Potential for bias in (sero)prevalence estimates when not accounting for test sensitivity and specificity: a systematic review

Abstract

Objectives The COVID-19 pandemic has led to many studies of seroprevalence. A number of methods exist in the statistical literature to correctly estimate disease prevalence or seroprevalence in the presence of diagnostic test misclassification, but these methods seem to be less known and not routinely used in the public health literature. We aimed to examine how widespread the problem is in recent publications, and to quantify the magnitude of bias introduced when correct methods are not used. Design: A systematic review was performed to estimate how often public health researchers accounted for diagnostic test performance in estimates of seroprevalence. Using straightforward calculations, we estimated the amount of bias introduced when reporting the proportion of positive test results instead of using sensitivity and specificity to estimate disease prevalence. Results Of the seroprevalence studies sampled, 78% (95% CI 72–82%) failed to account for sensitivity and specificity. Expected bias is often more than is desired in practice, ranging from 1–12%. Conclusions Researchers conducting studies of prevalence should correctly account for test sensitivity and specificity in their statistical analysis.