Probabilistic forecasts of trachoma transmission at the district level: A statistical model comparison

Abstract

The World Health Organization and its partners are aiming to eliminate trachoma as a public health problem by 2020. In this study, we compare forecasts of TF prevalence in 2011 for 7 different statistical and mechanistic models across 9 de-identified trachoma endemic districts, representing 4 unique trachoma endemic countries. We forecast TF prevalence between 1-6 years ahead in time and compare the 7 different models to the observed 2011 data using a log-likelihood score. An SIS model, including a district-specific random effect for the district-specific transmission coefficient, had the highest log-likelihood score across all 9 districts and was therefore the best performing model. While overall the deterministic transmission model was the least well performing model, although it did comparably well to the other models for 8 of 9 districts. We perform a statistically rigorous comparison of the forecasting ability of a range of mathematical and statistical models across multiple endemic districts between 1 and 6 years ahead of the last collected TF prevalence data point in 2011, assessing results against surveillance data. This study is a step towards making statements about likelihood and time to elimination with regard to the WHO GET2020 goals.

Keywords: Elimination; Forecasting; Model comparison; Trachoma.

Fig. 1

District level TF prevalence in each of the 9 districts between 1997 and 2010. Every set of two data points and one line indicates the prevalence data for that district. The red dashed like indicates the forecasted year of 2011.

Fig. 2

Forecast distributions of TF prevalence in 2011 for each of the 9 districts evaluated and for each of the 7 models analysed. Results from model forecasts are shown by a solid line and the true data for 2011 for each district is shown with a black dashed line. The colour of each line represents a different model as indicated in the legend.