Process-based bayesian modelling of imperfect diagnostic tests in the absence of a "gold" standard : applications to egg-count tests for helminth infections

Schistosomiasis and soil-transmitted helminthiasis are among the most prevalent infections worldwide with approximately 2 billion infected individuals that cause a considerable disease burden primarily in sub-Saharan Africa and South-East Asia. The World Health Organization has set a goal to achieve elimination as a public health problem by 2025 by provision of mass drug administration with praziquantel to school-aged children.
The decision of where and how often to treat is based on the prevalence of eggs in either a thick smear of stool or a urine sample. However, the diagnostic techniques that depend on the detection of eggs have a low sensitivity for light infections and therefore underestimate the ‘true’ prevalence of worms in the population. Missed light infections become increasingly important when approaching elimination of the diseases. Therefore, alternative diagnostic approaches have been evaluated that do not depend on the detection of eggs and hence, perform better for light infections.
Using alternative diagnostic approaches in practice calls for methods setting observations from different diagnostic methods in relation, ensuring consistent treatment decisions and impact assessment using historical data.
This work presents hierarchical Bayesian models developed to combine data from different diagnostic methods. They allow us to learn about the diagnostic sensitivity and specificity of various diagnostic methods and to determine thresholds for treatment decisions based on each of the methods that lead to consistent outcomes.