Rachel R. Bailey, MPH

To develop 3 computer simulation models to determine the potential economic effect of using intravenous (IV) antiviral agents to treat hospitalized patients with influenza-like illness, as well as different testing and treatment strategies.

 Stochastic decision analytic computer simulation model.

 During the 2009 influenza A(H1N1) pandemic, the Food and Drug Administration granted emergency use authorization of IV neuraminidase inhibitors for hospitalized patients with influenza, creating a need for rapid decision analyses to help guide use. We compared the economic value from the societal and third-party payer perspectives of the following 4 strategies for a patient hospitalized with influenza-like illness and unable to take oral antiviral agents: Strategy 1: Administration of IV antiviral agents without polymerase chain reaction influenza testing. Strategy 2: Initiation of IV antiviral treatment, followed by polymerase chain reaction testing to determine whether the treatment should be continued. Strategy 3: Performance of polymerase chain reaction testing, followed by initiation of IV antiviral treatment if the test results are positive. Strategy 4: Administration of no IV antiviral agents. Sensitivity analyses varied the probability of having influenza (baseline, 10%; range, 10%-30%), IV antiviral efficacy (baseline, oral oseltamivir phosphate; range, 25%-75%), IV antiviral daily cost (range, $20-$1000), IV antiviral reduction of illness duration (baseline, 1 day; range, 1-2 days), and ventilated vs nonventilated status of the patient.

When the cost of IV antiviral agents was no more than $500 per day, the incremental cost-effectiveness ratio for most of the IV antiviral treatment strategies was less than $10,000 per quality-adjusted life-year compared with no treatment. When the cost was no more than $100 per day, all 3 IV antiviral strategies were even more costeffective. The order of cost-effectiveness from most to least was strategies 3, 1, and 2. The findings were robust to changing risk of influenza, influenza mortality, IV antiviral efficacy, IV antiviral daily cost, IV antiviral reduction of illness duration, and ventilated vs nonventilated status of the patient for both societal and third-party payer perspectives.

 Our study supports the use of IV antiviral treatment for hospitalized patients with influenza-like illness.

	During the 2009 influenza A(H1N1) pandemic, the Food and Drug Administration granted emergency use authorization of peramivir because of a dearth of alternative drugs for patients unable to take oral or inhaled antiviral agents and because of limited safety and efficacy data for peramivir. Our economic model suggests the following:

		Intravenous antiviral agents are a cost-effective intervention for hospitalized patients with severe influenza-like illness.

		The use of PCR influenza testing to guide whether to initiate intravenous antiviral treatment is the most cost-effective strategy.

		The use of intravenous antiviral agents in this setting may save money and offer health benefits.

		Our findings should help guide policy makers and clinicians in the use of such novel treatments during other influenza seasons.

	The 2009 influenza A(H1N1) pandemic raised questions about the role of intravenous (IV) antiviral medications in treating hospitalized (ventilated and nonventilated) patients with influenza-like illness (ILI). Intravenous neuraminidase inhibitors had been under rapid development, with IV peramivir receiving emergency use authorization (EUA) from the Food and Drug Administration (FDA) in October 2009, after the US Secretary of Health and Human Services declared a public health emergency.

The primary motivation for EUA was a lack of alternative drugs for hospitalized patients with influenza A(H1N1) who were unable to take oral or inhaled antiviral agents. Standard FDA approval was not a viable option because efficacy and safety data were limited and the pandemic was already in full bloom. The EUA allowed healthcare providers to administer IV peramivir to patients hospitalized with confirmed or suspected influenza A(H1N1) only if they were unresponsive to or unable to take oral or inhaled antiviral agents.

 Contraindications included a history of severe allergic reaction to neuraminidase inhibitors.

	Antiviral medications are the only medications available to reduce the morbidity and mortality of individuals infected with influenza. Neuraminidase inhibitors are an important and widely used class of antiviral agents, the most commonly used being oral oseltamivir phosphate and inhaled zanamivir (both approved by the FDA in 1999).

 In influenza A and influenza B, an enzyme cleaves links between the infected host cell and the influenza virus envelope. This, in turn, allows the viruses that replicated in the host cell to be released to the rest of the body.

 By inhibiting viral replication and thereby limiting the number of viruses in the body, neuraminidase inhibitors could reduce the duration of illness and risk of mortality.

	Because IV peramivir was a novel drug, there were no available clinical trials among higher-risk groups such as pregnant women, pediatric patients, and older adults. It was also unclear how viral resistance to other neuraminidase inhibitors may translate to resistance to peramivir.

 Intravenous antiviral agents such as peramivir have several potential advantages. First, they offer an alternative route of administration, which is especially important for patients who cannot take medication by mouth (such as ventilated patients). Second, when heavy demand may deplete inventories of other antiviral agents such as oseltamivir and zanamivir, IV antiviral agents can serve as another available option. Third, there remains the possibility that strains resistant to other antiviral agents may not be completely resistant to newer antiviral agents such as peramivir, although evidence suggests  hat oseltamivir-resistant strains may also be resistant to peramivir.

	Intravenous antiviral agents have only recently emerged as potential treatment options, and questions remain about their economic value. Should they be reserved for intensive care unit patients or administered to all hospitalized patients with influenza who cannot take oral antiviral agents? What is a reasonable price for IV antiviral medications? How would the value of IV antiviral agents change with emerging resistance? Should patients be tested for influenza before the initiation of IV antiviral agents, or should IV antiviral treatment be initiated first, followed by confirmatory testing to determine whether treatment should be continued? Will the value be different for seasonal vs pandemic influenza?

	We developed 3 computer simulation models to estimate the potential economic effect of using IV antiviral agents to treat hospitalized patients with ILI, as well as different testing and treatment strategies. Various simulation runs explored seasonal and pandemic influenza scenarios and evaluated the  effects of varying patient age, probability of having influenza, ventilated vs nonventilated status of the patient, and the probability of different influenza outcomes such as mortality.

 shows the general structure of our Monte Carlo decision analytic computer simulation model, constructed using commercially available software (TreeAge Pro 2009; TreeAge Software, Williamstown, Massachusetts). Our model represented the economic value from the societal and thirdparty payer perspectives of the following 4 alternative strategies for a patient hospitalized with ILI and unable to take oral antiviral agents, as per the EUA1-3:

Strategy 1: Administration of IV Antiviral Agents Without Polymerase Chain Reaction (PCR) Influenza Testing. 

In this strategy, the patient received a full 5-day course of IV antiviral agents regardless of whether the patient actually had influenza. No PCR influenza testing was performed.

Strategy 2: Initiation of IV Antiviral Treatment, Followed by PCR Testing to Determine Whether the Treatment Should Be Continued.

 This strategy involved initiation of IV antiviral treatment for hospitalized patients with ILI and then performance of PCR testing for influenza, which required a 24-hour turnaround time for results. A negative test result prompted  discontinuation of IV antiviral treatment after a single dose.

Strategy 3: Performance of PCR Testing, Followed by Initiation of IV Antiviral Treatment If the Test Results Were Positive. 

For this strategy, IV antiviral agents would not be initiated until PCR testing was performed and results were available, delaying treatment for 24 hours. A positive test result prompted initiation of IV antiviral treatment.

 Strategy 4: Administration of No IV Antiviral Agents.

 This strategy involved giving no medical interventions as the treatment.

	Each simulated adult traveled through the decision tree pictured in the Figure and faced a probability draw (first-order trial) at each chance node. This draw was then compared with a value pulled from the probability parameter distribution to determine down which branch he or she traveled (second-order trials). The costs, utilities, and durations of each  resulting outcome also drew from their respective probability distributions (second-order trials). Each patient had a probability of having influenza (baseline, 10%).

 Test results depended on whether the patient actually had influenza, and the sensitivity and specificity of the test. Intravenous antiviral treatment had a probability (IV antiviral efficacy) of reducing the duration of illness by 1 day (ie, if the dice roll is less than the efficacy number, then the illness duration is reduced; if it is higher, then there is no effect) and the risk of mortality (ie, mortality is reduced by 1 minus IV antiviral efficacy).

	Scenarios from the third-party payer perspective considered only the direct costs of illness, while scenarios from the societal perspective included both direct and indirect costs of illness (ie, productivity losses from caregiver time determined by lost wages from time spent with the patient). Each simulation run sent 1000 simulated adults 1000 times through each model, for a total of 1,000,000 trials per scenario. During each run, each parameter drew from the relevant triangular or beta distributions. Table 1 lists the study sources of data inputs for the model.

	The following equation calculated the incremental costeffectiveness ratio (ICER) of each strategy vs the comparator strategy: Cost

	Each simulation run generated a mean ICER and a 95% confidence interval. A strategy was considered cost-effective if the ICER was less than $50,000 per quality-adjusted life-year (QALY).

 However, because there is some debate about the exact threshold for cost-effectiveness, we report the resulting ICER values so that individual readers may  ake their own determination of what constitutes a cost-effective intervention.

	The study used various inputs and their corresponding distributions for the probabilities, costs, durations, and utilities in the model. Hospital mortality from influenza drew from a beta distribution. All other variables drew from triangular distributions. A 3% discount rate converted all past and future costs to 2009 values.

	Because PCR testing is unavailable and is not part of standard care in many inpatient settings, our model did not include PCR test costs for the IV antiviral and no IV antiviral strategies.

 The hospitalization costs for these arms did not include PCR testing. The PCR test costs were added to the 2 strategies that included PCR testing.

	Effectiveness was measured in QALYs. Influenza caused QALY decrements for the duration of illness.

All QALY accruals were age adjusted based on the quality-of-life utility obtained by Gold et al.

Patients who survived accrued ageadjusted and discounted (3% discount rate) QALYs based on life-expectancy estimates from the Human Mortality Database; patients who did not survive did not accrue these QALYs.

	Probabilistic (Monte Carlo) sensitivity analyses explored all parameters simultaneously using all distributions examined. Sensitivity analysis varied these following key variables: the probability of having influenza (baseline, 10%; range, 10%-30%),9 IV antiviral efficacy (baseline, oral oseltamivir; range, 25%-75%), IV antiviral daily cost (baseline, $20 [oral oseltamivir]; range, $20-$1000), IV antiviral reduction of illness duration (baseline, 1 day; range, 1-2 days), patient age (baseline, 20 years; range, 20-60 years), and influenza mortality (baseline, seasonal influenza mortality; range, up to twice the seasonal influenza mortality for a pandemic scenario). These sensitivity analyses involved fixing the variable of interest and then allowing the rest to pull from the distributions examined. We evaluated seasonal and pandemic influenza scenarios, as well as IV antiviral treatment in ventilated and nonventilated patients.

 give the results at baseline, while varying IV antiviral efficacy (25%-75%) (ie, the proportion by which IV antiviral agents will reduce mortality) and IV antiviral daily cost from the third-party payer perspective for the seasonal and pandemic scenarios. Each table gives the ICER (compared with the most economically favorable option) for each strategy. Calculated ICER 95% confidence intervals demonstrated no overlaps, suggesting that the differences were significant.

	For the ventilated population under seasonal influenza conditions, the ICER of the IV antiviral treatment vs no treatment was less than $10,000 per QALY in most scenarios, as long as the IV antiviral cost was no more than $500 per day, except for strategy 2 (initiation of IV antiviral treatment, followed by PCR testing to determine whether the treatment should be continued). In fact, when IV antiviral cost was no more than $100 per day, all 3 IV antiviral strategies were even more cost-effective (ie, provided cost and health benefits) than no treatment, with the order from most to least cost-effective being strategies 3 (performance of PCR testing, followed by initiation of IV antiviral treatment if the test results are positive), 1 (administration of IV antiviral agents without PCR influenza testing), and 2 (initiation of IV antiviral treatment, followed by PCR testing to determine whether the treatment should be continued). As the risk of influenza increased from 10%, the antiviral strategies grew even more economically favorable. Similarly, increasing the IV antiviral illness reduction effect to 2 days further enhanced the cost-effectiveness of all IV antiviral strategies.

	These findings were robust to various sensitivity analyses. Pandemic conditions increased the economic value of only the IV antiviral strategies. The order of cost-effectiveness of the strategies did not change as long as IV antiviral treatment cost no more than $100 per day. Switching from the thirdparty payer perspective to the societal perspective captured additional cost savings for the IV antiviral strategies.

Articles

The US Food and Drug Administration granted emergency use authorization of intravenous neuraminidase inhibitors for patients with 2009 influenza A (H1N1), creating a need for economic studies.

Economic Model for Emergency Use Authorization of Intravenous Peramivir

During the 2009 influenza A(H1N1) pandemic, the Food and Drug Administration granted emergency use authorization of peramivir because of a dearth of alternative drugs for patients unable to take oral or inhaled antiviral agents and because of limited safety and efficacy data for peramivir. Our economic model suggests the following:

Intravenous antiviral agents are a cost-effective intervention for hospitalized patients with severe influenza-like illness. 

The use of PCR influenza testing to guide whether to initiate intravenous antiviral treatment is the most cost-effective strategy. 

The use of intravenous antiviral agents in this setting may save money and offer health benefits. 

Our findings should help guide policy makers and clinicians in the use of such novel treatments during other influenza seasons.

 To estimate the economic value of preoperative methicillin-resistant Staphylococcus aureus (MRSA) screening and decolonization for cardiac surgery patients.

Monte Carlo decision-analytic computer simulation model.

We developed a computer simulation model representing the decision of whether to perform preoperative MRSA screening and decolonizing those patients with a positive MRSA culture. Sensitivity analyses varied key input parameters including MRSA colonization prevalence, decolonization success rates, the number of surveillance sites, and screening/decolonization costs. Separate analyses estimated the incremental cost-effectiveness ratio (ICER) of the screening and decolonization strategy from the third-party payer and hospital perspectives.

Even when MRSA colonization prevalence and decolonization success rate were as low as 1% and 25%, respectively, the ICER of implementing routine surveillance was well under $15,000 per quality-adjusted life-year from both the third-party payer and hospital perspectives. The surveillance strategy was economically dominant (less costly and more effective than no testing) for most scenarios explored.

Our results suggest that routine preoperative MRSA screening of cardiac surgery patients could provide substantial economic value to third-party payers and hospitals over a wide range of MRSA colonization prevalence levels, decolonization success rates, and surveillance costs. Healthcare administrators, infection control specialists, and surgeons can compare their local conditions with our study's benchmarks to make decisions about whether to implement preoperative MRSA testing. Third-party payers may want to consider covering such a strategy.

	Routine preoperative methicillin-resistant Staphylococcus aureus (MRSA) screening of cardiac surgery patients could provide substantial economic value to third-party payers and hospitals over a wide range of MRSA colonization prevalence levels (>1%), decolonization success rates (>25%), and decolonization costs (up to $200 per patient).

		Healthcare administrators, infection control specialists, and surgeons can compare their local conditions with our study's benchmarks to make decisions about whether to implement preoperative MRSA testing.

		Third-party payers may want to consider covering such a strategy.

aureus is a common complication of cardiac surgical procedures. Postoperative 

 infections have been associated with significant morbidity, prolonged hospitalization, higher medical costs, an increased likelihood of readmission within 90 days postprocedure, and death.

 is the most common cause of sternal wound infections and the leading cause of infective endocarditis in the developed world.

 An increasing number of these infections are due to methicillin-resistant 

 (MRSA), resulting in greater morbidity and mortality than infections from methicillin-sensitive 

	Patients about to undergo cardiac surgery tend to have a number of factors predisposing them to postoperative MRSA infections, including older age, comorbid conditions (eg, congestive heart failure, chronic kidney disease, diabetes), and recent hospitalization.

 An estimated 30% of Americans are chronic or intermittent carriers of 

 and nasal MRSA colonization rates from 0.4% to 20.6% have been reported.

 Cardiac surgery procedures often are invasive, involving prolonged use of mechanical support devices and indwelling intravenous lines, and introduce foreign materials and tissues (eg, a pacemaker, a prosthetic valve) into the body.

	Preoperative screening and decolonization of cardiac surgery patients who are MRSA positive could help prevent postoperative MRSA infections and their sequelae.

 Studies have suggested that nasal MRSA colonization is associated with increased rates of postsurgical MRSA infections and poor outcomes, and that decolonization prior to surgery could lower rates of MRSA surgical-site infections and bacteremia.

 However, implementing routine screening and decolonization can be costly, and to date the economic value of such a strategy has not been fully ascertained. Moreover, the appropriateness of such a strategy may differ by rate of MRSA colonization prevalence, which varies geographically and temporally, as well as by individual patient risk factors.

	To estimate the economic value of preoperative MRSA screening and decolonization for cardiac surgery patients, we developed a stochastic decision-analytic computer simulation model. Sensitivity analyses varied key model parameters and allowed us to delineate how the cost-effectiveness of this strategy may vary by MRSA colonization prevalence, decolonization success rate, and surveillance/decolonization cost. The results of our model may help clinicians, hospital administrators, third-party payers, and other decision makers determine when and where to use this strategy.

 depicts the general structure of our Monte Carlo decision-analytic computer simulation model. This model, constructed using TreeAge Pro 2009 (TreeAge Software, Williamstown, MA), represents the decision of whether to perform preoperative MRSA screening (by means of a single anterior nares culture) and decolonization for those patients with a positive MRSA culture. The time course of each simulation run is the perioperative period for that patient’s cardiac surgical procedure.

	Each patient entering the model had a probability of being MRSA colonized based on the local prevalence of MRSA, but only those who were screened and tested positive underwent decolonization. All positive test results led to decolonization, regardless of true colonization status. Decolonization had a probability of success based on the chosen regimen’s efficacy. Successful decolonization eliminated a patient’s risk of MRSA infection throughout the perioperative period; unsuccessfully decolonized patients retained a probability of developing MRSA infection. Patients who developed a MRSA infection proceeded through the MRSA infection outcomes subtree (

). Patients entering this subtree had an independent probability of developing each of the following sequelae alone or in combination: mediastinitis, prosthetic valve endocarditis, septic shock, surgical-site infection, catheter-related bloodstream infection, and pneumonia. Returning to Figure 1, each individual with an invasive infection has a probability of death.

	In our model the probability of an outcome was not a point estimate, but rather the multiplicative product of the probabilities of all relevant events. For example, the probability of a postoperative MRSA infection was the product of a value drawn from the MRSA colonization probability distribution (range: 0.01 to 0.20) and a value drawn from the postoperative MRSA infection probability distribution (range: 0.0 to 0.4024). Therefore, the probability of a postoperative MRSA infection ranged from the product of the minimum values that can be drawn from each probability distribution (0%) to the product of the maximum possible values drawn from each probability distribution (8.048%).

	Each complication necessitates a distinct set of treatment procedures, outcomes, and costs. Surgical-site infections alone do not require additional procedures. Pneumonia requires a chest x-ray. Patients who develop mediastinitis have a probability of undergoing mediastinal exploration. Patients who develop prosthetic valve endocarditis may need a transesophageal or transthoracic echocardiogram, or require a valve replacement procedure. Septic shock necessitates central and arterial line insertion, as well as transesophageal and transthoracic echocardiograms.

	To estimate the probability of each clinical outcome, we conducted a PubMed search using the following key terms: [cardiac], [cardiac surgery], [methicillin-resistant 

], and [MRSA]. We limited the search to English-language articles published since 1999 and reviewed all abstracts returned by the search to determine their appropriateness. Case reports, case series, and studies that did not clearly report their population denominators were excluded. We included only studies that clearly characterized study populations and reported on the full set of clinical outcomes from that population (ie, the numerators were properly defined). Parameter distributions in the model reflect the range of values identified by the literature search.

	The median age of simulated patients was 65 years, consistent with the population undergoing cardiac procedures.

 All cost, probability, and utility values and their respective distribution parameters are shown in 

.  Gamma distributions were used for all cost variables. Hospitalization cost and length of stay (LOS) gamma distributions for each condition were approximated using mean and standard deviation values from the Healthcare Cost and Utilization Project Nationwide Inpatient Sample.

 All probability values drew from beta distributions, with the exceptions of diagnostic test sensitivity and specificity for MRSA, and the probability of developing pneumonia given a MRSA infection, which drew from a triangular distribution. Using a triangular distribution is appropriate when a parameter’s distribution is asymmetric and only the lower limit, mode, and upper limit are known. Such a distribution is shaped like a triangle with the lower limit at the far left of the base, the upper limit at the far right, and the mode at the point of the triangle. Model assumptions were made based on the convention for economic analyses of MRSA and previously published literature. A 3% discount rate was used to standardize all cost and utility measures into 2009 values.

	Our model measured effectiveness in quality-adjusted life-years (QALYs). An otherwise healthy 65-year-old had a baseline of 0.84 QALY per year of life.

 Patients who did not survive lost QALYs based on their projected life expectancy estimates from the Human Mortality Database.

Each simulation run consisted of 1000 hypothetical patients (each with a distinct set of characteristics) about to undergo cardiac surgery. These patients traveled through the model 1000 times each, accruing a distinct set of costs, utilities, and outcomes each time because of the probabilistic nature of the input parameters. This process generated a total of 1 million patient outcomes for each simulation run.

	For each simulation run, we estimated the incremental cost-effectiveness ratio (ICER), expressed in US dollars per QALY, of implementing the screening and decolonization strategy using the following formula:

	Although some debate exists about the ICER threshold below which an intervention is considered cost-effective, $50,000 per QALY is a commonly used cut point, and ICER values less than $20,000 per QALY offer strong economic support for the adoption of an intervention.

 Strategies that are both less costly and more effective are deemed economically dominant.

	Separate analyses evaluated the economic value of screening and decolonization from the following perspectives:

	• Third-party payer. This perspective accounted for the direct medical costs associated with each outcome.

	• Hospital. This perspective accounted for the additional hospital LOS associated with each outcome. Added LOS resulted in lost hospital bed days for the hospital.

 A method described by Graves et al converted lost hospital bed days into costs.

	Sensitivity analyses varied the values of key input parameters and established the model’s robustness. Specific analyses systematically varied MRSA colonization prevalence from 1% to 40%, and decolonization success rates from 25% to 100%. To simulate the effect of testing sites in addition to the anterior nares (eg, throat, axilla, perianal swabs), the cost of testing was varied up to $200. In addition, for each simulation run, probabilistic sensitivity analyses simultaneously varied all input parameters over the ranges listed in Table 1.

Routine preoperative MRSA screening of cardiac surgery patients could provide substantial economic value to third-party payers and hospitals under a wide range of circumstances.

Screening Cardiac Surgery Patients for MRSA: An Economic Computer Model

To determine how much should be invested each year to encourage and operationalize the administration of influenza vaccine to children before November and how late the vaccine should be offered each year.

Monte Carlo decision analytic computer simulation models.

The children’s influenza vaccination timing model quantified the incremental economic value of vaccinating a child earlier in the influenza season and the incremental cost of delaying vaccination. The children’s monthly influenza vaccination decision model evaluated the cost-effectiveness of vaccinating versus not vaccinating for every month of the influenza season.

Getting children vaccinated by the end of October rather than when they are currently getting vaccinated could save society between $6.4 million and $9.2 million plus 653 and 926 quality-adjusted life-years (QALYs) and third-party payers between $4.1 million and $6.1 million plus 647 to 942 QALYs each year. Decision makers may want to continue offering influenza vaccination to children at least through the end of December. Vaccinating with trivalent inactivated virus vaccine was more cost-effective than vaccinating with live attenuated influenza vaccine for every month.

Policymakers could invest up to $6 million to $9 million a year to get children vaccinated in September or October without expending any net costs.

	The 2009 influenza A(H1N1) pandemic, as well as past influenza seasons, demonstrated the feasibility of administering seasonal influenza vaccine to children by September or October.

		Getting children vaccinated by the end of October rather than when they are currently getting vaccinated could save society between $6.4 million and $9.2 million plus 653 and 926 quality-adjusted life-years (QALYs) and third-party payers between $4.1 million and $6.1 million plus 647 to 942 QALYs each year.

		Trivalent inactivated virus vaccine (TIV) would be more cost-effective than live attenuated influenza vaccine (LAIV) for every month.

		Administration of TIV would remain cost-effective through December, and administration of LAIV would remain cost-effective through November.

The Advisory Committee on Immunization Practices (ACIP) has identified children as a high-risk group for influenza and recommends that children (aged 6 months to 18 years) receive influenza vaccine each year.

 The 2009 influenza A(H1N1) (H1N1) pandemic highlights the importance of the timing of influenza vaccination as decision makers have endeavored to get vaccines to priority groups as soon as possible. While the presumption is that earlier vaccination is better so that a child is protected for a greater proportion of the influenza season, studies to date have not quantified how the economic value of vaccination may change with the timing of vaccine administration. Moreover, it is unclear how late in the influenza season vaccination still should be offered. Quantifying how the vaccine’s economic value changes with the timing of vaccination may be important for vaccine production, distribution, and administration planning.

The 2009 H1N1 pandemic demonstrated the feasibility of manufacturing, delivering, and administering seasonal influenza vaccine by September or October.

 Manufacturers and the supply chain were able to expeditiously produce and distribute seasonal influenza vaccine, despite manufacturing H1N1 influenza vaccine  concurrently.

have shown that in past years more than 70% of seasonal influenza vaccines have been available for administration by October. With evidence suggesting that earlier vaccination is possible, the next question is how much should be invested to ensure timely vaccination? The answer to this question depends in large part on the economic effect of delaying vaccination. Investment could occur all along the supply chain, from the very beginning (eg, investing to expand vaccine production capacity or to use antigen-sparing technology) to the very end (eg, investing to improve access to  immunization clinics, school-based programs, health fairs, and healthcare provider visits). To address these issues, we constructed the following 2 sets of computer simulation models for children aged 2 to 18 years: (1) the children’s influenza vaccination timing model, to quantify the incremental economic value  of vaccinating a child earlier in the influenza season and the incremental cost of delaying vaccination, and (2) the children’s monthly influenza vaccination decision model, to determine the cost-effectiveness of vaccinating versus not vaccinating a child for every month of the influenza season.

We used TreeAge Pro 2009 (TreeAge Software, Williamstown, MA) to develop 2 models (the children’s influenza vaccination timing model and the children’s monthly influenza vaccination decision model) that each fed into a pediatric influenza outcomes subtree. The models depicted 1 full influenza season, with the baseline scenario assuming a societal perspective (accounting for direct and indirect illness costs), and an additional scenario assuming the third-party  payer perspective (direct illness costs only). Details of the models are as follows.

Children’s Influenza Vaccination Timing Model

The children’s influenza vaccination timing model Monte Carlo decision analytic computer simulation model determined the incremental cost-effectiveness of administering influenza vaccine to a child at different months of the influenza season (September to June) (

). Each vaccinated patient has a probability of experiencing local adverse effects (requiring 1 day of ibuprofen treatment) or systemic adverse effects (requiring 3 days of ibuprofen treatment). Vaccination confers immunity 2 weeks after administration. Vaccinated patients have a decreased probability of contracting influenza if exposed. Exposure to influenza is a function of the time left in the influenza season. Patients who contract influenza then proceed through the pediatric influenza outcomes subtree.

Children’s Monthly Influenza Vaccination Decision Model

The children’s monthly influenza vaccination decision model characterized the decision of whether to vaccinate a patient in a specific given month (

). For example, if a patient is seen in February, is it still cost-effective to vaccinate the patient? Separate models were created for each month of the year from September to June, encompassing the influenza season. Similar to the children’s influenza vaccination timing model, patients who are vaccinated have chances of developing local or systemic adverse effects. Vaccination confers immunity after 2 weeks and reduces the probability of contracting influenza  if exposed. Exposure to influenza is a function of the time left in the influenza season. Patients who contract influenza then proceed through the pediatric influenza outcomes subtree.

For both sets of models, patients who develop influenza have probabilities of self-medicating with over-the-counter medications, visiting an outpatient clinic, or being hospitalized. Hospitalized patients have a probability of not surviving. Vaccination status affects these probabilities.

Trivalent Inactivated (Killed) and Live Attenuated Influenza Virus Vaccines

Different scenarios looked at trivalent inactivated virus vaccine (TIV) and live attenuated influenza vaccine (LAIV). This involved changing the costs and the efficacy variable values to match each vaccine, as summarized in 

The following equation computed the incremental costeffectiveness ratio (ICER) of vaccine administration in a given month versus another comparison month: (Cost of Vaccinating in Baseline Month − Cost of Vaccinating in Comparison Month)/(Effectiveness of Vaccinating in Baseline Month − Effectiveness of Vaccinating in Comparison Month).

To assess if vaccinating is cost-effective, the threshold used in our model was $50,000/quality-adjusted life-year (QALY), which is a frequently cited criterion for cost-effectiveness analysis in the United States.21 Based on nonparametric estimates by Shiroiwa et al,22 this US cost-effectiveness threshold was similar to previously reported thresholds.

National Investment Using the Children’s Influenza Vaccination Timing Model

Using the children’s influenza vaccination timing model, we then constructed a model to project the potential economic value (from the societal and third-party payer perspectives) of moving the current timing of pediatric vaccination to October throughout the United States. The US Census data (

) from 2008 provided the sizes of the different age groups as follows: 2 to 4 years (21,005,852 children), 5 to 9 years (20,065,249 children), 10 to 14 years (20,054,627 children), and 15 to 18 years (21,514,358 children). Vaccination coverage among children was obtained from previous studies23,24 by the Centers for Disease Control and Prevention (CDC).

Table 1 lists the models’ data inputs, their respective distributions, and their sources. The cost of vaccine was the average wholesale price of TIV and LAIV, respectively.

 The cost of over-the-counter medications comprised the average wholesale price of acetaminophen (triangular distribution).

 The cost of hospitalization for influenza was obtained from the National Inpatient Survey of the Healthcare Cost and Utilization Project (HCUP).

 The cost of death was derived from differences in invasive pneumococcal disease costs in adults discharged alive versus those not surviving, which varied from $3700 to $14,000 based on HCUP data.

 The cost of an outpatient visit was obtained from a systemic review article

 regarding the effectiveness of vaccinating healthy children. The cost of vaccine adverse effects consisted of the average wholesale price of ibuprofen, assuming 1 to 3 days of ibuprofen treatment.8 Productivity losses from an outpatient visit comprised 4 hours of lost median wages (incurred by a parent missing 4 hours of work to bring a child to the clinic).

Durations of clinic visits (in hours) and influenza (in days) were obtained from previous studies.

 systematic reviews provided probabilities of influenza and vaccine efficacy (in preventing influenza, hospitalization, and death). Probabilities of various influenza outcomes were obtained from a CDC study.

 Probabilities of vaccine adverse effects were obtained from a 2007 study.

All utility variables drew from triangular distributions and are expressed in QALYs; all other variables drew from normal and beta distributions.

 When possible, data inputs were obtained from published meta-analyses. The CDC27 monthly influenza surveillance data from 2000 to 2008 were used to generate the distribution of monthly influenza risk. Each simulation run randomly drew from 1 of 9 influenza seasons from 2000 to 2008.

A 3% discount rate converted costs and effectiveness measures into 2009 values. Effectiveness was expressed in QALYs. Each patient who experienced neither vaccine adverse effects nor influenza accrued 1.00 QALY.

 The occurrence of vaccine adverse effects, influenza, or hospitalization each caused relative QALY  reductions.

 Death resulted in a QALY loss based on the QALY-adjusted life expectancy of that patient’s age.

Sensitivity analyses varied variable values across the ranges listed in Table 1. We also looked specifically at how the age of the vaccine recipient (2-18 years) affected our results. This involved cohorts of the following age groups: 2 to 4, 5 to 9, 10 to 14, and 15 to 18 years. In addition, specific scenarios examined how much could be invested per child to encourage earlier vaccination with vaccination still remaining cost-effective.

Probabilistic (Monte Carlo) sensitivity analyses examined the effects of varying all variables simultaneously using all distributions in Table 1.

Children’s Influenza Vaccination Timing Model 

Each simulation run comprised sending 5000 simulated patients 1000 times through the model, for a total of 5 million trials. Simulations found September and October to be the optimal months for vaccination, dominating other months (ie, yielding lower costs and greater effectiveness than other months).

For TIV, delaying vaccination until November cost society an additional $0.13 to $0.26 and 0.000016 to 0.000018 QALY per patient and cost third-party payers an additional $0.08 to $0.16 and up to 0.000017 QALY per patient. For LAIV, delays were even more costly for society (an additional $0.18-$0.37 and 0.000024-0.000025 QALY per patient) and for third-party payers (an additional $0.10-$0.27 and 0.000024-0.000026 QALY per patient).

Various scenarios explored the effects of investing different amounts of money per child to facilitate October vaccination. For TIV, October vaccination remained cost-effective up to and including a $0.75 per child investment (the ICER from the societal perspective ranged from $27,901-$36,413 per QALY and from the third-party payer perspective ranged from $33,191-$39,490 per QALY). For LAIV, October vaccination remained cost-effective up to and including a $1.25 per child investment (the ICER from the societal perspective ranged from $36,538-$46,266 per QALY and from the third-party payer perspective ranged from $41,431-$48,282 per QALY). This reinforced that delaying vaccination for LAIV was worse than delaying vaccination for TIV.

 summarize the national investment results from the societal and third-party payer perspectives, respectively. The tables give the number of children who are currently being vaccinated in each month after October and then the potential value of moving all these vaccinations to before November. As summarized in Table 2, the potential total savings is between $6.4 million and $9.2 million from the societal perspective, suggesting that society could invest this much money into vaccinating children before November. In addition, this could save between 653 and 926 QALYs. As summarized in Table 3, the potential total savings for third-party payers is between $4.1 million and $6.1 million, with an accompanying 647 to 942 QALYs saved.

Each simulation run comprised sending 1000 simulated patients 1000 times through the model, for a total of 1 million trials. 

 summarizes the results from these simulation runs for each month. Using the $50,000 per QALY threshold, TIV remained cost-effective through December for society ($33,639-$44,511 per QALY) and for third-party payers ($32,823-$48,847 per QALY). After December, the ICER of vaccination exceeded $50,000 per QALY (the shaded area in Table 4). Results changed with LAIV. Vaccinating patients with LAIV remained cost-effective only through November for society ($28,988-$38,350 per QALY) and for third-party payers ($33,497-$41,038 per QALY). The ICER for LAIV vaccination quickly exceeds $50,000 per QALY in December and onward. By March, no vaccination dominates vaccination for TIV and for LAIV. In other words, vaccination not only costs more but also resulted in decreased effectiveness.

For every month, TIV is more cost-effective than LAIV. For example, in October from the societal perspective, TIV’s ICER ranged from $21,005 to $29,355 per QALY, while LAIV’s ICER ranged from $27,055 to $34,074 per QALY (Table 4). From the third-party payer perspective, this trend held as well: TIV’s ICER in October ranged from $26,507 to $32,327 per QALY, while LAIV’s ICER ranged from $31,387 to $38,337 per QALY.

Our results demonstrate how the economic value of influenza vaccine for children can vary substantially with the timing of vaccine administration and can rapidly decay with delays in vaccination. Earlier vaccination has the following potential advantages:

 A substantial number of influenza cases occur in November and December and can only be prevented if influenza vaccine is administered at least 2 weeks before this time (because influenza vaccine typically takes 2 weeks to achieve seroprotection).

Protecting Against Early Peaking Influenza Seasons.

 Some influenza seasons over the past decade have arrived earlier than usual. This makes earlier protection even more valuable.

 Our study may in some ways underestimate the value of earlier vaccination because it could achieve earlier local herd immunity and reduce local influenza risk, as suggested by findings of previous studies.

 In other words, immunizing a large proportion of children in a school before influenza enters the population may impede the spread of influenza and confer protection to unvaccinated children. Our model did not explicitly  model herd immunity because effects may vary significantly among populations.

 Another potential benefit of earlier vaccination is decreasing the length of time different locations (including clinics)  must store influenza vaccine. Inventory costs are not insignificant, especially in places with limited refrigerator space, and they vary significantly among locations.

Therefore, while efforts have focused on increasing vaccine coverage among children, focusing on earlier vaccination may be worthwhile as well. Although some  recent studies

 have focused on the importance of administering vaccine early during a pandemic, the literature to date is sparse on the issue of seasonal influenza vaccine timing. A study

 in Israel demonstrated that the effectiveness of influenza vaccine decreases with later administration during the influenza season. A multistate case-cohort study

 examined but did not focus on pediatric vaccination timing. The ACIP

 has advocated influenza vaccination in October or November. However, little is known about whether parents understand the effect of seasonal influenza vaccination timing or the extent to which clinics, local health departments, workplaces, and other vaccination locations are encouraging earlier vaccination.

Evidence suggests that earlier vaccination is feasible and would not result in decreased vaccine efficacy.

 Manufacturing, supply chain, and vaccine administration capacity did not prevent the expeditious rollout of seasonal influenza vaccine in August and September 2009. In addition, strain selection ypically occurs in January or February before the influenza season, leaving more than the typical 4 to 6 months required for traditional egg-based vaccine production, which suggests that promoting October vaccination will not adversely affect vaccine efficacy.

 The fact that historically more than 70% of seasonal influenza vaccines are available to be administered by October but more than 50% of children are vaccinated after October implies that the vaccination delays have not been due to inherent production timeline constraints.

 In other words, promoting earlier vaccination may not require earlier development or substantially strain distribution systems.

Decision makers (eg, policymakers, manufacturers, distributors, third-party payers, administrators, clinicians, and school officials) can take various steps to get vaccines to children earlier. They can increase access (eg, making hours and locations more convenient) to vaccination locations (eg, clinics or alternative locations), order vaccines further in advance, and initiate vaccination campaigns sooner. Educational and awareness programs may convince parents  to bring their children to vaccination locations more promptly. Parents and schools may be important targets. Parental availability can affect when their children are vaccinated. Working parents may struggle to find time to bring their children to physician offices. School-based vaccination campaigns could help vaccinate children who would otherwise wait until their next scheduled physician visits.

Investment can come in many forms such as supporting facilities and personnel to expand vaccination availability during off-hours and weekends, increasing refrigerator capacity, and conducting educational campaigns. Another possibility is financial incentives to patients or physicians to vaccinate earlier. However, results of prior studies

 have been equivocal about the potential effectiveness of such incentives.

Vaccinating children earlier (ie, September or October) can provide net economic benefit to society and to third-party payers. Vaccination of children remained cost-effective through December.

Rachel R. Bailey, MPH

Articles

Economic Model for Emergency Use Authorization of Intravenous Peramivir

Economic Model for Emergency Use Authorization of Intravenous Peramivir

Screening Cardiac Surgery Patients for MRSA: An Economic Computer Model

Economics of Influenza Vaccine Administration Timing for Children