The cost benefit model was designed to evaluate the costs, benefits, and net savings associated with influenza immunizations. It is used to illustrate the financial value of a program to a set of stakeholders over a specific time period. Sensitivity testing is performed to vary the user input and model assumptions to make sure the best model assumptions are used to generate the most reasonable the financial value projections. We used an actuarial approach because it allows for multiple interrelated assumptions around various aspects of risks for a set of contingent future events to determine the most reasonable time value of money for different stakeholders [14
]. The model provides users with a base set of assumptions and variable values derived from literature but with considerable ability to vary those assumptions interactively for specific populations. Key variables that affect outcomes are summarized in Table and organized into inputs, cost categories, and outputs. The majority of cost assumptions were taken from two studies: Molinari et al. [1
] and Prosser et al. [12
]. Both studies used Monte Carlo simulations based on the same retrospective analysis of influenza-related utilization over multiple years from a large database of insurance claims (Medstat Marketscan) to model costs of influenza.
Molinari et al. first used national epidemiological data and published studies to estimate the probability of four categories of influenza health outcomes: illness, outpatient visits, hospitalizations, and death. All probabilities were age and risk stratified. Next, Molinari and colleagues estimated direct and indirect costs associated with influenza. Most direct costs were based on 179,718 cases of influenza identified in the Medstat Marketscan database between 2000 and 2004. Indirect costs valued lost productivity due to illness or death. Monte Carlo simulations with sensitivity analysis yielded estimated cases and average costs (with 95% upper and lower confidence intervals). The probabilistic model predicted an estimated 30,151,934 annual cases that resulted in 21,354 [22,636, 43,507] outpatient visits, 3,131 [2,108, 4,511] inpatient days, 44,003 [25,694, 62,484] lost productivity days, and 611 [360, 953] undiscounted life years lost. Total direct costs and total economic burden in 2003 dollars was estimated as $10.4 billion [$4.1, $22.2] and $87.1 billion [$47.2, $149.5], respectively.
Molinari et al. did not project potential savings for averted cases due to vaccination; however, Prosser et al. built on Molinari’s cost assumptions to estimate cost savings of vaccination in various clinical settings, such as mass vaccination events, pharmacy and doctors’ offices. Prosser et al. noted that non-traditional settings, such as pharmacies, were highly cost effective. While their results provided general guidance for comparative cost efficacy, estimating cost efficacy for a particular population would require some effort. Hence, we built upon the foundation of both previous studies to build a decision support tool to model various immunization strategies.
1. Projection Year: Because of medical inflation, cost variables will differ by calendar year. The underlying assumptions and factors in the model are derived from different years and inflation factors are used to adjust the literature-based variables to project to the period chosen by the model user. Medical costs were adjusted by medical trend, while non-medical costs were adjusted by the consumer price index. The default projection year is the next (i.e., upcoming) influenza season.
2. Population demographics:
Risk of influenza and its complications varies by age. While some populations (e.g., employers) will be more heavily weighted towards younger persons as employers terminate their retiree benefits, other populations such as a Medicare health plans may be older. Therefore, the model allows users to tailor savings estimates to the specific demographics of the client population. Five age groupings are used in the model: ≤11 (children), 12–17 (youth), 18–49 (young adults), 50–64 (middle-age adults), and 65 and older (older adults). A sample population of 15,000 persons is provided as a default in the model with a typical age profile (13%, 13%, 40%, 20%, and 13%, by respective age group), based on a 2009 estimate of the U.S. census [15
3. Vaccination delivery channel:
By surveying non-traditional immunization providers, Prosser et al. [12
] found that the immunization costs were lower at mass vaccination events and pharmacies compared to traditional channel (TC) settings. To compare costs across channels, three input variables were needed: cost in traditional settings ($TC
), cost in non-traditional settings ($NTC
), and the proportion of persons using non-traditional channels (%NTC
). To provide a default value for traditional channels, we derived an average allowed charge per physician administered influenza vaccination ($TC
$82) from a large national dataset [Unpublished data from Solucia Consulting's proprietary database]. The model also allows users to adjust the expected utilization of non-traditional settings (default %NTC
0.50) and the cost of immunizations in such settings (default $NTC
4. Percent vaccinated:
As evidenced in the literature, not everyone who should get a flu shot actually gets one [16
]. Coverage rates not only remain below national targets but also vary substantially by age group. To account for this variation, the model permits users to estimate the percent of population who are successfully vaccinated by age group. The model provides default coverage rates by age group (< 11: 55.2%, 12–17: 55.2%, 18–49: 49%, 50–64: 49%, 65+: 72%) based on U.S. coverage during the 2009/2010 season [16
]. Alternatively, users can enter their own projected coverage distribution. Of note, this measure of adherence is distinct from vaccine effectiveness, which is discussed later in this section.
5. Economic assumptions
: Economic assumptions include productivity and wages. The model provides baseline numbers based on U.S. averages [15
], but allows the user to vary these assumptions to meet the needs of a specific population. To increase precision of productivity and death cost estimates, three inputs are required to calculate the average value of a lost workday: (a) Current Average Annual Wage, (b) Ratio of Benefits to Wages, and (c) Assumed Workdays per Year. The default assumption is an average annual salary of $40,000 with a 250 workday year increased by 30% to accommodate the value of employee benefits.
6. Benefit design:
By providing details of cost sharing of vaccination and other medical services, the model is able to allocate costs and benefits to employers and employees. The default values assume that the plan (as funded by the employer) covers 90% of the vaccination costs and 78% of general medical costs [17
]. When the benefit design for a particular population is known and different from the default value, the model can more accurately portray the distribution of cost and savings estimates to stakeholders.
7. Preventable cases:
The proportion of ILI cases that can be prevented by vaccination is contingent on the effectiveness of vaccine
and the incidence of disease
. Effectiveness of the seasonal influenza vaccine varies by year, depending on many factors including the infectivity, pathogenicity, and virulence of the viral strains and the antigenic match with the vaccine. Furthermore, effectiveness varies by age group (due to human behaviors and social networks as much as characteristics of the virus) or co-morbid risk factors (due to decreased immunity). Interconnected with effectiveness, incidence also varies by strain and by age group. Two published studies [12
] combined these concepts of vaccine effectiveness and influenza incidence into a rate of avoidable cases that was used in our model. The proportion of avoidable cases was higher in children aged 5–11 (5.5%) and persons aged 65 and older (5.4%) compared to youth aged 12–17
years (4.1%), young adults aged 18–49
years (4.6%), or middle-aged adults 50–64 (4.6%). For our model, estimated avoided cases among children under 5
years were assumed to be similar to children aged 5–11. These figures were based on non-pandemic years. To account for greater infectivity and severity during pandemic years, the model incorporates a pandemic risk factor, which is discussed in a later section.
8. Risk Level:
Individuals at higher risk for ILI and its complications are more likely to incur ILI-related expenses compared with individuals with lower risk. Several published studies stratified cost assumptions by risk category. Therefore, the model allows one of two immunization strategies based on risk level: (a) immunizing the entire population (i.e., universal vaccination) or (b) targeting at-risk individuals. The definition of high risk
followed Molinari et al. [1
], who defined high risk as the presence of at least one co-morbid condition, as listed in Advisory Committee on Immunization Practices (ACIP) guidelines [4
]. The risk profile by Molinari et al. estimated the proportion of high-risk individuals among age 17 and under, aged 18–49
years, 50–64 years, and 65 and older as 10.6%, 14.9%, 33.0%, and 51.2%, respectively. Although the Molinari profile was published between 1994 and 1997, this distribution was used because it remains consistent with 2010 CDC reported risk strata: 8.6% of youth (aged 12–17
years), 15.2% of young adults (aged 18–49
years), and 33.5% of middle-aged adults (aged 50–64
years). To be consistent with current ACIP guidelines, our model assumed all persons aged 65 and older were considered at high risk of influenza infection and complications [4
]. The difference between a population and high-risk strategy is the type of individuals targeted
for vaccination. Such strategies do not assume that all targeted individuals received
the vaccination; therefore, coverage rates were also applied to high-risk vaccination programs.
9. Risk of Pandemic:
During years with pandemic influenza, the value of vaccination is greater because the risk of exposure and infection is higher compared to non-pandemic years [13
]. Based on the available data from the last three pandemics, Crowe and colleagues [19
] estimated that there is a 2% probability of a pandemic (PrP
.02) in any given year, and pandemic flu incidence ranged from 9% to 35% , compared to 4% to 6% in non-pandemic years. Since incidence of pandemic influenza continues to vary with age, an attack rate adjustment (ARA) was used to tailor risk for the age distribution in the model:
To account for the potential of a pandemic in any given year, utilization costs were adjusted by a pandemic risk adjustment factor (PRA), whereby:
Influenza and its complications impose costs on patients, health plans, employers, and society. The literature identified six categories of potential costs associated with a diagnosis of influenza: inpatient hospitalization, outpatient visits, self-medication, death, and productivity (presenteeism and absenteeism).
Molinari et al. [1
] defined costs for the hospitalization episode to include inpatient costs as well as outpatient and pharmacy services incurred 2
weeks prior and 4
weeks after the inpatient stay. Using Molinari’s definition, Prosser et al. reported hospitalization costs by age and risk. For example, among persons without risk factors for influenza, hospitalization was most costly for persons who were aged 50–64
$23,281; Range: $20,887-$26,159) [12
]. This model used Prosser’s 2004 mean estimates, which were trend adjusted for the projected program year.
2. Outpatient visits:
As one of the most frequent costs of influenza, outpatient services were defined by Molinari et al. as visits to primary care providers, specialists, urgent care clinics, and emergency rooms for an ILI. In addition to the cost of the visit, any pharmacy and laboratory services within a 3-day window were also included to estimate total outpatient costs not associated with inpatient episodes. Like hospitalizations, the range of outpatient costs were age and risk dependent (range
$51-$765 per person with influenza) [12
Use of over-the-counter (OTC) medications is an early indicator of disease because many patients will use OTC either in lieu of or in addition to seeking medical care [20
]. The cost of the typical "flu basket" was used to provide an approximation of the costs of self-medication. The flu basket includes products that are sold along with purchase of an OTC flu medicine, such as cough drops, pain relievers, decongestants, or juice. A retrospective analysis of data from a large pharmacy chain was used to estimate the average value of the basket in 2010.
In the model, the costs associated with death were based on the value of a statistical life (VSL), which is a commonly used and comprehensive estimate to represent the “lost productivity as well as the intrinsic, or social, value place on human life” [1
]. To calculate the savings from death avoided due to influenza vaccination, costs associated with death (as reported by Spurr in Molinari et al.) [1
] were multiplied by the proportion of preventable deaths (as reported by Grosse in Prosser et al.) [12
5. Absenteeism and presenteeism:
In the United States, presenteeism and absenteeism due to influenza accounts for approximately 17 million lost workdays per year [21
is defined as days taken off work to care for influenza illness in oneself, a child, or other dependent. In the Molinari study [1
], estimated lost workdays ranged from 0.5-1.0
days in non-treated ILI, 1–7
days in outpatient-treated ILI, or 8–24
days in ILI requiring hospitalization. Absenteeism due to child’s illness ranged from 0.7-2.8
days per episode of flu. While this latter estimate seemed low, the estimate may be reasonable given that only one parent in a two-parent household would take time off.
was defined as decreased productivity due to workers with influenza continuing to be present in the workplace [22
]. Presenteeism was estimated from the average reduced work effectiveness days as reported by Nichol [23
Estimated days lost due to either absenteeism or presenteeism was transformed into costs by multiplying by the calculated average daily wage of the population, as discussed in the previous input section.
Our basic cost model estimates net savings as total costs avoided due to disease prevention from vaccination minus the total costs of vaccination. This calculation is comparable to that reported by Nichol et al. [23
] but provides cost output as a positive savings rather than negative costs.
1. Number of persons immunized (NPI)
was calculated from the input number of individuals in the population in an age category, multiplied by the expected coverage in each age category, and totaled across all age categories. As previously discussed, the default vaccination rates are based on national coverage estimates by age. When a high-risk strategy is selected (HR
1), NPI also accounts for the proportion of individuals who were estimated to be high risk by age group. Therefore:
subset of persons in the ith
age category, with the corresponding estimates of coverage (Vi
) and high risk (HRi
2. Expected total immunization cost (COSTS) for a non-traditional channel (NTC) such as pharmacies and a traditional channel (TC) such as doctor’s offices were calculated as a product of the unit cost of vaccination ($) in that channel multiplied by the count of number of persons vaccinated in each channel. Therefore, total costs were estimated as:
NPI * percent of immunizations in non-traditional channels.
These assumptions were discussed in the input section (3).
3. Total expected savings (SAVINGS) from avoided influenza cases are derived, for each risk and age category, from the product of (a) the number of persons immunized, (b) the estimated preventable cases per person immunized, (c) the unit of utilization per preventable case for each cost category, and (d) the unit cost for each cost category. For example, the equation for savings from avoided inpatient (IP) utilization is:
Parallel equations were constructed for outpatient, medication, and productivity savings. Total savings was the sum of these potential savings, in addition to estimated savings from averted deaths, which is discussed in the next section.
4. Stakeholder allocation: In the model, vaccination cost and savings are allocated to three stakeholders: payers (e.g., employers, health plans, or government payers), individuals (e.g., health plan members or employees and their dependents), and society (e.g., communities, families).
a. Total costs of vaccination were distributed to employers or employees based on the proportion of immunization cost sharing (‘Vaccination Cost Sharing % by Employers’). No costs associated with vaccination were allocated to society.
b. All savings from self-medication costs were allocated solely to the employee.
c. Depending on how labor is compensated, the cost of sick time may be borne by the employer as reduced productivity, passed to the employee through reduced wages (as in the case of hourly-paid workers), or passed to an insurer (when absence is insured, for example with short-term disability). In this model, savings from avoided absenteeism was applied to the employer category.
d. All savings from presenteeism are credited to employers.
e. avings from avoided outpatient and inpatient events are split between employers and members, based on the benefit plan design.
f. The cost savings from avoided death were shared by all three stakeholders. Of the total VSL, approximately 75% was allocated towards societal savings, whereas the remaining 25% was apportioned to employers and employees. For employer groups particularly, a number of members will be dependents or could be retirees. Therefore, savings from avoided death was applied only to the members who were also employees. Therefore, this allocation is dynamic and based on enrollment input. The model assumes that 40% of members were employees, although this proportion could be adjusted. Employers were allotted savings equal to the product of (a) twice the annual salary (SAL) of an employee, (b) the proportion of the population who were employees, and (c) the age-weighted count of preventable deaths. This figure accounts for the loss productivity in an employee’s role until a new employee is hired and fully trained. For an employee, or rather, for the family of an employee, the loss of life accounts for projected loss of future earnings as well as the value to the family unit. This amount was allocated as estimated VSL multiplied by the percent of VSL that was allocated to the present value of future earnings (% PVFE) minus the employer’s portion. The remaining portion of the VSL reflected loss of the intrinsic value of life to society. The equations for each of these portions are:
age-weighted unit preventable death * 2 * SAL * % members employed
($VSL * % PVFE) – Employer
$VSL – Employer’s portion – Employee portion
Aggregate cost and savings estimates across all stakeholders, as well as an aggregate of employer (payer) and employees (members) savings are provided.
5. Expected net savings equaled total SAVINGS – total COSTS (i.e., 3–2).
6. Expected net savings per vaccination (PV) was the net savings total divided by the number of persons immunized (i.e., 5 / 1).
7. Expected net saving per member per year (PMPY) was the net savings total divided by the number of enrolled members.