Pneumococcal Acquisition Among Infants Exposed to HIV in Rural Malawi: A Longitudinal Household Study

The prevalence of Streptococcus pneumoniae (pneumococcus) carriage is higher in adults who are infected with human immunodeficiency virus (HIV) than in adults who are not. We hypothesized that infants exposed to HIV become carriers of nasopharyngeal pneumococcus earlier and more frequently than infants who are not exposed to HIV. We compared infant pneumococcal acquisition by maternal HIV status and household exposure in Karonga District, Malawi, in 2009–2011, before the introduction of pneumococcal conjugate vaccine. Nasopharyngeal swabs were collected every 4–6 weeks in the first year of life from infants with known HIV-exposure status, their mothers, and other household members. We studied infant pneumococcal acquisition by maternal HIV status, serotype-specific household exposure, and other risk factors, including seasonality. We recruited 54 infants who were exposed to HIV and 131 infants who were not. There was no significant difference in pneumococcal acquisition by maternal HIV status (adjusted rate ratio (aRR) = 1.00, 95% confidence interval (CI): 0.87, 1.15). Carriage by the mother was associated with greater acquisition of the same serotype (aRR = 3.09, 95% CI: 1.47, 6.50), but the adjusted population attributable fraction was negligible (1.9%, 95% CI: 0.0, 4.3). Serotype-specific exposure to children under 5 years of age was associated with higher acquisition (aRR = 4.30, 95% CI: 2.80, 6.60; adjusted population attributable fraction = 8.8%, 95% CI: 4.0, 13.4). We found no evidence to suggest that maternal HIV infection would affect the impact of pneumococcal vaccination on colonization in this population.


Web Material Web Appendix 1: Duration of carriage
Duration of carriage was estimated for infants, mothers and other children <5 years. Duration of carriage could not be estimated for other adult household members, because the response rate was low. Episodes starting at the end of the follow up period were excluded, because duration of carriage could not be accurately determined. For mothers and children <5 years also events starting at the first observation were excluded, because duration of carriage could not be accurately determined. Differences in duration of carriage between infants, mothers, and other children <5 years, between serotypes, and between HIV-status of the mother were calculated using the Mann-Whitney U test.
Web Table 1 shows results for duration of carriage in infants, mothers and other children < 5years. Infants carried pneumococci of the studied serotypes (19F, 19A, 6B, 23F, 6A and 15B) on average for 52.5 days (median 39.5 days). Duration of pneumococcal carriage was shorter in mothers (mean 37.2 days, median 28.5 days, P=0.008) and other children <5 years (mean 38.9 days, median 28.5 days, P<0.001). No significant differences were found between the duration of carriage of the studied serotypes. There were no significant differences in duration of carriage by maternal HIV-status for the infant, mother or other children <5 years.

Web Appendix 2: Generalized Additive Mixed Model
The effects of infant age and seasonal and secular trends were studied with fitted generalized additive mixed models. For seasonal trend, parametric functions with different numbers of sin-cosine waves were examined. A model with three sin-cosine waves and fitted splines for infant age in days and secular trend (days since study onset) was found to provide the best fit: where y i,t is the carriage in infant i for each day of the year t, α and β terms are the regression coefficients for each sine and cosine function, f1 and f2 are smooth functions for infant age in days and secular trend (days since study onset), and u i,t is the infant-specific random effect. Log-binomial regression was used to report rate ratios. The individual-level variance was found to be negligible for infants (σ 2 <0.01); hence results from the non-mixed model are reported. Results are presented in Web Table 2

Web Appendix 3: Cox Proportional Hazards Model
Rates of infant pneumococcal acquisition were studied using counting process (Anderson-Gill) models. The counting process model is a simple extension of the Cox proportional hazards model where a subject is at continuous risk for an event and has the same baseline hazards function (1). We considered our study subjects to be at continuous risk of acquiring a new serotype because of the possibility of multi serotype carriage. Time at risk was calculated as the number of days between sampling dates. If an infant was missed on a sampling appointment, we estimated the start of the next period at risk as the midpoint between the two samples for which a time in days was available. We examined the proportional hazards assumption by testing the correlation coefficient between transformed survival time and the scaled Schoenfeld residuals. The proportional hazard assumptions were met by all variables apart from infant age, which has been omitted from the model presented below.
Results from the Cox proportional hazards model (Web Table 3) were similar to the log-binomial regression model (Table 1). A limitation of our Cox model is that failure to adjust for infant age could have led to residual confounding. Another limitation of the Cox proportional hazards model is that our sampling interval changed over time from four to six weeks. This could have led to an underestimation of hazard rates towards the end of the study period. A subanalysis including only time points with sampling intervals of four weeks found similar results, suggesting the change in sample intervals did not affect the results found.