Targeting trachoma control through risk mapping: the example of Southern Sudan

Abstract

Background: Trachoma is a major cause of blindness in Southern Sudan. Its distribution has only been partially established and many communities in need of intervention have therefore not been identified or targeted. The present study aimed to develop a tool to improve targeting of survey and control activities.

Methods/principal findings: A national trachoma risk map was developed using Bayesian geostatistics models, incorporating trachoma prevalence data from 112 geo-referenced communities surveyed between 2001 and 2009. Logistic regression models were developed using active trachoma (trachomatous inflammation follicular and/or trachomatous inflammation intense) in 6345 children aged 1-9 years as the outcome, and incorporating fixed effects for age, long-term average rainfall (interpolated from weather station data) and land cover (i.e. vegetation type, derived from satellite remote sensing), as well as geostatistical random effects describing spatial clustering of trachoma. The model predicted the west of the country to be at no or low trachoma risk. Trachoma clusters in the central, northern and eastern areas had a radius of 8 km after accounting for the fixed effects.

Conclusion: In Southern Sudan, large-scale spatial variation in the risk of active trachoma infection is associated with aridity. Spatial prediction has identified likely high-risk areas to be prioritized for more data collection, potentially to be followed by intervention.

Figure 1. Observed prevalence of active trachoma using data from population-based prevalence surveys, in children aged 1–9 years, Southern Sudan, 2001–2009.

Figure 2. Semivariograms related to risk mapping models for active trachoma in children aged 1–9 years, Southern Sudan, 2001–2009.

A) raw data and Person's residuals of: B) model 1; and C) model 2. Models 1 and 2 refer to the models presented in Table 1.

Figure 3. The median of the posterior distribution for predicted prevalence of active trachoma in children aged 1–9 years, Southern Sudan, 2001–2009.

Figure 4. Upper quartile of posterior distribution of predicted prevalence of active trachoma in children aged 1–9 years, Southern Sudan, 2001–2009.

Figure 5. Lower quartile of posterior distribution of predicted prevalence of active trachoma in children aged 1–9 years, Southern Sudan, 2001–2009.

Figure 6. Spatially structured residual variation in prevalence of active trachoma in children aged 1–9 years, Southern Sudan, 2001–2009, after accounting for rainfall, land cover and age and sex of survey participants.

Estimates were derived using a geostatistical random effect.

Figure 7. Predicted probability that prevalence of active trachoma is >10% in children aged 1–9 years, Southern Sudan, 2001–2009.

Estimates were obtained from posterior predictive distributions derived using model-based geostatistics.