Cost-Effectiveness and Public Health Effect of Influenza Vaccine Strategies for U.S. Elderly Adults

Jonathan M. Raviotta, MPH; Kenneth J. Smith, MD, MS; Jay DePasse, BS; Shawn T. Brown, PhD; Eunha Shim, PhD; Mary Patricia Nowalk, PhD; Richard K. Zimmerman, MD, MPH

Disclosures

J Am Geriatr Soc. 2016;64(10):2126-2131. 

In This Article

Methods

The cost-effectiveness and expected population outcomes of influenza vaccination strategies for U.S. adults aged 65 and older were estimated using a Markov state transition model. The model simulated identical hypothetical cohorts moving through influenza vaccination and infection health states over the course of a single influenza season. In the primary analysis, four strategies were compared: no vaccine, IIV3, IIV4, and high-dose IIV3. Quality-adjusted life years (QALYs) were used to measure duration and quality of life in cost-effectiveness comparisons. The analysis took a societal perspective, following reference case recommendations of the Panel on Cost-Effectiveness in Health and Medicine.[6] Consideration of the adjuvanted IIV3 in hypothetical scenarios was added in a secondary analysis because of uncertainties regarding its comparative effectiveness and its ultimate U.S. price.[5]

The Markov model began with all individuals unvaccinated in August and terminated in May after 10 monthly iterations over a single influenza season. Vaccination likelihood according to month was derived from the observed U.S. monthly influenza vaccination average frequencies,[7] which were 0 or greater for all 10 months and were the same for each vaccine type. The monthly likelihood of influenza infection was computed from the reported incidence from the 2009/10 through 2013/14 influenza seasons to approximate influenza frequency in unvaccinated individuals. Vaccinated and unvaccinated individuals were susceptible to influenza infection. For those who were vaccinated, influenza risk was set based on the effectiveness of the administered vaccine.[7] Greater effectiveness of IIV4 than of IIV3 was modeled as a relative increase in effectiveness, based on the average likelihood of uncovered influenza B from 1999/2000 through 2013/14.[8] Greater effectiveness of high-dose IIV3 than of IIV3 was calculated as an increase in relative effectiveness (0.242), based on randomized trial data.[3] It was assumed that vaccine-related adverse event risk was the same for all vaccine types, using available data for IIV3 adverse event incidence because data on the other vaccines was sparse.[3] Three outcomes from influenza infection were modeled: recovery, hospitalization and recovery, and death. The model excluded all other causes of death, assuming that vaccination strategy does not affect mortality from other causes, perhaps biasing against vaccination. Recovered individuals remained immune through the remainder of the influenza season.

Modeling identical cohorts assumes that all noninfluenza events will occur identically between modeled strategies and that risk associated with those events is homogenous within the cohort. Therefore, only events affected by vaccine selection will lead to differences among vaccine strategies. Differences in vaccine effectiveness that might occur because of comorbidities or immunosuppressive treatment within portions of the cohort were examined through varying vaccine effectiveness for the entire cohort over plausible ranges in sensitivity analyses.

Model parameters are shown in Table 1 . When possible, costs, utilities, and probabilities were selected from the most current and robust data sources, as noted. Vaccination coverage and influenza incidence were estimated as cumulative monthly likelihoods. These values were calculated by first computing the average monthly proportions of vaccine uptake and influenza incidence across the prior five influenza seasons and then defining a function to output a standardized cumulative monthly value for these parameters.[7] These functions were applied to the overall seasonal probabilities of vaccine receipt and of contracting influenza to produce values for each monthly Markov cycle. The model assumed that only one vaccine type was available in each strategy, which excludes patient or physician preference as a factor, and used identical values for vaccination uptake for all strategies. The model did not account for possible indirect (herd immunity) effects, which are expected to be low when elderly adults are vaccinated but, if present, could bias the analysis against vaccination strategies.

Influenza illness event probabilities were derived from estimates of U.S. influenza complications and mortality,[9,10] and a probability of nonhospitalized individuals with influenza seeking outpatient care was computed as a function of published age-specific influenza complication rates.[10] It was assumed that 100% of those seeking outpatient care would receive antivirals, and this value was varied widely in sensitivity analyses.[11] Age-specific influenza complication rates informed the estimation of time spent seeking or receiving care.[10] In those who sought outpatient care and were not hospitalized, one physician visit was assumed to have occurred.

Hospitalization due to influenza was modeled as a case-hospitalization rate, identical for all vaccines, and assumes that vaccination only effects influenza case rates, and vaccination-specific effects on influenza hospitalization do not affect hospitalization rates. This assumption is consistent with clinical trial and epidemiological data.[3,12] These influenza case-hospitalization assumptions were tested in sensitivity analyses by varying case-hospitalization rates simultaneously for all vaccines over broad ranges to include rates from recent data[3,12] and by examining differential vaccine protection from hospitalization.

In the cost-effectiveness calculation, effectiveness was tracked as a disutility value, representing the lost quality and duration of life from influenza vaccination and illness events.[10] QALYs lost because of influenza mortality were discounted at 3% per year. All vaccine-related adverse events were assumed to be minor and to result in a maximum of 1 day of lost quality of life.

Costs were obtained from relevant medical economic databases and scientific literature sources ( Table 1 ). Vaccine prices were determined according to the private sector costs of the 2014 Centers for Disease Control and Prevention (CDC) Adult Influenza Price List,[13] except for high-dose IIV3, which was estimated using its average wholesale price.[14] Oseltamivir was the selected antiviral treatment for individuals seeking medical treatment,[11] and the cost of a 40-count bottle of 200-mg ibuprofen tablets was assessed for each nonhospitalized vaccine adverse event.[15] All costs were inflated to 2014 levels based on the U.S. Consumer Price Index.

The expected population outcomes of each immunization strategy were computed for influenza cases, influenza-related deaths, and influenza-related hospitalizations, as calculated by the decision model, based on event frequency in the modeled cohort when base-case parameter values ( Table 1 ) were used. These cohort-based per-person event probabilities were then multiplied by the U.S. Census estimated 2013 U.S. population aged 65 and older[16] to obtain event frequency values.

In sensitivity analyses, each parameter was individually varied (one-way sensitivity analysis) across its listed range ( Table 1 ). Then all parameters were simultaneously varied in a probabilistic sensitivity analysis, in which values were randomly sampled, once per model iteration, from parameter-specific distributions for 5,000 iterations. Distributions were assigned to parameters based on data characteristics, parameter uncertainty, and methodological standards.[17] Probabilities, utilities, and vaccine effectiveness values were assigned beta distributions, costs were assigned gamma distributions, and counts used Poisson distributions. Exceptions include QALYs lost because of influenza death and days of influenza-related hospitalization, which used gamma distributions. Probabilistic sensitivity analysis results are presented using the commonly cited benchmark of $100,000 per QALY gained.[18,19]

processing....