Classification under uncertainty: data analysis for diagnostic antibody testing

Formulating accurate and robust classification strategies is a key challenge of developing diagnostic antibody tests. Methods that do not explicitly account for disease prevalence uncertainty therein can lead to significant errors. We present novel method leverages optimal decision theory address this problem. As preliminary step, we develop an analysis uses assumed conditional probability models measurement outcomes define (in the sense minimizing rates false positives negatives) domains. Critically, demonstrate how strategy be generalized setting in which unknown by either: (i) defining third class hold-out samples require further testing; or (ii) using adaptive algorithm estimate prior also provide examples recently published SARS-CoV-2 serology test discuss (e.g. associated with instrumentation) incorporated into analysis. find our new decreases error up decade relative more traditional methods based on confidence intervals. Moreover, it establishes theoretical foundation generalizing techniques such as receiver operating characteristics (ROC) connecting them broader field optimization.