Determinants of Visceral Leishmaniasis: A Case-Control Study in Gedaref State, Sudan

Background Improving knowledge on local determinants of visceral leishmaniasis (VL) is crucial to guide the development of relevant control strategies. This study aimed to identify individual and household level determinants of primary VL in 24 highly endemic villages of Tabarak Allah hospital’s catchment area, Gedaref State, Sudan. Methods From September 2012 to July 2013, in an unmatched case-control design, 198 patients with primary VL were compared to 801 controls free of VL symptoms and with a negative VL rapid test. Using random spatial sampling, controls were selected with a distribution of age, sex and village of residence proportionate to the distribution of the target population. Data were collected using a structured questionnaire. Results Children and men were at higher risk of VL. Reporting VL patient(s) in the household in the previous year was the strongest VL risk factor. In a multivariate analysis, VL risk increased with household size, sleep location (outside the yard, not in the farm), evening outdoor activities in the rainy season (playing, watching TV, radio listening), use of ground nut oil as animal repellent and of smoke of Acacia seyal as indoor repellent, presence of dogs in the yard at night, Acacia nilotica in the yard’s immediate surroundings and of a forest at eye range. VL risk appeared to decrease with the use of drinking water sources other than the village water tank, a buffer distance from the adjacent house yard, and with the presence of animals other than dogs in the yard at night. In contrast with previous studies, housing factors, mosquito-net use, black cotton soil, ethnicity, socioeconomic index, presence of Balanites aegyptica and Azadirachta indica in the yard were not independent VL determinants. Discussion and conclusion Although these results do not provide evidence of causality, they provide useful suggestions for guiding further intervention studies on VL preventive measures.


Introduction
Visceral leishmaniasis (VL) is a devastating illness with a high case fatality rate when left untreated. Gedaref state is the main endemic area of VL in Sudan where the detection figures vary largely between villages in relation with average rainfall and altitude [1]. Villages with higher incidence are clustered along the Atbarah and Rahad rivers [1], near the Ethiopian border. Since 2009, Médecins Sans Frontières (MSF) is offering diagnosis and treatment for VL patients in Tabarak Allah, near the Atbarah river (Fig 1).
In Eastern Sudan, VL is caused by the parasite Leishmania donovani and transmitted through the sandfly Phlebotomus orientalis [2,3]. P. orientalis sandfly populations tend to peak at the beginning of the rainy season [2,3] and are concentrated in areas with high densitiy of Acacia seyal (locally named "Taleh") and Balanites aegyptica trees ("Lalob" or "Higleeg") that grow on vertisols (black cotton soil) [4]. P. orientalis is believed to bite mainly outdoors, in the evening or in the early morning [4], though some populations of the vector may be more adapted to indoor biting [5]. Although some animals, especially dogs [6], have been shown infected by L. donovani in eastern Africa, their role in transmission is unclear and the disease is believed to be mainly anthroponotic [7,8]. In Sudan, post-kala-azar dermal leishmaniasis (PKDL) cases, corresponding to some 50% of treated cases, might act as a reservoir for parasites and play a role in human-to-human transmission [9,10]. Host factors increasing susceptibility to VL include malnutrition and HIV [11].
In the absence of a vaccine, VL preventive measures aim at reducing the parasite reservoirs (human and, where relevant, animal) and at limiting human exposure to sandflies notably through vector control [12]. Risk factors for VL transmission have been studied in Asia, where poverty and housing conditions appeared to consistently influence the risk of VL [13,14]. However, the important continental differences in terms of reservoir, ecology, parasites and vectors involved in VL transmission limit the generalizability of these results to eastern Africa. In Ethiopia [15][16][17], Kenya [18] and Uganda [19], factors such as proximity to dogs, sleeping outside, under an acacia tree, or in a thatch house, and a low socio-economic status were reported as possible risk factors for clinical VL. In Sudan, younger age, male gender, and ethnicity [20,21] appeared associated with an increased risk of VL while Azadirachta indica (locally named "Neem") [21] and use of bednets [15,22] appeared offering some protection. To date in Sudan, VL transmission and its possible individual VL risk factors remain poorly understood. However, further knowledge of local VL determinants is essential to design appropriate control activities. Therefore, we conducted a case-control study in endemic villages of Tabarak Allah area in order to identify individual and household level determinants of primary VL.

Study setting and target population
The study was conducted in the catchment area of the MSF Tabarak Allah Hospital, in Al-Gureisha locality, the main provider of VL treatment in the area. The study target area included the 24 villages (out of 45 villages) most affected by VL according to a survey on burden of VL undertaken in 2011 [23], and closest to Tabarak Allah (Fig 1). The population of the study area was estimated at 46,564 individuals in June 2011, with a VL incidence rate estimated at 9.8/ 1000 person-years from June 2010 to June 2011 [23]. The vegetation consists in dry savannah woodland (with Acacia seyal and Balanites aegyptica trees) combined with black cotton soil. The rainy season usually lasts from June to October.

Study design and sample size
This was an unmatched case-control study. For practical reasons, we expected to enrol cases and controls over a single dry season. Based on the VL incidence rate estimated from June 2010 to June 2011 [23] and accounting for patients that would be missed by the hospital-based passive recruitment, we expected to recruit a bit less than 300 cases. A sample size of 270 cases and 810 controls allowed to detect as significant an odds ratio (OR) of 1.6 with an expected 20% of exposure among controls and a power of 80% (α: 0.05, two-sided).

Sampling procedures
Cases were recruited in Tabarak Allah Hospital. All successive patients presenting clinical signs of VL (ie history of fever of two weeks or more; and either splenomegaly, lymphadenopathies or wasting) and who tested positive to a rk39-based rapid test (BioRad IT-Leish) [24] and/or a direct agglutination test (DAT; titre 1:6400) [25,26] and/or had a positive lymph node aspirate [27] were eligible for recruitment if they were at least 6 month-old and if they had been resident of a village of the target area since at least one year. Patients previously treated for VL were excluded. All eligible cases (or their caregivers for individuals younger than 15 years) were invited to sign a written informed consent for participating in a structured interview during a home visit in the month following discharge. HIV-co-infection was not systematically assessed but was expected to be low [28].
Controls were selected from the population with a distribution of age, sex and villages of residence proportionate to the distribution observed in a survey in 2011 [23]. Control houses were selected randomly using a geographic sampling procedure. The perimeter of the inhabited area of each village was recorded in the field using global positioning system (GPS) devices and saved as shapefiles. In each village perimeter, a pre-specified number of GPS random points, according to the 2011 population distribution, were generated. Control selection started in the nearest house from each point. If several houses were at the same distance, one house was selected at random (using a random number table). In each selected house, one individual of a pre-specified sex and age category (to fit the 2011 distribution) living in this house since at least one year, free of clinical symptoms of VL (including PKDL) and with no history of previous treatment for VL was randomly selected. After written informed consent, each potential control underwent a BioRad IT-Leish test. Individuals with a positive result were excluded. If the rapid test was negative, a structured interview was conducted. If there was no eligible control in the first house, the control selection was continued in the nearest one. In case of absence, a second visit was planned within the week. The recruitment started with the dry season on the 15 th of September 2012 and ended with the beginning of the rainy season on the 15 th of June 2013. We aimed to ensure a similar timing of recruitment for cases and controls. Interviews of controls took place between the 22 nd September 2012 and the 9 th of July 2013 and all villages were regularly visited over the all study period.

Data collection
Cases and controls were invited to answer a structured questionnaire on possible determinants of VL (S1 File). It was designed for an exploratory approach and it included data on demographic characteristics, socio-economic factors, personal medical history, travel history, housing conditions, daily activities (with a focus on evening activities), description of house yard and immediate surroundings including animal presence, and occurrence of VL in the house or the adjacent house. When relevant, different answers were allowed for the rainy and the dry seasons. The cases and controls were asked to describe their usual behaviour during the current dry season (for cases, before they developed VL symptoms) and the last rainy season.
Questionnaires were translated in Arabic and were piloted before the start of the interviews. The interviewers were trained to administer the questionnaires in a standardised way.

Data analysis
Data were double-entered and analysed using Stata 12 software (Stata Corporation, College Station, Texas, USA). Using a principal component analysis, a socioeconomic index including data of ownership among controls (radio, TV, bicycle, motorcycle, car/truck/tractor, mobile phone, donkey cart, and generator) was built. Each variable in the index was given a coefficient corresponding to eigenvectors of the first and second components, weighted for their respective contribution. The index was divided into quintiles.
Variables were grouped by predefined thematic sections (demographics, travel history, activities and sleeping habits, characteristics of the household, the house, house yard and surroundings) and by season. The thematic sections are presented in the supplementary material (S1 Table), where individual level and household level determinants are distinguished. In each section, the variables associated with VL (p<0.20) after adjustment for age, sex and village taken as a random effect (due to the high number of villages) were combined in mixed multivariate logistic regression models using a stepwise forward approach. Variables associated with VL (p<0.20) from the different sections were then combined to build one multivariate model with potential VL risk factors at the individual level and another multivariate model with household-level risk factors. Variables not associated with VL (p20%) were excluded. Finally all variables were combined into a multivariate logistic model in which variables significant at the 5% level were kept. An interaction between age and sex (likelihood ratio test for interaction p = 0.027) was kept in the model. The variables related to the presence of VL patients in the household or the adjacent house(s) were kept separate because they were likely on the causal pathway between VL and the other determinants.

Ethics statement
Authorization to conduct the study was obtained from the Gedaref Ministry of Health, Humanitarian Affairs Commission, and the village authorities. Ethical clearance was obtained from the MSF ethical review board and National Research Ethics Review Committee of the 'Health Research Council" (National Ministry of Health). All adult subjects provided informed consent, and a parent or guardian of any child participant provided informed consent on their behalf. Each individual signing the consent form to participate as a control was offered a mosquito net. Individuals with VL clinical symptoms were referred to Tabarak Allah for further diagnostic. Because of the possibility of a not yet clinical infection, asymptomatic individuals with a positive rk39 test were followed-up by phone during 6 months and encouraged to consult Tabarak Allah hospital in case they developed persistent fever in the months following the study.

Results
Of the 321 primary VL patients admitted in Tabarak Allah during the study period, 208 were coming from the study area, corresponding to a passive detection rate of 6.0/1000 person-years (ranging from 0 to 16.4 person-years across villages). Nine did not live in the same place during the past year and one patient died, resulting in 198 cases interviewed. Of the 849 eligible individuals selected as controls, 3 refused participation, 21 were absents despite two revisits, nine presented VL clinical symptoms and 15 were rK39 positive (of which 4 developed VL during the follow-up, 5 remained asymptomatic, and 6 were lost to follow-up). In total, 801 controls were included in the analysis.
Considering socio-demographic characteristics (Table 1), cases were younger than controls and more often male. A third of cases and controls belonged to the Massalit, other major ethnic groups being Zabarma, Fallata, Hausa (26.2% of controls) and Tama (20.1% of cases). Cases originated mainly from Berber Al-Fugera (29.8%) and Tabarak Allah (21.7%), compared to 21.4% of the controls. There was no evidence for a difference of socio-economic index category between cases and controls. Univariate analysis comparing all cases and controls characteristics is presented in the supplementary material (S1 Table).
After adjustment on age, sex, their interaction and village as random effect, there was no evidence (p0.20) for an association between VL and black cotton soil (surrounding the yard of 85.4% of the cases and 74.7% of the controls), proximity of water bodies, housing materials, use of mosquito nets, indoor spraying, travel history, day activities, and long term medical treatment. All variables that remained associated with VL (p<0.20) after adjustment are presented in the Table 2 with OR and 95% confidence intervals (95%CI). Among participants of schooling age or older, cases were more frequently attending school. Between sunset and bedtime, the majority of the participants stayed exclusively outdoor in the dry season (S1 Table) while, in the rainy season, 30.4% of controls and 26.8% of cases stayed indoor only (Table 2). In the rainy season, outdoor evening activities (playing, watching TV/radio listening) were associated with an increased risk of VL compared to staying indoor. Most participants reported to sleep within the house yard, although 2.9% of controls and 7.6% of cases reported to sleep outside the house yard but within the village (for example, some children slept at school). During the rainy season, 15.5% of controls and 8.6% of cases reported to sometimes sleep in the field/farm outside the village. In the dry season, about half of the participants were sleeping outside a room while, in the rainy season, more than 90% of all participants were sleeping inside a room. Cases belonged to bigger households than controls and tended to share the sleeping room with more people than controls. Presence of cracks in the room's wall (made of either strong mud, plastered or unplastered cane) tended to increase the risk of VL. Water sources were concordant in the dry and the rainy season (Cohen's kappa 0.75) and differed between cases and controls, the main source of drinking water being surface water for controls while cases tended to drink more often water from the village water tank. Trees tended to be more frequent in and around cases yards but, after adjustment, only the presence of Azadirachta indica and Acacia nilotica appeared more likely associated (p<0.05) with an increased risk of VL. Increasing distance between the house yard and the adjacent yard tended to reduce the risk of VL while the presence of a forest at eye range from the house appeared to increase the risk of VL. At night, in both seasons, animals (donkeys, goats, sheep, cattle and dogs) were reported in the majority of participants' yards. The presence of dogs in the yard at night was associated with an increased risk of VL while the presence of other animals in the yard appeared rather associated with a reduced VL risk. The use of smoke from Acacia seyal as indoor repellent in the rainy season was associated with an increased risk of VL. Also, repellent for animal (chemical, local tar, smoke from grass, natural oil) was frequently used but only the use of ground nut body oil was associated with an increased risk of VL.
The final multivariate model combining independent individual and household VL determinants is presented in the Table 3.
As shown in the Table 4, during the year prior the study, 7.4% of controls compared to 71.7% of cases had at least one household member who suffered from VL (p<0.001) and, among those, cases had more often 2 to 4 household members affected (21.8% versus 10.2%). Before that, VL was reported to have occurred in the households of about 74% of all participants. In the year prior the study, a patient with VL in an adjacent house was reported by 30.8% of the controls and in 42.6% of the cases. Before that period, VL was reported to have occurred in a neighbour of 95.6% of the controls and 90.2% of the cases. Adjusting for individual and household determinants did not affect the size of the associations between VL and previous VL patients in the household or an adjacent house (Table 4). When combining the variables related to VL patients in the household and the adjacent houses in a multivariate model, reporting a VL patient(s) in the household in the previous year remained the only independent VL determinant.

Discussion
In this study, living in the proximity of VL patients was the strongest VL risk factor. Three out of four cases reported having a household member sick with VL in the last year compared to less than one out of ten controls. This has also been reported in the Indian subcontinent [13,29] and for asymptomatic VL in Ethiopia [30]. It underlines the importance of studying  further the transmission dynamics between patients and their household contacts [31]. However, the role of PKDL as potential reservoir remains unclear. Indeed, reporting one or more household member(s) sick with VL in the past year was associated with a lower OR if the household member(s) also developed a rash compatible with PKDL. This might suggest that individuals with PKDL are less infectious than individuals with active VL but this might also result from misclassification of rashes as being PKDL although these were due to other diseases. As previously reported, the risk of VL was higher in males and the young age groups [21] and tended to increase with household size which logically increases the number of possible cases in the household [17]. However, this increased VL risk with the household size might have been overestimated due to the sampling procedures that tended to favour controls from smaller households (by selecting only one control per house). However, important overestimation is unlikely because weighting the final multivariate model to account for the number of eligible controls per house did not change the estimation of the OR nor its precision (S2 Table). By contrast [20,21], ethnicity was not associated with VL after adjustment despite the Fallata/ Hausa/Zabarma ethnic group being less represented among VL cases. This does not support the hypothesis of a genetic predisposition independent from environmental or behavioural factors. Most ethnic groups settled in this VL endemic area a long time ago possibly reducing the difference in terms of exposure and development of natural immunity.
Also in contrast, there was no evidence for an association between VL and neither the socioeconomic [15,19,32] nor housing factors [15,17]. Cracks in the walls, an important factor in Ethiopia [17], increased the risk of VL in adjusted but not in the multivariate analysis. This supports a predominant role of outdoor biting, as does the association with VL risk and outdoor evening activities (such as watching TV/listening to the radio or playing). This was more obvious during the rainy than the dry season during which most inhabitants remain unprotected outdoor in the evening. This outdoor exposure during the known biting-time and season of the sandflies might explain why bed-nets did not appear protective against clinical VL, adding to the mixed results reported in the literature [15,33]. Innovative preventive measures deserve further consideration. For example, it might be interesting to evaluate the effectiveness of targeted personal protection strategies, such as the use of adapted chemical repellent use or insecticide-impregnated clothing for household members of VL cases or to inhabitants of villages currently experiencing an increase in VL cases. Although the most frequent soil around the participants' yards, there was no association with VL and black cotton soil, whose cracks might offer breeding sites to sandflies [34]. The predominance of black cotton soil in the study area might have limited the possibility to find difference between cases and controls and, consequently, an association with VL. Sandy soil in or around the house yard appeared protective in the univariate analysis but not after adjustment. Soil composition is quite homogenous at village level, and in the yard, the soil is often covered with gravels or sand filling the cracks. This might possibly limit sandflies breeding opportunities and might explain sleeping outside the house yard as a VL risk factor. Also, a buffer space between house yards appeared to offer some protection, possibly against transmission from VL cases living in neighbouring yards.
Having a forest, a likely natural focus of sandflies, at eye range from the house yard appeared as a risk factor for VL as well as the presence of Acacia nilotica in its immediate surroundings. However, Balanites aegyptica, Acacia seyal [4,20] and Azadirachta indica [21], that have been reported associated with VL, were not in this study after adjustment. Only the use of smoke of Acacia seyal as indoor repellent appeared as an independent determinant of VL. During the time of the survey, about a third of the villages investigated were involved in planting Azadirachta indica or cutting Balanites aegyptica and Acacia seyal. Therefore reverse cause bias cannot be excluded because such activities might have been more intense around the houses of cases. Also, the surrounding vegetation, and microecological conditions influencing sandflies density, might have been different in the farm and the village. Consequently, sleeping in the field or the farm was associated with a decrease in VL risk, while the opposite was found in Ethiopia [17,35]. The association between having a dog in the yard at night and the risk of VL reinforces the role for dogs in sustaining VL transmission in eastern Africa [16,36]. Asymptomatic dogs have been found carriers of Leishmania in Sudan [37,38]. The presence of other animals in the yard at night was found protective. In Ethiopia, P. orientalis were shown to feed mainly on cattle [39] although none of the sandflies caught in that study was infected with Leishmania. Presence of animals or their ownership has alternatively been reported either as VL risk factor or protective factor [13,14,17,19,21]. Consequently, it is important to study locally the possible zoonotic transmission (especially with dogs) and to analyse the bloodmeal of sandflies caught in the house yards of the area. The use of ground nut oil as an animal repellent was associated with an increased risk of VL. The use of insecticide for livestock has also been reported as a VL risk factor in Kenya [19]. This suggests that ground nut oil either repel sandflies that consequently feed on another unprotected host, or actually attracts sandflies to the area. This finding needs confirmation but suggest that other animal repellents should be explored as alternative to ground nut oil.
The data were collected through a structured questionnaire by surveyors carefully trained to ensure standardization of its administration. However, a questionnaire might not always capture accurately the variation in population behaviours. Surveyors were not blinded of the participant status and cases were asked to recall the time period before the onset of the case VL symptoms. This might have contributed to inaccuracies and possible recall or observer bias. Also, because numerous variables were investigated and multiple statistical tests done, random findings cannot be excluded. Notably, the association observed between VL and the source of drinking water is unclear and might result from multiplicity of statistical testing or residual confounding due to imperfect adjustment on villages of residence.
Cases were recruited in the Tabarak Allah hospital. Patients with VL that did not attend this health facility were not included and this might have limited the comparability of cases to community control. However these likely represented only a small fraction of all VL cases occurring in the area where even habitants living in decentralised villages tend to present to Tabarak Allah hospital spontaneously [23]. Also, patients from nomadic population or seasonal workers were not included and this might limit the generalizability of the results. Although an important population at risk in some area [15], they represented less than 5% of the primary VL cases in Tabarak Allah during the study period.
Finally, this study compared VL clinical cases to non-infected controls. Consequently, it evaluated potential determinants for both infection and progression towards VL clinical disease but cannot disentangle these two aspects. In order to fully evaluate determinants for infection, we should also have included cases with subclinical infection. However, laboratory methods to identify subclinical infections are imperfect, which would have led to misclassification. Similarly, the duration of positive rK39 in subclinical infection is unknown and it is thus likely that individuals with past subclinical infection have been included as controls, possibly leading to an underestimation of the strength of some associations.
In conclusion, because of the exploratory approach, these results do not provide strong evidence of causality but still, they provide useful suggestions for guiding further intervention studies on VL preventive measures. Control strategies would benefit from further research on VL transmission within house yards and across villages. In the meantime, efforts should target households of patients diagnosed with VL.