Validation of a chart to estimate true Schistosoma mansoni prevalences from simple egg counts

Schistosoma mansoni egg counts by faecal examination vary considerably and are not very sensitive, so prevalences are underestimated. The distribution of egg counts can adequately be described by a stochastic model which distinguishes variation in counts between persons and variation in repeated counts within a person. Based on this model a pocket chart has been developed which predicts the proportion of individuals harbouring at least 1 S. mansoni worm pair – the ‘true prevalence’ – from a simple single survey prevalence and geometric mean egg count (using common duplicate 25 mg Kato–Katz smears). The current paper describes the validation of this chart by comparing predicted true prevalences with prevalences observed after 5–7 repeated Kato–Katz faecal examinations (Burundi), by examination of a large quantity of stool using the Visser filter (Brazil) or a selective sedimentation–filtration method (Surinam). Because 5–7 repeated examinations do not suffice to measure all infections, predictions have been made of the cumulative proportion positives over 5–7 surveys – the ‘approximate true prevalence’ – as well. After dividing the data into age groups, 12 different subsets were considered for validation. In all 12 cases, predicted true prevalences (or approximate true prevalences for the Burundi data) agree well with those observed. The overall agreement depends only slightly on the assumed relationship between worm numbers and mean egg counts, with a good fit for a productivity between 0·8 and 4·4 eggs per gramme faeces (EPG) per worm pair (WP). This interval includes the most plausible value from the literature, i.e. 1·0 EPG/WP, which has been applied in the initial pocket chart. These findings support the validity of the chart to predict true prevalences for a wide range of productivity assumptions, and reinforces the applicability of its underlying stochastic model to describe egg count variation. However, as predictions appear to vary importantly when using only part of the data, it is also concluded that the pocket chart never compensates for limited validity of initial single survey prevalences and geometric means in consequence of small sample sizes.