Study site

The GEMS sentinel health center for this study was St Elizabeth Mission Hospital in Lwak (henceforth referred to as Lwak Hospital), located in Rarieda sub-county, Siaya County (formerly Nyanza Province) in western Kenya. Lwak Hospital is the designated referral facility for population-based infectious disease surveillance (PBIDS) conducted in the surrounding 33 villages by the Kenya Medical Research Institute (KEMRI) and the U.S. Centers for Disease Control and Prevention (CDC) [30]. The area also falls within the KEMRI/CDC health and demographic surveillance system (HDSS) site in western Kenya [31]. The HDSS provides general demographic and health information including population age-structure, migration, fertility rates, birth and death rates, verbal autopsy, access and utilization of health care for approximately 220,000 inhabitants in 55,000 households. The primary economic livelihood is subsistence farming and fishing, and an estimated 70% of the population lived below the poverty line in 2003 [32]. The area is culturally homogeneous, with 95% of people being ethnically Luo [33]. Households in the PBIDS villages are clustered into compounds composed of related family units, with most compounds having between one and five family units [33]. Animal husbandry is common: 89% of compounds own at least one species of livestock or poultry, with 86% owning poultry (median flock size: 10), 49% cattle (median herd size: 4), 48% goats (median herd size: 4) and 18% sheep (median herd size: 3) (KEMRI/CDC HDSS data for 2008). Among compounds that own livestock, approximately one-half also own cats and/or dogs (International Emerging Infections Program–Zoonoses Project data for 2009). Rodents, including black rats (Rattus rattus), are also commonly present in and around houses in the PBIDS site [34].

GEMS

From January 31, 2008 through January 29, 2011, children 0–59 months old who sought care at selected sentinel health centers (including Lwak Hospital) and belonged to the HDSS population were screened for diarrhea. To be eligible for inclusion in GEMS, the diarrhea episode had to meet the case definition for MSD [29], which was three or more loose stools within the previous 24 h, with onset within the previous 7 days after a period of at least 7 diarrhea-free days, with one or more of the following: sunken eyes; loss of skin turgor; intravenous rehydration administered or prescribed; dysentery; or hospitalized with diarrhea or dysentery. Each GEMS site restricted enrollment to the first nine eligible cases per age stratum per fortnight. Three age strata were targeted: infants (0–11 months), toddlers (12–23 months), and children (24–59 months). For every enrolled case, one to three children without diarrhea were enrolled as controls. Controls were matched to individual cases by age (within 2 months of age for patients aged 0–23 months, and within 4 months of age for patients aged 24–59 months), sex, and residence (same or nearby village as patient). Potential controls were randomly selected from the KEMRI/CDC HDSS database and enrolled during a home visit within 14 days of the matched case. Potential controls who had diarrhea in the previous 7 days were ineligible. At enrollment, primary caregivers (parent or other caretaker) of cases and controls were interviewed to obtain demographic, epidemiological and clinical information. In addition, each case and control provided at least 3 g of fresh stool, which was submitted to the laboratory for identification of enteric pathogens using standard methods as described by Panchalingam et al. [28].

GEMS-ZED substudy

The GEMS-ZED substudy collected and analysed additional data on animal-related factors from a subset of GEMS case and matched control children with reported animal presence at home. From November 4, 2009 through February 4, 2011, all cases enrolled into GEMS at Lwak Hospital were screened for inclusion in the GEMS-ZED study. (Enrollment into GEMS continued for a short period after the official end date of January 29, 2011, during which time 3 case-control pairs were enrolled into GEMS-ZED. Data from the GEMS study [laboratory test results and wealth index] are not available for these 3 pairs.) Between zero and six cases per fortnight (median of two) were enrolled into GEMS at Lwak Hospital during the GEMS-ZED study period. Only cases and controls whose primary caregiver reported presence of animals (domestic animals as well as peridomestic wild rodents) at the child’s compound during the GEMS enrollment interview were considered eligible. For each eligible case, the first eligible GEMS-enrolled matched control was identified, resulting in one-to-one matching in the GEMS-ZED dataset. If no eligible child could be identified among the GEMS set of one to three matched controls, then the case was not enrolled into GEMS-ZED. Caregivers of eligible cases and controls were approached for enrollment into the GEMS-ZED study during a separate home visit that took place within 2 weeks of their enrollment into the GEMS study. Written informed consent for participation in the study was sought from the primary caregiver, as well as from the head of the compound of residence of each eligible child; only compounds in which both individuals provided consent were enrolled. Compounds were excluded if the child participating in GEMS had died subsequent to enrollment, or if no domestic animals were found to be resident (for example, if animals had died or were sold subsequent to GEMS enrollment).

Following enrollment, both the head of the compound and the child’s caregiver were interviewed using a standard questionnaire. The questionnaire consisted of two parts: the first part dealt with residence and husbandry of domestic animals in the compound (livestock, poultry, dogs and cats), as well as observations relating to the presence of rodents in and around the compound, and was asked of the person in the compound responsible for the management of animals (typically the head of the compound). The second part dealt with information specific to the participating child, relating to exposures to animals and their environment, and was asked of the child’s caregiver. A summary of the items included in the questionnaire is presented in S1 Table.

At the enrollment visit, fecal specimens were collected from a convenience sample of domestic animals resident at the compound. Specimens from a single species and age category (young, unweaned animals vs. older animals) were pooled together, with specimens from a maximum of five animals collected in a single pool, and a maximum of two pools per species and age category combination (i.e. a maximum of ten animals per species and age category combination were sampled from a compound). A previous study showed good agreement of bacterial culture results between individual and pooled fecal samples of five individuals per pool [35]. Between 3 and 10 g of feces were collected directly from the rectum of larger animals (cattle, sheep, goats and adult dogs). For smaller animals (cats and young dogs), three moistened cotton-tipped swabs were used to collect samples from the animal’s rectum and placed directly into transport media (two in modified Cary Blair and one in buffered glycerol saline); whole feces were not routinely collected from smaller animals.

For poultry, groups of birds of a single species (chickens or ducks) were confined overnight on a sheet of thick plastic. Owners were asked to confine approximately five birds per group, and not more than two groups of birds per species. Fecal specimens from a single pool of animals were mixed in a stool cup. Following thorough mixing of the pooled feces, two cotton-tipped swabs were inserted into the feces and then placed in a vial containing modified Cary Blair transport medium. A third swab was placed in a vial containing buffered glycerol saline. All specimen containers were clearly labelled and placed in a sealed bag in a coolbox with icepacks for transport to the laboratory.

Identification of potentially zoonotic enteric pathogens in animal specimens (Campylobacter jejuni, Campylobacter coli, non-typhoidal Salmonella, diarrheagenic E. coli, Cryptosporidium, Giardia, and rotavirus) was carried out using an identical protocol to that described for the human stool specimens tested in GEMS [28]. Briefly, bacterial agents were isolated and identified using conventional culture techniques. Three putative Escherichia coli colonies of different morphology types were pooled and analysed by multiplex PCR that detect targets for enterotoxigenic (ETEC), enteroaggregative (EAEC), enteropathogenic (EPEC), and enterohaemorrhagic E. coli (EHEC). The following gene targets defined each E. coli pathotype: ETEC (either eltB for heat-labile toxin [LT], estA for heat-stable toxin [ST], or both), ST-ETEC (either eltB and estA, or estA only), typical EPEC (bfpA with or without eae), atypical EPEC (eae without either bfpA, stx1, or stx2), EAEC (aatA, aaiC, or both), and EHEC (eae with stx1, stx2, or both, and without bfpA). Commercial immunoassays were used to detect rotavirus (ProSpecT Rotavirus kit, Oxoid, Basingstoke, UK), Giardia and Cryptosporidium spp. (TechLab, Inc., Blacksburg, VA, USA). Immunoassays were only performed on whole fecal specimens of adequate volume (≥ 3 g), and were therefore not completed for the majority of cat specimens, because volumes from this species were often inadequate.

To better understand the zoonotic potential, we genotyped Cryptosporidium parasites from immunoassay-positive animal fecal specimens. DNA was extracted from 0.5 ml of fecal specimens using a FastDNA SPIN Kit for Soil (MP Biomedicals, Santa Ana, CA). Cryptosporidium species present were differentiated by PCR-restriction fragment length polymorphism (RFLP) analysis of the small subunit (SSU) rRNA gene, and confirmed by DNA sequencing of the PCR products [36].

Data analysis

Data were analysed using R statistical software version 3.1.3 [37]. We used conditional logistic regression (clogit function applying the exact method in R package ‘survival’ [38]) with one-to-one matching to identify animal-related exposures that were significantly associated with MSD.

Exposure variables were screened for inclusion in the multivariable model using univariable conditional logistic regression. As part of the screening process, each exposure variable was evaluated for potential recoding. Husbandry-related variables for which values were conditional upon residence of the species in question were evaluated and recoded if this made biological sense. For example, the question “Do adult sheep enter the cooking area?’” was conditional on residence of adult sheep in the compound. If no adult sheep were resident, the response was recoded as “No–no adult sheep present” rather than a missing value, and compared against “No–adult sheep present but do not enter cooking area” and “Yes–adult sheep present and enter cooking area”. For these variables, the null state (species not resident) was taken as the reference level. Variables related to exposures of children to animals and their environments were kept as binary variables. For example, the question “Does the child play in an area of the compound where adult sheep defecate?” had one of two responses: ‘no’ if no adult sheep were resident in the compound or adult sheep were resident but the child did not play in the area where they defecated, and ‘yes’ if there were adult sheep resident and the child played in the area where they defecated. For categorical variables with four or more categories, we created new binary variables by combining categories based on frequencies. For example, the original four levels for frequency of observation of rodents or their excreta (never, seldom, often or daily) were dichotomised to never/seldom vs. often/daily. Both the original and new variables were tested in the univariable analysis. Continuous variables (e.g. number of chickens owned) were categorised into three categories [category 1: zero values; category 2: values greater than zero and less than or equal to the median value (excluding zeros); category 3: values greater than the median value (excluding zeros)]; both the original continuous variable and the new categorical variable were assessed in the univariable analysis. Variables with a significant number of missing values (>10% of observations) were discarded. Variables with a Wald test p-value greater than 0.2 on univariable analysis were excluded from further analyses. If both the original and recoded variable had a p-value below the threshold of 0.2, the one with the smaller p-value was retained.

After the univariable screening, we assessed collinearity between the selected exposure variables using condition indices (colldiag function in R package ‘perturb’ [39]). A condition index is a number ranging from 1 to infinity that is computed from data on a set of exposure variables–the higher the condition index, the greater the amount of collinearity [40]. The condition indices were investigated by calculating the variance decomposition proportion (VDP) for each condition index over 30, beginning with the largest. Exposure variables with a VDP >0.5 were considered potentially collinear. In cases where it made biological sense to do so, collinear variables were combined to create a new categorical variable. For example, the collinear variables “Chicken manure used in farm” and “Chicken manure used in the compound” were combined to create a variable “Chicken manure used”. When this did not make biological sense, or when the new variable still exhibited collinearity, the collinear variable with the higher univariable p-value was excluded. Remaining variables were taken forward for consideration in the multivariable conditional logistic regression model.

We compared main effects models using Akaike’s information criterion (AIC), whereby models with a smaller AIC are considered more optimal. We used a forward stepwise regression process to select exposure variables to retain in the model. Missing values were handled through multiple imputation (R package ‘mice’ [41]). Building of the main effects model was stopped when the addition of a variable resulted in an increase in the AIC. We assessed interactions between variables in the final main effects model by adding two-way interaction terms to the model and evaluating their effect on the AIC.

For evaluation of the final model, we identified outliers and influential pairs, using the transformation method described in [42] and applying a Bonferroni outlier test. We computed leverage values and delta β statistics to identify influential pairs (in R package ‘car’ [43]). To determine if these pairs were having an undue effect on the model, we refit the model with them omitted.

In GEMS, a wealth index quintile for households was generated by principle component analysis of thirteen household assets [26, 44]. The wealth index quintile was forced into the final model as an ordinal variable to evaluate the potential confounding effect of wealth.

Ethics statement

The GEMS protocol was approved by the KEMRI Scientific and Ethical Review Committee (protocol no. 1155) and the Institutional Review Board at the University of Maryland, School of Medicine, Baltimore, MD, USA. The Centers for Disease Control and Prevention, Atlanta, GA, USA, formally deferred to the IRB at the University of Maryland for review (protocol no. 5038). Written informed consent was obtained from the parent or primary caretaker of each participant before initiation of study activities. The GEMS-ZED study protocol was approved by the KEMRI Scientific and Ethics Review Unit (protocol no. 1572) and the CDC Institutional Review Board (protocol no. 5683). Written informed consent for participation in the study was provided by the parent or primary caretaker of each participant, as well as from the head of the compound of residence of each participant. Protocols for animal involvement were reviewed and approved by the KEMRI and CDC Institutional Animal Care and Use Committees (protocol no. SSC 1572 and 2088OREMULX, respectively). CDC IACUC protocols comply with the Animal Welfare Act (AWA) regulations promulgated by the United States Department of Agriculture (USDA) under Title 9, Code of Federal Regulations, Parts 1–3 as well as the Public Health Service Policy on Humane Care and Use of Laboratory Animals (PHS Policy) administered by the National Institutes of Health (NIH), Office of Laboratory Animal Welfare (OLAW). In Kenya, all vertebrates are protected under Cap 360 (the Prevention of Cruelty to Animals Act) (1963, revised 1983).

Laboratory results

We collected fecal specimens of acceptable quality for diagnostic testing from 2,174 domestic animals of eight species, resulting in a total of 691 pools (median of 5 and range of 1 to 10 pools per compound). Of these, 159 pools (23%) tested positive for one or more potentially zoonotic enteric pathogens (Campylobacter jejuni, C. coli, non-typhoidal Salmonella, diarrheagenic E. coli, Giardia, Cryptosporidium, or rotavirus). Test results for particular pathogens by host species and age group are given in Table 3. Species with the highest proportion of positive pools for particular pathogens were chickens for C. jejuni [18/231 (7.8%)] and non-typhoidal Salmonella [26/231 (11.3%)]; goats for C. coli [6/106 (5.7%)]; donkeys for diarrheagenic E. coli [1/12 (8.3%)]; dogs for Giardia [19/69 (27.5%)] and Cryptosporidium [4/69 (5.8%)]; and cattle for rotavirus [4/153 (2.6%)].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Test results for potential zoonotic enteric pathogens in pooled fecal samples collected from domestic animals resident in the homesteads of case and control children enrolled in the GEMS-ZED study of moderate-to-severe diarrhea in children in western Kenya.

https://doi.org/10.1371/journal.pntd.0005795.t003

Domestic animals from 45/73 (61%) compounds at which a child with MSD resided tested positive to one or more pathogens, compared with 44/73 (60%) compounds with a control child. There were no significant associations on univariable conditional logistic regression between the presence of particular pathogens in domestic animals residing in compounds, and MSD in the participating child from the compound (Table 4). When considering the children’s GEMS laboratory results, we found 21 instances in which the pathogen identified in the child was also identified in one or more species of domestic animals residing in the compound (Table 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Univariable conditional logistic regression results of pathogens identified in domestic animals resident in compounds of children with and without moderate-to-severe diarrhea enrolled in the GEMS-ZED study.

https://doi.org/10.1371/journal.pntd.0005795.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Instances in which a pathogen identified in a child was also identified in one or more species of domestic animals residing in the child’s compound.

https://doi.org/10.1371/journal.pntd.0005795.t005

Nineteen pooled specimens positive for Cryptosporidium spp. by immunoassay were analysed by PCR, including 14 pooled specimens from chickens, 4 from dogs, and 1 from calves. Among them, 7 chicken specimens and the bovine specimen generated the expected PCR products. RFLP analysis indicated the presence of C. meleagridis in 6 chicken specimens, C. bovis in one chicken specimen, and C. parvum in one bovine specimen. None of the canine specimens analysed were positive by PCR.