PropertyValue
is nif:broaderContext of
nif:broaderContext
is schema:hasPart of
schema:isPartOf
nif:isString
  • This research was conducted in two rural counties in Sichuan, China where schistosomiasis reemerged following the reduction of human and bovine infection prevalence below 1%, the Chinese Ministry of Health threshold for transmission control [12]. In 2007, we selected 53 villages in 3 counties where schistosomiasis had reemerged for a longitudinal study of social and environmental determinants of schistosomiasis reemergence [17]. One county was excluded from follow-up because it was severely impacted by the 7.9 magnitude earthquake that struck Sichuan May 12, 2008. The analysis presented here includes the 36 villages in the two counties followed through 2010. The names and exact locations of the study villages have been withheld to protect the privacy of study participants and promote candid reporting. In rural Sichuan, populations fluctuate due to rural-to-urban migration, as well as marriages, births and deaths. We therefore employed an open cohort design, conducting a census in 2007 and 2010 in order to identify all residents living in the study villages, age 6 and older. All individuals identified in each census were recruited for S. japonicum infection testing and household surveys. In 2007, we identified 2,891 eligible residents. In 2010, we identified 2,287 residents, including 1,875 identified in the previous census and 412 new residents. Of the 1,016 people identified in 2007 that were no longer residents in 2010, 760 had left their village for work, 203 had left to attend school, 33 had died and 20 had left for marriage. The research protocol was approved by the Sichuan Institutional Review Board and the University of California, Berkeley, Committee for the Protection of Human Subjects. All participants provided written, informed consent before participating in this study. All children provided assent and their parents or guardians provided written, informed permission for them to participate in this study. Everyone testing positive for S. japonicum was notified and provided treatment with 40 mg/kg praziquantel by the anti-schistosomiasis control station. Household interviews. In the summer of 2007 and 2010, the head of each household was invited to complete a structured interview about agricultural practices, sanitation access and socio-economic status (SES). Survey instruments were pilot tested in the study region and administered by trained public health workers fluent in the local dialect. We interviewed 1,156 households in 2007 and 951 households in 2010. Participants reported all crops planted in the past 12 months, the quantity of night soil applied to each crop and whether chemical fertilizers were used. The amount of night soil used by each household was calculated as the total quantity of night soil applied to all crops in the past 12 months. An additional metric of night soil use was also developed for supplemental analyses in light of evidence that improved toilet designs reduce the number of viable helminth eggs in effluent [18, 19]. Households with a working anaerobic biogas digester or triple compartment septic tank were classified as having improved sanitation, and night soil applied by such households was classified as coming from improved sanitation sources. Night soil applied by all other households was classified as coming from unimproved sources. SES was assessed using an asset-based approach. Household assets can provide a more stable estimate of long-term wealth than monetary income in agrarian regions, where income is episodic and includes agricultural products [20, 21]. The index included ownership of eight durable goods (car, motorcycle, tractor, computer, television, air conditioner, washing machine and refrigerator), reported by the head of household. Principal components analysis was used to create an aggregated SES score, deriving weights from the first principal component which explained 28% of the total variance between measures. Multiple imputation by chained equations was used to impute missing household survey data. Multiple imputation avoids biases due to exclusion of incomplete cases and reduces the variance introduced by the uncertainties of imputation through the generation of multiple datasets [22, 23]. The method assumes that data are missing at random—that missingness can be explained by the other, observed variables used in the imputation. We imputed the quantity of night soil used, the number of bovines owned by the household, SES score, whether the household had an improved toilet and the area of land cultivated by the household. Because the volume of night soil used and the number of bovines per household were highly right-skewed, predictive mean matching was used for the imputation of these variables. We imputed the missing data using the aforementioned variables, including observations from prior/future time points and county of residence, generating 20 imputed datasets. Imputed values were used for the 19% of households missing each of these survey measures, including the 482 households not interviewed and 13 households with incomplete survey data. S. japonicum infection testing. Everyone identified in each census was invited to submit three stool samples on three consecutive days in November/December of 2007 and 2010 for S. japonicum infection testing. Each sample was examined using the miracidium hatching test [24]. Briefly, 30 g of stool were suspended in aqueous solution, strained with copper mesh to remove large particles and strained with nylon mesh to concentrate schistosome eggs. The sediment was re-suspended in water and left undisturbed in a room with ambient temperatures between 28 and 30°C. Samples were examined two, five and eight hours after sample preparation for the presence of miracidia. One stool sample per person was also examined using the Kato-Katz thick smear procedure [25]. Three slides were prepared from each sample, using 41.7 mg stool per slide. Slides were allowed 24 hours to clear and were examined for S. japonicum eggs by trained technicians using a dissecting microscope. Stool samples were delivered to county laboratories daily for analysis. A person was classified as infected if any test was positive for S. japonicum. Estimation of village-level night soil use: Because night soil use by one household may impact the S. japonicum infection risk of other village residents, we wanted to include both village- and household-level night soil use in our statistical models. However, using a village-level measure that is an average of all household-level measures results in a given individual’s household night soil use appearing in the model twice—once as part of the average, and again as an individual-level variable. This endogeniety leads to theoretical challenges. Using an outcomes based causal framework, we typically define the effect of an exposure, X, on an outcome as the difference in mean outcomes when the population is uniformly at X = a vs. X = b (where a and b are any combination of exposure levels, only one of which is observable) for some target population with a specific distribution of confounders [26]. But the overlap of household and village-level variables makes it difficult to evaluate changes in one without changes in the other. To avoid this problem, we defined village-level night soil use as the average amount of night soil applied by all households in the village excluding the index household. This allows for theoretical considerations of the effect of changes in village-level night soil use holding an individual’s household-level night soil use constant and vice versa. We employed a two-step approach to evaluate the relationship between night soil use and human infection. First, we examined the association between village-level night soil use and human S. japonicum infection using a multi-level, fixed-effect logistic regression model, modeling village-level night soil use as a categorical variable to allow for non-linear relationships between the explanatory variable and the outcome. Tests for trend were conducted by treating the categorical variable as ordinal. Models were run separately for each study year. Potential confounding variables were selected a priori based on prior evidence and the plausibility of a relationship with both the outcome of interest and night soil use. Models adjusted for participant age, county, household night soil use, bovine ownership, village bovine density (the mean number of bovines per household), area of land cultivated by the household in the past year, agricultural intensity in the village (mean area cultivated in the past year per household), household SES score and village SES (mean household SES score). Village SES, bovine density and agricultural intensity were estimated separately for each household, excluding the index household, as described above. Occupation was not included as a potential confounder, as greater than 95% of adults in the region reported their occupation as farmer in 2007 [17]. Given the large set of potential confounders and the uncertainties in variable selection, we defined a reduced set of variables that we strongly suspected to be confounders (age, sex, county and household-level night soil use) and ran all models twice, once using the reduced set and once using the full set of confounders. We accounted for correlation within villages using generalized estimating equations and exchangeable working correlation, calculating robust variance estimates [27]. Second, we evaluated the potential impact of interventions to reduce the use of night soil on schistosomiasis. To do this, we estimated a parameter akin to attributable risk using a population intervention model approach [28, 29] that we call the intervention attributable ratio. The intervention attributable ratio is defined as E(YA)/E(Y), where E(YA) is the expected prevalence of infection in the study population if night soil use were eliminated, and E(Y) is the observed prevalence of infections at the observed levels of night soil use. For the purpose of the model, we assume that all infections were acquired recently (a reasonable assumption given the frequency of mass and targeted chemotherapy in this population), that our measure of night soil use is relevant to the infection risk period, and that the observed statistical associations between night soil use and schistosomiasis prevalence represent a causal relationship. We explore the strengths and limitations of these assumptions in the discussion. G-computation was used [28–30]. We fit a fixed-effect logistic regression model assuming an independent correlation structure to allow for a population-level prevalence estimate. Based on the results of the first analysis, we included the limited set of potential confounders, modeled each year separately and modeled village night soil use as a continuous variable. This model was used to calculate infection probabilities for each individual surveyed when household and village night soil use were reduced to zero. Inference was estimated by bootstrapping: the population was sampled with replacement by village to obtain a 36 village population, the model was re-fit, E(Y) was estimated as the observed infection prevalence in the resampled population and E(YA) was estimated as the predicted infection prevalence when household and village night soil use were set to zero, as above, repeating this procedure 1,000 times. The 2.5th and 97.5th percentile values used to estimate the 95% confidence interval. All analyses were conducted using Stata version 12 (College Station, TX).
rdf:type