Study design and procedures
Full details of the study population, design, and outcomes were previously described Kephaet al. [15]. The trial was conducted in 23 purposively selected schools between January 2013 and October 2014 in Bumula District, Bungoma County, Western Kenya. All children in classes 1–6 (typically aged 5–15 years) with informed consent from a parent or legal guardian were asked to provide a single stool sample, which was examined in duplicate for the presence of hookworm, A. lumbricoides, and T. trichiura eggs using the Kato-Katz method.
The trial originally recruited 1 505 children with detectable STH infections and 841 uninfected children [15]. Enrolled children were randomly assigned to one of two treatment groups, either (i) a single dose of 400mg albendazole (Zentel®, GlaxoSmithKline South Africa, Cape Town) at baseline and a single 250mg dose of vitamin C (Cosmos Limited, Nairobi) at 4, 8, and 12 months, or (ii) a single dose of 400mg albendazole every 4 months for 12 months. Cross-sectional surveys investigating the participating children’s infection status and intensity (egg counts) were carried out at baseline, and 7, 11, and 15 months.
Anthropometric and nutritional status data
At baseline, each child’s weight was measured to the nearest 0.1kg using an electronic balance, and height was measured to the nearest 0.1cm using a portable fixed base stadiometer. Hemoglobin concentration was assessed using a hemoglobin photometer (HemoCue® Hb 201+ System, Ångelholm, Sweden). Anthropometric indices for nutritional status at baseline included z-scores of height for age (HAZ), weight for age (WAZ), and body mass index for age (BMIZ), and were calculated using the World Health Organization (WHO) AnthroPlus software Stata macro for children aged 5–19 years [16]. Age was self-reported as it was logistically difficult to collect exact birth dates, and because there were doubts over precision a mid-year age was assumed. Children were classified as stunted, underweight, or thin if their HAZ, WAZ, and BMIZ scores were below -2 standard deviations from the reference median. To investigate potential influences of assuming the mid-year age, we conducted a sensitivity analysis using the lowest and highest possible exact ages of children (e.g. 8.0 and 8.9 for the midpoint age of 8.5).
Household data
At enrolment, a household questionnaire was administered to parents/guardians to collect information on the construction materials of their houses (wall, floor, and roof); sources of fuel; mobile phone ownership; and level of education of the household head. These factors were used to generate a wealth index based on a principal component analysis (PCA) [17], which was then divided into two groups (poor and less poor) based on median (see Additional file 2: Table S1). Household-level access to water and sanitation was assessed by direct observation and included information on source of drinking water and presence of a pit latrine.
School data
School-level data on WASH were collected by interviewing the head teacher or deputy head teacher, and by visual inspection, using questionnaires and checklists developed for a previous study in Kenya [18]. Conditions of school sanitation facilities were assessed based on observed ‘cleanliness’ of the latrine, presence of visible feces, excessive smell, and excessive flies combined by PCA (Additional file 2: Table S1). The ratio of children per latrine was determined as an indicator for access to sanitation, and was calculated by dividing the number of enrolled children by the number of latrines available in the school. We also asked the head teacher about the source of drinking water, availability of water, and availability of soap and handwashing facilities near latrines.
School locations were mapped using a handheld eTrex 20 global positioning system (Garmin Ltd., Olathe, KS, USA). Estimates of land surface temperature, aridity index, enhanced vegetation index, elevation, and normalized difference vegetation index were determined for each school after averaging the values of covariates within 1-km catchment area. A detailed description of sources and the pre-process of environmental data are provided in Additional file 2, Section 2.
Statistical analysis
Quantification of treatment impact
The impact of repeated (4-monthly) treatment was assessed based on data from 579 children, who were infected at baseline with any STH and presented at the 15-month follow-up point. The proportions of children with hookworm, A. lumbricoides, and T. trichiura infections together with 95% confidence intervals (CIs) were calculated at baseline and at the 15-month follow-up point using binomial regression analysis. Clustering of infection by school was taken into account by estimating clustered robust standard errors as children in the same school may have similar risk of STH infection compared with children from different schools. Intensity of infection was measured as eggs per gram (epg) of feces, and the arithmetic mean epg with 95%CIs was estimated using negative binomial regression taking school clustering into account. Reductions in infection levels between baseline and the 15-month follow-up point were investigated using a mixed-effects logistic model (for infection status) and a negative binomial regression model (for intensity of infection), with the individual outcomes at the two time points treated as a repeated measures outcome and a random intercept for schools. Additionally, relative reductions were calculated as the percentage difference between the proportions of children infected, or the mean intensity at baseline and at the 15-month follow-up point.
To assess the efficacy of repeated (4-monthly) treatment in relation to a single annual treatment (as delivered through the Kenyan national deworming program), treatment success was also quantified among the single treatment group (562 children) using the same statistical methods as outlined above. To assess comparability between children in the two study arms, summary statistics were calculated for all individual, household, and school-level characteristics pertaining to the children, by treatment group. To quantify the added benefit of repeated (4-monthly) treatment compared to standard deworming, absolute differences in the proportion of residual infections at 15 months between the treatment groups were estimated together with 95% CIs using prtest in Stata. (STATA Corp, College Station, Texas, USA) This approach was chosen as, due to trial randomization, any difference in baseline infections between the treatment groups would be due to chance. Therefore, a difference in reduction over 15 months can be directly observed in residual infections.
Predisposition to infection
Predisposition to infection was investigated based on data collected from 1,141 children in the 4-monthly treatment group at baseline and at the 7-, 12-, and 15-month follow-up points. Two non-parametric rank correlation tests (Spearman’s rank and Kendall’s Tau) were used for the pairwise comparison of infection intensities in children in all observation rounds from baseline to the 15-month follow-up point for hookworm and A. lumbricoides. For both tests, we rejected the null hypothesis (absence of predisposition) if P-values were ≤ 0.05.
Factors associated with residual infection following repeated treatment
Residual infections assessed 3 months after the delivery of the fourth treatment dose provide an indication of either rapidly occurring reinfections or a lack of parasite clearance after treatment. The analysis was based on 579 children in the 4-monthly treatment group; outcomes were the proportion and intensities of hookworm and A. lumbricoides infections at the last follow-up (15 months after the baseline assessment). Risk factors associated with hookworm and A. lumbricoides infections at 15 months were investigated using mixed-effects regression models (logistic regression for infection status and negative binomial regression for intensity) with a random school intercept. All statistical models were adjusted for baseline individual infection (infection status for logistic regression and intensity of infection for negative binomial regression). We first investigated associations by including one covariate at a time (unadjusted analysis). Variables with significant associations (P ≤ 0.05, based on a likelihood ratio test) were combined into an adjusted regression model, which was then reduced to a final model using a backwards variable selection approach, eliminating one variable at a time based on the highest P-value and retaining only variables in which P ≤ 0.05 (adjusted analysis).
Ethical considerations
The study was approved by the Kenya Medical Research Institute Ethics Review Committee (SSC 2242), the London School of Hygiene and Tropical Medicine (LSHTM) Ethics Committee (6210), and the Makerere School of Public Health Institutional Review Board (IRB00005876).
Written informed consent was obtained from a parent or guardian of each child, and assent was sought from children before enrolment into the study. A questionnaire was administered to parents/guardians to collect information on household socioeconomic characteristics.