Dynamic Modeling of Cost-effectiveness of Rotavirus Vaccination, Kazakhstan

Comparison of projected rotavirus vaccine health effects and costs over 20 years showed its use to be cost-effective. Cost-effectiveness of Rotavirus Vaccination Program

the model to the weekly, age-stratified RVGE hospitalizations in children <5 years of age collected from two sentinel hospitals in Kazakhstan from 2007-2009 (6,7).

Model Fitting and Model Selection
We fitted the model output of severe rotavirus gastroenteritis (RVGE) in children < 5 years old to hospital sentinel data from Kazakhstan, assuming that the pediatric hospitalizations represented a time-independent proportion of the modeled severe infections. The sentinel study was conducted during 113 weeks from week 44 in 2007 through week 52 in 2009, and included a total of 1,012 children hospitalized with rotavirus-associated gastroenteritis.
We utilized a maximum likelihood approach, assuming that the hospital admissions in age group and at time point were Poisson-distributed, with a mean equal to the corresponding incidence of severe infections predicted by the model. The likelihood function was given by: (1.1) The parameters to be fitted, , included the proportion of severe RVGE cases represented in the sentinel hospitalization data , an infectivity parameter and seasonal forcing and phase described by: (1.2) We assumed that children <3 years of age had a higher risk of acquiring infection because rotavirus has a fecal-oral transmission route, and young children have a tendency to put fingers and objects in the mouth. This effect was implemented by fitting parameters for higher transmission to 0-7-month-old children, To account for uncertainty in the duration of complete immunity following infection, we fitted the model by varying this time period from 6 to 12 months, and we varied the infectiousness in "later" infections relative to that of the primary infection between 1/5 -1/10. All models were scored using the Akaike Information Criterion (Technical Appendix Table 1): where k is the number of fitted model parameters and is the maximized likelihood.
Five of the candidate models had support , and they were all used to make predictions according to their Akaike weights: The fitted models reproduced well the age distribution of pediatric hospitalizations in the sentinel data (Technical Appendix Figure 2, panel A), although with a slight tendency to both over-and underestimate cases in children < 1 year, and 1-year old children, respectively. The models predicted a seasonal peak in late January (week 4). The bimodal seasonal pattern indicated in the sentinel data (Technical Appendix Figure 2, panel B), which may relate to climatic differences between the central and south region of Kazakhstan where the sentinel hospitals are situated, were not reproduced in the models because seasonal forcing was incorporated using a simple sinusoidal forcing function.

Model Calibration
We calibrated the number of deaths and hospital admissions from the incidence of severe dh pp young children with severe RVGE die and become admitted to a hospital, respectively. All children <5 years of age with severe RVGE were assumed to seek outpatient medical care or be hospitalized or both, and based on local data, we assumed that 80% of hospitalized children sought medical care prior to admission (8). Lastly, we estimated the number of homecare episodes from the predicted incidence of mild infections : ( outcomes from 2012-2031 in the base case (Scenario A) and the sensitivity analyses (Scenario B-C).

Direct Effects
We used the method suggested by Atkins et al. (10,11) to estimate the direct effects of vaccination. In this approach, the direct effects are estimated from the long-term annual number of vaccinated children who would have been infected had vaccination not been implemented. This number is calculated as follows: where is the vaccine coverage, is the vaccine efficacy and denotes the incidence in the absence of vaccination at day . The expression is summed over all age groups between 2 to 16 months of age, assuming that vaccination is effective from the first dose and that vaccination protects against infection for a mean duration of 12 months following the last vaccine dose scheduled at 4 months of age. This way of calculating the direct effects is identical to the vaccine effect predicted by a static cohort model (10,11).

Vaccine efficacy
Vaccine efficacy against any infection was calculated as: where is the sero-conversion rate and is the relative susceptibility against infection in vaccinated children (Technical Appendix  B  low  high  B  low  high  B  low  high  B  low  high  2012  25  23  27  1 483  1 380  1 587  10 584  6 344  14 831  44 897  22 396  67 506  2013  74  68  82  4 375  4 046  4 708  31 216  18 594  44 011  132 145  65 615  199 599  2014 138 127  151  8 135  7 544  8 732  58 049  34 669  81 631 Technical Appendix Figure 3. Epidemiologic impact of rotavirus vaccination in children <5 years in Kazakhstan with a 2-year mean vaccine protection. A) Estimated daily incidence of severe RVGE (base case) with introduction of rotavirus vaccination in January 2012 in the five candidate models; B) Estimated daily incidence of mild RVGE (base case) with introduction of the rotavirus vaccination in January 2012 in the five candidate models using a similar color scheme as shown in legend A); C) Yearly age-specific incidence of severe RVGE pre-vaccination (white) and 10 years post-vaccination (gray); Yearly age-specific incidence of mild RVGE pre-vaccination (white) and 10 years post-vaccination (gray); E) Relative incidence of severe RVGE with vaccination compared with the expected incidence without vaccination; the blue curve shows the mean relative incidence with lower and upper bounds predicted by the synthesis of dynamic models, including both direct and indirect effects, while the red curve shows the relative incidence predicted from a static cohort model incorporating only the direct effects (Supplemental Section S.3); F) Relative incidence of mild RVGE with vaccination compared with the expected incidence without vaccination; the blue curve shows the mean relative incidence with lower and upper bounds in the synthesis of dynamic models, while the red curve shows the relative incidence predicted by a static cohort model.