Epidemiologic and economic impact of pharmacies as vaccination locations during an influenza epidemic

Introduction: During an influenza epidemic, where early vaccination is crucial, pharmacies may be a resource to increase vaccine distribution reach and capacity. Methods: We utilized an agent-based model of the US and a clinical and economics outcomes model to simulate the impact of different influenza epidemics and the impact of utilizing pharmacies in addition to traditional (hospitals, clinic/physician offices, and urgent care centers) locations for vaccination for the year 2017. Results: For an epidemic with a reproductive rate (R0) of 1.30, adding pharmacies with typical business hours averted 11.9 million symptomatic influenza cases, 23,577 to 94,307 deaths, $1.0 billion in direct (vaccine administration and healthcare) costs, $4.2–44.4 billion in productivity losses, and $5.2–45.3 billion in overall costs (varying with mortality rate). Increasing the epidemic severity (R0 of 1.63), averted 16.0 million symptomatic influenza cases, 35,407 to 141,625 deaths, $1.9 billion in direct costs, $6.0–65.5 billion in productivity losses, and $7.8–67.3 billion in overall costs (varying with mortality rate). Extending pharmacy hours averted up to 16.5 million symptomatic influenza cases, 145,278 deaths, $1.9 billion direct costs, $4.1 billion in productivity loss, and $69.5 billion in overall costs. Adding pharmacies resulted in a cost-benefit of $4.1 to $11.5 billion, varying epidemic severity, mortality rate, pharmacy hours, location vaccination rate, and delay in the availability of the vaccine. Conclusions: Administering vaccines through pharmacies in addition to traditional locations in the event of an epidemic can increase vaccination coverage, mitigating up to 23.7 million symptomatic influenza cases, providing cost-savings up to $2.8 billion to third-party payers and $99.8 billion to society. Pharmacies should be considered as points of dispensing epidemic vaccines in addition to traditional settings as soon as vaccines become available.


Introduction
As getting people vaccinated as early as possible in an epidemic is crucial to mitigating the impact of an influenza epidemic [1,2], pharmacies may represent an important resource to increase reach and capacity of vaccine distribution in the event of a novel epidemic. Over the last century, four influenza pandemics have caused significant morbidity and mortality globally [3,4]. During the 2009 H1N1 pandemic, the federal government allotted vaccines to states based on population size, with each state determining where to administer their vaccine supply [5,6]. The relative epidemiological and economic benefits of using a single vaccination channel (traditional locations, such as doctor offices and hospitals) versus multiple channels (traditional locations plus alternative locations, such as pharmacies) to administer influenza vaccine in the event of an epidemic caused by a novel virus have not been determined.
Pharmacies are increasingly identified as key partners in public health, contributing to expanded patient access and emergency preparedness [7][8][9][10]. As of 2009, pharmacists in every US state are trained to vaccinate in some capacity [11][12][13]. However, despite nearly 86% of the population living within five miles of a pharmacy, 28.2% of adults and 4.9% of children received their seasonal influenza vaccine from a pharmacy in 2017 [14,15]. Pharmacies offer unique advantages that many traditional locations do not; they have expanded evening and weekend hours, provide vaccinations without an appointment, and are located in close proximity to patients, increasing access and convenience of immunization delivery [16][17][18][19][20][21][22]. To estimate the benefits of utilizing pharmacy locations in addition to traditional locations (e.g., doctor offices and hospitals) for immunization in the event of an influenza epidemic caused by a novel virus, we used the Public Health Influenza Laboratory agent-based model and the FluEcon clinical and economic outcomes model to simulate the spread of influenza and the impact of vaccination under varying conditions. representations for six types of vaccination locations across the 48 continuous states: 5720 hospitals, 51,560 clinic and physician offices, 4659 urgent care clinics, and 61,202 pharmacies (21,781 large retailers and 39,421 others).
Each virtual agent in PHIL represents a person with a set of characteristics (e.g., age, sex, race, socioeconomic status) and is assigned to a geographically explicit household and, depending on age, workplace or school. PHIL simulates the movement of each agent from place to place (i.e., from household to work, from school to community, etc.) and the interactions between agents each day at each location. These interactions potentiate the transmission of influenza (Appendix Table A1).
Each agent could be in one of four mutually exclusive influenza states: (1) susceptible (S, not infected with influenza and able to become infected), (2) exposed (E, infected with influenza, but not able to transmit to others), (3) infectious (I, infected and able to transmit to others), or (4) recovered/immune (R, not infected and unable to become infected). All agents start in the 'S state'. An influenza epidemic was seeded by randomly infecting 1000 agents on day one. The model advances in discrete one-day time steps, where agents interact with one another based on their movement between households, communities, workplaces, and schools. With each point of contact, agents in the 'I state' have a probability of transmitting influenza to agents in the 'S state', with those agents subsequently moving to the 'E state'. An agent remains in the 'E state' for the latent period duration, before moving to the 'I state', where they are able to transmit to others. In the 'I state', 33% are asymptomatic, and are half as infectious as a symptomatic agent [32,33]. We assume 50% of all symptomatic agents stay home from school or work (2.5 and 1.5-5 days, respectively; Appendix Table A2), thereby limiting contact with other agents. The likelihood of influenza transmission between agents varies based on where contact occurs (Appendix Table A1), the reproductive rate (R0) (i.e., average number of secondary cases generated by one infectious case), the duration of infectiousness, and agent infectivity (i.e., if agent is symptomatic or asymptomatic). If an agent is vaccinated, the agent will have a probability of moving to the 'R state' based on the vaccine efficacy. We assume protection is complete and immediate (i.e., individuals move to the 'R state' on the day of vaccination).
Each agent's likelihood of seeking vaccination is based on the distance they are willing to travel from their household to be vaccinated. Each day, an agent will look for a vaccination location within this set distance (i.e., radius) that varies for rural and urban areas; if no vaccination locations are within this radius, that agent is not eligible for vaccination. If more than one location is within the radius, an agent will randomly pick one location to visit on that day and will be vaccinated if there are enough doses available at that location. Age and gender limits were applied to specialized physician offices: women's (only women), pediatric (0-18 years old), and geriatric (60 years and older) physician offices and clinics. Following current state regulations, only those of a certain age could be vaccinated in pharmacy locations. The number of people vaccinated each day at a given location depends on the location-specific daily vaccination rate. This rate is determined from number of persons that a nurse or pharmacist could vaccinate per hour (for pharmacies this was normalized by the weekly number of prescriptions filled to account for size) and the number of hours that location is open (Appendix Table A2). We assume hospitals and clinic and physician offices are open five days per week, urgent care centers could vaccinate six days per week, and pharmacies are open seven days per week.

FluEcon
Using PHICOR's previously published FluEcon model [2,34], we translate the number of vaccinated persons and influenza infections from PHIL into health outcomes and their corresponding costs from the third-party payer and societal perspectives. Each symptomatic case has probabilities of seeking ambulatory care, being hospitalized, or dying from influenza. Each of these is associated with costs and health effects. The third-party payer perspective includes all direct costs (i.e., vaccination, ambulatory care, hospitalization). Societal costs include direct and indirect (i.e., productivity losses due to absenteeism and mortality) costs. Hourly wage for all occupations [35] serves as a proxy for productivity losses. Productivity losses for mortality result in the net present value of missed lifetime earnings based on annual wage [35] and years of life lost based on his/her life expectancy [36]. Health effects are measured in quality-adjusted life years (QALYs) and calculates QALYs lost (i.e., accounting for reductions in health effects due to influenza and/or death). Each person accrues QALY values based on age-dependent healthy QALY value attenuated by the influenza-specific utility weight for their illness duration. Death results in the loss of the net present value of QALYs for the remainder his/her lifetime.
For each scenario, we calculate the incremental cost-effectiveness ratio (ICER) and costbenefit, as follows:

Data inputs and sources
Appendix  [42]. The distance agents are willing to travel to be vaccinated is based on the distance persons traveled for medical or dental care from the National Household Travel Survey [43]. Minimum age for vaccination at pharmacy locations vary by state (ranging from 7 to 18 years), to be conservative we assumed the same minimum age of across all pharmacies. All costs, clinical probabilities, and durations came from the scientific literature or nationally representative data sources (e.g., Healthcare Cost and Utilization Project [44], CMS Physician Fee Schedule [45]). When available, all costs and probabilities are age-specific. All costs are 2017 $US, converted using a 3% discount rate.

Scenarios and sensitivity analyses
Our baseline scenario distributes vaccines only through traditional locations, while various experimental scenarios distributed vaccines through pharmacies in addition to traditional locations. We simulated vaccination in pharmacies two ways: (1)  . These scenarios assume that each person seeking vaccination will be vaccinated (i.e., there are enough doses available to cover the population).
Each experiment consisted of 30 realizations in PHIL and Monte Carlo simulations of 1000 trials in FluEcon, varying each parameter throughout their ranges. Results are reported as mean and 95% credibility interval (CrI). Sensitivity analyses varied the R0 (1.30-1.63), the time between epidemic start and vaccine availability (1-28 days), and the probability of mortality (seasonal estimates and four times these values to simulate more virulent circulating strains of influenza [46]). We also evaluated the impact of various locations (traditional only, pharmacies only, and all locations) being able to vaccinate persons faster (i.e., increasing daily vaccination rate such as by adding additional staff per location) and the number and types of pharmacies that increase their hours (all large retail pharmacies and 20% of all other pharmacies). Additional scenarios evaluated the impact of limiting the number of doses to 50% of the needed supply, with doses distributed to locations based on the volume of people they could vaccinate in a day, so that larger locations received more vaccines. Other sensitivity analyses used total compensation (where wage represented 68.2% of the total value [47]) for productivity losses and included age-specific future medical costs [48], where the NPV of future medical costs were subtracted for those who die and accounted for the fraction of individuals that incur that cost at each age. We varied the total cost of the vaccine and its distribution in the cost-benefit analysis.
We ran these sensitivity analyses to account for variations in scenarios, such as epidemic severity, with R0 values within the range of reported values for past pandemics and seasonal influenza [49] and different mortality rates. Varying the availability of vaccines account for different situations like the epidemic starts elsewhere and US has forewarning, or we have future production, vaccine stockpiles (i.e. it is already available), or various timing for virus identification and vaccine distribution. Scenarios limiting the supply can account for stockpile depletion.

Impact of extending pharmacy hours
Extending the hours of all large retailers and 20% of all other pharmacies (15,407 total with increased hours) resulted in 105.1 million vaccinations (Fig. 1) and averted 12.9 million symptomatic influenza cases compared to vaccination in only traditional locations. Costsavings totaled $1.0 billion and $4.1-50.3 billion from the third-party payer and societal perspectives, respectively (Table 1). Extending the hours in all pharmacies resulted in 107.0 million vaccinations and averted 13.0 million cases, was economically dominant compared to traditional only and typical hours from both perspectives and resulted in cost-benefits (Fig. 3).
Extending hours in selected pharmacies garnered societal cost-savings of up to $20.1 billion (R0 = 1.30) and $31.8 billion (R0 = 1.63) when considering employee compensation and future medical costs.

Discussion
Our study shows that during an influenza pandemic, including pharmacies as vaccination locations could avert a substantial number of symptomatic influenza cases, deaths, and costs.
Our results show the value of vaccination during a novel influenza epidemic depends on the number of vaccination locations, as the impact increases with additional sites (i.e., pharmacies). This is consistent with the observation that proportions of the population do not have ready or convenient access to health clinics and hospitals. Vaccine administration rate also plays an important role. Health clinics and hospitals may not have the capacity to vaccinate enough people early enough during the epidemic. Pharmacies have the potential advantage of being more focused on dispensing medications and vaccines during an epidemic without having the same range of services that a traditional location needs to provide [8]. Consequently, the rate at which people can be immunized and capacity may be proportionately higher in pharmacies. Finding ways to further increase pharmacy vaccination rates, such as training pharmacy technicians to vaccinate patients or creating special queues, could be important and may further increase the value of pharmacies.
Pharmacies have potential advantages as immunization sites, including numerous locations in closer proximity to residential neighborhoods, extended operating hours seven days a week, and ability to serve individuals on a walk-in basis, including those without an established healthcare provider [16,18,22,50]. Another advantage is improved access, especially to those residing in medically underserved areas [17,21,51]. Our results show that expanded access and convenience of pharmacy vaccination increases vaccination coverage (33.7% vs. 23.8% when including all pharmacies open typical hours). Additionally, pharmacies may be able to administer vaccinations at a reduced cost compared to traditional locations [52,53].
However, there are limitations to pharmacies. Although pharmacists are authorized to administer vaccinations, there may be policies which limit that authority. In the case of an epidemic, protocols may limit the ability for pharmacies to vaccinate. For example, stateimposed age restrictions limits pharmacists' ability to vaccinate children. If age restrictions are lowered or lifted during an epidemic, pharmacy locations may further increase vaccination access, especially for those under the age of 18 in areas with few traditional locations. Additionally, vaccination in traditional locations can serve as a point entry for healthcare services that are not offered by pharmacies. Furthermore, some insurance providers do not reimburse for pharmacist-provided services.
There is ongoing national and state-level work to incorporate pharmacies into pandemic planning and the acceptability of pharmacists to serve as immunizers, with many pharmacies participating in a Centers for Disease Prevention and Control (CDC)-led effort along with government and private sectors [54]. Additionally, the current national pandemic plan lists continuing to work with pharmacies to improve operations for vaccine distraction as a key action [55]. Thus, quantifying the potential value of pharmacies as vaccination locations can help a number of decision makers determine how to best leverage pharmacies in the event of an epidemic. According to Fitzgerald et al., underestimating the value of pharmacies is one of the biggest gaps in pandemic vaccine program planning [14]. Information on the resulting cost-savings can help policy makers and other officials determine how much can be invested into distributing and allocating vaccines to pharmacies. This also gives third-party payers a better sense of how to structure reimbursements for vaccines administered in other such locations. This information shows traditional locations that adding alternative locations is one way to reduce their burden. Reducing the vaccination workload of traditional locations may allow them to have a greater focus on patient care. It informs people there are more vaccination options out there that may be more convenient. It also shows pharmacies the value of providing this service; they can use this information to make the case for receiving epidemic vaccines and make decisions on program planning.
Models are simplifications of real life and cannot account for every possible event or outcome. The course of an actual epidemic may not conform to our model data and assumptions. Our scenarios assumed a novel virus for which there would be no residual immunity. Existence of residual immunity, such as what occurred during the 2009 pandemic and any other factors that may reduce transmission could result in lower attack rates. Moreover, measured attack rates, such as those previously reported [56], may not always represent actual attack rates. Our data inputs were derived from sources of varying rigor and quality; thus, our results may change as better data become available. For example, we used prescription fill rate to estimate pharmacy size and subsequent vaccination rate. If this proxy is inaccurate, the value of pharmacies would fluctuate based on with the number of vaccines pharmacies could administer (i.e., fewer doses reduces value). Our model did not account for individuals' potential vaccination location preference and assumes that vaccination likelihood was not dependent on type and number of locations in the area (besides the age/ gender limits imposed on specialty practices). However, the presence of clinicians may be associated with the probability of receiving vaccines [57]. Thus, this preference may reduce the value of adding pharmacies if a person is only willing to be vaccinated by their own physician. Our scenarios made vaccines available at the same time, regardless of location. However, during the 2009 H1N1 pandemic, vaccines were not made available in pharmacies until much later than traditional locations [58]. A delay in vaccine availability for pharmacies would lower the incremental benefit of including pharmacies. Our model focuses only on traditional locations that have an established channel for distributing vaccinations that may vaccinate during an epidemic and does not include other potential locations (e.g., workplaces and schools), as these locations require special set up, such as a new distribution chain. As our study focuses on vaccination and determining if adding locations would be helpful, we did not evaluate the distribution of antivirals (which could also affect the spread of influenza) nor did we include mass social distancing measures, such as school closures. Given our study focus, a delay in vaccine availability for all locations would reduce the overall value of vaccination (i.e., epidemic not mitigated to the same degree) but would still show value of adding pharmacies.
Furthermore, we made several assumptions that may impact the value of vaccinations. For example, we assumed an equal chance of visiting any of the eligible vaccination location types (if all types are available to a person), as capturing preference for each person is complicated. We also assumed each location type will have an epidemic vaccine supply. However, not all locations may be willing to offer vaccines or be willing to offer them to non-patients. For example, not everyone could be vaccinated in all locations (we may not account for all limitations for each location) -hospitals may not offer the vaccine to the general public, only vaccinating its patients and employees; physicians' offices may not offer the vaccine to those who are not current patients. This assumption is conservative as it overestimates the value of vaccination in general, overestimates the benefits of traditional locations, and underestimates the value of pharmacies.

Conclusions
Administering vaccines through pharmacies in addition to traditional locations in the event of an epidemic can increase vaccination coverage and mitigate up to 23.7 million symptomatic influenza cases, providing cost-savings up to $2.8 billion to third-party payers and $99.8 billion to society. Pharmacies should be considered as points of administering epidemic vaccines in addition to traditional settings as soon as vaccines become available.