05 to detect differences of 0.11 log10

in cytokine responses for exposures with two equal-sized categories [19]. The objective of this observational analysis was to determine socio-demographic, maternal and infant factors learn more associated with cytokine responses following BCG and tetanus immunisation. Socio-demographic factors were maternal age, maternal education (categories none, primary, secondary or tertiary), household socioeconomic status (a six-level score based on building materials, number of rooms, items owned) and location of residence (by zone, Fig. 1). Maternal factors were the three commonest maternal helminth infections (hookworm, Mansonella perstans, Schistosoma mansoni), maternal asymptomatic malaria parasitaemia (Plasmodium falciparum) and maternal immunisation status (absence or presence of a maternal BCG scar; Raf targets number of documented doses of tetanus immunisation during pregnancy). Infant factors were gender, birth weight, anthropometric scores at age one year (weight-for-age, height-for-age and weight-for-height [27]), infant malaria (current, asymptomatic malaria on the day of the assay; number of documented clinical malaria episodes in the preceding year) and HIV status (based on maternal and infant serology, and infant PCR at age six weeks: unexposed, exposed-uninfected, or

infected). Cytokine responses showed skewed distributions, with a disproportionate number of zero values, as has commonly been observed for immunoepidemiological data and, in particular, for the use of whole blood stimulation and cytokine response assays [28], [29] and [30]. Results were transformed to log10(cytokine concentration + 1) and analysed by linear regression using

bootstrapping with 10,000 iterations to estimate standard errors Sitaxentan and bias-corrected accelerated confidence intervals [29]. Regression coefficients and confidence limits were back-transformed to express results as ratios of geometric means. Crude associations were first examined. The following strategy was then employed to investigate multivariate associations. A simple hierarchical causal diagram was developed (Fig. 2). Socio-demographic factors were considered as potential confounders for the relationship between each exposure and cytokine response, and maternal co-infections (malaria parasitaemia and helminths) were considered as potential confounders for each other and for infant exposures. Treatment with albendazole was considered as a potential effect modifier for maternal hookworm and M. perstans infections, and treatment with praziquantel for S. mansoni infection. Infant co-infections were considered as potential confounders for infant anthropometric exposures.