Vaccination Parameters

We modeled the effect of introducing the 2-dose rotavirus vaccine Rotarix in the childhood immunization program in Kazakhstan. We implemented vaccination in the model assuming that the vaccine was effective from the first dose at 2 months, similar to other modeling studies (15). We chose a 2-dose rotavirus vaccine versus a 3-dose product because it may be more feasible in practice to achieve high coverage for a vaccine requiring fewer doses. Rotarix demonstrated 96% efficacy against rotavirus gastroenteritis (RVGE) hospitalizations in clinical trials and 90% field effectiveness against hospital admissions in high-income European countries (26). A lower field effectiveness range of 76%–79% was reported from the middle-income countries in Latin America (26–28). Because of lack of clinical trials of rotavirus vaccines in countries like Kazakhstan that are in transitional economies, it is difficult to predict the vaccine performance in these settings. On the basis of the aforementioned findings and our own assumptions, we applied a vaccine efficacy of 80% (range 72%–86%) against severe RVGE and 58% (range 51%–64%) against mild RVGE. The vaccine efficacies were varied by varying the proportions of RVGE infection, and severe RVGE infection in vaccinated children (Technical Appendix). We did not adjust the vaccine efficacy for specific rotavirus genotypes because the strains circulating in Kazakhstan are globally common (29). Pre- and post-licensure data from developing settings indicate that vaccine protection may wane in the second year of life (30,31). We conservatively assumed vaccine protection to be 1 year, commencing after administration of the last dose at 4 months of age; that is, children were assumed to be fully protected on average until 16 months of age. However, studies from industrialized settings demonstrated high vaccine efficacy through 3 years of life (32,33). Therefore, we increased duration of vaccine protection to 2 years in a separate scenario analysis.

The vaccination program in our model was hypothetically initiated on January 1, 2012, with a linear buildup of vaccine coverage during the first 6 months. After this period, the vaccination coverage was assumed to be constant at a fixed level. In Kazakhstan, rotavirus vaccine would be administered concomitantly with the diphtheria–tetanus toxoid–pertussis (DTP) vaccine. The reported coverage for 3 doses of DTP in Kazakhstan is 99% (7), although a recent study suggests that only 76% of children 12–60 months of age receive all 3 doses of the DTP vaccine without delay (34). Considering age restrictions for the administration of rotavirus vaccines, coverage for the rotavirus vaccine may be lower than for other traditional vaccines administered under the World Health Assembly Expanded Programme on Immunization (35) because vaccination may not always be on time. We therefore applied 90% coverage in the base case, but varied coverage between 80% and 100% to explore the effect of this parameter on the cost-effectiveness of vaccination. Similarly to other studies, we did not consider an increased risk for intussusception or any other adverse events after rotavirus vaccination (2).

Disease Outcomes

We calculated the numbers of rotavirus-associated deaths, hospitalizations and outpatient visits in children <5 years from the modeled incidence of severe RVGE (I_s). We assumed that all children with severe RVGE require outpatient care or hospital care, and on the basis of local data, we modeled that 80% of children who were hospitalized with acute diarrhea sought medical care before admission (8). We calculated the numbers of rotavirus homecare episodes (without health care encounters) from the modeled incidence of mild RVGE (I_m), whereas the number of rotavirus-associated deaths and hospitalizations was calibrated to the 2009 estimates of 68 (95% CI 63–74) deaths and 4,007 (95% CI 3,740–4,274) hospitalizations based on a recent study from Kazakhstan (8) (Technical Appendix).

Model Uncertainty and Scenario Analyses

We considered uncertainty related to natural history parameters (Table 1), model calibration, and vaccine efficacy and uptake. The role of adults in rotavirus transmission is a key uncertain factor (15). Because no sentinel data in this age group were available, we varied the infectiousness of later rotavirus infections relative to that of the primary infection between 1/5 and 1/10 (14,15). We also varied the mean duration of complete immunity after rotavirus infection from 6 to 12 months. All models were scored according to how well they fit with the sentinel data, adopting a likelihood-based approach by using the Akaike information criterion. In total, 5 candidate models (Technical Appendix Table 1) had support and were simulated, both with and without vaccination. For each model, we calculated the yearly numbers of avoided health outcomes resulting from incidence difference with and without vaccination implemented (Technical Appendix Table 2).

In the economic analysis, we took a weighted average of the incidence differences using Akaike weights (Technical Appendix). We modeled several different scenarios to account for uncertainty in the calibration process, in vaccine efficacy and vaccine uptake (Table 2). In each scenario, we calculated a weighted model as described above. We analyzed a base case (most likely), best-case, and worst-case scenario to account for uncertainty instead of adopting a probabilistic approach because data on vaccine efficacy and rotavirus-associated health outcomes in Kazakhstan are lacking or sparse. In the base case, we used mean estimates for both vaccine efficacy and calibration values. The best-case scenario was based on the highest vaccine efficacy (86% against severe RVGE and 64% against mild RVGE), in combination with the upper bounds of estimated health outcomes (more events to prevent). The worst-case scenario incorporated the lowest vaccine efficacy (72% against severe RVGE and 51% against mild RVGE) and lower estimates of health outcomes (fewer events to prevent). Vaccine coverage was set to 90% in the base case. Scenarios A and B were constructed as described above, by using coverage of 80% and 100%, respectively (Technical Appendix Tables 3, 4). Finally, in Scenario C we extended the vaccine protection period to 2 years in line with data from industrialized settings demonstrating high vaccine efficacy through 3 years of life (32,33). To estimate indirect or herd protection, we compared predictions of the dynamic model with those of a static cohort model, as was previously suggested (8) (Technical Appendix).

Economic Parameters and Cost-effectiveness Analysis

Direct and indirect costs associated with rotavirus disease were recently estimated in a cost-of-illness study of RVGE in Kazakhstan (8). These costs included direct health care and non–health care costs and indirect costs associated with productivity losses due to the work absenteeism of caregivers and rotavirus-related deaths. For this analysis, cost estimates in 2009 US dollars were inflated to 2012 values by using the consumer price index (Table 3). In the absence of a market price for the rotavirus vaccine in Kazakhstan, we used the 2010 price of pneumococcal vaccine (US $43.00 per dose) purchased by the government. Because the vaccine price is a key determinant of cost-effectiveness, we performed various sensitivity analyses with the price ranging from US $1.00 (assuming program price for traditional Expanded Programme on Immunization vaccines) to US $60.00 per dose (considering a price of pneumococcal vaccine that was the most recent vaccine introduced in the program in Kazakhstan). The program costs included the costs of vaccine doses needed to vaccinate the yearly birth cohorts with 2 doses, a 10% vaccine wastage, and an additional US $267,300 to cover the costs of upgrading the cold chain for rotavirus vaccine in the first year of introduction. A 10% loss from vaccine waste was based on published estimates (36,37). The yearly costs of maintaining the cold chain and the costs of training health personnel were estimated in consultation with the Kazakh Ministry of Health. We assumed that rotavirus vaccination does not incur additional costs to parents because it will be administered concomitantly with other vaccines included in the national immunization program.

We assessed the long-term effect of rotavirus vaccination over a 20-year horizon. All costs and health outcomes were discounted at a rate of 3.0% per year. We conducted cost-effectiveness analyses from the health care system’s perspective (including only direct medical costs) and the societal perspective (including indirect costs) using life-years gained as a measure of benefit. We estimated a break-even price for the rotavirus vaccine, in which the total health care costs of the vaccination program were equal to the expected cost savings for the health care system. All results are expressed as mean values with a range to represent realistic vaccination outcomes given the uncertainty in the epidemiologic model. Lacking actual data on uncertainty in the parameter values, we could not express uncertainty in terms of statistical distributions, so we chose to use 1-way sensitivity analyses.

Base Case

Figure 1

Figure 1. . Projected epidemiologic effect of rotavirus vaccination in children <5 years of age in Kazakhstan. A) Estimated daily incidence of severe RVGE (base case scenario) with introduction of rotavirus vaccination in...

Our model projects that the introduction in Kazakhstan of routine rotavirus immunization with 90% coverage and a mean duration of vaccine-induced protection of 1 year would reduce the incidence of severe and mild RVGE in children <5 years of age within the first year (Figure 1, panels A, B). After ≈4 years of administration of the vaccine program, the dynamics of rotavirus would stabilize, and infection would occur with yearly oscillations. Before the start of vaccination, the peak incidence of RVGE would be among children <12 months of age. The highest incidence of severe and mild disease postvaccination is found during the second and third years of life, respectively (Figure 1, panels C, D). The age shift is predicted to occur within 3 years of vaccination. The yearly peak is predicted to be delayed by 14–20 weeks compared with the epidemic peak timing without vaccination (Technical Appendix Figure 4).

During 20 years of vaccination, the predicted incidence of severe RVGE would be reduced by 74% (base case range 64%–80%) of prevaccine levels. The incidence of mild RVGE would be reduced by 56% (range 45%–64%) compared with incidence among unvaccinated children (Figure 1, panels E, F). Our model predicts substantial indirect or herd protection conferred by rotavirus vaccination. The indirect effects account for ≈40% (range 25–33% in relative terms) of the reduction in the projected incidence of severe RVGE, whereas 60% (range 0.28–0.38 in relative terms) of the incidence drop in mild RVGE would be caused by a reduced circulation of rotavirus. Our model projects that over 20 years, a vaccination program with 90% coverage would prevent 881 (range 776–1,004) deaths, 51,891 (range 46,094–57,971) hospital admissions, 370,268 (range 211,825–541,919) outpatient clinic visits, and 1.345 (range 0.641–2.112) million homecare episodes. These values correspond to ≈74% (range 70%–77%) averted deaths, hospitalizations, and outpatient clinic visits and 55% (range 53%–58%) averted homecare episodes compared with the values predicted without vaccination. In that time period, 54,784 (range 48,304–62,442) undiscounted life-years are saved (Table 4).

Figure 2

Figure 2. . Projected cost (US $) per life-year gained over a 20-year time period (2012–2031) after introduction of rotavirus vaccination in Kazakhstan, according to purchasing price of 1 vaccine dose.

In the base case, the net undiscounted program costs would be US $530.7 million; the net costs when accounting for cost savings would be US $372.8 (range $325.0–$416.4) million (Table 4). These results would imply a cost of US $23,892 per life-year saved (i.e., incremental cost-effectiveness ratio of US $23,892) in a health care perspective (range $20,557–$27,573) and US $18,044 in a societal perspective ($13,854–$22,779); both estimates were discounted at 3%. Figure 2 shows the cost per life-year gained as a function of the vaccine price per dose in a 20-year perspective. At a cost of US $2.78 (range $2.01–$3.62), the additional cost of the vaccination program would be entirely offset by the cost savings to the health care system corresponding to the medical break-even price.

Scenario Analyses

Varying the vaccination coverage between 80% and 100% (scenarios A and B) did not substantially influence the cost-effectiveness ratio (Table 4). For example, increasing the coverage to 100% generated only a moderate decrease in the cost per life-year gained (US $23,658 in a health care perspective and US $17,775 in a societal perspective); a marginal change in the cost-effectiveness ratio was also observed when vaccine coverage was decreased to 80%. We included 6 years without vaccination to explore any carryover effects after vaccination is discontinued. The model predicts that such effects are small because infection rates return to prevaccine levels quickly.

Lastly, in scenario C we simulated a 90% vaccine coverage assuming 2-year mean vaccine protection (Table 5; Technical Appendix Table 5; Technical Appendix Figure 3). These results suggest a cost of US $22,579 per life-year saved in a health care perspective and US $16,775 in a societal perspective; the medical break-even price was estimated at a cost of US $ 2.95 per dose. Compared with the base case (Table 4), assuming 2 years of vaccination protection reduced the cost per life-year saved by 5.5% in a health care perspective and 7.4% in a societal perspective.