17 - Important Assumptions

As with conventional methods for estimating prevalence, pooled testing methods and bayesian methods for estimating prevalence depend on a number of important assumptions. Violation of these assumptions will result in biased estimates. Key assumptions are that:

• the outcome is assumed to follow a binomial distribution - clustering or overdispersion of the positive outcome can cause substantial bias in the resulting estimate;
• the health status of each individual is independent of the status of others, both within and between pools;
• sampling of the population is by simple random sampling;
• individual samples are allocated into groups (pools) by random selection;
• sample size is small relative to the population being sampled;
• assumed values for sensitivity and specificity are appropriate;
• dilution of individual samples by pooling has no effect on sensitivity or specificity estimates (or the estimates used take any dilution effects into account);
• samples are assumed to be mixed homogeneously in the pools and any sub-samples taken for testing are equally representative of all of the individuals contributing to each pool;
• all pools represent the same number of individuals (except for the variable pool-size method);
• assumed prior distributions for prevalence, sensitivity and specificity for Bayesian methods are appropriate; and
• assumed true prevalence for sample size calculations and simulations is appropriate.