Cost-effectiveness of a pentavalent rotavirus vaccine in Japan

Abstract

Objective:

To evaluate the impact of universal vaccination with a pentavalent rotavirus vaccine (RV5) on the healthcare burden and costs associated with rotavirus gastroenteritis (RGE) in Japan.

Methods:

The model included a hypothetical cohort of 1,091,156 children followed for their first 5 years of life. In the absence of universal vaccination, there were 19 deaths, 78,000 hospitalizations, and 678,000 outpatient visits due to RGE. The efficacy of RV5 is based on international clinical trial data, which was similar to the efficacy observed in clinical trials conducted in Japan. The primary outcome measure is the cost per quality-adjusted-life-year (QALY) gained. In the base case, the QALY loss per 1000 RGE episodes included 2.2 for children and 1.8 per parent.

Results:

Universal vaccination is projected to reduce hospitalizations by 92%, outpatient visits by 74%, and work-loss days by 73%. For the base case analysis, the total vaccination cost was ¥26 billion. The estimated reduction in medical costs was ¥16 billion. Of 2500 QALYs gained with the vaccination program, approximately half are directly attributed to the child. In the base case analysis, the incremental cost-effectiveness ratio (ICER) for vaccination vs no vaccination is ¥4 million and ¥2 million per quality-adjusted life year (QALY) gained from the healthcare payer and societal perspectives, respectively. The ICERs are ¥8 million and ¥4 million if parental disutilities are excluded.

Key limitation:

The QALY decrements for children and parents were evaluated using different instruments, and the QALY decrements do not vary based on episode severity. Given the interdependence between children and their parents, excluding parental disutilities may under-estimate the impact of RGE.

Conclusion:

Universal vaccination with RV5 in Japan is projected to have a substantial public health impact and may be cost-effective from both the payer and societal perspectives if parental disutilities are included in the cost-effectiveness ratios.

Keywords::

• Cost-effectiveness
• Vaccines
• Rotavirus

Introduction

Rotavirus is the most common cause of severe gastroenteritis in infants and young children worldwide. Prior to the availability of safe and effective vaccines, virtually all children were known to acquire a rotavirus infection by 5 years of age regardless of socioeconomic status1. Rotavirus gastroenteritis (RGE) often results in diarrhea, vomiting, fever, and/or lethargy lasting from a few days up to a couple of weeks. Severe episodes of RGE are more likely to involve a combination of these symptoms which may lead to dehydration if untreated2,3. Rotavirus associated deaths are common in developing countries where access to care is limited4, and are responsible for greater than 450,000 deaths per year in children under the age of 55. Although deaths are rare in industrialized countries, the healthcare burden of RGE-related hospitalizations and emergency department visits is substantial6–8. In addition to the healthcare burden, the impact that RGE has on the family can be considerable9–11.

In Japan, rotavirus is known to be associated with substantial healthcare utilization and medical costs. Prior to the introduction of rotavirus vaccines, RGE was associated with over 50% of all hospitalizations for acute gastroenteritis in children younger than 5 years of age. In this age group, it is estimated that RGE is responsible for 78,000 hospitalizations12 and 800,000 outpatient visits13.

Two rotavirus vaccines are currently available in Japan, the monovalent vaccine (Rotarix, GlaxoSmithKline, Research Triangle Park, SC) and the pentavalent rotavirus vaccine (RotaTeq, Merck Sharp & Dohme Corp, West Point, PA). Both products have been shown in clinical trials and post-vaccine introduction studies to significantly reduce rotavirus disease when widely adopted14–16. Although RGE is not associated with a high rate of mortality in developed countries, universal rotavirus immunization has been shown to be a cost-effective health intervention in many economic analyses due to reductions in healthcare utilization and parental work loss17–28. Rotavirus vaccines have been recommended as part of the infant vaccination schedule in many countries and are recommended by the World Health Organization20,29–34.

Historically, the Japanese government only provided reimbursement for a small number of mandatory vaccines due to concerns about vaccine safety in general35. All other vaccines were regarded as voluntary and, overall, coverage rates were low. Recently, the Vaccination Subcommittee of the Health Sciences Council was formed to re-evaluate the National Immunization Program. Within this past year, seven vaccines have received a recommendation to be added to the list of mandatory vaccines36.

The objective of this study is to describe the projected public health and economic impact of the pentavalent rotavirus vaccine (RV5) in Japan and provide Japanese decision-makers with information to support inclusion of rotavirus vaccination in the National Immunization Program as a mandatory vaccine. Inclusion in the National Immunization Program will result in government funding for rotavirus vaccination and should increase coverage. In the Sato et al.19,37 cost-effectiveness analysis for rotavirus vaccination which was previously published in Japan, the number of hospitalizations estimated was based on a combination of prospective and retrospective data. However, research has shown that retrospective studies under-report the burden of rotavirus because physicians generally do not test for RGE38–40. The Sato et al. study also included QALY decrements derived from the EQ-5D based on a study of physicians in the UK who served as proxies for children with RGE41. Recent research has shown that physicians tend to over-estimate the health status of their patients, and this could lead to an under-estimate of the QALY decrements, particularly for acute conditions42,43. The cost-effectiveness analysis presented here gives more weight to published data based on prospective hospital surveillance studies12, and uses a different approach to determining the decrement in QALYs for episodes of rotavirus gastroenteritis.

Methods

Model structure

A Markov model developed in Microsoft Excel® (Version 2003) tracks two hypothetical birth cohorts with and without rotavirus vaccination during the first 5 years of life because infants may develop multiple rotavirus infections. The model consists of monthly cycles through 6 months before they are fully vaccinated and then quarterly cycles through year 5.

As shown in Figure 1, the model allows for movement between different health states over each Markov cycle. Transition probabilities reflect the likelihood of moving from one health state to another44. The different health states reflect whether the child has ever experienced a rotavirus infection, the number of prior rotavirus infections, the serotypes causing the episodes, and whether the child survives or dies from the illness. The model allows children to contract up to two symptomatic episodes of RGE after which all subsequent infections are assumed to be asymptomatic. This assumption is consistent with a study conducted in Mexico that illustrated the natural history of rotavirus infection in a cohort of 200 children45. This study was unique because study personnel collected stool specimens from all children born in the catchment area during the study period, regardless of symptoms, every 2 weeks for the first 2 years of life. Almost all other studies are conducted in a specific healthcare setting after symptoms develop45,46. The transition probabilities or the likelihood that the child will move from one health state to another, reflect the age-specific probability of developing first and second episodes of rotavirus and the percentage of episodes that are symptomatic.

Figure 1. Description of the RV5 Markov model.

Analytic perspectives and outcome measures

This cost-effectiveness analysis is conducted from the healthcare payer and societal perspectives based on the recommendations of the Panel on Cost-Effectiveness47. The healthcare perspective is limited to direct medical care costs and the cost of the vaccination program. The societal perspective encompasses all RGE-related costs including travel to healthcare facilities and parental work days missed to care for sick children. The outcome measure is the cost per quality adjusted life year (QALY) gained by universal vaccination compared to no vaccination. The benchmark used to determine the cost-effectiveness of universal vaccination is consistent with the recommendations of the Vaccine Evaluation Sub-Committee of the Health Science Council Vaccination Committee in Japan36. Vaccines are considered cost-effective if the ICER is ¥5 million per QALY gained or less from the payer perspective.

Model inputs

General overview

Table 1 lists the key model inputs for the base case scenario along with a range of estimates for one-way sensitivity analyses and the distributions on which the probabilistic sensitivity analysis was based. All medical and non-medical care costs are presented in Japanese Yen (¥) adjusted for inflation to 2011 dollars using the Consumer Price Index for Japan48. All costs and QALYs were discounted at 3% per annum in future years. RV5 is administered as a 3-dose vaccine at 2, 3 and 4 months of age, which is the schedule recommended by the Japan Pediatric Society49.

Table 1. Model inputs used in analysis.

Parameter	Base case value (range for one-way sensitivity analysis)	Distribution/parameter values for PSA^a	Source
Birth cohort	1,091,156	Not included	Vital Statistics for Japan 200848
Resource utilization inputs for unvaccinated cohort
Deaths	19 (9–29)	Beta^b	Base Case: Kawai et al.57, Black et al.56; one-way: same as base case; PSA: Nakagomi et al.12
		r = 244 (# with events)
		n−r = 178 (# without events)
Hospitalizations	78,000 (43,100–96,022)	Beta^b r = 244 (# with events)	Base Case: Nakagomi et al.12; one-way: same as base case; PSA: Nakagomi et al.12
		n−r = 178 (# without events)
Office visits	678,218 (632,870–687,428)	Beta^b	Base Case: Sato et al.19; one-way: same as base case; PSA: Yokoo et al.13
		r = 1684 (# with events)
		n−r = 7672 (# without events)
Average hospital length of stay	5 days	Not included	Base case: Nakagomi et al.12
Distribution of healthcare contacts by age	<1 year: 14% 2 years: 43% 3 years: 18% 4 years: 18% 5 years: 7%	Dirichlet <1 year: 14% 2 years: 43% 3 years: 18% 4 years: 18% 5 years: 7%	Base Case: Yokoo et al.13; PSA: same as base case; used non-informative prior distributions
Vaccine efficacy
Between the first and second dose
Deaths and hospitalizations	88%	Log-Normal^c	Base Case: Dennehy et al.52; PSA: same as base case
		ln(RR) = −2.081
		se ln(RR) = 1.061
Office visits	55%	Log-Normal^c	Base Case: unpublished results from REST; PSA: same as base case
		ln(RR) = −0.630
		se ln(RR) = 0.438
Between the second and third dose
Deaths and hospitalizations	94%	Log-Normal^c	Base Case: Dennehy et al.52; PSA: same as base case
		ln(RR) = −2.888
		se ln(RR) = 0.7264
Office visits	69%	Log-Normal^c	Base Case: unpublished results from REST; PSA: same as base case
		ln(RR) = −2.300
		se(ln(RR) = 0.742
In the first year after the 3rd dose
Deaths and hospitalizations	96%	Log-Normal^c	Base Case: Vesikari et al.14; PSA: Boom et al 54.
		ln(OR) = −2.210
		se(ln(OR) = 0.485
Office visits	86%	Log-Normal^c	Base Case: Vesikari et al.14; PSA: same as base case
		ln(RR) = −1.973
		se(ln(RR) = 0.294
In the 2nd–5th years after the 3rd dose
Deaths and hospitalizations	96%	Log-Normal^c ln(OR) = −2.120	Base Case: Vesikari et al.14 (assumes efficacy for severe disease is maintained after 1st year of vaccination); PSA: Boom et al.54
		se(ln(OR) = 0.485
Office visits	78%	Log-Normal^c ln(RR) = −1.973 se(ln(RR) = 0.294	Base Case: Vesikari et al.14 (assumes a 10% drop in vaccine efficacy after the first year of vaccination); PSA: same as base case
In the first year after the 2nd dose with only two doses
Deaths and hospitalizations	81%	Log-Normal^c	Base Case: Boom et al.54; PSA: same as base case
		ln(OR) = −1.66
		se(ln(OR) = 0.7393
Office visits	69%	Log-Normal^c ln(OR) = −1.66 se(ln(OR) = 0.7393	Base Case; Dennehy et al.52 (assumes same efficacy as between 2nd and 3rd dose with three-dose regimen); PSA: derived from Boom et al.54 data (assumes VE is 10% of VE for hospitalizations and ED visits)
In the 2nd–5th years after the 2nd dose with only two doses
Deaths and hospitalizations	73%	Log-Normal^c ln(OR) = −1.66 se(ln(OR) = 0.7393	Base Case: Boom et al.54 (assumes a 10% drop in vaccine efficacy after the first year of vaccination); PSA: Boom et al.54
Office visits	62%	Log-Normal^c ln(OR) = −1.66 se(ln(OR) = 0.7393	Base Case: Dennehy52 Boom et al.54 (assumes a 10% drop in vaccine efficacy after the first year of vaccination); PSA: derived from Boom et al. data (assumes VE is 10% of VE for hospitalizations and ED visits)
In the first year after the 1st dose with only one dose
Deaths and hospitalizations	69%	Log-Normal^c ln(OR) = −1.171 se(ln(OR) = 0.489	Base Case: Boom et al.54; PSA: same as base case
Office visits	55%	Log-Normal^c ln(OR) = −1.171 se(ln(OR) = 0.489	Base Case; Dennehy et al.52 (Assumes same efficacy as between 1st and 2nd dose with three-dose regimen); PSA: derived from Boom et al.54 data (assumes VE is 10% of VE for hospitalizations and ED visits)
In the 2nd–5th years after the 1st dose with only one dose
Deaths and hospitalizations	62%	Log-Normal^c ln(OR) = −1.171 se(ln(OR) = 0.489	Base Case: Boom et al.54 (assumes a 10% drop in vaccine efficacy after the first year of vaccination); PSA: Boom et al.54
Office visits	49%	Log-Normal^c ln(OR) = −1.171 se(ln(OR) = 0.489	Base Case: Dennehy et al.52, Boom et al.54 (assumes a 10% drop in vaccine efficacy after the first year of vaccination); PSA: derived from Boom data (assumes VE is 10% of VE for hospitalizations and ED visits)
Parental work loss
Days of work loss for episodes involving deaths and hospitalizations	2.5 days	Gamma^d α = 2.77 β = 1.7	Base Case: Kawamura58; Van der Wielen et al.71
Days of work loss for episodes involving outpatient visits	2.5 days	Gamma^d α = 1.028 β = 4.321	Base Case: Kawamura58; Van der Wielen et al.71
Days of work loss for episodes not requiring healthcare	2.5 days	Gamma^d α = 1.028 β = 4.321	Base Case: Kawamura58; Van der Wielen et al.71
Proportion of parents who missed work
Percentage of working parents who missed work for episodes involving deaths/hospitalizations	56%	Empiric^e Range: 39–91%	Base Case: Mast et al.7; PSA: Van der Wielen et al.71
Percentage of working parents who missed work for episodes involving OP visits and no healthcare	21%	Empiric^e Range: 20–64%	Base Case: Mast et al.7; PSA: Van der Wielen et al.71
Vaccine coverage	94% 3 doses, 3% 2 doses, 3% 1 dose	Not included	Base Case: WHO Vaccine Preventable Monitoring System; available at: http://apps.who.int/immunization_monitoring/en/globalsummary/countryprofileselect.cfm
Costs of care and costs of vaccination (in ¥ 2011)
Hospital day	26,923 (21,538–32,307)	Log-Normal ln(µ) = 10.2 ln(se) = 0.13	Base Case: Diagnosis Procedure Combination55; one way: Assumes ±20% of base case estimates; PSA: Sato et al.19
Office visit	13,664 (10,931–16,397)	Log-Normal ln(µ) = 9.522 ln(se) = 0.13	Base Case: Sato et al.19; one way: Assumes ±20% of base case estimates; PSA: Sato et al.19
Day of missed work	12,717	Not included	Base Case: Kawamura58
Direct non-medical care for hospitalizations or deaths	8505	Not included	Base Case: Kawamura58
Direct non-medical outpatient visits and no healthcare	3611	Not included	Base Case: Kawamura58
Price per dose	5700	Not included
Administration fee per dose	2500 (1000–3500)	Not included	Base Case: Chunichi Shimbu, August. 28, 2012, and city of Nagoya http://www.city.nagoya.jp/kenkofukushi/page/0000034344.html60,61.
QALY decrements for
episodes involving deaths and hospitalizations	0.0058 (0.0022–0.0133)	Gamma α = 105.88 β = 0.00006	Base Case: Brisson et al.10; one-way: same as base case: PSA: same as base case
episodes involving OP visits	0.0058 (0.0022–0.0133)	Gamma α = 105.88 β = 0.00006	Base Case: Brisson et al.10; one-way: same as base case; PSA: same as base case
episodes involving no healthcare	0.0058 (0.0022–0.0133)	Gamma α = 105.88 β = 0.00006	Base Case: Brisson et al.10; one-way: same as base case; PSA: same as base case

^aThe choice of probability distribution for each parameter included in the probabilistic sensitivity analysis is in accordance with the recommendations in Briggs et al.63. The probability distributions chosen depend on the constraints imposed by the parameter being estimated as well as the data available from various published studies.

^bThe number of deaths and healthcare contacts by type attributable to rotavirus gastroenteritis (RGE) is often estimated as a percentage of all acute gastroenteritis (AGE) events. The Beta distribution was used to vary the proportion of all AGE events due to RGE.

^cThe log-normal distribution was used to vary the risk ratios which were then converted to relative rate reductions using the formula Relative Rate Reduction = 1 – Risk Ratio.

^dThe Gamma distribution is often used to represent the uncertainty in count (Poisson) data. In the absence of information specific to Japan, we used estimates of the mean and variance from the REVEAL study71 to vary the number of parental work days missed based on the type of healthcare utilization required.

^eIn the absence of information specific to Japan as well as the mean and variance of the proportion missing work we randomly sampled from an empirical distribution based on the REVEAL study71.

The size of the birth cohort in Japan, based on recent census data, is 1,091,15650. Rotavirus incidence is assumed to vary by age. Age-specific incidence of all infections (both symptomatic and asymptomatic) is calculated based on the observed age distribution of outpatient visits13, with the majority of cases occurring during the first 2 years of life. Secondary infection rates are calculated assuming that 80% of those who get a primary infection are at risk of getting a second infection45. The model then estimates the percentage of all infections within each age group that are symptomatic.

Vaccine efficacy estimates used in this analysis are from published clinical trial data, the Rotavirus Efficacy and Safety Trial (REST), which includes an evaluation of vaccine efficacy based on healthcare utilization end-points14. REST was conducted in 11 countries and enrolled over 70,000 infants who were followed through the first 2 years of life. Although Japan was not one of the countries included in REST, the efficacy estimates for RV5 from a recent clinical trial conducted in Japan were similar to REST51. Since the Japanese study did not include healthcare resource utilization end-points, data from REST was used to populate the model.

Since there are no efficacy data against death, the effect of vaccination on hospitalizations is also applied to RGE deaths. The efficacy prior to completion of the 3rd dose shown in Table 1 is consistent with post-hoc analyses of the clinical trial data from REST52. Given that there is very little data from the trial regarding the efficacy against RGE for those receiving only one or two doses, efficacy assumptions for hospitalizations and deaths for an incomplete series are based on a recent non-randomized study53,54. Efficacy for genotypes other than G1–G4 and G9 is assumed to be 50% of the efficacy for the G1–G4 and G9 genotypes.

Base case scenario for a single cohort of children born in Japan

For the base case scenario we rely on a prospective surveillance study conducted by Nakagomi et al.12 for the number of rotavirus hospitalizations as well as the cost of a hospitalization for children under five in Japan55. The number of deaths is based on the number of deaths due to acute gastroenteritis and the proportion of those events that are due to rotavirus56,57. The number of outpatient visits and the cost of a visit are consistent with Sato et al.19. Finally, the non-medical and indirect costs including the number of work loss days associated with episodes of rotavirus gastroenteritis are derived from a study conducted by Kawamura58. The percentage of all RGE episodes that are caused by G1, G2, G3, G4, and G9 rotavirus genotypes is assumed to be 91.4%59. Vaccination coverage for infants in the vaccinated cohort is assumed to follow coverage rates consistent with published rates of DTP vaccination (94% for all three doses, 3% for two doses, and 3% for one dose).

The QALY data is based on a Canadian study where the QALY decrements for RGE were based on the Health Utilities Index 2 for children and the EuroQol 5D for parents10,11. RGE mainly affected two domains in children—emotion and pain/discomfort. The main EQ-5D health domains affected in parents of children with RGE were usual activities, pain/discomfort, and anxiety/depression. The QALY lost per 1000 cases of RGE in children ≤ 3 years of age was estimated to be 2.2 based on the HUI2 and 1.8 for each parent based on the EQ5D. In total there were 5.8 QALYs lost per 1000 RGE cases assumed in the base case scenario or 0.0058 for each episode of rotavirus gastroenteritis or ∼2.2 days on an annual basis10. The QALYs lost due to the morbidity associated with RGE were then added to the life years lost associated with each RGE death estimated in the base case scenario.

The total cost of the 3-dose RV5 series is based on the catalogue price (¥5700 per dose) plus administration fee (¥2500 per dose) for a total series cost of ¥24,60060,61.

Analyses based on children born in five hypothetical municipalities

A set of analyses were also run to examine the public health and economic impact of RV5 on five hypothetical municipalities in Japan. Birth cohorts of 1000, 3000, 5000, 10,000, and 20,000 were evaluated.

Sensitivity analyses

Various one-way and probabilistic sensitivity analyses (PSA) were performed to check the robustness of model results. One-way sensitivity analyses were performed for: frequency and cost of RGE hospitalizations and outpatient visits, number of RGE deaths, vaccine administration fee, and QALY loss assumptions because these parameters were likely to be the most influential. We also conducted a sensitivity analysis to examine the impact of adding minor adverse events as reported in REST. The only statistically significant difference in adverse events between placebo recipients and vaccinees was a slightly higher rate of diarrhea and vomiting in the first week after the first dose62.

The PSA included efficacy of the three-dose series, efficacy of incomplete regimens, number of deaths, hospitalizations, and outpatient visits attributable to RGE, mean number of work loss days per RGE episode, proportion of working parents who miss work, the QALY loss per RGE episode, medical care costs, and the age-specific distribution of healthcare contacts by type. Table 1 provides detail on the data sources and the distributions assumed for each of the parameters included in the PSA. The model was run 1000 times, and, for each simulation, each PSA variable was set to a different value drawn from its assigned probability distribution. Probability distributions were chosen according to recommendations from a previous study that indicated that different distributions are most appropriate depending on the type of data from which the probabilities are derived63.

Results

Base case scenario for a single cohort of children born in Japan

The model predicts that in the absence of vaccination there would be 606,285 symptomatic episodes of RGE in a single birth cohort over 5 years with rotavirus. Universal vaccination would avert 71,985 hospitalizations, 503,933 outpatient visits, and 281,412 parental work loss days. This results in reductions of 92%, 74%, and 73% for hospitalizations, outpatient visits, and parental work loss days, respectively.

In the absence of vaccination, RGE-related costs are expected to be ¥26,043,595,288, including ¥18,860,561,942 in direct medical costs, of which 53.1% are hospitalization costs (Table 2). If a universal vaccination program were implemented, the ¥26,037,164,472 in vaccination costs would be offset by lower costs to treat rotavirus, which would be ¥2,857,625,197 in direct medical costs or ¥5,043,846,303 if indirect costs are included.

Table 2. Public health and economic impact of a universal vaccination program vs no vaccination program for the base case scenario.

	No vaccination program	Universal vaccination program	Net difference	Percentage change
Events
Deaths	19	1.5	−17.5	−92.1
Hospitalizations	78,000	6015	−71,985	−92.3
Outpatient visits	678,218	174,285	−503,933	−74.3
Work loss days	386,566	105,154	−281,412	−72.8
Rotavirus cases	606,285	250,268	−356,016	−58.7
Costs
Medical care costs	18,860,561,942	2,857,625,197	−16,002,936,745	−84.8%
Hospitalizations	10,018,388,395	599,959,427	−9,418,428,968	−94.0%
Outpatient visits	8,842,173,546	2,257,665,769	−6,584,507,777	−74.5%
Direct non-medical costs	2,481,374,583	898,370,476	−1,583,004,108	−63.8%
Indirect (work days lost)	4,701,658,763	1,287,850,630	−3,413,808,133	−72.6%
Societal costs	26,043,595,288	5,043,846,303	−20,999,748,985	−80.6%
Cost of vaccination	0	26,037,164,472	26,037,164,472	100%
Net direct medical care costs	18,860,561,942	28,894,789,669	10,034,227,727	53.2%
Net societal costs	26,043,595,288	31,081,010,775	5,037,415,481	19.3%
Quality-Adjusted Life Years (QALYs)	3925	1425	−2500	−63.7%
Children with rotavirus	1836	567	−1269	−69.1%
Parents of children with rotavirus	2089	858	−1231	−58.9%

Of the 2500 QALYs gained with the vaccination program, 1269 are directly attributed to the effects on the child including 516 life years lost due to the 17 RGE deaths avoided. The ICERs would be ¥4,014,001 and ¥2,015,122 per QALY gained from the payer and societal perspectives, respectively (Table 3).

Table 3. Cost-effectiveness results for universal vaccination vs no vaccination in a single birth cohort in Japan from birth through 5 years of age (in ¥).

Cost-effectiveness outcome	Payer perspective	Societal perspective
Cost per case avoided	28,185	14,149
Cost per hospitalization avoided	139,394	69,979
Cost per QALY gained^a	7,907,648	3,969,823
Cost per QALY gained^b	4,014,001	2,015,122

^aOnly includes the QALYs reported for the children with rotavirus derived from the HUI2.

^bIncludes the QALYs reported for the children with rotavirus as well as their parents.

Analyses based on children born in five hypothetical municipalities

The results that show the public health and economic impact from the perspective of five municipalities with different birth cohort sizes are shown in Table 4. The table provides the number of hospital admissions and outpatient visits as well as direct medical and societal costs for the vaccination and no vaccination strategies.

Table 4. Public health and economic impact of universal RV5 for five birth cohorts in municipalities of varying sizes.

	Events		Costs
Birth cohort size	Hospitalizations	Outpatient visits	Net direct medical care costs	Net societal costs
1000
No vaccination program	71	622	17,284,936	23,867,894
Universal vaccination program	6	160	26,480,897	28,484,480
Net difference	66	462	9,195,961	4,616,586
3000
No vaccination program	214	1865	51,854,809	71,603,681
Universal vaccination program	17	479	79,442,691	85,453,439
Net difference	197	1386	27,587,882	13,849,758
5000
No vaccination program	357	3108	86,242,681	119,339,468
Universal vaccination program	27	799	132,404,485	142,422,398
Net difference	330	2309	45,979,804	23,082,930
10,000
No vaccination program	715	6216	172,849,363	238,678,936
Universal vaccination program	55	1597	264,808,970	284,844,796
Net difference	660	4619	91,959,607	46,165,860
20,000
No vaccination program	1430	12,431	391,180,299	477,357,872
Universal vaccination pogram	110	3195	529,617,940	569,689,591
Net difference	1320	9236	183,919,215	92,331,720

Sensitivity analyses

Of the one-way sensitivity analyses explored, the model was found to be most sensitive to QALY loss per episode followed by vaccine administration fee, number of hospitalizations, and hospitalization cost. The model was least sensitive to number of outpatient visits, outpatient visit costs, and number of deaths. The results of the one-way sensitivity analyses from the medical care and societal perspectives are shown in Figure 2(a and b).

Figure 2. Results of one-way sensitivity analysis: impact of variations in key parameters on ICER, (a) payer perspective; (b) societal perspective: base case value (range).

If we conservatively assume the same disutility associated with an episode of rotavirus gastroenteritis for the adverse events involving diarrhea or vomiting, the ICERs would increase by ∼¥136,000 and ¥68,000 from the healthcare or payer and societal perspectives, respectively.

If all of the base case assumptions are maintained but parental disutilities are excluded from the QALY calculation, the ICERs would be ¥7,907,648 and ¥3,969,823 from the payer and societal perspectives, respectively.

The PSA results are shown in Figure 3. Each point in the figure represents incremental costs and QALYS for vaccination vs no vaccination from the societal perspective and the diagonal line indicates the ICER threshold of ¥5 million per QALY gained. Of the 1000 PSA simulations, 99% fall below the 5 million per QALY gained line.

Figure 3. Results of probabilistic sensitivity analysis, cost effectiveness plane, societal perspective. The solid line from the origin represents the cost-effectiveness threshold in Japan ¥5 million yen and each point represents the ICER results from each of the simulations in the PSA.

Discussion

The results of this study indicate that RV5 has the potential to be a cost-effective public health intervention in Japan, when evaluated from both the societal and payer perspectives. Although there is no formal published threshold in Japan at which health interventions are assumed to be cost-effective, a recent report has indicated that interventions with ICERs below ¥5 million would be considered cost-effective from the healthcare payer perspective36.

The economic impact of rotavirus vaccines in Japan has also been described in a previous study19. This analysis estimated that implementation of rotavirus vaccines on a national level would be considered cost-effective in Japan at ¥9 million and ¥900,000 per QALY gained from the payer and societal perspectives, respectively.

There are several model assumptions and data inputs that differ between the two studies. Our model includes inputs for rotavirus-associated hospitalizations generated from a prospective study12, while the previous analysis used a combination of hospitalization estimates from prospective and retrospective studies12,37. A recent comparison of RGE surveillance data from prospective and retrospective studies indicated that, in the US and Hong Kong, rotavirus healthcare utilization estimates generated from studies utilizing retrospective study designs may under-estimate the true burden of rotavirus disease38,39. Prospective study designs that are able to capture and test all cases of acute gastroenteritis may provide a more complete understanding of the burden of RGE. This difference in hospitalization estimates accounts for part of the difference in results between our study and the previous cost-effectiveness study.

Sato et al.19 also used a different source for the QALY decrement assumptions. The QALY decrements were derived from the EQ-5D based on a study of physicians in the UK who served as proxies for the children, but physicians often over-estimate the health status of their patients. Use of physician judgement as a proxy for patient preferences may, therefore, under-estimate the decrement in QALYs42,43. Recent studies have recognized the problems associated with applying the existing guidelines from the Panel on Cost-Effectiveness for pediatric studies64–66. Generating health state preferences and subsequent QALY decrements for temporary health states in children is quite complex and there is not currently a single accepted methodology. Although the use of proxies may potentially introduce bias into the analyses, they are inevitable when young infants are involved. Some researchers have recommended a family model to elicit preferences for conditions affecting very young children in view of the interdependence between young children and their parents67–69. We believe the approach taken in this study is consistent with these recent recommendations. Although the QALY decrements for the children and their parents were collected separately using two different instruments, the exclusion of parental disutilities is likely to under-estimate the impact of rotavirus gastroenteritis. The estimates used in our analysis are based on prospective surveys of parents and caregivers of children with RGE over a 2-week period, so they may more accurately reflect the quality-of-life impact on the child and family than the physician-proxy survey. We chose a mid-range value for QALY loss per RGE episode and varied the QALY loss assumptions in sensitivity analysis over the range cited in the Brisson et al.10 publication to account for the different elicitation instruments and inclusion of QALY loss for caregivers. As Brisson et al. noted, the QALY decrements derived for rotavirus are reasonable given the QALY decrements reported for other childhood preventable vaccines.

Several other parameters differed between the two models, which may impact differences in results including percentage of working parents projected to miss work, and costs associated with parental work loss. Vaccine cost in our model is greater than the previous model. Healthcare cost inputs for our model and the previous publication, as well as vaccine efficacy estimates, are similar.

The generation of a probabilistic sensitivity analysis is a significant contribution of this analysis that was not completed as part of the previous published analysis. The probabilistic sensitivity analysis allows for the assessment of the impact of parameter uncertainty on model results. The PSA results indicate that, despite significant uncertainty in some parameter estimates such as QALY decrements, the qualitative conclusions that may be drawn from this analysis are robust to model assumptions. RV5 would be considered a cost-effective health intervention under a wide range of model assumptions.

This analysis indicates that universal vaccination with RV5 is projected to have a substantial impact on RGE burden of disease and would be considered a cost-effective healthcare intervention from the payer perspective, as well as the societal perspective, but the economic benefits extend beyond medical care costs saved. When evaluated from the societal perspective, it is projected that parental work loss associated with RGE episodes would also be substantially reduced by universal vaccination. The ICER values generated in this analysis are well within the range of other recently adopted childhood vaccines in Japan70. Exclusion of parental disutilities is likely to under-estimate the impact of rotavirus and decrease the estimated value of a vaccination program. In order to reduce the substantial burden of rotavirus-related disease in Japan, efforts to improve rotavirus vaccine coverage and reduce financial barriers should be considered and implemented in the near future.

Transparency

Declaration of funding

This study was funded by Merck Sharp & Dohme Corp, a subsidiary of Merck & Co., Inc.

Declaration of financial/other relationships

All authors have disclosed that they are employed by and own stock or stock options in Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Whitehouse Station, NJ, USA. JME Peer Reviewers on this manuscript have no relevant financial relationships to disclose.