A Critique of Methods of Sampling and Reporting Pathogens in Populations of Fish

Abstract

We used maximum likelihood methods to estimate an observed or apparent prevalence for a pathogen in a pooled sample of fish and here provide the program code for such calculations using commonly available statistical software. To illustrate the characteristics and variability of prevalence estimates from pooled samples, we explored the relationships among pathogen prevalence, sample size, and method of pooling samples. We calculated the average width of confidence intervals and the mean square error of the prevalence estimator for samples from populations with pathogen prevalence ranging from 1% to 90% using several pooling strategies for samples of 30 and 60 fish. As an illustration, we calculated the confidence interval and apparent prevalence of Myxobolus cerebralis in samples of fish from Utah screened with pooled sampling strategies. When all pools were positive, the apparent prevalence was 100%, but the bounds of the confidence interval ranged from 8% to 100%. Interpretations of data sets that are based only on the results for positive pools may be misleading, as the percentage of pools that is positive when any single pool scores negative is higher than the maximum likelihood estimate of apparent prevalence. The confidence intervals bounding estimates were generally smaller when larger numbers of groups were used and samples had few fish per pool. In populations with higher prevalence, the use of pooled samples significantly enlarges the confidence interval of estimates.