Estimating size of at-risk population

To estimate the number of people living in areas with likely JEV transmission we used several different spatial datasets representing factors associated with the maintenance of the zoonotic JEV cycle and subsequent spillover to humans. To identify areas with suitable habitat for JEV transmission we used a fine-scale map of the extent of lowland rice production in Asia modeled from satellite data by the International Rice Research Institute (IRRI) [9] (S2 Fig). This dataset was combined with three spatial datasets of various waterbodies: (1) a map of seasonal and permanent wetlands in the tropics and subtropics at a 231x231 m spatial resolution generated using a hydrological model and satellite-derived estimates of soil moisture to represent water flow and surface wetness [10], (2) HydroLakes, a dataset of all global lakes with a surface area of at least 10 ha [11], and (3) HydroRivers, a dataset containing all global rivers that have a catchment area of at least 10 km² [12]. A combined map of the areas classified as either wetland, lake, river, or under rice cultivation was aggregated at a 1 km² resolution, so that people living within a small buffer adjacent to wetlands or rice cultivation would potentially be considered at-risk (S1 Fig). This map was used as a baseline for suitable JEV transmission. The number of people living in these areas was calculated using maps of human population density at a 1 km² resolution for 2020 from Worldpop [13]. All maps were created using country boundaries from the geoBoundaries Comprehensive Global Administrative Zones (CGAZ) dataset [14].

To further refine our estimates, we incorporated the modeled spatial distribution of the main JEV mosquito vector, Culex tritaeniorhynchus, based on environmental suitability at a 5x5 km resolution [15]. Population-at-risk estimates were calculated for all rice and wetland areas where the probability of occurrence for C. tritaeniorhynchus was ≥ 0.25. This relatively low threshold was chosen to ensure that areas with at least a moderate probability of occurrence were not excluded as potential at-risk areas if all of the other risk factors were present. Although C. tritaeniorhynchus is considered the main vector for JEV, the virus has been isolated from over thirty other mosquito species [5]. In addition, epidemiological data from several locations, such as Bali, Indonesia, suggest that JEV transmission intensity and spillover to humans can be high even where environmental suitability for C. tritaeniorhynchus is predicted to be low. Areas of low environmental suitability for C. tritaeniorhynchus with known circulation of JEV appear to be concentrated in the hotter, tropical regions of countries such as Malaysia and Indonesia. An earlier attempt at modeling the spatial distribution of C. tritaeniorhynchus did predict higher probabilities of occurrence in hotter, wetter tropical areas [16] compared to the updated distribution model of Longbottom et al. [15]. In addition, a recent analysis of several arboviruses transmitted by Culex species, including two flaviviruses, found that the optimal temperatures for viral transmission were between 23 − 26°C [17]. Therefore, to capture areas where the environmental suitability for C. tritaeniorhynchus may be underestimated, or where conditions are ideal for transmission by other mosquito species, we also included areas where the annual mean minimum temperature was ≥ 20°C and annual rainfall exceeded 150 cm, because these thresholds captured the known locations of JEV occurrence in Indonesia, Malaysia, and the Philippines where the predicted environmental suitability for C. tritaeniorhynchus is low (S2 Fig).

Due to the strong association between JEV transmission to humans and the presence of domestic pigs, it is likely that areas with high domestic pig densities are capable of experiencing high transmission intensities. However, relatively high JE incidence rates have been observed in areas of Bangladesh [18] and Indonesia [19] that have few pigs because the local population is majority Muslim. In addition to pigs, domestic fowl (particularly ducks) have been associated with JEV transmission [5] and they can be present in high densities in the regions of Bangladesh, and on the Indonesian islands of Sumatra and Java, where JE cases have been reported despite the lack of local domestic pig populations. To best represent a more conservative estimate of the size of the at-risk population, we calculated the number of people living in areas with the appropriate habitat type (rice and wetland), and suitable environmental conditions for the mosquito vector, where the combined density of pigs and ducks exceeded a certain threshold. Estimated cattle, pig, chicken, and duck population densities at a 1x1 km resolution were taken from the Global Distribution of Livestock dataset [20]. The initial threshold for JE suitability was a combined local density of 2 pigs or ducks per square km, with sensitivity to this threshold value tested by ranging the threshold from 1–10 per sq km (S3 Fig). The size of the population at risk of infection was identified as those living in areas with a combined pig and duck population above the minimum threshold, a land use type of wetland or rice production, and environmental conditions suitable for the mosquito vector. In addition, as part of our sensitivity analysis we used an upper estimate for the size of the population at risk based on only the habitat type (wetland or rice production). Because we do not consider human or mosquito movement in our model, and small water bodies or rice farms are likely missing from our dataset, areas that were not classified as “at-risk” might still experience some JEV transmission. In particular, individuals living immediately adjacent to “at-risk” areas might still experience some risk due to travel to, or mosquito movement from, “at-risk” areas and small pockets of transmission could occur near small rice farms or bodies of water not captured in our dataset. Therefore, rather than there being a strict risk/no-risk dichotomy, areas identified as “at-risk” should be considered areas of higher risk or transmission.

Validation of at-risk areas

To validate our estimates of JEV-risk areas we conducted a search of PubMed for all reports of JE occurrence or incidence between 2000 and 2019. Studies were included in our analysis if they reported the location (typically a village or town) of a suspected or confirmed JE case or death since 2000 (excluding reports that only reported cases or deaths by administrative area or hospital). This search yielded 117 unique locations from 17 published studies (see S4 Table for locations and references). If location coordinates were not provided in the study, latitude and longitude were estimated from the center of the village, town, or city using Google Maps. Because the location coordinates are not precise estimates of where infection occurred we established either a 1.5km or 5km buffer around each location and then estimated what fraction of the area within these buffers were classified as at-risk areas. As a comparison, we randomly sampled 1000 locations 100 times from the countries included in our analysis and calculated the fraction of at-risk area surrounding these locations.

Epidemiological and vaccination data

To estimate the FOI in each JE-endemic country at the national level, we conducted a non-systematic review of publications that reported age-specific JE incidence data. The search was restricted to studies reporting either confirmed or probable JE cases, excluding studies that only reported acute encephalitis cases (AES) because the fraction of AES attributable to JE varies considerably both spatially and temporally [1]. Details of the studies included in our analysis are provided in S1 Table. For countries where at least one study was available prior to the introduction of vaccination, any studies that were conducted after the introduction of vaccination were excluded to avoid having to estimate both FOI and vaccination coverage simultaneously. As a result, we excluded incidence data from 2007 in Nepal [21] and after 2001 from Sarawak, Malaysia [22].

For several countries (China, India, Japan, Malaysia, Sri Lanka, South Korea, Taiwan, Thailand, and Vietnam), incidence data was only available for time periods after the introduction of JEV vaccine into at least part of the study area, so for these countries we simultaneously estimate both FOI and vaccination coverage. Initial estimates of vaccination coverage in these countries were obtained from literature sources identified by Quan et al. [23] and routine vaccination coverage estimates provided by WHO-UNICEF joint reporting [24]. A list of references used to generate preliminary vaccination estimates for each study are provided in S1 Table. Population age distributions for each study were assumed to match the national age distributions in the UN World Population Prospects demographics dataset [25].

Estimating the force of infection

In an unvaccinated population, the FOI can be estimated from either serological or case data using a catalytic model [7]. For a constant FOI (λ), the proportion of the population that remains susceptible at age a is calculated using the catalytic model equation: (1)

Due to the lack of longitudinal studies on JE transmission dynamics, we assumed that λ does not vary with age. While there are likely to be age- and sex-specific risks of exposure to JEV, age differences in FOI rates are likely to be smaller than those observed in many vaccine-preventable childhood diseases where assortative mixing among school-age children leads to significant differences in age-specific FOIs. A constant FOI can be directly estimated from seroprevalence data by assuming that s(a) is equivalent to the fraction of seronegative individuals at age a.

To estimate the FOI from age-structured incidence data we must estimate both the probability that an individual remains uninfected until age a and the probability that an uninfected individual of age a will become infected at that age. For a population that is stratified into M age groups, with age group j spanning in age from to , the proportion of the population that will become infected in age group j, Γ_j, is the integral of the probability of being susceptible at age a multiplied by the FOI: (2)

If a portion of the population has been vaccinated, then the probability that an individual remains susceptible to infection at age a from Eq (1) is modified to: (3) where α is the vaccine efficacy and υ_a is the probability of being vaccinated by age a. This equation assumes that vaccine-derived immunity does not wane over time. The proportion of individuals infected at age a based on λ and the proportion who have not been previously infected or vaccinated is represented as: (4) We assume that the proportion of individuals who are infected in age group j spanning in age from to can then be calculated as (5) where υ_j is the vaccination coverage for age group j.

The FOI was estimated for each study using an MCMC approach. The observed vector of JE cases per age class in year t, I_t, is a vector of length equal to the number of reported age classes, where each element, I_j,t is the number of cases observed in age class j. I_t was modeled as (6) where Ω_t is a vector of probabilities that depends on the proportion of age class j that is infected in year t, Γ_j,t, and the proportion of the study population that is in age class j in year t, θ_j,t, (7)

For studies conducted in populations that have been partially vaccinated, the proportion of age group j that is infected, Γ_j,t, depends on both λ and vaccination coverage rates for age class j in year t, υ_j,t. In addition to estimating λ, we also estimate the vaccination coverage for each age class. Initial values for υ_j,t were obtained from the data sources listed in S1 Table. We assumed that individuals of age a in year t were vaccinated as infants with a probability equal to the reported routine vaccination coverage level () in year t − a. Vaccination campaigns occurring between year t − a and year t were included if an individual was in the target population so that (8) where is the reported coverage level in the target population for campaign x.

Study-specific force of infection (λ) values were estimated by fitting Eqs (2), (6) and (7) in the absence of vaccination and Eqs (4), (5), (6) and (7) in the presence of vaccination. We used a non-negative, truncated Normal(0, 10) prior for λ. To incorporate sufficient uncertainty in the priors for the vaccination coverage parameters, υ_a,t, we used an informative prior following the approach detailed by Quan et al. [23]. The value was the initial estimate of the vaccination coverage in age class a derived from reported vaccination coverage levels at the national level (see S1 Table for data sources). The uncertainty in the vaccination information is represented by setting ϕ = 5. Model fitting was done using a Bayesian framework via a Markov chain Monte Carlo (MCMC) approach implemented in STAN using the ‘rstan’ version 2.18.2 package in R [26]. The models were run with four chains of 10,000 iterations each and a 50% burn-in period. Smaller step sizes for the sampling algorithm were set by increasing the adapt delta parameter from the default of 0.8 to 0.99. In addition, the maximum tree depth was increased from 10 to 15. Convergence was assessed using the Gelman-Rubin convergence diagnostic, R_c [27]. Posterior predictive checks were performed by comparing the empirical data to data simulated from the posterior parameter distributions for each dataset. Posterior predictive data was simulated at each iteration, k, of the MCMC, with . At each iteration, the total observed number of cases, I_t, was compared to the total predicted number of cases, . This test statistic was used to calculate a Bayesian p-value , which indicates whether the distribution of the model-generated data was more extreme than the observed data [27].

Factors associated with the force of infection

Following the estimation of country-specific FOIs from age-specific incidence data, we conducted a regression analysis to determine whether any study-specific or country-specific variables were associated with FOI. The study-specific variables included in the analysis were the start and end years of the study, and whether the study was conducted in an area with a history of JE vaccination. Country-specific variables included the estimated 2020 population size [13], the regional classification and incidence level (very low, low, medium, medium-high, high) from Campbell et al. [1], national per capita gross domestic product (GDP) [28], total land area, the proportion of land area classified as urban [29], the proportion of land area under rice cultivation [9], the mean environmental suitability and the total amount of suitable habitat for C. tritaeniorhynchus [15], and the national population sizes and mean population densities of domesticated cows, pigs, ducks, and chickens [20]. In addition, we also included our baseline and high-end estimates of the percentage of the population living in at-risk areas as potential explanatory variables. We examined the Pearson correlation coefficient between the median FOI estimate and each covariate to identify potential explanatory variables. We then performed model selection via a best subsets regression approach using the ‘leaps’ package in R [30]. The total land area was highly correlated with population size, the population sizes of domestic cows, pigs, ducks, and chickens, the amount of area under rice cultivation, and the total amount of habitat for C. tritaeniorhynchus. Therefore, of these variables only total population size and the size of the pig population were retained as a potential explanatory variable during model selection as these two variables were only moderately correlated (r = 0.60) to each other, but were highly correlated with several of the other variables. We conducted an exhaustive search of all possible models with up to ten explanatory variables and used BIC values to select the best model.

Estimating annual JE burden and vaccine impacts

The annual number of JEV infections for each country was calculated from the posterior FOI estimates from that country and the size of the national at-risk population. For countries where the FOI (λ) had been estimated from multiple studies we combined the posterior distributions of λ from each study into a single pooled distribution. For countries where we were unable to estimate a country-specific FOI we used a pooled distribution of the λ posterior distributions from all 29 studies. To capture variability in the FOI estimates the infection model was run 1000 times with random sampling from the FOI distribution. In the absence of vaccination, the number of infections at age, a, is calculated by multiplying the age-specific probability of infection Γ_a by the number of at-risk individuals in the age class. Estimates of age-specific population sizes were made by assuming that the age distribution of the at-risk population matched the overall national age distribution for the appropriate year from the UN World Population Prospects 2019 demographic estimates [25]. The age-specific population estimates are available yearly, capturing demographic changes over the study time period.

The effects of vaccination on the number of JEV infections can be estimated by adjusting the age-specific probability of infection, Γ_a, to include the vaccination rate at age a, υ_a, as in Eq (4). The efficacy of a single dose of the SA 14–14–2 JEV vaccine is estimated to be 99.3% (94.9–100%) [31]. Although we estimated vaccination coverage for several study sites these estimates were not used to calculate JE burden from 2010–2019. These estimates were for the period of each study, which for some studies only partially overlaps our study period, and don’t necessarily reflect changes in vaccination coverage over the past decade. In addition, the estimates were aggregated into a handful of age classes and don’t capture within age-class variations in coverage levels. For our burden estimation process we instead used annual, age-specific vaccination coverage estimates from the literature (see S2 Table for references) and routine vaccination coverage estimates from WHO-UNICEF joint reporting [24].

Calculating the number of JE cases and deaths from the estimated number of JEV infections requires estimating the proportion of infections that are asymptomatic or result in only mild symptoms. The majority of human JEV infections are asymptomatic, with fewer than 1% of people infected developing clinical disease [32]. The ratio of asymptomatic infections to JE cases has been estimated as 270:1 in children aged 5–9 [33] and 300:1 for people under 40 [34], with other estimates ranging from 50:1 to 1000:1 [5]. We assume that the distribution of encephalitis cases, C_a, observed from I_a JEV infections in age class a can be modeled with a binomial distribution C_a ∼ Binomial(I_a, ρ_c), with a symptomatic probability of ρ_c. The symptomatic probability, ρ_c was estimated by using the ‘optim’ package in R to fit a gamma distribution to the asymptomatic:symptomatic infection ratio (A:S), assuming a median A:S of 300:1 with a 95% CI of 100:1 to 750:1. The estimated gamma distribution had a shape parameter = 3.577 and a rate parameter = 0.0108. This distribution produces an A:S ratio with a median of 295:1 (95% CI: 83:1 to 717:1). The symptomatic probability was then calculated as ρ_c = S/(S + A). The mortality rate for severe encephalitis cases is believed to be between 10–40% [5]. The number of deaths from JE in age class a, D_a, was assumed to follow D_a ∼ Binomial(C_a, ρ_m), with ρ_m ∼ Beta(α = 1.88, β = 3.50). This distribution produces a median mortality estimate of 0.329 (95% CI: 0.050–0.741).

Country-specific estimates of the annual number of infections, cases, and deaths were calculated by running our model 1000 times with random draws from the FOI (λ), symptomatic ratio (ρ_c), and mortality rate (ρ_m) distributions. Model simulations were started in 1950 and run through 2019 to ensure that the model approximated baseline population immunity levels from 2010 to 2019. In order to assess the impact of vaccination these estimates were generated for two scenarios: (1) using the reported vaccination coverage levels for 1950–2019, and (2) a counterfactual scenario with vaccination coverage levels set to 0 for all countries in order to estimate the impact of vaccination. Vaccination impacts were calculated as the number of cases and deaths averted for all ages during the calendars years from 2010 to 2019, as opposed to estimating the impact of vaccination for particular birth cohorts or individuals vaccinated during 2010–2019 over their entire lifetimes.

At-risk population size estimates

Overall, 1,543.1 million (range: 1292.6–2190.9 million) people are estimated to live in an area suitable for endemic JEV transmission (Fig 1). This represents 37.7% (range: 31.6–53.5%) of the 4,095.5 million people living in countries with endemic JEV transmission. China had the largest number of individuals living in a likely at-risk area with 630.3 (range: 625.7–713.5) million, followed by India with 402.1 (range: 220.6–691.4) million and Bangladesh with 130.2 (range: 129.5–153.0) million (Table 1). Bangladesh had the highest percentage of its population living in at-risk areas (78.5%; range: 78.1–92.3%), followed by Vietnam (61.5%; range: 60.8–77.3%). The other countries with greater than 50% of the population living in at-risk areas were Cambodia (52.2%; range: 48.4–74.0%) and Nepal (51.9%; range: 29.5–64.0%). Several countries had fewer than 10% of the population living in likely at risk areas: Timor-Leste (5.5%; range: 4.2–30.5%), Pakistan (7.1%; range: 2.8–38.6%), and Sri Lanka (9.6%; range: 3.4–43.3%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Map of the areas estimated to be suitable for endemic JEV transmission.

Grey represents land areas classified as either wetland or rice cultivation and was used as the upper bound for the extent of the at-risk population. Dark blue represents the lower bound on the extent of the at-risk population and only includes suitable habitat areas that have a combined density of domestic pigs and ducks of at least 10 per square km. Light + dark blue combined represents the default extent of the at-risk population and includes areas with wetland or rice cultivation that also have suitable environmental conditions for the mosquito vector population and a combined density of at least 2 domestic pigs and ducks per square km. Uncolored areas of the map are not considered as at-risk for endemic JEV transmission. The base map layer was generated using the geoBoundaries Comprehensive Global Administrative Zones (CGAZ) dataset available at https://github.com/wmgeolab/geoBoundaries/raw/main/releaseData/CGAZ/geoBoundariesCGAZ_ADM0.zip.

https://doi.org/10.1371/journal.pntd.0009385.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Estimates of the population at risk of JE in millions (percent of total population).

Baseline estimate is for areas with wetlands or rice cultivation, suitable environmental conditions for the vector and at least 2 domestic pigs or ducks per square km. The low end estimate increases the required density of pigs or ducks to 10 per square km and the moderate estimate assumes a threshold of 1 per square km. The high estimate is for all areas classified as wetlands or rice cultivation.

https://doi.org/10.1371/journal.pntd.0009385.t001

Of the 117 locations with published JE reports from 2000–2019, 98 (83.8%) were classified as at-risk within a 1.5km buffer using our default scenario and 110 (94.0%) were classified as at-risk within a 5km buffer (S42–S47 Figs). All 117 locations were classified as at-risk based on our broader estimate of at-risk areas, which only considered the presence of wetlands or rice cultivation. The seven locations that were not classified as at-risk by our baseline model were all adjacent to a water body in Khammouane Province, Laos, but the area was not classified as suitable habitat for the main JEV vector species. However, these JE cases were recorded in 2007–2008 and these villages were impacted by the establishment of a reservoir above the Nam Theun 2 Dam in 2010, so current environmental conditions may differ from those at the time of the study. Within the 1.5km buffered locations the mean fraction of grid cells classified as at-risk was 0.58 (IQR: 0.21–0.94), and within the 5km buffer the mean fraction was 0.58 (IQR: 0.31–0.91). In comparison, only 37.5% (95% CI: 34.6–40.1%) and 49.9 (95% CI: 46.5–52.4%) of random locations were classified as at-risk within a 1.5km or 5km buffer. For these randomly selected locations, the mean fraction of grid cells classified as at-risk was 0.17 (95% CI: 0.17–0.19) within the 1.5km buffer, and 0.17 (95% CI: 0.16–0.19) within the 5km buffer.

Force of infection estimates

The median annual force of infection (FOI) across all studies was 0.098 (95% range: 0.012–0.354). The FOI ranged widely between countries, with median estimates for FOI ranging from a low of 0.011 (95% CrI: 0.004–0.017) in Japan to 0.286 (95% CrI: 0.125–0.437) in Indonesia (Fig 2). For countries with multiple studies, the FOI estimates from different studies often varied considerably resulting in a bimodial or more complex combined FOI distribution. Countries with significant between-study estimates in FOI included China, India, Indonesia, Malaysia, Nepal, and the Philippines (S1 Table). Our model did not generate predicted cases that were more extreme than the numbers of observed cases—as indicated by Bayesian p-values between 0.1 and 0.9—for any of the datasets included in the analysis (S4–S30 Figs). Several potential explanatory variables were weakly associated with our FOI estimates. Higher FOI estimates were associated with larger population sizes (r = 0.29), less urban area (r = -0.27) and a lower per capita GDP (r = -0.27), but also lower historical incidence according to Campbell et al. [1] (r = -0.36). The best-fit model contained only two explanatory variables and had low explanatory power (BIC = 1.11; adjusted R² = 0.21), while the second-best model contained four explanatory variables and had a ΔBIC = 0.78 (BIC = 1.89; adjusted R² = 0.30). Several additional models with two to five explanatory variables had ΔBIC values of <2. The best-fit model included a negative association with the incidence according to Campbell et al., and a negative association with the proportion of area under rice cultivation. The next-best model included the same two covariates, plus a negative association with GDP and a positive association with the mean density of chickens.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Posterior force of infection (FOI) estimates.

Posterior force of infection (FOI) estimates for each country where age-specific incidence data was available. For countries where the FOI was estimated independently from more than one data source, the posterior FOI estimates for each study have been combined into a single distribution.

https://doi.org/10.1371/journal.pntd.0009385.g002

Vaccination coverage estimates

The posterior estimates of vaccination coverage levels from our FOI model often differed considerably from the prior estimates derived from published studies or WHO estimates (S31–S41 Figs). In particular, the estimated vaccination coverage in the oldest age groups (typically ages 60+) was higher than the prior estimate for several studies that included older adults. These differences in vaccination coverage estimates among the oldest age groups could result from prior estimates that were too low because vaccinations that occurred several decades ago were missing from the data sources we used. Our model would also overestimate vaccination coverage in older individuals if transmission intensity was higher in the past. The posterior estimates of vaccination coverage in younger children were lower than our prior estimates for studies conducted in Japan and Malaysia (S35 and S38 Figs). In India, South Korea, and Taiwan vaccination coverage estimates in younger children were higher than posterior estimates (S33–S40 Figs).

Estimates of Japanese encephalitis burden and vaccination impacts

Based on the baseline number of people at risk of infection, the mean number of JE cases decreased from 81,258 (95% CI: 25,437–273,640) in 2010 to 56,847 (95% CI: 18,003–184,525) in 2019 (Fig 3A). The mean number of JE deaths decreased from 29,520 (95% CI: 3,334–112,498) in 2010 to 20,642 (95% CI: 2,252–77,204) in 2019. In the absence of any vaccination we estimate there would have been 99,049 (95% CI: 30,834–326,817) cases and 36,026 (95% CI: 3,295–137,198) deaths in 2019, a small decline from an estimated 101,471 (95% CI: 31,577–334,267) cases and 36,913 (95% CI: 3,998–139,979) deaths in 2010. This small decline in the absence of vaccination resulted from demographic trends in several countries, with declining birth cohort sizes over the decade leading to fewer JEV infections in young children. Between 2010 and 2019, vaccination prevented 314,793 (95% CI: 94,566–1,049,645) cases and 114,946 (95% CI: 11,421–431,224) deaths. India had the largest estimated JE burden in 2019 with 22,219 (95% CI: 7,048–70,933) cases and 8,081 (95% CI: 887–30,432) deaths (Fig 4). The next highest estimated burdens were in Bangladesh and China with 9,983 (95% CI: 3,106–32,841) and 7,692 (95% CI: 1,240–27,797) cases and 3,631 (95% CI: 389–13,916) and 2,754 (95% CI: 189–11,146) deaths respectively (Figs 4 and 5). In the absence of vaccination we estimate that China, rather than India, would have had the highest JE burden in 2019 with 33,925 (95% CI: 10,604–110,749) cases and 12,336 (95% CI: 1,374–46,553) deaths. From 2010–2019, we estimate that vaccination had the largest impact in China, with 204,734 (95% CI: 74,419–664,871) cases and 74,893 (95% CI: 8,989–286,239) deaths prevented. Taiwan and Malaysia had the largest percent reductions in JE burden due to vaccination, with 91.2% and 80.1% reductions in JE cases respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Annual Japanese encephalitis (JE) cases from 2010–2019.

(A) Annual JE cases from 2010 to 2019 with vaccination (blue) or under a counterfactual scenario where vaccination coverage in all countries was zero (red). Estimates based on default at-risk population size. (B) Estimates of annual JE cases from 2010 to 2019 using three different estimates of the total size of the at-risk population. Darker bands represents the interquartile range (IQR) and lighter bands represents the 95% confidence intervals (CIs).

https://doi.org/10.1371/journal.pntd.0009385.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. National-level Japanese encephalitis (JE) cases from 2010–2019.

Annual national-level JE cases from 2010 to 2019 with vaccination (blue) or under a counterfactual scenario where vaccination coverage in all countries was zero (red). Darker bands represents the interquartile range (IQR) and lighter bands represents the 95% confidence intervals (CIs).

https://doi.org/10.1371/journal.pntd.0009385.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. National-level Japanese encephalitis (JE) cases from 2010–2019.

Annual national-level JE cases from 2010 to 2019 with vaccination (blue) or under a counterfactual scenario where vaccination coverage in all countries was zero (red). Darker bands represents the interquartile range (IQR) and lighter bands represents the 95% confidence intervals (CIs).

https://doi.org/10.1371/journal.pntd.0009385.g005

Using the lower estimate for the size of the population at risk of JEV infection leads to an estimated 42,872 (95% CI: 13,344–136,418) cases and 15,552 (95% CI: 1,749–59,232) deaths in 2019 compared to the estimate of 56,847 (95% CI: 18,003–184,525) cases and 20,642 (95% CI: 2,252–77,204) deaths using our default estimate for the size of the at-risk population (Fig 3B). With the highest estimate for the size of the at-risk population, we estimate a mean total of 91,403 (95% CI: 28,989–294,667) cases and 33,201 (95% CI: 3,626–125,682) deaths in 2019. India, Bangladesh, and China were the countries with the first, second, and third highest estimated JE burdens in 2019 using the default and low at-risk population size estimates; however, with the high estimate of the at-risk population size, Indonesia had the second highest burden ahead of Bangladesh and China (S3 Table).

Conclusion

By estimating the FOI experienced per susceptible individual in different JE-endemic countries, we were able to estimate the total number of JE cases and deaths under different vaccination scenarios over the past decade. By coupling these FOI estimates with estimates of the size of the at-risk population from a spatial analysis of the risk factors associated with JEV transmission, we were able to estimate the JE burden even for countries where JE is currently under-reported. Comparing our estimates of at-risk population sizes based on knowledge of the environmental factors associated with JE endemicity to a prior estimate of the population living in JE-endemic regions [1] highlights the importance of refining our understanding of where JE is endemic and spillover to humans occurs regularly. Understanding the sub-national variation in infection risk will be key to targeting vaccination campaigns towards the most vulnerable populations. By modeling current JE incidence against a counterfactual scenario of incidence in the absence of vaccination, we estimated the progress each country has made towards reducing their JE burden through vaccination over the past decade. We have also identified which countries could benefit the most from the introduction or expansion of vaccination coverage.