Non-specific effects of BCG and DTP vaccination on infant mortality: An analysis of birth cohorts in Ghana and Tanzania

Background: Vaccines may induce non-specific effects on survival and health outcomes, in addition to protection against targeted pathogens or disease. Observational evidence suggests that infant Baccillus Calmette-Guérin (BCG) vaccination may provide non-specific survival benefits, while diphtheria-tetanus-pertussis (DTP) vaccination may increase the risk of mortality. Non-specific vaccine effects have been hypothesized to modify the effect of neonatal vitamin A supplementation (NVAS) on mortality. Methods: 22,955 newborns in Ghana and 31,999 newborns in Tanzania were enrolled in two parallel, randomized, double-blind, placebo-controlled trials of neonatal vitamin A supplementation from 2010 to 2014 and followed until 1-year of age. Cox proportional hazard models were used to estimate associations of BCG and DTP vaccination with infant survival. Results: BCG vaccination was associated with a decreased risk of infant mortality after controlling for confounders in both countries (Ghana adjusted hazard ratio (aHR): 0.51, 95% CI: 0.38–0.68; Tanzania aHR: 0.08, 95% CI: 0.07–0.10). Receiving a DTP vaccination was associated with a decreased risk of death (Ghana aHR: 0.39, 95% CI: 0.26–0.59; Tanzania aHR: 0.19, 95% CI: 0.16–0.22). There was no evidence of interaction between BCG or DTP vaccination status and infant sex or NVAS. Conclusion: We demonstrated that BCG and DTP vaccination were associated with decreased risk of infant mortality in Ghana and Tanzania with no evidence of interaction between DTP or BCG vaccination, NVAS, and infant sex. Our study supports global recommendations on BCG and DTP vaccination and programmatic efforts to ensure all children have access to timely vaccination. Clinical trials registration: Ghana (Australian New Zealand Clinical Trials Registry (ANZCTR): ACTRN12610000582055) and Tanzania (ANZCTR: ACTRN12610000636055) This prospective cohort study used data from two parallel, randomized double-blind, placebo- controlled trials of neonatal vitamin A supplementation conducted in Ghana (Australian New Zealand Clinical Trials Registry (ANZCTR): ACTRN12610000582055) and Tanzania (ANZCTR: ACTRN12610000636055) [22], [23]. The assessment of whether the effect of vitamin A on mortality was modified by DTP was included a priori in the original trial protocol [24]. The analysis of the association BCG and DTP with mortality was a posthoc analysis. The trials were conducted from 2010 to 2014 and were implemented with a similar protocol [24]. Briefly, infants were eligible for the trials if they were able to feed orally and families planned to live in the study area for 6 months. In Ghana, liveborn neonates were recruited in 7 contiguous districts in central, rural Ghana. In Tanzania, pregnant mothers were recruited from antenatal clinics, labor wards, or population-based pregnancy surveillance in the Dar es Salaam and Morogoro regions [22], [23]. All participants provided written consent for participation. In Ghana and Tanzania, eligible infants were randomized at birth to either placebo or single-dose oral capsules of 50,000 IU of vitamin A. Over the course of one year, infants were visited at home by field interviewers, and data on vaccination dates were collected from immunization records and if immunization records were unavailable, vaccination status was collected through caregiver interview. Research staff verified the information for all infant deaths and performed a verbal autopsy. In Ghana, follow up visits occurred monthly until 12 months of age. In Tanzania, follow up visits occurred at 1, 3, 6, and 12 months. All study protocols have been described in detail elsewhere [24]. The study protocols were approved by the Ghana Health Service Ethical Review Committee, the ethics committees of the Kintampo Health Research Centre, Ghana, the London School of Hygiene and Tropical Medicine, UK, the Institutional Review Boards of the Harvard T.H. Chan School of Public Health, Ifakara Health Institute, Medical Research Coordinating Council of Tanzania, and by the WHO Ethical Review Committee. We first examined the potential non-specific effects of BCG and DTP vaccination by assessing the association between vaccination and survival using time-varying Cox proportional hazard models. The underlying time scale was days of age. Schoenfeld residuals were plotted to assess the proportional hazards assumption. Participants contributed person-time to unvaccinated and vaccinated periods, and vaccination status was updated over time, at the time the vaccine was received. This has been referred to as retrospective updating and was chosen to minimize exposure misclassification and selection bias. This method in comparison to other methods used for this type of data are described further in supplemental figures S1-S3 and tables S1-S2. Because participants were only eligible for the DTP vaccine after 30 days of life, our analysis of DTP vaccination included only infants who had survived to 30 days. Additionally, the time origin was set at 30 days because analyzing DTP from birth introduces potential immortal time bias since deaths occurring before eligibility would be attributed to being unvaccinated. The adjusted models included the birthweight and sociodemographic confounders collected at birth. For the Ghana data, we modeled continuous birthweight with cubic splines with knots at 1.5, 2, 2.5, and 3.5 kg. We also included: head of household (mother, father, grandmother, grandfather, other), household religion (Christian, Muslim, None, Other), maternal age (<20, 20–24, 25–29, 30–34, 35–39, ≥40), maternal education (None, Primary, Secondary, Post-secondary), multiple or singleton birth, number of living children in household (0, 1, 2, 3 + ), number of children in household who have died (0, 1, 2, 3 + ), place of birth (Compound, Facility, Other), site ID (1–4), wealth quintile, delivery type (vaginal or caesarean), and maternal megadose of vitamin A. For the Tanzania models, we included the same variables, but sites were Ifakara and Dar es Salaam and educational categories were limited to none, primary, or secondary and higher. Missing covariate values were imputed in the Ghana and Tanzania datasets. The primary analysis was restricted to those with complete information on whether they received BCG for the BCG analysis or whether they received their first DTP vaccination for the DTP analysis. Vaccine information was considered missing unless there was either a yes or no on the vaccination card or confirmation by caregiver interview. Analyses for BCG and DTP were conducted separately, and data was only excluded if the relevant vaccine was missing information. Sensitivity analyses to test the effect of different assumptions regarding missing data are described below. We next examined the potential interaction between neonatal vitamin A supplementation (NVAS) and BCG or DTP vaccination and examined the three -way interaction between vitamin A supplementation, vaccination, and infant sex. To do this, we fit time-varying cox proportional hazard models for the effect of vitamin A status on survival allowing for interaction by vaccination status. Similar to the analysis of vaccination alone, vaccine status was allowed to vary with time and models were adjusted for the same baseline covariates to control for confounding between vaccination and survival. Interaction p-values were calculated using likelihood ratio tests comparing the model with the interaction of vitamin A and vaccination to the model without the interaction term. We also conducted stratified analyses by sex to evaluate whether there were sex-specific interactions between NVAS and vaccination status. We tested whether the three-way interaction p-value was statistically significant using a likelihood ratio test comparing a model with the three-way interaction for vitamin A or placebo, vaccination status, and sex to a model with only two-way interactions between vitamin A or placebo, vaccination status, and sex. Finally, we examined whether there was effect modification of the association between DTP and survival by a previous BCG vaccination. All analyses were conducted in R 3.5.1, and p-values<0.05 were considered statistically significant. We conducted sensitivity analyses to investigate how robust the results were to missing vaccination data, particularly for infants who died. We investigated three scenarios for the missing vaccination data that represented the extreme range of possibilities: 1, assuming all participants with missing vaccination status or date of vaccination were vaccinated on day 0 for BCG or day 30 for DTP (contributing to only vaccinated person-time); 2, assuming all participants missing vaccination status or vaccination date data were vaccinated at the mode of 1 day for BCG in Ghana and Tanzania and 45 days and 31 days for DTP in Ghana and Tanzania, respectively (contributing to both unvaccinated and vaccinated person-time); 3, assuming none of the participants who were missing vaccination data were ever vaccinated during follow-up (contributing to only unvaccinated person-time) (Supplemental Table S3).