HIV-exposed children account for more than half of 24-month mortality in Botswana

We performed the largest study of 24-month child mortality in Botswana since national scale-up of PMTCT programs. Using cell phone follow up, we were able to achieve a loss-to-follow-up (LTFU) rate of 0.5 % at 24 months. We confirmed previous findings that even if uninfected, HIV-exposed children have decreased survival compared with HIV-unexposed children and found very high mortality among HIV-infected children, with little improvement from the early ART era [5]. These two groups combine to account for more than half of deaths among children under two in Botswana. Mortality among HIV-unexposed children in Botswana was low compared with regional estimates [24], despite breastfeeding durations that fell short of WHO recommendations [25].

Higher mortality among HEU and HIV-infected children may relate to lack of breastfeeding, but our analysis could not untangle the impact of infant feeding method from the impact of HIV-exposure on child survival given extreme collinearity. Although the risk of death for the small number of HIV-exposed breastfed children was lower than those who formula fed, this comparison was underpowered and not statistically significant. We found HIV-infected mothers who breastfed more commonly had markers of lower socioeconomic status compared with HIV-infected mothers who did not breastfeed (data not shown). We would therefore expect HIV-exposed breastfed babies to have higher mortality and so our effect estimate may be biased toward the null. In sum, although we suspect that lack of breastfeeding may have contributed to the higher mortality seen among HEUs, our data were not able to conclusively demonstrate this association.

Mortality among HIV-infected children remains similar to that seen in 2003–2006, near the beginning of the rollout of pediatric ART in Botswana [5]. Our findings were somewhat lower than cumulative 24-month mortality (54 %) reported from a pooled analysis of African countries, most of which had higher overall childhood mortality rates [13]. Previous studies in Botswana have demonstrated very high mortality risk among HIV-infected infants who formula feed from delivery [15], and this was likely a contributor in our cohort as well. Although we did not collect data on ART use among HIV-infected children, we know that children were not initiated prior to diagnosis at their 6-week PCR. Given the emerging data to supporting long-term benefits of early ART initiation [26, 27], our study provides support for HIV testing at birth (in addition to standard six week testing) particularly in areas where formula feeding is common and an HIV diagnosis within days of birth may allow for appropriate counseling to initiate or maintain breastfeeding for these high-risk children.

Infectious diseases were the primary cause of death for all exposure groups, with diarrheal diseases and respiratory diseases most common. Seasonality of birth also appeared to play a role, with children born during the rainy season experiencing higher likelihood of death. The reasons for this association are unknown but could be due to a more contaminated water supply from the rains, food scarcity in the period before the harvest (May/June), or decreased support in the home because family members are often needed for ploughing in this time period. In our study, 88 % of children who died from a cause other than trauma or accident in Botswana had contact with the healthcare system shortly before their death. Therefore, interventions to improve inpatient treatment, recognize at-risk children, and improve diagnostics among those with diarrhea and respiratory symptoms could also improve outcomes.

Our study has several strengths, including a large and geographically diverse sample, 99.5 % follow-up at 24 months and a study design with minimal impact on child outcomes. However, our study likely underestimates early neonatal mortality. All infants enrolled in this study were discharged from the hospital alive and proportionately fewer were very preterm (37 weeks GA) and very low birthweight (2500 g) than in the general Botswana population [28, 29]. Although we did not exclude sick infants, mothers of the most critically ill neonates may have been less likely to participate in a study. Extrapolating from other studies of infant outcomes at our study sites [28, 29], we estimate that early in-hospital mortality missed by our study was approximately 1 %; because our sample was otherwise representative, we believe that our mortality estimates may have underestimated overall 24-month mortality in the general population by this amount. Also, while we attempted to control for differences between HIV-infected and HIV-uninfected mothers, unmeasured sociodemographic factors may confound the association between mortality and HIV-exposure status. Our study was also limited in that we were unable to conclusively know the HIV-infection status on most infants who died before their six-week HIV PCR.