Impact of Rotavirus Vaccine Introduction in Children Less Than 2 Years of Age Presenting for Medical Care With Diarrhea in Rural Matlab, Bangladesh

Abstract Background Following the conclusion of a human rotavirus vaccine (HRV) cluster-randomized, controlled trial (CRT) in Matlab, Bangladesh, HRV was included in Matlab’s routine immunization program. We describe the population-level impact of programmatic rotavirus vaccination in Bangladesh in children <2 years of age. Methods Interrupted time series were used to estimate the impact of HRV introduction. We used diarrheal surveillance collected between 2000 and 2014 within the 2 service delivery areas (International Centre for Diarrhoeal Disease Research, Bangladesh [icddr,b] service area [ISA] and government service area [GSA]) of the Matlab Health and Demographic Surveillance System, administered by icddr,b. Age group–specific incidence rates were calculated for both rotavirus-positive (RV+) and rotavirus-negative (RV–) diarrhea diagnoses of any severity presenting to the hospital. We used 2 models to assess the impact within each service area: Model 1 used the pre-vaccine time period in all villages (HRV– and control-only) and Model 2 combined the pre-vaccine time period and the CRT time period, using outcomes from control-only villages. Results Both models demonstrated a downward trend in RV+ diarrheal incidences in the ISA villages during 3.5 years of routine HRV use, though only Model 2 was statistically significant. Significant impacts of HRV on RV+ diarrhea incidences in GSA villages were not observed in either model. Differences in population-level impacts between the 2 delivery areas may be due to the varied rotavirus vaccine coverage and presentation rates to the hospital. Conclusions This study provides initial evidence of the population-level impact of rotavirus vaccines in children <2 years of age in Matlab, Bangladesh. Further studies are needed of the rotavirus vaccine impact after the nationwide introduction in Bangladesh.

Globally, an estimated 13 000 deaths due to rotavirus diarrhea occur annually in children <5 years of age, with most of the burden in sub-Saharan Africa and Asia [1]. While diarrhea-associated mortality rates have decreased globally in the last decade, the burden of rotavirus diarrhea remains substantial in low-income settings [2]. In 2006, 2 rotavirus vaccines were introduced worldwide: GlaxoSmithKline's human rotavirus vaccine (HRV; Rotarix) and Merck's pentavalent rotavirus vaccine (PRV; RotaTeq). Large, multi-site, randomized, controlled trials (RCTs) of both vaccines in Africa demonstrated moderate vaccine efficacy (VE) against severe rotavirus diarrhea during the first year of life [3,4]. As of August 2018, 96 countries, of which 46 are Gavi-eligible, have introduced rotavirus vaccines into their regional or national immunization programs [5]. In the World Health Organization (WHO) Africa region, 74% of countries have introduced rotavirus vaccination. Studies in sub-Saharan Africa have shown statistically significant rotavirus vaccine effectiveness and population-level impacts against all-cause and rotavirus diarrhea in children <5 years of age within 2-3 years of the initiation of routine use [6][7][8][9][10][11][12][13][14].
Despite the WHO recommendation for rotavirus vaccine use worldwide, only 18% of countries in the WHO southeast Asia region have introduced a rotavirus vaccine [5]. Limited data on vaccine effectiveness and population impacts may have slowed the introduction of rotavirus vaccines in Asia [15]. The only multi-site RCT of PRV in Asia demonstrated moderate vaccine efficacy against severe rotavirus gastroenteritis in the first 2 years of life (Bangladesh VE 42.7%, 95% confidence interval [CI] 10.4-63.9; Vietnam VE 63.9%, 95% CI 7.6-90.9; combined VE 51.0%, 95% CI 12.8-73.3) [16]. In Bangladesh, this RCT included half of the Matlab villages (International Centre for Diarrhoeal Disease Research, Bangladesh [icddr,b] service areas).
To evaluate the effectiveness of HRV on rotavirus diarrhea in Asia, a 2-year cluster-randomized trial (CRT) was conducted in all villages in Matlab, Bangladesh, beginning in 2008 [17]. The overall effectiveness, which assesses the overall reduction in the incidence of acute rotavirus diarrhea, regardless of vaccination status, was 29.0% (95% CI 11.3-43.1) in children <2 years of age. This study provided initial evidence of the potential population impact of routine rotavirus vaccine use in Bangladesh. After the CRT, HRV was provided for routine use among infants in all Matlab villages between March 2011 and September 2014.
To evaluate the population-level impact of HRV in Matlab, Bangladesh, during the 3.5 years of routine use following the CRT, we examined trends in the rotavirus-positive (RV+) and rotavirus-negative (RV-) diarrhea incidence rates of any severity presenting to Matlab Hospital between February 2000 and September 2014.

Study Setting
The study utilized diarrheal surveillance data collected among children <2 years of age residing in villages of the Matlab Health and Demographic Surveillance System (HDSS), administered by the icddr,b, and presenting to Matlab Hospital [18]. The HDSS is divided into the icddr,b service area (ISA; 67 villages) and the government service area (GSA; 75 villages). The icddr,b provides ISA villages with child and maternal health intervention programs and the Bangladesh Ministry of Health and Family Welfare provides GSA villages with the government standard of care. The HDSS maintains a census and registration of vital events, including internal and external migration.

Immunization Records
The HDSS also maintains immunization records through a formal record-keeping system. In the ISA villages, community health workers maintain vaccination records, and in the GSA villages, community health workers check vaccination cards or ask mothers if the card is missing.

Diarrheal Surveillance
Matlab Hospital is the central diarrhea treatment facility for the Matlab HDSS population. This study includes data from children <2 years of age. The incidence rate for presentations to Matlab Hospital of all-cause diarrhea among children from GSA villages has historically been about half of the incidence rate for presentations from ISA villages [17]. Stool specimens are collected from all patients presenting with diarrhea (3 or more loose stools per 24 hours) to Matlab Hospital. The samples are tested for group A rotavirus VP6 antigens using a solid-phase, sandwich-type enzyme immunoassay (Prospect, Oxoid Diagnostics Ltd, Hampshire, United Kingdom).

Statistical Analysis
Interrupted time series, using segmented regression models, were used to estimate the impact of the rotavirus vaccine introduction in Matlab, Bangladesh, among children <2 years of age [19]. The monthly incidence rates of RV+ and RV-diarrhea were examined separately, by age group (0 to <12 months, 12 to <24 months, and combined [0 to <24 months]). The incidence rates were calculated for RV+ and RV-diarrhea with the number of events presenting to Matlab Hospital per month as the numerator and the monthly population at risk, using HDSS census estimates, as the denominator.
Due to varied rotavirus vaccine coverage and baseline diarrheal incidences, analyses were conducted separately for the ISA and GSA villages.
Among the ISA villages, the pre-vaccine time period was defined as We used 2 models to estimate the impact of HRV use on RV+ and RV-diarrhea incidence rates. Model 1 was defined a priori, while Model 2 was defined after examining the count data. Model 1 and Model 2 differ by both the baseline period used as the referent category and the types of villages included (HRV-and/or control-only). In both models, a generalized linear model was fit to the time-series data, assuming a negative, binomial distribution due to over-dispersion of the data [20]. Calendar months were included in each model to account for seasonality, and a sequential, monthly term for every month over the entire time period was included to account for secular trends. The natural log of the monthly population at risk was included in the model as the offset term. The Breusch-Godfrey test identified some autocorrelation; therefore, 95% CIs were estimated using Newey-West heteroskedastic-and autocorrelation-consistent variance estimators, with a lag of 2 [19,21]. The estimates of the coefficients for each time period were exponentiated to estimate incidence rate ratios (IRRs), compared to the referent category.
In Model 1, within the ISA and GSA areas separately, the corresponding pre-vaccine time period was used as the referent category. Villages randomized as both HRV and control-only were included in the analysis. To estimate the IRRs and corresponding 95% CIs, the time periods corresponding to the RCT, CRT, and each of the 3.5 years of routine HRV use were modeled with separate indicator variables. This is a conservative model, which directly compares incidence rates in February 2000-February 2007 (ISA villages) and February 2000-October 2008 (GSA villages) to the years of routine HRV use, starting in April 2011 in all ISA and GSA villages.
In the secondary analysis (Model 2), only the villages randomized as control-only during the CRT were used. Within the ISA and GSA regions, the pre-vaccine and CRT time periods were combined in the referent category. The time period corresponding to the RCT was excluded. To estimate the IRRs and corresponding 95% CIs, each of the 3.5 years of routine HRV use were modeled with separate indicator variables. This approach directly compared incidence rates in February 2000-March 2011, excluding the RCT time period, to the years of routine HRV use, starting in April 2011 in those ISA and GSA villages randomized as controls.
The monthly vaccine coverage was estimated as the proportion of children 6 to <52 weeks old receiving each HRV dose within regions of Matlab, Bangladesh. Analyses were completed using Stata version 14 (Stata Corporation, College Station, TX). This study was approved by the ethical review committee of icddr,b in Bangladesh and the Fred Hutchinson Cancer Research Center. Figure 1 show RV+ and RV-counts and average incidence rates over time in the GSA and ISA villages, using the study populations used for Models 1 and 2.

Rotavirus vaccine was not available in Matlab between February 2000 and March 2007. Between April 2007 and
March 2009, 568 infants in ISA villages were randomized to receive PRV and 568 infants were randomized to placebo as part of a multi-site RCT [16]. In the stratified HRV CRT in both ISA and GSA areas, villages were randomized to 2 doses of HRV at 6 and 10 weeks of age or randomized as observed, control-only villages [17]. In the GSA villages, the CRT started in November 2008, and in the ISA villages, the CRT started in April 2009. Follow-ups and vaccinations during the CRT occurred in both the ISA and GSA villages through March 2011. Through a donation of vaccines post-CRT, HRV was provided routinely starting in April 2011. After September 2014, the rotavirus vaccine was unavailable.
HRV vaccine coverage levels among children <1 year of age changed during the study period ( Figure 2). During the CRT, both the ISA and GSA villages showed similar vaccine coverage levels. Following the CRT, the coverage level among age-eligible children in ISA villages was maintained at between 65-80%, while GSA villages decreased to 42% at the end of the study period.
Observed and predicted RV+ diarrhea counts in ISA and GSA villages for both models demonstrated a satisfactory model fit ( Supplementary Figures 1-2).

Diarrhea Incidence Trends: International Centre for Diarrhoeal Disease Research, Bangladesh, Service Area Villages
Using Model 1, with the pre-vaccine time period as the referent category, RV+ diarrhea rates increased during the RCT period and the CRT period in both age groups in ISA villages (Table  3; Figure 3). During periods of routine HRV use, there was a downward trend that was not statistically significant in RV+ diarrhea incidences after each additional year of vaccine use. During the entire 3.5 years of routine use, there was no meaningful decrease in RV+ diarrhea rates in 0-to <12-month-old children (IRR 0.72, 95% CI 0.39-1.33) or 12-to <24-monthold children (IRR 0.91, 95% CI 0.46-1.83). Using Model 2, combining the pre-vaccine time period and the CRT time period in the reference category and using control-only villages, there was a downward trend in the RV+ diarrhea incidence rates after each additional year of routine HRV use in both age groups (Table 4; Figure 3). During 3.5 years of routine HRV use, there was a statistically significant, 41% decrease in RV+ diarrhea rates in 0-to <12-month-old children (IRR 0.59, 95% CI 0.43-0.80), a 35% decrease in 12-to <24-month-old children (IRR 0.65, 95% CI 0.42-1.02), and a statistically significant, 39% decrease in children 0 to <24 months of age (IRR 0.61, 95% CI 0.45-0.82).
In Model 1, RV-diarrhea rates increased during the RCT period and the CRT period in both age groups. During periods of routine HRV use, there was an increased risk of RVdiarrhea in 0-to <12-month-old children (IRR 1.59, 95% CI 1.09-2.31) and no meaningful change in 12-to <24-monthold children. In Model 2, there were no statistically significant changes in RV-diarrhea rates during periods of HRV routine use.

Diarrhea Incidence Trends: Government Service Area Villages
Using Model 1, with the pre-vaccine time period as the referent category, the incidence of RV+ diarrhea increased during the CRT period in 0-to <12-month-old children, but did not meaningfully change in 12-to <24-month-old children (Table 5; Figure 4). During periods of routine HRV use, there was an upward trend in the RV+ diarrhea incidence after each additional year of vaccine use in 0-to <12-monthold children, but no clear trends in 12-to <24-month-old children. During 3.5 years of routine use, there was no statistically significant change in the incidences of RV+ diarrhea Table 1   in 0-to <12-month-old children (IRR 1.25, 95% CI 0.78-2.01) or in 12-to <24-month-old children (IRR 1.00, 95% CI 0.52-1.92). Using Model 2, there was a downward trend in the RV+ diarrhea incidence after each additional year of routine HRV use in both age groups (Table 6; Figure 4). However, during 3.5 years of routine HRV use, there was no meaningful change in the RV+ diarrhea rate in either age group. In Models 1 and 2, there were no statistically significant changes in RV-diarrhea rates during periods of HRV routine use.

DISCUSSION
Our study demonstrates a decreasing trend in RV+ diarrhea incidences among children <2 years of age from ISA villages presenting to Matlab Hospital during 3.5 years of routine HRV use. Using a conservative model to estimate pre-vaccination rotavirus diarrhea trends (Model 1), the results were not statistically significant. However, by restricting the analysis to control-only villages, we gained an additional 2 years of recent, pre-vaccine time to model baseline trends (Model 2), and found a statistically significant, 39% reduction in RV+ diarrhea rates in children 0 to <24 months of age. No significant impact of HRV on the RV+ diarrhea incidence among children from GSA villages was observed using either model. Differences in the population-level impacts between ISA and GSA villages are likely due to lower HRV coverage and lower reported diarrhea incidences in GSA areas, compared to ISA villages. Our study also examined changes in the rate of RV-diarrhea as a control outcome, with the assumption that HRV introduction should have no significant impact on RV-diarrhea [22]. In Model 1, using only the pre-vaccine period in the referent category, we observed an increasing trend in both RV+ and RV-diarrhea rates in children 0 to <24 months of age in ISA villages during the RCT and CRT time periods. While other interventions or unmeasured biases may have influenced the all-cause gastroenteritis incidence, we believe this increase was due to changes in health-care-seeking behaviors due to the RCT. During the RCT, field staff visited the homes of infants enrolled in the study to remind parents to bring their child to the hospital for episodes of diarrhea [16]. A change in community health-care-seeking behavior is the most likely explanation, as there was no significant change in all-cause diarrhea in the corresponding time period in the GSA villages, where no RCT took place (Figure 1), and no specific pathogen was identified as a cause of the increase in all-cause diarrhea. The most conservative model to estimate the HRV impact (Model 1) modelled the RCT and CRT time periods separately and directly compared the pre-vaccine time period to the years of routine HRV use in both ISA and GSA villages. However, if increased  health-care-seeking behaviors were sustained, results from Model 1 would underestimate the population-level impact of HRV.
In the secondary analysis (Model 2), both to increase power and to include relevant health-care-seeking behaviors to estimate the baseline incidence, we restricted the analysis to those villages randomized as control-only during the CRT period, and assessed the impact of routine HRV use on diarrhea over time. The referent category combined the pre-vaccine time period and the CRT time period. These models showed a significant impact of routine HRV use on RV+ diarrhea rates in 0-to <24-month-old children in ISA villages, but not in GSA villages. RV-diarrhea rates did not significantly change over time using this model. Notably, both models showed a decreasing trend in RV+ diarrhea in ISA villages during sustained HRV coverage. This analysis demonstrates the importance of using the appropriate baseline incidences and underlying trends in time-series analyses.
Despite the potential differences in health-care-seeking behavior over time, our results are similar to the RCT and CRT conducted in Matlab, Bangladesh, with the greatest impact of rotavirus vaccine on children 0 to <12 months of age. To our knowledge, no other population-level impact analyses have been reported in Asia with rotavirus diarrhea as the outcome, though a study in the Philippines saw a 60% (95% CI 55-64%) reduction in all-cause diarrhea hospitalizations within 4 years after rotavirus vaccine introduction [23]. Similar time-series analyses conducted 2-3 years after rotavirus introduction found a 49% (95% CI 32-63%) decrease in rotavirus diarrhea in <5-year-old children in Ghana [12], a 54% (95% CI 33-69%) decrease in rotavirus diarrhea in <1-year-old children in Malawi [11], a 33% (95% CI 25-41%) reduction in rotavirus diarrhea in <5-year-old children in Botswana [14], and a 38% reduction in rotavirus positivity among children <5 years old in Zambia [10]. Longterm impacts were also observed in Ghana [24] and Zambia [25]. Importantly, in these studies, >90% vaccine coverage   for 1 or 2 doses of rotavirus vaccine were reported within 1 year of vaccine introduction. In our study, the maximum, 2-dose HRV coverage of 68% was attained in the ISA villages during the second year of routine use. Our study has limitations. As in any time-series analysis, our study may have been confounded by other interventions or other unmeasured factors associated with RV+ diarrhea and the timing of the vaccine introduction. However, our confidence in the impact of HRV is increased, because no meaningful changes in RV-diarrhea incidences were observed. Second, while the Matlab HDSS database shows lower vaccine coverage in GSA areas, coverage may be underestimated or inaccurate due to the lack of recording on health cards in this region and potential reliance on maternal reports. Though measured with the same potential biases, during the study period, the average coverage for 3 doses of Diphtheria-Pertussis-Tetanus (DTP3) was 97% in ISA villages and 91% in GSA villages [26]. Third, with the available data, we were unable to assess the impact of the rotavirus vaccine on severe rotavirus diarrhea, as indicated by a Vesikari score ≥11, which is the outcome used in rotavirus vaccine clinical trials.
This study provides initial evidence of the population-level impact of rotavirus vaccines in children <2 years of age in regions of high vaccine coverage in Matlab, Bangladesh. Pecenka et al [27] estimated that, with a Gavi subsidy in Bangladesh, the averted cost/disability adjusted life year (DALY) ratio ranged between $58/DALY and $142/DALY, indicating a highly cost-effective vaccine, given 94% coverage of DTP3 in Bangladesh [27,28] In our study, during the pre-vaccine period, rotavirus was detected in 34.5% of diarrhea cases in children <5 years of age presenting to Matlab Hospital. Other regions of Bangladesh show an average of 64% of diarrhea instances being due to rotavirus in children <5 years of age [29]. With sustained vaccine coverage and a considerable nationwide burden of rotavirus diarrhea, larger impacts of HRV on rotavirus gastroenteritis are likely to be observed long-term in Bangladesh. This may provide additional evidence to influence other countries in the region to introduce the rotavirus vaccine.  Abbreviations: CI, confidence interval; CRT, cluster-randomized controlled trial; GSA, government service area; IRR, incidence rate ratio; RV-, rotavirus negative; RV+, rotavirus positive; YR, year.