Bolivia is a lower-middle income country in South America with an annual birth cohort of about 263 000 and a gross national income of $1699 (£1092; €1284) per capita in 2009.13 The Bolivian Ministry of Health added RV1 to the routine childhood immunization schedule in August 2008, recommending two doses of RV1 for all children in Bolivia at 2 and 4 months of age. From March 2010 to June 2011 we did a case-control evaluation at six hospitals in four of the largest cities in Bolivia (La Paz, El Alto, Cochabamba, and Santa Cruz) to assess the effectiveness of RV1 against hospital admissions for rotavirus. These hospitals are ministry hospitals that were selected on the basis of WHO guidelines for rotavirus surveillance that recommend selecting hospitals that admit more than 250 children for gastroenteritis each year.14 These six hospitals were estimated to have 19% of all hospital admissions for diarrhea among children before the introduction of vaccine.

We defined cases as children admitted to the hospital overnight for treatment of acute diarrhea, defined as at least three loose stools in a 24 hour period. Inclusion criteria were onset of diarrhea less than 14 days before the hospital visit; a rotavirus positive stool sample during the first 48 hours of admission (to avoid nosocomial infection); and eligibility to receive at least one dose of RV1, defined as being born after June 1, 2008 and being at least 8 weeks of age when admitted to hospital. We excluded cases when we were unable to contact a parent or care-taker to obtain consent, identify three hospital controls, or verify vaccination status through parental card or vaccination registry. To identify case patients, we did active hospital based surveillance 24 hours a day in the emergency department and inpatient wards. Bulk stool specimens were collected within 48 hours of admission. Specimens were stored at 2-8°C before transfer to the national laboratory on a weekly basis during the first nine months of the study and to a local laboratory during the last six months of the study. Rotavirus testing was done with a commercially available enzyme immunoassay (ProSpecT ELISA, Oxoid, UK). Specimens were stored frozen at −70°C until they were shipped to the Centers for Disease Control and Prevention, Atlanta, GA, USA for genotyping analysis. Genotyping was done on samples with sufficient stools, as described by Hull et al.15

We assessed effectiveness by using two groups of controls: children admitted to hospital for conditions other than diarrhea (that is, hospital controls) and children with rotavirus negative diarrhea (that is, test negative controls). For non-diarrhea hospital controls, inclusion criteria were seeking care in the emergency department or being admitted to the same hospital as the case for an acute illness unrelated to diarrhea or a vaccine preventable condition (measles, mumps, rubella, diphtheria, pertussis, tetanus, tuberculosis, hepatitis B); being born within 30 days of the case’s date of birth; and being eligible to receive at least one dose of RV1, defined as being born after June 1, 2008 and being at least 8 weeks of age when admitted to hospital. We excluded controls when we were unable to contact a parent or care-taker to obtain consent, identify three hospital controls for each case, or verify vaccination status through parental card or vaccination registry. After a rotavirus case was identified, we routinely queried emergency department and hospital admission logs daily during the subsequent two weeks to identify three consecutive hospital controls. All efforts were made to capture the child during the hospital visit to avoid logistical challenges of home visits and potential loss to follow-up. We also sought to assess vaccine effectiveness by using children with rotavirus negative diarrhea as controls (test negative controls). Test negative controls were those children who were enrolled during the surveillance for rotavirus diarrhea but tested negative for rotavirus by enzyme immunoassay.

We conducted face to face interviews with parents of case patients and hospital controls during the hospital visit. After written informed consent had been given, we obtained information on vaccination history, demographics, socioeconomic factors, history of breast feeding, and medical history. For cases, we also gathered information on clinical characteristics, treatment, and course of illness. We selected variables on the basis of recommendations from a WHO guideline document on studies of the effectiveness of rotavirus vaccines.16 The primary objective of the study was to assess differences in antecedent exposure to the full series (two doses versus zero) among cases compared with non-diarrhea controls and compared with test negative controls.

We obtained vaccination history from the parent and considered it confirmed if the parent showed a vaccination card with the date of vaccination, the type of vaccine used, and the name of the child. If parents reported any vaccination but did not possess a card, we obtained confirmation by review of vaccine cards at the clinic where the child was reportedly vaccinated. We identified vaccination records at the clinic on the basis of the participant’s name, sex, and date of birth. We obtained a photocopy of the vaccination record for cases and controls, and, after data entry into an electronic database, we verified all RV1 vaccination dates against this record.

Using a precision based approach,17 we estimated that we needed a total of 170 case patients to compute a vaccine effectiveness of 60% with a confidence limit width of 30%, using a matched design with a control to case ratio of three to one and vaccine coverage of 50%. We enrolled a total of 400 case patients to allow for subgroup analyses including effectiveness of partial vaccination, strain specific effectiveness, and effectiveness stratified by age. Because we did not specifically calculate sample sizes for the subgroup analyses (strain specific and age stratified vaccine effectiveness), we did a post hoc power analysis and present vaccine effectiveness results when expected power using χ² exceeded 80% at a significance level of 0.05 given the observed number of cases and controls for each of the secondary outcomes.

To minimize bias associated with differential surveillance and diagnosis, we used a standard WHO recommended case definition for severe gastroenteritis at all surveillance sites and laboratory confirmed diagnosis of rotavirus with a validated enzyme immunoassay with high sensitivity and specificity. For non-diarrhea controls, we excluded children with diseases not preventable by rotavirus vaccine because they would be less likely to receive rotavirus vaccine than the source population from which the cases arose. Information bias was minimized by blinding coordinators who verified vaccination records from knowledge of case or control status and the study hypothesis. Efforts to determine vaccine status were similar between cases and controls.

Our primary aim was to calculate the vaccine effectiveness of two doses of RV1 against hospital admission for rotavirus. To assess for a potential gradient in protection by severity, we analyzed for vaccine effectiveness against rotavirus diarrhea with a clinical severity score of at least 11 and at least 15 on a 20 point Vesikari scoring scale that was used in the RV1 clinical trials.4

We firstly did bivariate analyses to assess for differences in indicators of socioeconomic condition between rotavirus case patients and the two groups of controls to identify potential confounders or biases for the association between RV1 vaccination and rotavirus disease. We used the Wilcoxon rank sum test or χ² test to assess differences.

We constructed two separate logistic regression models for non-diarrhea and test negative controls to calculate odds ratios with associated 95% confidence intervals.18 For both models, we considered cases and controls to be vaccinated with the respective number of doses (one or two) if the most recent dose was administered 14 days before the case patient’s hospital visit (the reference date). Children who received two doses did not contribute to the one dose vaccine effectiveness analysis, and children who received one dose of vaccine did not contribute to the two dose vaccine effectiveness analysis. For non-diarrhea controls, we used a conditional logistic regression model to estimate the crude odds ratio, because these controls were matched to case-patients by hospital and date of birth (±30 days). To estimate a crude odds ratio with test negative controls comparable to the crude odds ratio for non-diarrhea controls generated through a matched analysis, we used an unconditional logistic regression that included hospital, age (in months), and month/year of birth in the base model, because test negative controls were unmatched with regard to age and hospital during the design phase of the study. For both control groups, we then assessed for confounding by using multivariate modeling. To the base models, we included all additional variables with P<0.20 in the bivariate analyses. We then used a hierarchical backward elimination approach to select the variables in the final model individually, excluding those variables at a significance level of P>0.05.19 For substantive reasons, we retained age in months, month/year of birth, and hospital for test negative controls. To assess for potential clustering by hospital, we included hospital as a random effect in the regression models, but it did not alter the model outcomes.

We did subgroup analyses to assess protection from partial dose vaccination (that is, one dose of RV1), strain specific protection, and protection among children 6-11 months of age compared with those aged 12 months or over. We assessed for interaction by age (6-11 months versus >11 months of age) and the prevalent strains by including an interaction term for age and vaccination and for strain type and vaccination in the model.

Finally, to assess the potential for bias in our estimates of effectiveness, we did a “bias indicator” analysis to examine whether two doses of RV1 provided protection against cases of diarrhea that tested negative for rotavirus with non-diarrhea controls, under the hypothesis that significant vaccine effectiveness against test negative diarrhea would be due to residual confounding in the non-diarrhea controls. For this analysis, we compared vaccination rates among rotavirus negative diarrhea cases and non-diarrhea controls and adjusted for age, hospital, and month/year of birth by using unconditional logistic regression.

We estimated the adjusted odds ratio by using the exponential of the coefficient for the vaccination variable in the model. We calculated the 95% confidence interval for the adjusted odds ratio by using the standard error of the coefficient,18 and we subsequently calculated vaccine effectiveness as (1−adjusted odds ratio)×100%. Statistical significance was designated as P<0.05. We used SAS statistical software (version 9.2) for analyses.

We approached a total of 451 case patients, 1247 non-diarrhea controls, and 817 test negative controls. Of these, we excluded 51 (11%), 47 (4%), and 99 (12%), respectively, and the final analysis included 400 case patients, 1200 non-diarrhea controls, and 718 test negative controls (fig 1). In comparison with non-diarrhea controls, case patients were more likely to be male and attend day care but less likely to have chronic underlying illness, higher level maternal education, and telephones and computers in their home (table 1). Test-negative controls were somewhat more similar to case-patients but also were more likely to be male and attend day care and less likely to have higher level maternal education and computers in their homes.

Vaccine records were confirmed for all participants in the analysis. Adherence to the age recommendations was good; only 10% of the children were vaccinated outside the recommended age windows of 2 and 4 months of age (fig 2).

We identified no difference greater than 10% between the crude and adjusted estimates of vaccine effectiveness for either the primary or secondary analyses in the study (tables 2, 3, and 4). The adjusted vaccine effectiveness of a full series of two doses of RV1 against hospital admission for rotavirus was 77% (95% confidence interval 65% to 84%) with non-diarrhea controls and 69% (54% to 79%) with test negative controls (table 2). One dose of RV1 also provided significant protection of 56% (32% to 72%) with non-diarrhea controls and 36% (0% to 59%) with test negative controls.

Of the 400 case patients admitted to hospital for rotavirus diarrhea, 373 (93%) had rotavirus diarrhea with a Vesikari score of 11 or greater and 191 (48%) had a Vesikari score of 15 or greater. Protection was similar against each of the severity outcomes in the study. With non-diarrhea controls, RV1 provided protection of 76% (64% to 84%) for a Vesikari score of 11 or above and 74% (54% to 85%) against a score of at least 15. When we used test negative controls, vaccine effectiveness was 69% (53% to 79%) for Vesikari score of 11 or above and 62% (37% to 88%) for a severity score of at least 15 (table 2).

Of the 400 cases, 295 had sufficient stool samples for genotyping. Commonly detected strains included G9P[8] (n=107; 36%), G2P[4] (74; 25%), G3P[8] (52; 18%), and G9P[6] (23; 8%). Others were non-typeable (24; 8%) or other sparsely detected strains (15; 5%). Vaccine effectiveness ranged from 59% to 93% against four different strains, without any significant interaction by type of strain (P=0.70). With non-diarrhea controls, strain specific effectiveness was 85% (69% to 93%) and 93% (70% to 98%) against the partially heterotypic G9P[8] and G3P[8] strains and 69% (14% to 89%) and 87% (19% to 98%) against the fully heterotypic G2P[4] and G9P[6] strains (table 3). With test negative controls, strain specific effectiveness was 80% (60% to 90%) and 74% (22% to 91%) against the partially heterotypic G9P[8] and G3P[8] strains and 59% (7% to 78%) and 80% (37% to 94%) against the fully heterotypic G2P[4] and G9P[6] strains.

We found no significant difference in effectiveness of RV1 against hospital admission for rotavirus between the two age groups when using non-diarrhea controls (P=0.42) or test negative controls (P=0.35) (table 4). For non-diarrhea controls, vaccine effectiveness was 77% (51% to 89%) for children aged 6-11 months compared with76% (59% to 86%) for those aged 12 months or over. For test negative controls, vaccine effectiveness was 64% (34% to 80%) for children aged 6-11 months compared with 72% (52% to 86%) for those aged 12 months or over. Similarly, effectiveness was high for both age groups when we restricted the analysis to hospital admissions with severity scores of 11 or greater and to admissions related to the most prevalent strain, G9P[8].

In the bias indicator analysis, RV1 did not confer protection against non-rotavirus diarrhea cases in comparison with non-diarrhea controls for one dose (9%, −33% to 37%) or two doses (9%, −27% to 24%) of RV1.

In Bolivia, RV1 provided higher protection (~71-77%) against outcomes of similar severity than RV5 provided in the low income setting of Nicaragua (50%).10 In fact, effectiveness for RV1 in Bolivia was similar to that for RV1 in El Salvador (76%),24 a slightly more developed lower-middle income country in Central America, and comparable to RV1 efficacy for first two years of life in the large clinical trial from 10 Latin American countries (80%).2 In this trial, the efficacy of RV1 in Nicaragua was 78% (18 to 96) through two years of life.25 The differing efficacy by vaccine type might reflect a chance occurrence, as data on effectiveness of RV5 in Latin America are available only from a single country.10 Furthermore, in clinical trials, the efficacy of RV1 in the first year of life in Malawi (49%) was comparable to that of RV5 in low and lower-middle income countries in Africa (48%) and Asia (39%).6 7 26 The possibility of inter-study differences in case severity might also explain some of the variation in vaccine effectiveness; however, all studies applied comparable WHO recommended case definitions and Vesikari severity scores, which should have minimized this effect. As rotavirus vaccines are introduced in additional low and lower-middle income countries in Africa and Asia, further assessments of effectiveness of both RV1 and RV5 in these settings to compare the performance of the two vaccines will be important.

In clinical trials from Africa and in some post-licensure studies from low socioeconomic settings, protection from both RV5 and RV1 seemed to be lower among children older than 1 year compared with those aged under 1.6 24 27 28 We did not see this phenomenon in Bolivia, which is consistent with the sustained effectiveness of 79-83% seen through two years of life in the large RV1 clinical trial in high and middle income countries from Latin America.2 In addition, two studies in impoverished populations in Brazil and Australia have suggested that RV1 (G1P[8]) effectiveness against fully heterotypic strains (such as G2P[4]) might diminish more rapidly than that against homotypic strains, but confidence bounds were too wide for a meaningful conclusion.29 30 In Bolivia, we saw sustained protection against partially heterotypic G9P[8] strains, which is consistent with findings from the large Latin America pre-licensure RV1 trial, but we also did not have sufficient power to assess duration of protection against fully heterotypic G2P[4] strains. Although these findings are encouraging, our data should be interpreted with some caution as the study was not specifically designed to evaluate effectiveness among specific age groups. Such questions could be better assessed as the vaccine program matures and additional older children are vaccinated.

Some limitations must be considered. Estimates of vaccine protection obtained through observational studies hinge upon differences in exposure to vaccination between cases and controls and thus are subject to biases relating to recording and ascertainment of exposure. The availability of documentation of vaccination status differed between case patients (93%), test negative controls (88%), and non-diarrhea controls (98%) in our study and might have biased our estimates of effectiveness to some extent. The control population might be enriched with non-vaccinees if controls with missing vaccine history were more likely to be vaccinated than were cases with missing history, which might have increased our estimates of effectiveness. If controls with missing vaccine history had lower vaccination rates than cases with missing history, effectiveness estimates might be biased towards null. In addition, each of the control groups might have its own specific biases. For example, a false positive enzyme immunoassay result for rotavirus could lead to misclassification of a test negative control as a case. If the vaccine offers any protection, this would enrich the case population with vaccinees thus falsely lowering estimates of effectiveness. However, test negative controls have previously proved to be good controls in studies of rotavirus vaccine effectiveness when compared with community or non-diarrhea controls,29 31 32 33 and the similarity in estimates of effectiveness with both control groups in our study provides further reassurance.

Although some differences in demographics and socioeconomic indicators existed between the cases, non-diarrhea controls, and test negative controls, we did not identify any substantial differences between the crude and adjusted estimates of vaccine effectiveness. However, some residual unmeasured confounding could still be present. Bias related to health seeking behavior is possible, but the test negative diarrhea controls are likely to have similar healthcare seeking patterns as rotavirus diarrhea cases and thus to be less prone to this potential bias than are non-diarrhea controls. In addition, our bias indicator analysis did not identify any significant bias with the non-diarrhea controls. Lastly, if fecal shedding of RV1 leads to horizontal transmission of vaccine virus and resultant immune response in unvaccinated contacts of vaccine recipients, as was shown in a clinical trial in the Dominican Republic, this might decrease the attack rates of disease in unvaccinated children and decrease estimates of RV1 effectiveness as measured through observational studies.34

In conclusion, our study provides evidence for protection by RV1 vaccination against hospital admissions for rotavirus and severe rotavirus disease caused by four different rotavirus strains during the first two years of life in a low income setting in Latin America. Attainment of significant protection from one dose of RV1 also is encouraging for target populations in low and lower-middle income settings, who can develop severe and fatal rotavirus disease during the first few months of life. Further research is needed to refine our understanding of the heterogeneous immune response to rotavirus vaccines in low socioeconomic settings. Potential strategies to improve vaccine efficacy that warrant investigation may include altering the age at rotavirus immunization to avoid the negative influence of circulating maternal antibodies, decoupling rotavirus vaccines and oral polio vaccine, and adding additional doses in the routine schedule or as a booster with the measles dose at the end of infancy. Although understanding the reasons for lower efficacy is important, it should not hinder use of currently available rotavirus vaccines. Our study provides compelling data favoring broader use of rotavirus vaccine in low income settings to reduce the burden of severe and fatal rotavirus disease among children.