The methodology of this study has been described elsewhere [33]. In brief, laboratory confirmed RV related pediatric hospitalizations occurring in four participating hospitals (three general hospitals, one tertiary care center) during a five-year period (December 2005 to November 2010) were retrospectively studied. RV underreporting was subsequently assessed by hospital-based active surveillance during the 2011 RV season in the same four hospitals [33].
A comprehensive chart review was performed for each case extracting data on RV disease course, healthcare resource utilization and patient’s medical history to identify conditions potentially associated with increased clinical vulnerability, such as prematurity of <36 weeks gestational age and/or LBW (<2,500 grams) and complex chronic conditions. Complex chronic conditions were those that (1) are expected to last longer than 12 months and (2) involve either several different organ systems or one organ system severely enough to require specialty pediatric care and hospitalization. This classification characterizes a group of patients with increased healthcare needs and mortality [34–36]. We further classified complex chronic conditions into those with a congenital origin (that is, severe congenital pathology) and those with onset later in life.
Prevalence rates for prematurity/LBW and severe congenital pathology were also derived for the Dutch infant population from national disease and birth registries, covering 96% of the Dutch infant population [37, 38]. In addition, a nested case–control study was performed to investigate if the same conditions increased the risk of nosocomial RVGE compared to otherwise healthy age-matched hospital controls [see details in Additional file 1].
Healthcare resource utilization, assessed at the individual patient level, was used for cost calculations, adapting standard cost prices and charges [see Additional file 1: Table S1] [39, 40]. Costs included hospitalization days, preceding emergency department visits, contact isolation precautions [41] and ambulance transportation. For nosocomial RVGE costs for RV related excess hospitalization days were used [33], or, when hospitalization was not prolonged, isolation costs and RVGE related diagnostic and therapeutic costs. This study was approved by institutional review board of the University Medical Centre Utrecht.
The prevalence of prematurity/LBW and severe congenital pathology among RV hospitalizations compared to the general infant population were used to compute Risk Ratios (RR). To account for the clustered study design and for oversampling of tertiary-care hospitalizations compared to their national share in pediatric hospitalizations (20%), weighted prevalence estimates were calculated with variance estimated using Taylor series linearization [42].
Rates of ICU admission and RV-related deaths were compared between RV patients with and without potential high-risk conditions by computing RR. Length of stay or excess hospitalization days in the case of nosocomial RVGE and healthcare costs were compared by t-test, using the arithmetic mean despite the usually skewed distribution. The arithmetic mean is considered most informative in evaluations designed to have an impact on medical policy, because it is the total disease burden that is important [43].
Any of the assessed risk factors (prematurity/LBW or congenital pathology) that were associated with increased risk of RV hospitalization, nosocomial RVGE, RV-related death and/or increased length of stay were included to determine eligibility for targeted vaccination, excluding those who suffered from severe immunodeficiencies, in whom RV vaccination is contra-indicated [44].
Differential RV hospitalization rates were calculated for eligible and ineligible children from RVGE numbers adjusted for underreporting using weighted estimation [42]. Similarly, weighted mean hospitalization costs for community-acquired and nosocomial RV infections were calculated among eligible and ineligible children.
Analyses were performed using R software, version 1.14.1.
We used an age-structured, discrete time-event, stochastic multi-cohort model of the Dutch population, as previously described by Mangen et al. [45] to investigate the cost-effectiveness of adding RV vaccination to the Dutch infant immunization program under two scenarios: (1) universal vaccination and (2) targeted vaccination of high-risk infants.
Strategies were compared assuming an annual birth cohort of 180,000 infants, equivalent to the 2010 Dutch birth cohort. The effect of vaccination was modeled as a reduction in RVGE and associated health outcomes in vaccinated compared to non-vaccinated children between 0 and 15 years old with RV disease risk stratified by age and time since vaccination [see Additional file 1: Figure S1]. Time steps of one month were used for ages 0 to 11 months and of one year thereafter. Effects were modeled over a time-horizon of 20 years with year one being the start of either vaccination program. The model was adapted to simulate targeted vaccination by splitting the population into a vaccination eligible and an ineligible fraction. We assumed no effect on adult RV infections from any of the infant vaccination strategies.
Estimates of RV infection rates, outpatient healthcare visits and related direct and indirect healthcare costs for different age-groups were derived from existing epidemiological sources as previously described by Mangen et al. (Table 1) [45]. Hospitalization rates and costs for children eligible and ineligible for targeted vaccination and for combined groups were derived from our multi-center observational study.
Mortality due to RV was determined by combining data from the multi-center observational study and from another study in a Dutch tertiary-care hospital. In this study RV-related mortality was determined for all children who had died within three weeks of confirmed RV infection between 2000 and 2006. Conservative estimates were used for national mortality figures, assuming that fatal cases exclusively occurred among RV hospitalizations at tertiary-care centers without underreporting.
An expert panel of four pediatricians was consulted to determine years of life lost (YLL) accountable to RV infection among observed fatal cases (Table 1). This approach was used to take into account the reduced life expectancy in children with complex chronic conditions.
We used Quality Adjusted Life Years (QALYs), the product of the health-state utility and the length of time in that state, to weigh losses as a result of RV episodes requiring different levels of healthcare, similar to those used in previous cost-effectiveness analyses [9, 10, 45]. QALY losses due to RV mortality were based on YLL estimates for observed fatal RV cases.
European vaccine efficacy data were used for age-specific vaccination effects (Table 2) [53–56, 63]. Linear waning immunity was assumed for the third, fourth and fifth year post-vaccination, and zero protection thereafter. Rotavirus genotype distribution in the Netherlands is comparable to the observed genotype distribution in European vaccine efficacy trials. Overall, G1P [8] is the dominant strain and G2P [4], G3P[8], G4P [8] and G9P [8] are common co-circulating strains with year-to-year variability in strain distribution [52]. We assumed 88% adherence to vaccination recommendations for both universal and targeted RV vaccination, the observed current vaccine coverage in neighboring Belgium where universal RV vaccination was implemented in 2007 [64], and used coverage rates from 65% to 97% in sensitivity analysis.
Vaccine costs for universal RV vaccination were based on Rozenbaum et al. who assumed that tender processes lower vaccine prices by almost 50% (€75 per vaccine course) compared to the current free market price [10]. Targeted vaccination was assumed to generate price reductions of 25% (€100 per vaccine course). We also included scenarios with the free market price for both vaccination strategies. We assumed vaccine doses would be administered during routine immunization clinic visits at a standard application fee of €6.44 per vaccination [45].
Indirect vaccination effects (herd-immunity) among unvaccinated children were considered as part of the sensitivity analysis (Table 1) [58–60]. No herd-immunity was assumed in the case of targeted vaccination, as vaccine coverage was considered too low for herd-immunity to occur.
Our primary perspective was that of the healthcare provider and a societal perspective was included in sensitivity analysis taking non-healthcare costs into account, updated from Mangen et al. with additional data on parental work loss due to RV hospitalizations in children [65]. All costs were converted to 2011 Euros. A 3% discount rate for costs and benefits was used in base-case scenarios according to World Health Organization (WHO) guidelines [66]. Other discount rates, including those recommended for Dutch health economic evaluations, were used in sensitivity analysis [39]. Although there is no consensus on a cut-off point for good value for resources, we present our results in the context of commonly cited thresholds per QALY of $50,000 equivalent to €35,000 [67, 68]. This amount is approximately equal to the Dutch Gross Domestic Product per capita in 2011, the recommended threshold for highly cost-effective interventions by the WHO [66]. In addition, we used the unofficial threshold of €20,000/QALY commonly applied in the Netherlands for preventive healthcare interventions.
Total net healthcare costs for either vaccination strategy compared to no vaccination are reported (cost-analysis) as well as incremental cost-effectiveness ratios (ICERs), representing costs per QALY gained comparing either strategy with no vaccination. Strategies were considered cost-effective if they generated ICER’s less than a willingness-to-pay threshold of €35,000/QALY from the healthcare provider perspective. An ICER below €20,000/QALY was considered highly cost-effective. As additional scenario analysis, we also calculated incremental costs and QALY’s gained for universal RV vaccination compared to targeted vaccination.
The simulation model was built in Microsoft Excel using add-in software @Risk, version 5.5 (Palisade). Results are presented as means and 95% confidence interval (CI) of simulated results, based on 10,000 iterations. Parameters were varied simultaneously in probabilistic sensitivity analyses, performing random draws from distributions. Distributions were chosen based on parameter characteristics and level of certainty. Input parameters and their distributions with corresponding information source are presented in Table 1. In addition, we performed one-way sensitivity analysis to determine variables which were most influential on model results.
