The Impact of Improved Water, Sanitation, and Hygiene on Oral Rotavirus Vaccine Immunogenicity in Zimbabwean Infants: Substudy of a Cluster-randomized Trial

Abstract Background Oral vaccines have lower efficacy in developing compared to developed countries. Poor water, sanitation, and hygiene (WASH) may contribute to reduced oral vaccine immunogenicity. Methods We conducted a cluster-randomized 2 × 2 factorial trial in rural Zimbabwe. Pregnant women and their infants were eligible if they lived in clusters randomized to (1) standard of care (52 clusters); (2) improved infant feeding (53 clusters); (3) WASH: ventilated improved pit latrine, 2 hand-washing stations, liquid soap, chlorine, infant play space, and hygiene counseling (53 clusters); or (4) feeding plus WASH (53 clusters). This substudy compared oral rotavirus vaccine (RVV) seroconversion (primary outcome), and seropositivity and geometric mean titer (GMT) (secondary outcomes), in WASH vs non-WASH infants by intention-to-treat analysis. Results We included 801 infants with documented RVV receipt and postvaccine titer measurements (329 from 84 WASH clusters; 472 from 102 non-WASH clusters); 328 infants with prevaccination titers were included in the primary outcome. Thirty-three of 109 (30.3%) infants in the WASH group seroconverted following rotavirus vaccination, compared to 43 of 219 (19.6%) in the non-WASH group (absolute difference, 10.6% [95% confidence interval {CI}, .54%–20.7%]; P = .031). In the WASH vs non-WASH groups, 90 of 329 (27.4%) vs 107 of 472 (22.7%) were seropositive postvaccination (absolute difference, 4.7% [95% CI, –1.4% to 10.8%]; P = .130), and antirotavirus GMT was 18.4 (95% CI, 15.6–21.7) U/mL vs 14.9 (95% CI, 13.2–16.8) U/mL (P = .072). Conclusions Improvements in household WASH led to modest but significant increases in seroconversion to RVV in rural Zimbabwean infants. Clinical Trials Registration NCT01824940.

Rotavirus diarrhea causes approximately 215 000 deaths annually, predominantly in sub-Saharan Africa and south Asia [1].
A key preventive strategy is oral rotavirus vaccination, which had been introduced in 96 countries by October 2018 [2]. However, oral vaccines consistently underperform in settings where most diarrheal deaths occur. For example, oral rotavirus vaccine (RVV) protective efficacy against severe rotavirus gastroenteritis in Europe is 98% [3], compared to 39% in Ghana, Kenya, and Mali [4], and seroconversion to RVV is consistently lower in developing compared with developed countries [5].
Multiple intestinal factors have been implicated in oral vaccine underperformance [6], including enteropathogens [7], diarrhea [8], and perturbations in the commensal microbiota [9], which may decrease the effective titer of vaccine virus. Additionally, environmental enteric dysfunction, a chronic condition of altered gut structure and function [10] might prevent vaccine uptake and attenuate the mucosal immune response to vaccine antigens [11,12]. Fecal contamination due to poor water, sanitation, and hygiene (WASH) may contribute to this intestinal pathology and reduced oral vaccine immunogenicity.
Improving oral vaccine performance would reduce morbidity and mortality from enteric disease, but most interventions have shown no benefit [13]. No trials have evaluated the impact of WASH on oral vaccine performance. We hypothesized that a WASH intervention could enhance immune responses to RVV in early infancy. We conducted a nested substudy within the Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial to test this hypothesis.

Study Population
SHINE was a 2 × 2 factorial, cluster-randomized trial across 2 districts in rural Zimbabwe, which tested the independent and combined effects of improved WASH and improved infant and young child feeding (IYCF) on child length-for-age and hemoglobin at 18 months of age. The trial design, procedures, and outcomes have been reported elsewhere [14,15]. In brief, pregnant women were enrolled between November 2012 and March 2015 from clusters randomized to 1 of 4 arms: standard of care, IYCF, WASH, or combined IYCF and WASH. WASH households had a ventilated improved pit latrine [16] and 2 hand-washing stations [17] installed within 6 weeks of enrollment (approximately 20 weeks' gestation), and received monthly liquid soap, water chlorination solution (WaterGuard, Nelspot, Zimbabwe), an infant play space, and mat. A subgroup of infants underwent longitudinal specimen collection at 1, 3, 6, 12, and 18 months of age [18].

Rotavirus Substudy
The rotavirus substudy was a prespecified objective in the SHINE trial protocol; the analysis plan was published prior to unblinding (https://osf.io/ad9zr/). In May 2014, midway through the trial, oral rotavirus vaccine (Rotarix) was introduced in Zimbabwe and administered with oral polio vaccine at 6 and 10 weeks of age. Vaccination was undertaken at local clinics and not overseen by the trial; national rotavirus vaccination coverage in 2015-2016 was 87%-91% [19]. Dates of vaccine receipt were recorded through review of child health cards, with each child's final vaccination status categorized as complete (2 doses), incomplete (1 dose), or no vaccine (zero doses).
Infants were eligible for the substudy if they were human immunodeficiency virus (HIV) unexposed, live-born after introduction of rotavirus vaccine, and had at least 1 stored plasma sample (minimum 100 uL) collected before 6 months of age. Available plasma for each infant was allocated to preand postvaccination time-points, based on the documented date of receipt of the first dose of RVV. Infants were excluded from analysis if they had no postvaccine titer measurement, had missing data on vaccination, or had not received at least 1 dose of rotavirus vaccine.
The primary outcome of this substudy was the proportion of infants with RVV seroconversion, defined as a postvaccine plasma concentration of antirotavirus immunoglobulin A (IgA) ≥20 U/mL in infants who were seronegative (<20 U/mL) prevaccination [20]. Secondary outcomes were the proportion of seropositive infants (defined as postvaccine titer ≥20 U/mL) and geometric mean titer (GMT). Antirotavirus IgA is the most widely used marker of vaccine response or natural infection [5]. We measured antirotavirus IgA titers in cryopreserved plasma using an enzyme-linked immunosorbent assay, as previously described [21]. The assay lower limit of detection was 7.5 U/ mL. Laboratory assays and data analysis were conducted blind to trial arm.
Our sample size calculation was based on published seroconversion rates from African rotavirus vaccine trials of 47.2%-57.1% [22]. Assuming a coefficient of variation of 0.25, 360 infants per group would provide 80% power to detect a 15% increase in RVV seroconversion between WASH and non-WASH groups. These calculations assumed an arithmetic mean of 5 infants per cluster and a harmonic mean of 3 per cluster (to adjust for variability in cluster size).

Ethical Considerations and Data Availability
The original SHINE trial and this laboratory substudy were approved by the Medical Research Council of Zimbabwe and the John Hopkins School of Public Health Committee on Human Research. Written informed consent was obtained from all mothers prior to enrollment in the trial.
The full protocol and statistical analysis plan for the SHINE trial are available at https://osf.io/w93hy. Data collected for the SHINE trial will be made publicly available as individual participant data with an accompanying data dictionary. The data will be housed and made accessible to the global research community through Clinical Epidemiology Database Resources (ClinEpiDB) (http://ClinEpiDB.org) at the University of Pennsylvania. This platform is charged with ensuring that epidemiological studies are fully anonymized by removing all personal identifiers and obfuscating all dates per participant through application of a random number algorithm to comply with the ethical conduct of human subjects research. Researchers must agree to the policies and comply with the mechanism of ClinEpiDB to access data housed on this platform.

Statistical Analysis
Analyses were intention-to-treat. We used generalized estimating equations accounting for within-cluster correlation, containing 2 dummy variables representing the WASH and IYCF interventions, unadjusted for other covariates, and with an exchangeable working correlation structure [23]. Because there was no statistical interaction between interventions, we compared all outcomes between the combined WASH and IYCF + WASH arms ("WASH group") and the combined standard of care and IYCF arms ("non-WASH group"). We used a log-binomial specification to estimate relative risks (RRs) for dichotomous outcomes. For antirotavirus GMT, we used a log-normal censored regression model (Tobit), with left censoring at 7.5 U/mL. We used other methods to adjust for within-cluster correlation including multinomial and ordinal regression models with robust variance estimation, and Somers' D for medians.
Covariates for adjusted analyses were selected using a 2-step procedure. First, in bivariate analysis we identified baseline covariates with an important association with the outcome (for dichotomous outcomes: P < .2, or RR >2.0 or <0.5; for continuous outcomes: P < .2, or difference >0.25 standard deviation). Next, selected covariates were entered in a multivariable regression model using a forward stepwise selection procedure. In both models, we included season of birth, because it influences exposure to wild-type rotavirus, and IYCF, to account for the other randomization in the factorial trial design.
A per-protocol analysis examined the impact of WASH on RVV immunogenicity when behavior-change modules were delivered at high fidelity [15]. We also undertook several sensitivity analyses. First, we excluded children who were seropositive prevaccination and presumed to have acquired wild-type rotavirus. Second, we restricted timing of immunogenicity measurements to tighter windows, which we defined as 0-14 days before the first dose of vaccine (for prevaccine titer) and 21-60 days after the last dose of vaccine (for postdose titer), based on previous studies [24]. Finally, a prespecified subgroup analysis by infant sex was planned if the interaction terms for the primary outcome were statistically significant.
All statistical analyses were performed using Stata version 14 (StataCorp, College Station, Texas) and Prism version 7 (GraphPad Software, La Jolla, California) software.

Baseline Characteristics
Among 5280 enrolled mothers, there were 3989 HIV-unexposed live births (Supplementary Figure 1). Of these, 882 fulfilled the inclusion criteria for this substudy and had antirotavirus IgA measured; 81 were subsequently excluded from analysis because they had no record of RVV receipt (n = 18) or no postvaccine titer measurement (n = 63) ( Figure 1). This analysis therefore included 801 infants (329 WASH infants from 84 clusters; 472 non-WASH infants from 102 clusters). All infants had postvaccine titers measured and were included in the secondary outcomes (seropositivity and GMT); a subset of 328 infants had both prevaccine and postvaccine titers measured and were included in the primary outcome (seroconversion) (Supplementary Figure 2). Baseline characteristics of infants and their mothers were generally well balanced (  trial population of HIV-negative births, although infants in this substudy were less likely to be preterm (Supplementary Table 1).

Intervention Delivery and Uptake
There was high fidelity of intervention delivery (Supplementary  Figure 2). Findings remained similar after adjustment and in the per-protocol analysis (Supplementary Table 3).

Secondary Outcomes
In  Figure 2). Among infants who received both doses of RVV, differences between groups were marginally increased ( Figure 2); findings remained similar after adjustment and in the per-protocol analysis (Supplementary Table 4).  16.4 weeks' gestation. In some instances, WASH hardware was therefore introduced before the baseline visit, leading to an apparent imbalance in household WASH characteristics between arms.

Sensitivity Analyses
We conducted sensitivity analyses to account for variation in the timing of pre-and postvaccine titer measurements, and for baseline seropositivity due to wild-type rotavirus infection. Among 142 infants who had prevaccine and postvaccine titers measured within a more restricted window, 15 Table 6). Finally, in a prespecified subgroup analysis, there was no significant interaction between infant sex and RVV seroconversion (P = .52).

DISCUSSION
Improvements in household WASH in rural Zimbabwe increased infant seroconversion to RVV. Among infants in the non-WASH group, approximately 20% seroconverted to the RVV, compared to >30% in the WASH group. Both the direction and magnitude of effect remained highly consistent after adjustment and in multiple sensitivity analyses. Together, these findings suggest that environmental health improvements could help to increase oral vaccine immunogenicity. However, our findings also highlight the poor seroconversion to RVV even after substantial household WASH improvements, showing that other approaches will be required to realize the full benefits of oral vaccination in developing countries.
Our finding that WASH modestly increased RVV seroconversion provides some insight into the mechanisms underlying poor responses to RVV. It is plausible that conditions of poor WASH perturb the intestinal milieu, thereby reducing the immunogenicity of oral vaccines [6]. For example, carriage of enteropathogens has been associated with reduced immune responses to RVV [25]. Environmental enteric dysfunction may impair oral vaccine processing during transit through the small intestine, although its contribution to oral vaccine underperformance remains uncertain [26]. Some studies have found differences in the composition of the intestinal microbiota between responders and nonresponders to RVV [9]. The WASH intervention may therefore confer benefits for oral vaccine immunogenicity via several mechanisms. Future data on enteropathogen carriage and the microbiota from SHINE will provide important mechanistic insights. Alternatively, WASH may confer benefits for oral vaccination by reducing exposure to wild-type rotavirus infection. Fewer seropositive infants at baseline in the WASH group would increase susceptibility to vaccination as the first exposure to rotavirus. Nonetheless, after excluding baseline seropositive infants, our inferences remained unchanged, suggesting that this is not the only mechanism underlying our findings.
Numerous trials have explored interventions to augment oral vaccine immunogenicity. Some have aimed to promote gut health and mucosal responsiveness, including probiotics [27], antibiotics [28], and zinc [29], but overall these have shown no effect [30]. Changes to the formulation or schedule of oral vaccines are the only strategies with evidence of benefit for oral vaccine immunogenicity [30]. The modest effect size from the WASH intervention in our study is comparable with reported effect sizes following alterations in vaccine scheduling. However, WASH has the advantage of already being a priority intervention to reach Sustainable Development Goal targets, whereas changes to vaccination schedules may be difficult to implement. Whether the finding of increased seroconversion with WASH translates into improved vaccine efficacy cannot be addressed in the current study. However, a recent observational study in Malawi showed that reductions in all-cause mortality following RVV introduction were greater in areas also receiving intensive sanitation improvements, suggesting potential synergy between interventions [31].
The promising difference in vaccine responses with WASH is an important proof of concept but is tempered by the scale of oral vaccine underperformance. RVV seroconversion in the WASH group was 39% at best (Supplementary Table 5), which is still far lower than in developed countries (67%-98%) [5]. It is, however, consistent with recent data from other developing countries: RVV seroconversion was 24% in Malawi [32] and 27% in Bangladesh [33]. Taken together, there are important implications of these findings. First, our WASH interventions may have been insufficiently effective and further improvements in vaccine immunogenicity may be gained through more transformative WASH approaches. The household-level elementary WASH interventions implemented in SHINE were similar to those commonly accessible in rural areas of developing countries. However, we have recently shown that these interventions had no impact on stunting, anemia, or diarrhea [15], similar to other trials [34,35]. We have therefore argued that more effective approaches may be required to achieve the desired public health benefits of improved WASH. Alternatively, factors not addressed by WASH may contribute to the immunogenicity gap, such as genetic variation [36], interference from maternal antibodies, and micronutrient deficiencies. Other approaches that address these factors may be required. This is the first study to explore the impact of a WASH intervention on oral vaccine responses, leveraging the cluster-randomized design of SHINE and the introduction of RVV in Zimbabwe. There was high reported uptake of WASH interventions, which were based on extensive formative research. We prespecified all outcomes and ran laboratory assays and statistical analyses blind to intervention arm to reduce bias. However, there are some important limitations. First, SHINE was not a vaccine trial and we therefore did not control vaccine administration or timing of pre-and postvaccine titer measurements. This may have contributed to the overall lower rates of seroconversion than is typical of published RVV trials in southern Africa [22]. However, >98% of infants received at least 1 dose of RVV and the median time-point of immunogenicity measurement was 28 days postvaccination, with no significant difference between WASH and non-WASH groups. In addition, when we restricted analyses to infants who had titers measured during a narrower window, our inferences remained unchanged. Second, although we exceeded our sample size, the final number of infants in the primary outcome analysis (which required both prevaccine and postvaccine plasma samples) was smaller than planned and the precision of our seroconversion estimates was likely reduced. Finally, antirotavirus IgA is a limited correlate of protection from disease [24], particularly in developing countries [33]. Nevertheless, it remains the preferred measure of RVV response and is likely to be biologically meaningful given the enormous differences in seroconversion rates between developing and developed countries [5].
Oral rotavirus vaccination has substantially reduced diarrheal mortality globally [37]; however, its underperformance in developing countries limits its full potential. Improvements in the efficacy of RVV would confer major public health benefits. We show here that improvements in household WASH significantly increased RVV seroconversion. However, the increase was modest and seroconversion remained low overall, showing the enormous challenge of oral vaccine underperformance. More effective WASH interventions and alternative strategies are likely required  to further improve the performance of oral vaccines so that all children in developing countries can benefit from their impact.