Active trachoma and community use of sanitation, Ethiopia

Abstract Objective To investigate, in Amhara, Ethiopia, the association between prevalence of active trachoma among children aged 1–9 years and community sanitation usage. Methods Between 2011 and 2014, prevalence of trachoma and household pit latrine usage were measured in five population-based cross-sectional surveys. Data on observed indicators of latrine use were aggregated into a measure of community sanitation usage calculated as the proportion of households with a latrine in use. All household members were examined for clinical signs, i.e. trachomatous inflammation, follicular and/or intense, indicative of active trachoma. Multilevel logistic regression was used to estimate prevalence odds ratios (OR) and 95% confidence intervals (CI), adjusting for community, household and individual factors, and to evaluate modification by household latrine use and water access. Findings In surveyed areas, prevalence of active trachoma among children was estimated to be 29% (95% CI: 28–30) and mean community sanitation usage was 47% (95% CI: 45–48). Despite significant modification (p < 0.0001), no pattern in stratified ORs was detected. Summarizing across strata, community sanitation usage values of 60 to < 80% and ≥ 80% were associated with lower prevalence odds of active trachoma, compared with community sanitation usage of < 20% (OR: 0.76; 95% CI: 0.57–1.03 and OR: 0.67; 95% CI: 0.48–0.95, respectively). Conclusion In Amhara, Ethiopia, a negative correlation was observed between community sanitation usage and prevalence of active trachoma among children, highlighting the need for continued efforts to encourage higher levels of sanitation usage and to support sustained use throughout the community, not simply at the household level.


Introduction
It has been estimated that, as a result of trachoma, approximately 1.2 million people are blind and a further 1.7 million have low vision. 1 Globally, trachoma remains the leading infectious cause of blindness. In 2009, an estimated 40.6 million people had active trachoma and 8.2 million had trichiasis -i.e. the blinding stage of the disease. 2 About 77% of those living in trachoma-endemic areas of the world are to be found in 29 of the countries in the World Health Organization's (WHO's) African Region, and Ethiopia is the country most affected by trachoma worldwide. 3 Trachoma is caused by ocular infection with a bacterium: Chlamydia trachomatis. Inflammation attributable to repeat infections during childhood constitutes the disease's active stage. This inflammation may then lead to scarring of the conjunctiva and trichiasis. In trichiasis, the eyelashes rub and damage the cornea, causing pain and, eventually, blindness. Trachoma is predominantly found in resource-poor, rural communities in low-income countries. 4,5 By afflicting some of the most deprived people in the world, it leads to disability, dependency and further poverty. 6 A pilot programme for trachoma control, begun in four districts of the Amhara region in north-west Ethiopia, was scaled-up so that the programme covered the whole of Amhara by 2007. 7,8 This ongoing programme, based on WHO's SAFE strategy, was set in Amhara because this region has the highest burden of active trachoma within Ethiopia. 9 The SAFE strategy is a comprehensive WHO strategy -based on the available relevant biological and epidemiological evidence -to treat and prevent trachoma. It combines four measures: (i) surgery for the correction of trichiasis; (ii) antibiotics, given in mass drug administrations, to reduce the infection reservoir in the community; (iii) facial cleanliness, to reduce transmission; and (iv) environmental improvements -e.g. control of flies through sanitation and improved access to water for hygiene -for further reductions in the potential for transmission. 10,11 The presence of latrines or other facilities for the disposal of human faeces is understood to have an indirect beneficial effect on the risk of trachoma because it reduces the access of Musca sorbens flies -a probable vector of C. trachomatis -to potential breeding sites. [12][13][14] As flies can easily move throughout an area, however, a few scattered latrines may have little impact on trachoma in that community -or even in households with latrines. 6 Effective fly control through sanitation requires not only high levels of latrine access but also consistent latrine use throughout the community. 15 Programmes to improve sanitation and/or control trachoma typically measure changes in sanitation at regional or national level. If they do investigate latrine coverage at community level at all, they tend to record latrine access rather than use. The role of latrine use -at community level -in trachoma control requires elucidation. 16 We believed that, rather than latrine access at household level, the proportion of households in a community with latrines in use would be a stronger indicator of the effectiveness of fly control and levels of exposure to C. trachomatis. 15 We hy-Objective To investigate, in Amhara, Ethiopia, the association between prevalence of active trachoma among children aged 1-9 years and community sanitation usage. Methods Between 2011 and 2014, prevalence of trachoma and household pit latrine usage were measured in five population-based crosssectional surveys. Data on observed indicators of latrine use were aggregated into a measure of community sanitation usage calculated as the proportion of households with a latrine in use. All household members were examined for clinical signs, i.e. trachomatous inflammation, follicular and/or intense, indicative of active trachoma. Multilevel logistic regression was used to estimate prevalence odds ratios (OR) and 95% confidence intervals (CI), adjusting for community, household and individual factors, and to evaluate modification by household latrine use and water access. Findings In surveyed areas, prevalence of active trachoma among children was estimated to be 29% (95% CI: 28-30) and mean community sanitation usage was 47% (95% CI: [45][46][47][48]. Despite significant modification (p < 0.0001), no pattern in stratified ORs was detected. Summarizing across strata, community sanitation usage values of 60 to < 80% and ≥ 80% were associated with lower prevalence odds of active trachoma, compared with community sanitation usage of < 20% (OR: 0.76; 95% CI: 0.57-1.03 and OR: 0.67; 95% CI: 0.48-0.95, respectively). Conclusion In Amhara, Ethiopia, a negative correlation was observed between community sanitation usage and prevalence of active trachoma among children, highlighting the need for continued efforts to encourage higher levels of sanitation usage and to support sustained use throughout the community, not simply at the household level.
pothesized that higher community sanitation usage would be associated with a lower prevalence of active trachoma.

Study overview and subjects
Between 2011 and 2014, the Amhara Regional Health Bureau conducted trachoma-impact surveys in various areas of Amhara. These surveys were designed to provide population-based estimates of trachoma prevalence, quantify uptake of trachoma control efforts and estimate the proportions of households with water and sanitation access. 17 A districtknown locally as a woreda -only became eligible for surveying when at least five annual rounds of mass administrations of azithromycin had occurred. Each survey was conducted at least six months after the last such antibiotic administration in the target district.
For the present study, we combined data collected in trachoma-impact surveys that were conducted in the South Gondar zone in June-August 2011, 18,19 in the North Gondar and West Gojjam zones in May-June 2012, in eastern Amhara between December 2012 and January 2013, in western Amhara in June-July 2013, and in eastern Amhara in January-February 2014.
All five surveys used multistage cluster random sampling to estimate the district-level prevalence of trachomatous inflammation-follicular. Villages -known locally as gott -represented the smallest administrative unit with available population data and were the primary sampling units. Within each target district, villages were selected, from a geographically-ordered listing, using probability-proportional-to-size sampling. Within each selected village, smaller administrative units of approximately 40 households -i.e. household clusters that were locally called development teams -were used as segments for a modified segment design. 20,21 Such clusters were listed and given identification numbers upon arrival in a study village with the assistance of a designated village representative, who then drew a number from a hat to select the cluster to be surveyed. The entire village was surveyed if it consisted of 40 or fewer households.
Community information was collected in interviews with village leaders. In each selected cluster, all residents who gave verbal consent were examined for clinical signs of trachoma, according to WHO guidelines. 17 Each eye was examined separately, by a trained trachoma grader using a 2.5x binocular loupe, for the presence or absence of all five clinical signs of the simplified trachoma grading system. 22 Heads of household were interviewed for demographic and socioeconomic information as well as knowledge and practices regarding trachoma, hygiene, sanitation and water. Visual inspections were made of household latrines and hand-washing stations. Responses were recorded electronically using tablet computers with Swift Insights software (The Carter Center, Atlanta, USA). Questionnaires at community, household and individual levels were linked. 23

Measures
The exposure variable -community sanitation usage -was calculated as the proportion of households within the cluster with a latrine with evidence of use. A latrine was considered to be in use if there was a defined path to it and faeces were observed in the pit. 24 For trachoma, the outcome variable was a dichotomous measure, based on WHO's simplified grading scale, for the absence/presence of active trachoma -i.e. absence/presence of trachomatous inflammation -follicular and/or intense. 17

Analyses
We used multilevel logistic regression to estimate the association between the proportion of households in each cluster with a latrine in use and active trachoma among children aged 1-9 years -accounting for dependence of observations nested within households and clusters. Multilevel analysis can assess the influence of area-level effects -e.g. community sanitation usage -on an individual outcome in addition to between-group and within-group variability. 25 These variance-based measures provide a useful complement to standard measures of association for the analysis of contextual effects. 26 Accounting for study design and unequal selection probabilities, means and proportions were estimated, with 95% confidence intervals (CI), across categories of community sanitation usage. Generalized linear mixed models were fitted, specifying random intercepts for cluster and households nested within clusters. Models were estimated using adaptive quadrature with eight integration points, and robust standard errors (SE) were requested to account for clustering within districts. Sampling weights, based on the inverse selection probability for cluster, household and individual, were incorporated. Individual and household weights were scaled to sum to the household and cluster sample size, respectively. 27 After scaling weights, to restrict analysis to the subpopulation of interest, participants who were not aged 1-9 years were assigned an individual weight of 0.0001. 28 Our treatment of the exposure variable as a categorical measure -rather than a linear or quadratic measure -was based on the results of a preliminary assessment that considered both fit and interpretability. An empty model was fitted to measure between-cluster and between-household variance. 29 Potential confounders were identified, from the community-level, household and individual covariates recorded in the surveys (Table 1), based on the results of a literature review, an evaluation of directed acyclic graphs, 31,32 univariable analyses and initial unweighted modelling. We used a sequential modelling approach to explore confounding -indicated by change in exposure estimates -and changes in residual variance. All models controlled for survey round, to account for year and possible betweensurvey variations in the method. Most surveys were conducted during the rainy season, so an assessment of seasonality was not possible. We calculated intraclass correlation coefficients by converting individual-level and arealevel components of variance to the same scale -using the latent variable method -and median odds ratios (ORs) were calculated as measures of residual variance on the OR scale. 29,33 To evaluate multiplicative modification of the effect of community sanitation usage on active trachoma, by household latrine use and water access, we used likelihood ratio tests, after controlling for covariates included in the fully-adjusted model. We calculated summary measures of association with CI from the weighted means of stratum-specific log ORs. 34,35 Models included all observations with information available for included covariates. Initial unweighted modelling was conducted in SAS 9.4 (SAS Institute, Cary, United States of America). All described analyses were conducted

Ethical approval
The protocols for the surveys that were our data sources were approved by Emory University's Institutional Review Board and the Amhara Regional Health Bureau. Our secondary analysis was exempt from additional review.

Results
Of 56 Table 1 summarizes overall, and by category of community sanitation usage, the community, household and individual characteristics of children aged 1-9 years with eye examination results. In general, compared with other children, those in communities with relatively low sanitation usage had indicators of poorer hygiene, more impoverished and rural living conditions and less education and health care. In terms of mean household counts of wealth indicators, communities with the lowest category of sanitation usage appeared much poorer than communities with the highest category of such usage ( Table 1).
Levels of exposure to mass administrations of antibiotic for trachoma control and levels of prevention knowledge mirrored patterns of community sanitation usage. For example, children in communities with the lowest category of sanitation usage were less likely to have ever received antibiotics during mass administrations than children in communities with the highest category of such usage ( Table 1). The median number of times that all community residents had reportedly received such antibiotics was lower in communities with the lowest category of sanitation us-age than in communities with the highest category of usage. Some trachoma prevention knowledge was reported by more than half of households (Table 1) but such knowledge was less frequently reported in communities with relatively low sanitation usages. Where latrines were present, latrine usage was high. Among households, an estimated 52% (95% CI: 50-53) owned a pit latrine, the primary form of sanitation recorded. Of these pit latrines, 93% (95% CI: 92-93) were classified as in use based on observation. The mean reported age of latrines was 2.59 years (95% CI: 2.49-2.70), based on data from three of the surveys that recorded such data.
We fitted several models, described as models 1-6, to examine the association between community sanitation usage and active trachoma -sequentially controlling for selected community, household and individual factors and adjusting for survey (Table 2). In terms of the available covariates, populations for all six models were similar. In model 1, which only adjusted for survey round, community sanitation usages of 60 to < 80% and ≥ 80% were associated with lower prevalence odds of active trachoma compared with usage of < 20%. Adding adjustments for child's age and sex (model 2) and then household water access and latrine use (model 3) did not meaningfully change the estimated ORs. After inclusion of household wealth indicators and education (model 4), an aggregated community wealth measure (model 5) and a community measure for the median number of times that a mass antibiotic administration had been received (model 6), the pattern remained the same but was attenuated towards null (Table 2).
In the empty model, which excluded any covariates, residual variances between households and between clusters were estimated to be 0.68 (SE: 0.08) and 2.22 (SE: 0.20), respectively. Intraclass correlation coefficients for the same household and for different households in the same cluster were calculated as 0.47 and 0.36, respectively. In this study, median ORs indicate the extent, always greater than or equal to 1,  Ref. Ref.
Ref. Results were weighted to account for the unequal probabilities of selection and down-weight survey participants excluded from the assessment of trachoma prevalence because they were aged < 1 year or > 9 years. Model 1 was adjusted for survey round. Model 2 was like model 1 but also adjusted for sex and age of the child-with age centred at 5 years. Model 3 was like model 2 but also adjusted for household access to bathing water within a journey of < 30 minutes and household latrine with evidence of use. Model 4 was like model 3 but also adjusted for household access to electricity, household education, household ownership of mobile phone, radio and/or television and iron roof. Model 5 was like model 4 but also adjusted for community mean household wealth. Model 6 was like model 5 but also adjusted for the median number of times surveyed residents in a community had reportedly received a mass drug administration against trachoma.

Random effects
b Usage was measured as the proportion of households with a latrine with evidence of use.
to which the individual probability of active trachoma was determined by cluster and household levels. In a comparison of children of different households from different communities, the median OR was calculated to be 5.07. In a comparison of children of different households from the same community, the corresponding ratio was 2.19. Variance did not meaningfully change across additional models ( Table 2). Residual heterogeneity between clusters was of greater relevance than community sanitation usage. In model 6, median ORs indicated that residual heterogeneity between children of different communities reflected, on average, a 5.82-fold increase in the individual odds of active trachoma. The magnitude of association between community sanitation usage and active trachoma was found to vary significantly by household latrine use and water access (P < 0.0001). As no clear pattern in stratified OR estimates and CIs was discerned, summary estimates, weighted by population in the strata of household latrine use and water access, are reported by category of community sanitation usage in Table 3.

Discussion
Our study shows that increasing the proportion of households in a community with latrines in use may be protective against active trachoma among children aged 1-9 years, independent of whether a child's household had a latrine in use or better access to water and controlling for potential confounders. There was no clear evidence of multiplicative modification of the effect of community sanitation usage on active trachoma by household latrine use and water access. Multilevel analysis, which allowed estimation of residual variation between communities and households, indicated the importance of additional contextual factors -beyond community sanitation usage and other measures that we included in our models -that may have more influence on an individual's propensity for active trachoma.
Studies in the Gambia identified the fly, M. sorbens -that breeds in openlydeposited faeces of humans and other mammals but not in pit latrines -as an insect vector of trachoma, clarifying the relationship between faeces in the environment and the disease. [12][13][14] Subsequently, in a randomized controlled trial, fly catches from children's eyes Ref.
CI: confidence interval; OR: odds ratio; Ref.: reference group; sOR: summary odds ratio. a Because the household has no latrine or has a latrine considered to be unused. b Summary OR weighted by total population within strata of household water and sanitation access, adjusted for age, sex, household education, household wealth items, community mean sum of household wealth indicators, median number of times surveyed residents in a community had reportedly received a mass drug administration against trachoma and survey round.

Research
Community sanitation usage and trachoma, Ethiopia William E Oswald et al. and mean active trachoma prevalence were reduced through latrine provision -but not by a statistically significant amount. 36 In another randomized controlled study -designed to measure the effect of latrine promotion on reemergence of trachoma after a mass administration of antibiotic -there was no evidence of a significant relationship between increased latrine provision and prevalence of active trachoma or C. trachomatis infection in children because there was no rapid re-emergence of infection in either study arm. 37 In a later cohort analysis of the communities that had received a single mass azithromycin distribution and promotion of latrine usage, it was found that, for each 10% increase in the proportion of household latrines with evidence of use 12 months after baseline, there was a 2.0% decrease (95% CI: 0.2-3.9) in the community prevalence of ocular C. trachomatis infection over the subsequent year. 38 However, no corresponding decrease in the prevalence of active trachoma was observed, 38 perhaps because the follow-up period was too short to allow the beneficial impact of a cleaner living environment on the occurrence of active trachoma to become apparent. 39 Based on available data, most latrines observed in the surveys we used had been in place for more than 12 months.
In the control of neglected tropical diseases, the relative importance of hygiene, sanitation and water components, and of household sanitation compared with community sanitation, remains to be established. 16,40 Improvements in access to water supplies could lead to increased facial cleansing, one of the four components of the SAFE strategy, by increasing the quantity of water available at household level. However, we did not identify any pattern of difference in the association between community sanitation usage and active trachoma by household water access and latrine usage, despite statistically significant interaction.