, Brown JS, Kulldorff M (2013) Medical product safety surveillance: how many databases to use? Epidemiol Camb Mass 24:692–699
- 36. Petousis-Harris H, Poole T, Turner N, Reynolds G (2012) Febrile events including convulsions following the administration of four brands of 2010 and 2011 inactivated seasonal influenza vaccine in NZ infants and children: the importance of routine active safety surveillance. Vaccine 30:4945–4952
- 37. Lee GM, Greene SK, Weintraub ES, Baggs J, Kulldorff M, et al. (2011) H1N1 and Seasonal Influenza Vaccine Safety in the Vaccine Safety Datalink Project. Am J Prev Med 41:121–128
- 38. Tse A, Tseng HF, Greene SK, Vellozzi C, Lee GM (2012) Signal identification and evaluation for risk of febrile seizures in children following trivalent inactivated influenza vaccine in the Vaccine Safety Datalink Project, 2010–2011. Vaccine 30:2024–2031
- 39. Dodd CN, Romio SA, Black S, Vellozzi C, Andrews N, et al.. (2013) International collaboration to assess the risk of Guillain Barré Syndrome following Influenza A (H1N1) 2009 monovalent vaccines. Vaccine. doi:10.1016/j.vaccine.2013.06.032.
- 40. Salmon DA, Proschan M, Forshee R, Gargiullo P, Bleser W, et al. (2013) Association between Guillain-Barré syndrome and influenza A (H1N1) 2009 monovalent inactivated vaccines in the USA: a meta-analysis. Lancet 381:1461–1468
- 41. Bass FM (1969) A New Product Growth for Model Consumer Durables. Manag Sci 15:215–227.
- 42. Peres R, Muller E, Mahajan V (2010) Innovation diffusion and new product growth models: A critical review and research directions. Int J Res Mark 27:91–106
- 43. Geroski P (2000) Models of technology diffusion. Res Policy 29:603–625
- 44. Keeney RL, Raiffa H (1976) Decisions with multiple objectives: preferences and value tradeoffs. New York: Wiley. 569 p.
- 45. Böhmer MM, Walter D, Falkenhorst G, Müters S, Krause G, et al. (2012) Barriers to pandemic influenza vaccination and uptake of seasonal influenza vaccine in the post-pandemic season in Germany. BMC Public Health 12:938
- 46. Bish A, Yardley L, Nicoll A, Michie S (2011) Factors associated with uptake of vaccination against pandemic influenza: a systematic review. Vaccine 29:6472–6484
- 47. Poland GA (2010) The 2009–2010 influenza pandemic: effects on pandemic and seasonal vaccine uptake and lessons learned for seasonal vaccination campaigns. Vaccine 28 Suppl 4: D3–D13
- 48. Horney JA, Moore Z, Davis M, MacDonald PDM (2010) Intent to receive pandemic influenza A (H1N1) vaccine, compliance with social distancing and sources of information in NC, 2009. PloS One 5:e11226
- 49. Keeney RL (2013) Foundations for Group Decision Analysis. Decis Anal. Available: http://da.journal.informs.org/content/early/2013/03/13/deca.2013.0265. Accessed 21 August 2013.
- 50. Keeney RL, Nau R (2011) A theorem for Bayesian group decisions. J Risk Uncertain 43:1–17
- 51. Blyth CC, Richmond PC, Jacoby P, Thornton P, Regan A, et al. (2014) The impact of pandemic A(H1N1)pdm09 influenza and vaccine-associated adverse events on parental attitudes and influenza vaccine uptake in young children. Vaccine 32:4075–4081
- 52. Bults M, Beaujean DJMA, Richardus JH, van Steenbergen JE, Voeten HACM (2011) Pandemic influenza A (H1N1) vaccination in The Netherlands: parental reasoning underlying child vaccination choices. Vaccine 29:6226–6235
- 53. European Medicines Agency (2010) Benefit-risk methodology project: Work package 2 report: Applicability of current tools and processes for regulatory benefit-risk assessment. London, UK: European Medicines Agency. Available: http://www.ema.europa.eu/ema/index.jsp?curl=pages/special_topics/document_listing/document_listing_000314.jsp&mid=WC0b01ac0580223ed6. Accessed 12 June 2014.
- 54. Gold MR, Stevenson D, Fryback DG (2002) HALYS and QALYS and DALYS, Oh My: similarities and differences in summary measures of population Health. Annu Rev Public Health 23:115–134
- 55. Armstrong PK, Dowse GK, Effler PV, Carcione D, Blyth CC, et al. (2011) Epidemiological study of severe febrile reactions in young children in Western Australia caused by a 2010 trivalent inactivated influenza vaccine. BMJ Open 1:e000016
- 56. Yih WK, Sandhu S, Nguyen MD, Zichittella L, McMahill-Walraven CN, et al.. (2013) Accessing the Freshest Feasible Data for Conducting Active Influenza Vaccine Safety Surveillance. Mini-Sentin. Available: http://www.mini-sentinel.org/work_products/PRISM/Mini-Sentinel_PRISM_Active-Influenza-Vaccine-Safety-Surveillance-Protocol.pdf. Accessed 12 June 2014.