Abstract
Prevalence is a common epidemiological measure for assessing soil-transmitted helminth burden and forms the basis for much public-health decision-making. Standard diagnostic techniques are based on egg detection in stool samples through microscopy and these techniques are known to have poor sensitivity for individuals with low infection intensity, leading to poor sensitivity in low prevalence populations. PCR diagnostic techniques offer very high sensitivities even at low prevalence, but at a greater cost for each diagnostic test in terms of equipment needed and technician time and training. Pooling of samples can allow prevalence to be estimated while minimizing the number of tests performed. We develop a model of the relative cost of pooling to estimate prevalence, compared to the direct approach of testing all samples individually. Analysis shows how expected relative cost depends on both the underlying prevalence in the population and the size of the pools constructed. A critical prevalence level (approx. 31%) above which pooling is never cost effective, independent of pool size. When no prevalence information is available, there is no basis on which to choose between pooling and testing all samples individually. We recast our model of relative cost in a Bayesian framework in order to investigate how prior information about prevalence in a given population can be used to inform the decision to choose either pooling or full testing. Results suggest that if prevalence is below 10%, a relatively small exploratory prevalence survey (10-15 samples) can be sufficient to give a high degree of certainty that pooling may be relatively cost effective.