The Effectiveness of National Expanded Program on Immunization With Hepatitis A Vaccines in the Chinese Mainland: Interrupted Time-Series Analysis

Background The high prevalence of hepatitis A delivered a blow to public health decades ago. The World Health Organization (WHO) set a goal to eliminate viral hepatitis including hepatitis A by 2030. In 2008, hepatitis A vaccines were integrated into the Expanded Program on Immunization (EPI) in China to alleviate the burden of hepatitis A, although the effectiveness of the EPI has not been well investigated. Objective We aimed to evaluate the intervention effect at both provincial and national levels on the incidence of hepatitis A in the Chinese mainland from 2005 to 2019. Methods Based on the monthly reported number of hepatitis A cases from 2005 to 2019 in each provincial-level administrative division, we adopted generalized additive models with an interrupted time-series design to estimate province-specific effects of the EPI on the incidence of hepatitis A among the target population (children aged 2-9 years) from 2005 to 2019. We then pooled province-specific effect estimates using random-effects meta-analyses. We also assessed the effect among the nontarget population and the whole population. Results A total of 98,275 hepatitis A cases among children aged 2-9 years were reported in the Chinese mainland from 2005 to 2019, with an average annual incidence of 5.33 cases per 100,000 persons. Nationally, the EPI decreased the hepatitis A incidence by 80.77% (excess risk [ER] –80.77%, 95% CI –85.86% to –72.92%) during the study period, guarding an annual average of 28.52 (95% empirical CI [eCI] 27.37-29.00) cases per 100,000 persons among the target children against hepatitis A. Western China saw a more significant effect of the EPI on the decrease in the incidence of hepatitis A among the target children. A greater number of target children were protected from onset in Northwest and Southwest China, with an excess incidence rate of –129.72 (95% eCI –135.67 to –117.86) and –66.61 (95% eCI –67.63 to –64.22) cases per 100,000 persons on average, respectively. Intervention effects among nontarget (ER –32.88%, 95% CI –39.76% to –25.21%) and whole populations (ER –31.97%, 95% CI –39.61% to –23.37%) were relatively small. Conclusions The EPI has presented a lasting positive effect on the containment of hepatitis A in the target population in China. The EPI’s effect on the target children also provided a degree of indirect protection for unvaccinated individuals. The continuous surveillance of hepatitis A and the maintenance of mass vaccination should shore up the accomplishment in the decline of hepatitis A incidence to ultimately achieve the goal set by the WHO.


Provincial-level-specific model
We used a time-series quasi-Poisson regression to examine the province-specific effectiveness of the Expanded Program on Immunization (EPI).The formula of the model is as shown: where denotes the reported monthly number of hepatitis A cases in the PLADs i and the tth month under study (i = 1, 3,..., 31; t = 0, 1, 2, 3,..., 179); the logarithm of population was used as an offset; is a dummy variable with 0 and 1 indicating the pre-intervention and post-intervention period, respectively; − is the time point when the EPI initiated in the PLADs i; ( − ) × indicates the time-varying change in the effectiveness of the EPI [1]; ( , = 3) denotes natural cubic spline of monthly average temperature with 3 dfs; ℎ is a categorical variable of calendar months; denotes reported public health emergencies of hepatitis A (ie, five or more hepatitis A cases occurring within one week in the same collective unit according to historical reports or ≥20 cases occurring within one month) [2]; ,−1 is an autoregressive term of residuals at lag 1 to adjust for auto-correlation if necessary.
The models of the nontarget population and the whole population have the same structure as the target children model, except that interaction items are not considered.

Effect estimators
In the target population analysis, the provincial-specific excess risk (ER) of the hepatitis A incidence in the PLAD i and the time point t was expressed as: = [( 2 + 3 × ) − 1] × 100%, and the corresponding 95% confidence intervals (CI) were estimated.By assuming the regression coefficients followed a multivariate normal distribution, the time-varying term also follows a multivariate normal distribution, which is as follows: 2 + 3 × = ~2 (, ) , where B is a joint normal distribution of 2 and 3 , with a mean vector of and a covariance matrix of ; T = [1, t] is a constant matrix of t×2 dimensions.In non-target population analyses and whole population analyses, the ERs of the EPI was performed using the following formula: = [( 2 ) − 1] × 100%.
In the target population analysis, the excess incidence rate (EIR) in the province i was estimated as: terminal point of the study period, respectively.Then, we use Monte Carlo simulation to estimate the 95% empirical CIs (eCIs) of EIR, because it is difficult to deduce the analytical formula for confidence intervals of EIRs [3].Firstly, we took random samples B (j) , with a mean of and a covariance of , of the original parameters 2 and 3 derived from the regression model.Consequently, the distribution of EIR can be recalculated empirically.An 95% eCI could be defined with the 2.5th and 97.5th percentiles of the sampling distribution.For instance, 10,000 sets of B (j) (j = 1, 2,..., 10,000) were sampled from the multivariate normal distribution, calculating 10,000 sets of EIRs.The 2.5th and 97.5th percentiles (i.e. the 250th and 9,750th EIRs from the smallest to the largest) of these 10,000 EIRs were regarded as upper and lower bound of 95% eCI.
The calculation of EIR for the nontarget population model and the whole population model are identical to that for the target population model.The average population in the formula of EIR is based on the corresponding population group included in the model.

Sensitivity analyses
We undertook sensitivity analyses to examine the robustness of our results.First, we applied a natural cubic spline function with 3 dfs instead of calendar months to capture the seasonality of hepatitis A incidence.Second, we investigated the potentially nonlinear intervention effect over time by replacing ( − ) × with ( − , = 3) × .Finally, we consider a transition period from the implementation of the EPI to 2010 and excluded this period from the analyses [4,5], because the EPI was initially limited to several cities with the heavy burden of hepatitis A in several PLADs and did not cover the entire PLAD until 2010 [6].We replaced different dfs in the natural cubic spline of monthly average temperature.c The model only contains 3 terms (ie, the time, the indicator variable of intervention, and the interaction term between the 2 variables mentioned above).a For public health emergencies without a clear date or a place of the occurrence, we consider that there was a hepatitis A public health emergency in that province in that year.PLAD: provincial-level administrative division.

Supplemental tables and figures
In Zhejiang, there were 1 public health emergency of hepatitis A in 2007 and one in 2009 [39].c A total of 13 public health emergencies were reported from 2005 to 2020 in Henan [40,41].

b
The lag equal to 1 represents that the model of the corresponding PLAD includes the residual regression term./ indicates the model without adding an autoregressive term of the residuals.The criterion for adding a residual lag term is whether the model residuals existed first-order partial autocorrelation in the main analysis.

b A natural cubic spline function with 3
dfs to calendar months was applied to capture the seasonality of hepatitis A incidence.cThe potentially nonlinear intervention effect over time was investigated by applying a natural cubic spline function with 3 dfs to the time point when the EPI initiated in each PLADs.d A transition period from the implementation of the EPI to 2010 was excluded this period from the analysis.e ER: excess risk.

Figure S1 .
Figure S1.The partial auto-correlation functions of residuals from preliminary analyses for 30 provincial-level administrative divisions without an autoregressive term of residuals.

Figure S2 .
Figure S2.The partial auto-correlation functions of residuals from preliminary analyses for 30 provincial-level administrative divisions with an autoregressive term of residuals.

Figure S3 .
Figure S3.Monthly standardized hepatitis A incidence in the Chinese mainland.The standardized incidence was calculated as the original incidence divided by the maximum incidence during the study period.(A) Monthly standardized hepatitis A incidences of 31 provincial-level administrative divisions; (B) Monthly standardized hepatitis A incidences of different years in the Chinese mainland.

Figure S4 .
Figure S4.Monthly incidence of hepatitis A among children aged 2-9 years in seven regions of Chinese mainland from 2005 to 2019.The shadow represents the approximate period of the intervention.(A) Northeast China; (B) Northern China; (C) Northwest China; (D) Eastern China; (E) Central China; (F) Southern China; (G) Southwest China..

Figure S5 .
Figure S5.Excess risks of hepatitis A among children aged 2-9 years associated with the Expanded Program on Immunization in 30 provincial-level administrative divisions.The shadow represents the 95% CIs of excess risks.

Figure S6 .
Figure S6.Excess risks and excess incidence rates of hepatitis A among children aged 2-9 years in different subgroups.In FigureS6A-L, the dark-colored point estimates and confidence intervals represent the combined effect of PLADs where the value of the variable is greater than or equal to the median and the light-colored point estimates and confidence intervals represent the combined effect of PLADs where the value of the variable is less than the median.In FigureS6 M-N, the dark-colored point estimates and confidence intervals represent the combined effect of PLADs integrating Inactivated vaccines and, the light-colored point estimates and CIs represent the combined effect of PLADs integrating live attenuated vaccines.(A) Urbanization rates; (B) Urbanization rates; (C) GDP per capita; (D) GDP per capita; (E) The number of hospitalization beds per 1,000 persons; (F) The number of hospitalization beds per 1,000 persons; (G) Average incidence of hepatitis A before the implementation of EPI; (H) Average incidence of hepatitis A before the implementation of EPI; (I) The proportion of children; (H) The proportion of children; (K) Illiteracy rates; (L) Illiteracy rates; (M) The type of vaccines; (N) The type of vaccines.Abbreviation: EPI, Expanded Program on Immunization; PLADs, provincial-level administrative division; GDP, Gross Domestic Product.

Figure S7 .
Figure S7.Excess risks of hepatitis A incidence associated with the Expanded Program onImmunization in sensitivity analysis with the replacement of seasonality control.The shadow represents the 95% CIs of excess risks.

Figure S8 .
Figure S8.Excess risks of hepatitis A incidence associated with the Expanded Program onImmunization in sensitivity analysis with the non-linear trend.The shadow represents the 95% CIs of excess risks.

Figure S9 .
Figure S9.Excess risks of hepatitis A incidence associated with the Expanded Program onImmunization in sensitivity analysis with the transition period.The shadow represents the 95% CIs of excess risks.

Table S1 .
Details on the Expanded Program on Immunization information in the Chinese mainland.

Table S2 .
Hepatitis A cases before and after the intervention of the Expanded Program on

Table S3 .
Values of quasi-Akaike information criterion in different modeling strategy.
a PLAD: provincial-level administrative division.b

Table S4 .
Description on hepatitis A public health emergencies in Chinese mainland.

Table S5 .
Autoregressive term of model residuals in the main analysis.

Table S6 .
Yearly hepatitis A cases and incidence among children aged 2-9 years in seven regions of Chinese mainland from 2005 to 2019.
a The unit of the hepatitis A incidence is cases per 100,000 persons.

Table S7 .
Excess risks of hepatitis A incidence associated with the Expanded Program on Immunization in the Chinese mainland.
a EPI: Expanded Program on Immunization.

Table S8 .
Average annual excess cases of hepatitis A associated with the Expanded Program on Immunization among the target population.
a EPI: Expanded Program on Immunization.

Table S8 .
Average annual excess cases of hepatitis A associated with the Expanded Program on Immunization among the target population.continued.
a EPI: Expanded Program on Immunization..

Table S9 .
Average annual excess incidence of hepatitis A associated with the Expanded Program on Immunization among the target population.
a EPI: Expanded Program on Immunization.

Table S9 .
Average annual excess incidence of hepatitis A associated with the Expanded Program on Immunization among the target population.continued.
a EPI: Expanded Program on Immunization..

Table S10 .
Excess risks of hepatitis A incidence among children aged 2-9 years associated with the Expanded Program on Immunization in 7 regions of the Chinese mainland.
a EPI: Expanded Program on Immunization.b ER: excess risk.

Table S10 .
Excess risks of hepatitis A incidence among children aged 2-9 years associated with the Expanded Program on Immunization in 7 regions of the Chinese mainland.continued.
a EPI: Expanded Program on Immunization.b ER: excess risk.

Table S11 .
Average annual excess incidence of hepatitis A associated with the Expanded Program on Immunization among the nontarget population.

Table S11 .
Average annual excess incidence of hepatitis A associated with the Expanded Program on Immunization among the nontarget population.continued.

Table S12 .
Average annual excess incidence of hepatitis A associated with the Expanded Program on Immunization among the whole population.
a EPI: Expanded Program on Immunization.

Table S12 .
Average annual excess incidence of hepatitis A associated with the Expanded Program on Immunization among the whole population.continued.
a EPI: Expanded Program on Immunization.