Spatially regulated designs of incidence surveys can match the precision of classical survey designs whilst requiring smaller sample sizes: the case of snakebite in Sri Lanka

Abstract

AbstractBackgroundSnakebite envenoming is a neglected tropical disease. Data from the worst affected countries are limited because conducting epidemiological surveys is challenging. We assessed the utility of inhibitory geostatistical design with close pairs (ICP) to estimate snakebite incidence.MethodsThe National Snakebite Survey (NSS) in Sri Lanka adopted a multistage cluster sampling design, based on population distribution, targeting 1% of the country’s population. Using a simulation-based study, we assessed predictive efficiency of ICP against a classical survey design at different fractions of the original sample size of the NSS. We also assessed travel distance, time taken to complete the survey, and sensitivity and specificity for detecting high-risk areas for snake envenoming, when using these methods.ResultsA classical survey design with 33% of the original NSS sample size was able to yield a similar predictive efficiency. ICP yielded the same at 25% of the NSS sample size, a 25% reduction in sample size compared to a classical survey design. ICP showed >80% sensitivity and specificity for detecting high-risk areas of envenoming when the sampling fraction was >20%. When ICP was adopted with 25% of the original NSS sample size, travel distance was reduced by >40% and time to conduct the survey was reduced by >75%.ConclusionsThis study showed that snakebite envenoming incidence can be estimated by adopting an ICP design with similar precision at a lower sample size than a classical design. This would substantially save resources and time taken to conduct epidemiological surveys and may be suited for low resource settings.Key messagesInhibitory geostatistical design with close pairs (ICP) for incidence surveys can match the precision of classical survey designs at lower sample sizes.The ICP design showed a lower predictive variance than classical design, indicating ICP designs were able to produce more reliable predictions.The ICP design showed a lower time to complete the survey than the classical sampling method.Although primary sampling units in the ICP design maintain a minimum distance between two units, ICP did not increase travel distance compared to the classical survey design.Resource requirements and time to complete surveys can be reduced without increasing the distance to travel by adopting ICP design for epidemiological surveys.