Using historical BL data to guide our selection of controls by age and geography, we found that many children attend local health facilities suggesting that it may be feasible to enrol controls at local health facilities. However, substantial difference in the mean age of BL cases versus the mean age of children found at local health facilities was observed. Specifically, 92% of children at local health facilities were aged 0-5 years. This percentage is 2.4 times higher than the proportion of children aged 0-5 years in the general population in Uganda [22
] and 2.6 times higher than children with historical BL in this region [23
]. Based on these differences in age pattern, local health facilities may be appropriate sites for recruiting controls for young BL cases, but they may not be appropriate sites for recruiting controls for older BL cases. Theoretically, one could enrol older controls at local health facilities by prolonging the period of enrolment at the facilities, but such a plan requires one to assume that the few older children who are found at local health facilities during the extended period of enrollment are representative of older children in the general population. Because EMBLEM is being conducted in rural areas, we worry that prolongation of enrollment would lead to a "Hawthorne bias" [24
]. A Hawthorne bias occurs when subjects modify behavior because of actions by the investigator.
Our study also noted that malaria, which is a central exposure under investigation, is highly prevalent in children attending health facilities, especially young children. Given the absence of reliable data on malaria prevalence in children at health facilities versus in the general population, the high prevalence of malaria in children at local health facilities is a concern. Alternative methods of sampling controls were considered. We considered school-based controls, but rejected them because children in Uganda start attending school at age 7 years, thus pre-school children would be excluded. We considered using other cancers as controls, similar to previous case-control studies of BL in Africa [7
], but this idea is not feasible in the EMBLEM study area because other cancers are rare and not well diagnosed. We considered hospital controls, but rejected this idea after reviewing the hospital admissions charts and finding that hospital controls were much younger (mean age, 2 years) than BL cases, and come from villages near the hospital than did children with BL. Finally, we considered neighbourhood controls, as has been used in previous studies [9
]. We were concern about overmatching and lack of mechanism to verify compliance with selection procedures during implementation.
Given these results, the study design was refined to obtain population-based controls sampled directly from their homes in the village. For practical reasons, the controls will be obtained from representative 100 villages randomly selected from all villages in the study region. The villages were selected after stratifying by proximity to water defined as village having a boundary within or ≥ 500 m of a swamp, river, or lake, and by rural and urban status, defined according to population density from the national census [22
]. Stratification was necessary because malaria transmission and BL risk correlated with proximity to water and urban status [27
]. Children in households selected randomly from a household list are approached and matched according to a pre-specified age and sex eligibility criteria will be invited to participate in EMBLEM as controls.
Our current study does not conclusively evaluate the validity of population versus facility-based controls. Thus, we included a pilot component where both population and facility-based controls will be obtained in 12 villages to examine this issue. Specifically, population pilot controls (PPC) selected from the village (90 per village) will be compared to health centre pilot controls (HCPC) enrolled at the HCII (30 per health facility serving the village). These controls will be compared for distribution of asymptomatic, mild malaria, and malaria-resistance genotypes to determine whether one group is invalid and, if so, to what extent.
To conclude, we compared and contrasted age and sex patterns of historical BL patterns with patterns of children attending local health facilities to gain insights about feasibility and appropriateness of local facilities as site to enrol controls for BL. We noted substantial differences in the age distribution of BL cases versus children at local health facilities leading us to conclude that local health facilities might not be suitable for enrolling controls for older BL cases. To address this concern, the study design was refined to use population-based controls from representative number of villages. A pilot component was included to investigate bias that may be related to using different controls.