Epidemiological and economic impact of pandemic influenza in Chicago: Priorities for vaccine interventions

The study objective is to estimate the epidemiological and economic impact of vaccine interventions during influenza pandemics in Chicago, and assist in vaccine intervention priorities. Scenarios of delay in vaccine introduction with limited vaccine efficacy and limited supplies are not unlikely in future influenza pandemics, as in the 2009 H1N1 influenza pandemic. We simulated influenza pandemics in Chicago using agent-based transmission dynamic modeling. Population was distributed among high-risk and non-high risk among 0–19, 20–64 and 65+ years subpopulations. Different attack rate scenarios for catastrophic (30.15%), strong (21.96%), and moderate (11.73%) influenza pandemics were compared against vaccine intervention scenarios, at 40% coverage, 40% efficacy, and unit cost of $28.62. Sensitivity analysis for vaccine compliance, vaccine efficacy and vaccine start date was also conducted. Vaccine prioritization criteria include risk of death, total deaths, net benefits, and return on investment. The risk of death is the highest among the high-risk 65+ years subpopulation in the catastrophic influenza pandemic, and highest among the high-risk 0–19 years subpopulation in the strong and moderate influenza pandemics. The proportion of total deaths and net benefits are the highest among the high-risk 20–64 years subpopulation in the catastrophic, strong and moderate influenza pandemics. The return on investment is the highest in the high-risk 0–19 years subpopulation in the catastrophic, strong and moderate influenza pandemics. Based on risk of death and return on investment, high-risk groups of the three age group subpopulations can be prioritized for vaccination, and the vaccine interventions are cost saving for all age and risk groups. The attack rates among the children are higher than among the adults and seniors in the catastrophic, strong, and moderate influenza pandemic scenarios, due to their larger social contact network and homophilous interactions in school. Based on return on investment and higher attack rates among children, we recommend prioritizing children (0–19 years) and seniors (65+ years) after high-risk groups for influenza vaccination during times of limited vaccine supplies. Based on risk of death, we recommend prioritizing seniors (65+ years) after high-risk groups for influenza vaccination during times of limited vaccine supplies.


Introduction
The Advisory Committee on Immunization Practices (ACIP) recommends seasonal influenza vaccination annually for individuals aged 6 months and older without contraindications to prevent and control seasonal and pandemic influenza [1]. They update information on the dosage for children, antigenic composition and influenza vaccine products. While the ACIP recommendations for 2015-2016 influenza season partially account for risk of transmission, such as influenza immunized individuals caring for immunosuppressed persons are recommended to avoid contact with such persons for 7 days after vaccination, they do not address prioritization of influenza vaccination among subpopulations [2]. For the 2009-2010 influenza pandemic season, the ACIP recommended seasonal influenza vaccination for children above 6 months, adolescents and adults with a focus on individuals at higher risk of influenza complications, or are close contacts of persons at higher risk [3]. In February 2010, ACIP expanded the recommendation of annual influenza vaccination to any person aged above 6 months who does not have contraindications to vaccination, taking effect from the 2010-2011 influenza season. It took months to develop and distribute the 2009 H1N1 influenza vaccine. Similar scenarios of delay in vaccine introduction with limited vaccine efficacy and limited supplies are not unlikely in future influenza pandemics. Understanding and analysis of these challenging scenarios through computational modeling and simulation to improve influenza prevention and control programs is the primary motivation of this study.

Prioritization of influenza vaccine intervention
Evidence on the epidemiological and economic impact of vaccination for all age and risk groups from the societal standpoint assists in prioritization of influenza vaccine intervention, especially when vaccine supplies are limited, and minimize the direct cost of clinical care for influenza related health outcomes and indirect cost of productivity loss due to workplace absenteeism. While some studies have analyzed the direct epidemiological and economic impact of vaccine intervention strategies on controlling influenza pandemics [4][5][6][7][8][9][10][11][12][13], other studies have analyzed both the direct and indirect epidemiological and economic impact of influenza vaccination [14][15][16]. There are also prior studies that focused on the prioritization of vaccination and other interventions among people in different age groups [17][18][19]. This study adds to the evidence of prior studies by using a detailed agent-based model for estimating the direct and indirect effects of epidemiological and economic impact of vaccine-based interventions. The objective of the vaccine interventions is to minimize deaths, hospitalizations, outpatient visits, and the number of ill people who do not seek medical care.

Direct epidemiological and economic effects
Direct effect is due to the immune protection gained by effectively vaccinated individuals, and indirect effect is due to blocking of the influenza transmission by vaccinated individuals to susceptible individuals in their social network. Cost effectiveness of influenza vaccination for 65 + years [5], healthy working adults [6,7], and children [9,20] with a focus on direct effects have been studied. Prosser et al. evaluate the economic impact of 2009 pandemic influenza vaccine intervention for all age and risk groups [8]. They infer that vaccination of the subpopulation with a high risk of developing influenza related complications in each age group is cost saving, and vaccination of the healthy subpopulation in each age group is cost effective. Other studies have inferred that vaccine administration during previous and potential pandemics produces health benefits in terms of number of averted influenza cases and related health outcomes [10][11][12]. These studies included the direct cost of hospitalizations, outpatient visits, and deaths, and included the related costs of vaccine production and administration, and lost productivity. Depending on the risk and age group of the subpopulations, geographic region, and analytic methodology, the vaccine interventions may or may not be cost effective [21,22].

Indirect epidemiological and economic effects
Indirect effects account for the indirect protection due to vaccine intervention. Effectively vaccinated individuals who develop protective immune response to the prevalent influenza strains, cut off transmission pathways to secondary and subsequent individuals. The indirect effect of vaccinating school children has been found to be significant, due to their high connectivity in the social network and significance of their transmission pathways to their households and community [23][24][25][26][27] [14].
The epidemiological benefits and economic costs estimated by taking into account only the direct effect is relatively conservative, in comparison to taking into account both the direct and indirect effects. We improve the fidelity and robustness of the cost-benefit estimates to facilitate optimal prioritization of our vaccine interventions among different age and risk groups.

Study objective
Meltzer et al. estimate the potential net value of different vaccination strategies, and identify vaccination priorities for different age and risk groups during an influenza pandemic [4]. A Monte Carlo based static model is used to estimate the costs and benefits due to the direct effect of vaccine interventions in the United States. We focus our study on similar influenza related health outcomes, risk levels and age groups as Meltzer's study. We use an agent-based dynamic model to estimate the direct and indirect epidemiological and economic impact of vaccine interventions during an influenza pandemic in Chicago, and assist in the assessment of vaccine intervention priorities.

Public health significance
Population dynamics play an important role in influenza pandemic planning and response. Influenza vaccination not only protects effectively vaccinated individuals who develop a protective immune response from contracting influenza, but also prevents the spread of influenza in the social contact network of people by breaking the transmission chain. To optimally allocate limited resources, it is important to inform decision makers and public health officials about both the direct and indirect effects of influenza vaccine interventions.

Ethics statement
The Institutional Review Board at Virginia Tech has given ethics approval (IRB exempt) for the research conducted in this study.

Dynamic agent-based modeling of Chicago synthetic population
The Chicago metropolitan area is a major urban area in the United States, and had high influenza incidence during the 2009 H1N1 influenza pandemic [28]. We analyzed the impact of vaccine-based interventions on pandemic influenza in Chicago, using the population distribution of 9,047,574 people from the census data [29]. The disease diffusion occurs on a  collocation based synthetic social contact network for Chicago, based on dynamic agent-based modeling [30][31][32]. We generated the synthetic population and estimated the social contact network in Chicago through population synthesis, activity assignment, location choice and contact estimation, as illustrated in Table 1 [30,33]. The social contact network simulated the movement of individuals throughout the city and estimated the contact times between individuals based on their simultaneous presence at a location.

Influenza transmission dynamics
The transmission dynamics of an influenza-like-illness in the population is simulated using the susceptible-exposed-infectious-recovered (SEIR) epidemiological model on this synthetic social contact network of Chicago. Each person in the model is in one of the following four health states at any time: susceptible, exposed, infectious, and removed. A person is in the susceptible state until he becomes exposed. If a person becomes exposed, he remains exposed for the duration of the latent period, during which he is not infectious. At the end of the latent period, an exposed person becomes infectious and remains infectious for the duration of the infectious period. A person in the infectious state will probabilistically transmit the disease, based on the transmission rate, to any of his contacts who are in the susceptible state. A proportion of infectious individuals are asymptomatic, and there is a reduction in probability of transmission by an asymptomatic infectious person in comparison to a symptomatic infectious person to a susceptible individual. After the infectious period, the infectious person becomes recovered (or removed). Transmissibility is the probability of transmission per minute of contact with a symptomatic infectious person and is set to 0.00008, 0.00009, and 0.0001 to calibrate the simulation for the moderate, strong and catastrophic influenza pandemic scenarios respectively, with attack rates of 11.73%, 21.96% and 30.15% respectively. The simulation parameters for the social contact network and influenza dynamics are illustrated in Table 2. We estimate the direct and indirect effects of vaccine interventions on influenza pandemics of moderate, strong and catastrophic severities, in comparison to the base case scenario of no vaccine intervention.
Influenza related health outcomes, risk levels and age groups Influenza related health outcomes for the infected individuals are death, hospitalization, outpatient visits, and ill but not seeking medical care. The risk levels are high and non-high, and the age groups are 0-19 years, 20-64 years and 65+ years. Based on pre-existing medical conditions, influenza infected individuals may be at a high or non-high risk of experiencing influenza related health outcomes. The distribution of the four influenza related health outcomes among the high and non-high risk cases in the three different age groups is based on the study by Meltzer et al. [4]. Table 1. Synthetic social network of Chicago. Synthetic population of Chicago is generated and a social contact network is estimated through the following four steps.

Population synthesis
Synthetic representation of each household in Chicago metropolitan area is created that is statistically identical to US census data when aggregated to a block group level.

Activity assignment
Each synthetic person in a household is assigned a set of activities to perform during the day, along with the times when the activities begin and end, as given by activity or time-use survey data.
Location choice An appropriate real location is chosen for each activity for every synthetic person based on data such as land use data or Dunn and Bradstreet location data.

Contact estimation
Each synthetic person is deemed to have made contact with a subset of other synthetic people simultaneously present at the same location. This gives rise to the synthetic social contact network

Base case scenario of no vaccine intervention
For the base-case scenario of no vaccine intervention, three different severities of an influenza pandemic were simulated using the dynamic model: moderate influenza with 11.73% attack rate, strong influenza with 21.96% attack rate, and catastrophic influenza with 30.15% attack rate. We use the dynamic model to simulate the epidemic curves for these 3 attack rates for the Cost of influenza vaccine $28.62 [38] Medical costs and productivity losses of influenza-related health outcomes among the different age groups See base-case scenario of no vaccine intervention, based on the average incidence from 25 replicates (see Fig 2 and Table 3). The simulation timeline of the influenza pandemics are in accordance with prior experiences of influenza pandemics in the United States [34]. S1 Appendix describes the risk space of transmissibility and clinical severity for the pandemic scenarios, as defined by the framework for assessing epidemiologic effects of influenza epidemics and pandemics by Reed et al [35,36].

Vaccine intervention
Effectiveness of influenza vaccines varies between 10% to 60% [37]. We analyzed the impact of the vaccine intervention scenario of 40% efficacy and 40% compliance for all age and risk groups, following Meltzer et al. [4]. It took months to develop and distribute the 2009 H1N1 influenza vaccine, and similar scenarios of delay in vaccine introduction, limited vaccine efficacy and limited supplies are not unlikely in future influenza pandemics. Thereby, we analyze delays in the implementation of the vaccine intervention with limited efficacy and compliance rates.

Fig 2. Influenza incidence (average number of new cases per day) during the pandemic for no vaccine intervention and vaccine intervention scenarios.
The epidemic curves illustrate influenza incidence without and with vaccination intervention for the catastrophic, strong and moderate influenza pandemic scenarios. The number of cases is the average of new cases over 25 simulations. Higher attack rates cause the earlier, more severe, and shorter pandemic duration, compared to the less severe but longer pandemics. The vaccination intervention is applied 15 days after the start of pandemic and implemented for 60 days. The vaccine intervention scenarios are simulated at 40% efficacy and 40% compliance for all age and risk groups in the dynamic agent-based model. https://doi.org/10.1371/journal.pcbi.1005521.g002

Priorities for influenza vaccine interventions
Direct epidemiological effect of vaccine intervention using static model Static model is used to estimate the direct benefit of influenza vaccination, that is, vaccination only protects effectively vaccinated individuals who develop protective immune response, but does not account for preventing influenza transmission from effectively vaccinated individuals to their social contact network. Using the simulation results of the base-case scenario of no vaccine intervention from the dynamic model, the influenza attack rates of moderate, strong and catastrophic pandemic scenarios are decreased by the proportional impact of the vaccine intervention at 40% coverage and 40% efficacy. Thereby, the influenza attack rates in the 3 age group sub-populations are decreased by 16% (40% efficacy Ã 40% compliance) in each of the three pandemic scenarios (see Table 4). Table 3. Pandemic cost per capita, attack rate, and reproduction number for different severities of pandemic influenza in the base case scenario of no vaccine intervention. Pandemic cost per capita is the average cost of influenza related health outcomes among infected individuals for death, hospitalization, outpatient visit, and ill but not seeking medical care. The attack rate is the proportion of population infected by influenza during the influenza pandemic. Reproduction number is the number of secondary cases caused by the index case in a susceptible population.

Direct and indirect epidemiological effects of vaccine intervention using dynamic model
We simulated the vaccine intervention scenarios at 40% efficacy and 40% compliance for all age and risk groups in the dynamic agent-based model. The vaccine intervention is initiated 15 days after the start of the pandemic and is carried out for 60 days. The dynamic model simulates the diffusion of influenza on the population in Chicago. It takes into account the indirect effect of limiting disease diffusion by vaccinated individuals, who develop protective immune response and cut off transmission pathways to secondary and subsequent individuals. The influenza attack rates for the 3 age groups in moderate, strong and catastrophic pandemic scenarios are estimated (see Table 4).

Vaccine cost
The cost of influenza vaccine is estimated to be $28.62, and includes the clinical personnel, non-clinical personnel, and all overhead costs [38]. Direct medical costs and indirect productivity losses were estimated from a prior study, and are presented in Table 5 [39][40][41][42].

Pandemic cost estimation
Based on Meltzer's study [4], we developed a decision tree that includes the probability distribution of an influenza case experiencing the influenza related health outcomes of death, hospitalization, outpatient visits, and ill but not seeking medical care, and the cost associated with these health outcomes among the different age and risk groups (see Fig 3). All costs have been adjusted to 2015 US$ (see Table 5). We used this decision tree to estimate the cost due to influenza related health outcomes among the different age and risk groups. This cost estimation process is conducted in all the three scenarios: base case scenario of no intervention using  dynamic model, vaccine intervention scenario using static model, and vaccine intervention scenario using dynamic model. Within each of these scenarios, for each pandemic severity (moderate, strong and catastrophic), we compute the pandemic cost, pandemic cost per capita, net benefits, and return on investment, as illustrated in Table 6 (also, see Tables 3 and 4). The pandemic cost is the total cost associated with the health outcomes of influenza cases and the cost of vaccination, and pandemic cost per capita is the average pandemic cost per person. The net benefits are the difference in cost due to improved health outcomes from vaccination and the vaccination cost. Return on investment is the gain in net benefits relative to the vaccination cost.
Patients with pre-existing medical condition have a high risk of experiencing severe influenza related health outcomes. The probability of each health outcome is assigned an uniform or triangular distribution [4]. For the uniform distribution, the lower and upper rate are presented; for triangular distribution, the lower, most probably, and higher rates are presented.
https://doi.org/10.1371/journal.pcbi.1005521.g003 Table 6. Computation of pandemic cost, pandemic cost per capita, net benefits and return on investment. The formulations to compute pandemic cost, pandemic cost per capita, net benefits and return on investment are presented below for the scenarios of without and with vaccine intervention. Pandemic cost is the total cost associated with the health outcomes of influenza cases and the cost of vaccination, and pandemic cost per capita is the average pandemic cost per person. The net benefits is the difference in cost due to improved health outcomes from vaccination and the vaccination cost. Return on investment is the gain in net benefits relative to the vaccination cost.

Metrics No vaccine intervention Vaccine intervention
Cost of influenza related health outcomes (Cost of influenza related health outcomes) + (Vaccination cost) Per capita (Cost of influenza related health outcomes) Per capita (Cost of influenza related health outcomes) + Per capita (Vaccination cost) Pandemic cost per capita Benefits -Costs (Benefits from reduction in the cost of influenza related health outcomes due to reduction in influenza cases after vaccine intervention)-(Vaccination cost)

Net benefits Vaccination cost
Return on investment is the gain in net benefits relative to the vaccination cost, that is, dollars saved per $1 investment in vaccine intervention

Perspective of economic evaluation
We conducted economic evaluation from the medical and productivity perspective, and includes the direct cost of clinical care for influenza related health outcomes incurred by the health care provider and indirect cost of productivity loss incurred by the patient. To extend this analysis to a societal perspective, costs incurred by the federal government in vaccine distribution, vaccine coverage monitoring, vaccine effectiveness monitoring, vaccine safety monitoring, health communication, and national coordination and technical assistance [43], productivity loss of volunteers in the influenza vaccine campaign, and costs of global influenza surveillance for vaccine strain selection will need to be included, which are beyond the scope of this study.

Simulation replicates
The values of the simulation parameters and their sources are shown in Table 2. Each influenza pandemic scenario in the agent-based model is simulated 25 times. The costs of influenzarelated health outcomes among the different age and risk groups are estimated using the decision tree (Fig 3). The agent-based model is executed through SIBEL [44], a web-based tool to conduct epidemiological disease studies based on realistic social network simulation, and the influenza-related health outcome estimation using decision tree and cost-benefit analysis is executed through the R software for statistical computing and graphics [45].

Sensitivity analysis
We conducted univariate sensitivity analysis for vaccine compliance, vaccine efficacy and vaccine start date, and their impact on attack rates and return on investment for catastrophic, strong, and moderate influenza pandemic scenarios with no vaccine intervention (base case), and with vaccine intervention in static model (direct effect) and dynamic model (direct + indirect effects).

Base case scenario of no vaccine intervention
The pandemic cost per capita is $678.10, $486.67 and $255.18 for catastrophic, strong, and moderate influenza scenarios respectively (see Table 3). The attack rate is 30.15%, 21.96% and 11.73% for catastrophic, strong, and moderate influenza scenarios respectively. The reproduction number is 1.19, 1.13 and 1.06 for catastrophic, strong, and moderate influenza scenarios respectively. The pandemic cost per capita is positively correlated with attack rate and reproduction number, with the highest in catastrophic influenza scenario followed by the strong and moderate influenza scenarios.

Vaccine interventions
The vaccine intervention is simulated at 40% compliance and 40% efficacy, using the static model and the dynamic model. The vaccine intervention decreases the pandemic cost per capita, attack rate and reproduction number in the catastrophic, strong and moderate influenza pandemic scenarios in both the static and dynamic models. Fig 4A, 4B and 4C illustrate the comparison of pandemic cost per capita, attack rate and reproduction number in the catastrophic, strong and moderate influenza pandemic scenarios with and without vaccine intervention. In the catastrophic influenza pandemic scenario with vaccine intervention, the pandemic cost per capita, attack rate and reproduction number are $370.56, 16.34% and 1.09 respectively in the dynamic model, while they are $581.09, 25.33% and 1.15 respectively in the static model (see Table 4). In the strong influenza pandemic scenario with vaccine intervention, the pandemic cost per capita, attack rate and reproduction number are $90.81, 3.90% and 1.02 respectively in the dynamic model, while they are $420.28, 18.45% and 1.11 respectively in the static model. In the moderate influenza pandemic scenario with vaccine intervention, the pandemic cost per capita, attack rate and reproduction number are $14.85, 0.16% and 1.00 respectively in the dynamic model, while they are $225.83, 9.85% and 1.05 respectively in the static model.
Molinari et al estimated the annual economic impact (medical costs and productivity loss) of seasonal influenza in the United States to be $87.0673 billion (95% CI: $47.2153, $149.5086) in 2003 with the vaccine intervention [40], which relates to an inflation adjusted cost per capita of $392.24 (95% CI: $212.71, $673.54) in $2015. We estimated the pandemic cost per capita with no vaccine intervention to be $678.10, $486.67 and $255.18 (in $2015) for catastrophic, strong, and moderate influenza scenarios respectively, and with vaccine intervention to be $370.56, $90.81 and $14.85 respectively.

Direct and indirect effects on return on investment
While the vaccine interventions are cost-beneficial in both the dynamic and static models, the return on investment is relatively higher in the dynamic model due to the combined impact of direct and indirect effects, in comparison to the static model which includes only the direct effect (see Fig 5 and Table 7).

Prioritization of vaccine intervention
Vaccine prioritization criteria includes risk of death, total deaths, net benefits, and return on investment. Table 8 shows the values for risk of death, total deaths, net benefits, and return on investment of high and non-high risk groups among the 0-19, 20-64, 65+ years subpopulations for the catastrophic, strong and moderate influenza pandemic scenarios. The prioritization criteria of risk of death, total deaths, net benefits, and return on investment assist in the decision making process for vaccine prioritization among different age and risk groups, as shown in Table 9. In the catastrophic influenza pandemic scenario, the risk of death among the high-risk 65+ years subpopulation is the highest at 392.18 deaths per 100,000 influenza cases, while it is the lowest among the non-high risk 0-19 years subpopulation at 6.63 deaths per 100,000 influenza cases. In the strong influenza pandemic scenario, the risk of death among the high-risk 0-19 years subpopulation is the highest at 281.08 deaths per 100,000 influenza cases, while it is the lowest among the non-high risk 0-19 years subpopulation at 5.04 deaths per 100,000 influenza cases. In the moderate influenza pandemic scenario, the risk of death among the highrisk 0-19 years subpopulation is the highest at 157.85 deaths per 100,000 influenza cases, while it is the lowest among the non-high risk 20-64 years subpopulation at 2.65 deaths per 100,000 influenza cases.   Table 7. Pandemic cost, net benefits and return on investment. Pandemic cost is the total cost associated with the health outcomes of influenza cases and the cost of vaccination. Net benefits are the difference in cost due to improved health outcomes from vaccination and the vaccination cost. Return on investment is the gain in net benefits relative to the vaccination cost, that is, dollars saved per $1 investment in vaccine intervention.  Table 8. Risk of death, total deaths, net benefits and return on investment for different age and risk groups in the catastrophic, strong, and moderate influenza pandemic scenarios. Risk of death is estimated based on the number of influenza related deaths per 100,000 subpopulation for the specific age and risk groups. Total deaths is estimated based on the proportion of influenza related deaths for the specific age and risk groups among total influenza related deaths. Net benefits are the difference in cost due to improved health outcomes from vaccination and the vaccination cost. Return on investment is the gain in net benefits relative to the vaccination cost, that is, dollars saved per $1 investment in vaccine intervention.

Catastrophic influenza pandemic
Age and risk group Risk of death (per 100,000 cases) Priorities for influenza vaccine interventions lowest among the non-high risk 0-19 years subpopulation at 0.031. In the strong influenza pandemic scenario, the proportion of total deaths among the high-risk 20-64 years subpopulation is the highest at 0.45 while it is the lowest among the non-high risk 0-19 years subpopulation at 0.033. In the moderate influenza pandemic scenario, the proportion of total deaths among the high risk 20-64 years subpopulation is the highest at 0.447, while it is the lowest among the non-high risk 0-19 years subpopulation at 0.036. Fig 6C illustrates the prioritization criteria for the vaccine intervention based on net benefits. In the catastrophic influenza pandemic scenario, the net benefits among the high-risk 20-64 years subpopulation is the highest at $1201.38 million, while it is the lowest among the non-high risk 65+ years subpopulation at $68.80 million. In the strong influenza pandemic scenario, the net benefits among the high risk 20-64 years subpopulation is the highest at $1462.52 million, while it is the lowest among the non-high risk 65+ years subpopulation at $79.71 million. In the moderate influenza pandemic scenario, the net benefits among the high risk 20-64 years Table 9. Prioritization of influenza vaccine intervention. Prioritization of influenza vaccine intervention among different age and risk groups based on different criteria: risk of death, total deaths, net benefits, and return on investment. a Risk of death is estimated based on the number of influenza related deaths per 100,000 subpopulation for the specific age and risk groups. Risk of death is the highest among the high risk 65+ years subpopulation in the catastrophic influenza and it is the highest among high risk 0-19 years subpopulation in the strong, and moderate influenza pandemic scenarios. b Total deaths is estimated based on the proportion of influenza related deaths for the specific age and risk groups among total influenza related deaths. The proportion of influenza related deaths is the highest among the high risk 20-64 years subpopulation in the catastrophic, strong, and moderate influenza pandemic scenarios. c Net benefits are the difference in cost due to improved health outcomes from vaccination and the vaccination cost. Net benefits are the highest among the high risk 20-64 years subpopulation in the catastrophic, strong, and moderate influenza pandemic scenarios. d Return on investment is the gain in net benefits relative to the vaccination cost, that is, dollars saved per $1 investment in vaccine intervention. Return on investment is highest among the high risk 0-19 years subpopulation in the catastrophic, strong and moderate influenza pandemic scenarios.

Net benefits
subpopulation is the highest at $864.83 million, while it is the lowest among the non-high risk 65+ years subpopulation at $43.02 million.

Fig 6. Prioritization of influenza vaccine intervention.
Prioritization of influenza vaccine intervention among different age and risk groups based on different criteria: risk of death, total deaths, net benefits, and return on investment. Fig 6A: Risk of death is estimated based on the number of influenza related deaths per 100,000 subpopulation for the specific age and risk groups. Risk of death is the highest among the high risk 65+ years subpopulation in the catastrophic influenza and it is the highest among high risk 0-19 years old among strong, and moderate influenza pandemic scenarios. Fig 6B: Total deaths is estimated based on the proportion of influenza related deaths for the specific age and risk groups among total influenza related deaths. The proportion of influenza related deaths is the highest among the high risk 20-64 years subpopulation in the catastrophic, strong, and moderate influenza pandemic scenarios. Fig 6C: Net benefits are the difference in cost due to improved health outcomes from vaccination and the vaccination cost. Net benefits are the highest among the high risk 20-64 years subpopulation in the catastrophic, strong, and moderate influenza pandemic scenarios. Fig 6D: Return on investment is the gain in net benefits relative to the vaccination cost, that is, dollars saved per $1 investment in vaccine intervention. Return on investment is highest among the high risk 0-19 years subpopulation in the catastrophic, strong and moderate influenza pandemic scenarios.

Sensitivity analysis
We conducted univariate sensitivity analysis for vaccine compliance, vaccine efficacy and vaccine start date, and their impact on attack rates and return on investment for catastrophic, strong, and moderate influenza pandemic scenarios. Vaccine compliance, vaccine efficacy and vaccine start date impact on attack rate. Fig  7 illustrates the univariate sensitivity analysis for vaccine compliance rates of 10%, 40%, 60% and 80% (Fig 7A), vaccine efficacy rates of 10%, 20%, 30%, 40%, 50% and 60% ( Fig 7B) and vaccine start dates after epidemic onset of day 15, day 30, day 60 and day 90 (Fig 7C), and their impact on attack rates for catastrophic, strong, and moderate influenza pandemic scenarios with no vaccine intervention (base case) and vaccine intervention (static and dynamic models).
We observe a negative correlation between vaccine compliance and attack rate, negative correlation between vaccine efficacy and attack rate, and positive correlation between vaccine start date and attack rate. The relative impact in the dynamic model is higher due to the combined benefits of direct and indirect effects of the vaccine intervention, in comparison to the static model with only the direct effect of the vaccine intervention.
Vaccine compliance impact on return on investment. Fig 8 illustrates the univariate sensitivity analysis for vaccine compliance rates of 10%, 40%, 60% and 80%, and their impact on return on investment for catastrophic (Fig 8A), strong ( Fig 8B) and moderate (Fig 8C) influenza pandemic scenario in the static (direct benefit) and dynamic (direct + indirect benefits) models. We observe a negative correlation between vaccine compliance and return on investment in the dynamic model, indicating that the return on investment is higher with vaccine introduction and decreases with vaccine compliance but also beneficially decreases the attack rate. In the static model, return on investment remains stable for the varied rates of vaccine compliance, with relatively higher return on investment in the catastrophic followed by strong and moderate pandemic influenza scenarios.
Vaccine efficacy impact on return on investment. Fig 9 illustrates the univariate sensitivity analysis for vaccine efficacy rates of 10%, 20%, 30%, 40%, 50% and 60%, and their impact on return on investment for catastrophic (Fig 9A), strong ( Fig 9B) and moderate (Fig 9C) influenza pandemic scenario in the static and dynamic models. We observe a positive correlation between Univariate sensitivity analysis for vaccine compliance rates of 10%, 40%, 60% and 80%, and their impact on return on investment for catastrophic (Fig 8A), strong (Fig 8B) and moderate (Fig 8C) influenza pandemic scenario in the static (direct benefit) and dynamic (direct + indirect benefits) models.
https://doi.org/10.1371/journal.pcbi.1005521.g009 vaccine efficacy and return on investment in the dynamic and static models, with relatively higher return on investment in the dynamic model compared to the static model.
Vaccine start date impact on return on investment. Fig 10 illustrates the univariate sensitivity analysis for vaccine start dates after epidemic onset of day 15, day 30, day 60 and day 90, and their impact on return on investment for catastrophic (Fig 10A), strong ( Fig 10B) and moderate (Fig 10C) influenza pandemic scenario in the static and dynamic models. We observe a negative correlation between vaccine start date and return on investment in the dynamic and static models, with relatively higher return on investment in the dynamic model compared to the static model.

Direct and indirect epidemiological and economic effects of vaccine intervention
Direct effect is due to the immune protection gained by effectively vaccinated individuals, and indirect effect is due to blocking of the influenza transmission by vaccinated individuals to susceptible individuals in their social network. The static model provides a conservative estimate of the epidemiological and economic benefits of influenza vaccine intervention by accounting for only the direct effect. The dynamic model provides a comprehensive estimate of the epidemiological and economic benefits of influenza vaccine intervention by accounting for both the direct and indirect effects.
The vaccine intervention has a higher probability of effectively vaccinating individuals who will have otherwise being infected in the absence of the vaccine intervention in more severe pandemic scenarios (such as catastrophic influenza). This is due to relatively higher attack rates and higher proportion of population at risk of influenza infection in comparison to less severe pandemic scenarios (such as moderate influenza). Thereby, the impact of the direct effect decreases from catastrophic, strong to moderate influenza pandemic scenarios (see Fig 5).
The vaccine intervention has a lower probability of breaking transmission pathways in more severe pandemic scenarios (such as catastrophic influenza), because the transmission network is densely connected in comparison to sparsely connected transmission networks in less severe pandemic scenarios (such as moderate influenza). Thereby, the impact of the indirect effect increases from catastrophic, strong to moderate influenza pandemic scenarios (see Fig 5).  (Fig 10A), strong (Fig 10B) and moderate (Fig 10C) influenza pandemic scenario in the static (direct benefit) and dynamic (direct + indirect benefits) models. https://doi.org/10.1371/journal.pcbi.1005521.g010 Pandemic cost per capita, attack rate and reproduction number are relatively lower in the dynamic model due to the combined impact of direct and indirect effects, in comparison to the static model which includes only the direct effect, in the catastrophic, strong and moderate influenza pandemic scenarios. While the vaccine interventions are cost-beneficial in both the dynamic and static models, the return on investment is relatively higher in the dynamic model in comparison to the static model.

Prioritization of vaccine interventions
We analyzed vaccine prioritization criteria based on risk of death, total deaths, net benefits and return on investment for the high and non-high risk groups among 0-19, 20-64 and 65+ years subpopulations. The risk of death is the highest among the high-risk 65+ years subpopulation in the catastrophic influenza, and it is the highest among the high-risk 0-19 years subpopulation in the strong and moderate influenza pandemic scenarios. The proportion of total deaths is the highest among the high-risk 20-64 years subpopulation in the catastrophic, strong and moderate influenza pandemic scenarios. The net benefits are the highest among the high-risk 20-64 years subpopulation in the catastrophic, strong and moderate influenza pandemic scenarios. The return on investment is the highest in the high-risk 0-19 years subpopulation in the catastrophic, strong and moderate influenza pandemic scenarios.
The proportion of total deaths and net benefits measure the epidemiological and economic impact respectively, and are dependent on the absolute size of the different risk and age group subpopulations. Risk of death and return on investment measure the epidemiological and economic impact respectively, and are independent of the absolute size of the different risk and age group subpopulations. Based on risk of death and return on investment, high-risk groups of the three age group subpopulations are recommended for prioritization of influenza vaccine intervention. Also, the vaccine intervention is cost-beneficial for all age and risk groups.

Targeted vaccination
The attack rates among the children (0-19 years) are higher than the attack rates among the adults (20-64 years) and seniors (65+ years) in the catastrophic, strong, and moderate influenza pandemic scenarios, as illustrated in Table 4. This can be attributed to their larger social contact network and homophilous interactions in schools. Thereby, if we target children for vaccination, there will be higher reduction among the children as well on the overall attack rate in the general population, as also illustrated in prior studies by Hodgson et al [16], Ferguson et al [17], and Germann et al [18]. Also, as shown in Table 9, high risk children have the highest return on investment from the vaccine intervention.

Public health implications
The dynamic model provides improved estimates of the epidemiological and economic benefits of vaccine interventions in comparison to a static model, by accounting for both the direct and indirect effects. These comprehensive estimates assist in prioritization of vaccine interventions among subpopulations of different risk and age groups, especially during influenza pandemics with limited availability of vaccines. Decision makers can use the dynamic model simulations to compare the epidemiological and economic impact of using different prioritization criteria of influenza vaccine interventions among different risk and age group subpopulations, thereby optimizing allocation of limited resources and improving evidence-based public health policy and practice.
Based on risk of death and return on investment, high-risk groups of the three age group subpopulations can be prioritized for vaccination, and the vaccine interventions are cost saving for all age and risk groups. The attack rates among the children are higher than among the adults and seniors in the catastrophic, strong, and moderate influenza pandemic scenarios, due to their larger social contact network and homophilous interactions in school. Based on return on investment and higher attack rates among children, we recommend prioritizing children (0-19 years) and seniors (65+ years) after high-risk groups for influenza vaccination during times of limited vaccine supplies. Based on risk of death, we recommend prioritizing seniors (65+ years) after high-risk groups for influenza vaccination during times of limited vaccine supplies.

Modeling implications
We used an agent-based individual model in this study to estimate the direct and indirect epidemiological and economic impact of vaccine interventions during an influenza pandemic in Chicago, similar to related studies [14,46,47]. Alternatively, a population level compartmental model can also be used to conduct this study, similar to related studies [15,48,49]. While agent-based individual models add heterogeneity in contact patterns between individuals in comparison to homogeneous mixing in compartmental models, it will be valuable to compare the vaccine intervention priorities derived from these two modeling methods in future studies.

Limitations
We used a mean estimate of $28.62 (inflation adjusted to 2015 US dollars) for the cost of influenza vaccine, and did not include the range and uncertainty in vaccination costs by location and size of clinical practice. While beyond the scope of this study, this analysis can be extended to additional studies for a range of vaccine compliance and efficacy values at different attack rates of influenza pandemics in different rural and urban areas of the United States and at the country level, to infer objective prioritization criteria for influenza vaccine interventions among different risk and age groups.