Abstract
Purpose :
The World Health Organization (WHO) tracks the prevalence of Trachomatous Inflammation-Follicular (TF) in children aged 1-9 as part of its guidelines for treatment intervention and elimination of trachoma as a public health concern. Grading is subjective and variable, but conjunctival photographs allow for multiple graders to evaluate the same photograph. Trachoma trials have used single graders or a consensus (median) of 3 graders. Here, we use photographs from the multiple graders of the WUHA trial to estimate the sensitivity and specificity of graders and the true village-level prevalence from the baseline of the WHUA trial.
Methods :
In WHUA, 10 graders graded conjunctival photographs from 40 communities. We assumed an underlying true grade and performed a latent class analysis (LCA) to estimate each grader’s sensitivity and specificity for that true grade, as well as each community’s true prevalence.
Results :
The estimate of the mean sensitivity for true clinical activity was 67% (55%-80%), and mean specificity was 75% (58%-85%). The true prevalence correlated with prevalence from a single grader or a consensus of 3 graders (r>.99). The LCA estimate of true prevalence was consistently higher than consensus prevalence because the average sensitivity was lower than specificity for the 10 graders.
Conclusions :
Trachoma grading is known to be inconsistent, but a latent class analysis can account for the sensitivity and specificity of each grader and be used to estimate the true prevalence of clinical activity in any community. In high prevalence areas, using a single grader may systematically underestimate this true prevalence of clinical activity, suggesting some areas are worse off than previously thought. A consensus of 3 graders is closer to the estimated true prevalence, but still lower than the estimated true prevalence. Surveys could consider having multiple photographic graders, or if not feasible, interpret their single grade surveys accordingly.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.