Four overlooked errors in ROC analysis: how to prevent and avoid

Diagnostic tests are frequently applied within clinical practice to assist with disease diagnosis, differential diagnosis, disease grading and prognosis evaluation. Receiver operating characteristic (ROC) curve analysis is one common approach for analysing discriminative performance of a diagnostic test, where it can determine the optimal cut-off value with the best diagnostic performance.1 However, as a majority of clinicians are non-statisticians, several errors have been observed in clinical research when applying ROC curves. These errors may be misleading in the selection of diagnostic tests and disease diagnosis, thus adding to patient burden. To address these errors, clinicians do not need a deep understanding of the intricate mathematical formulas of ROC analysis, but should develop basic knowledge and skills to prevent or avoid commonly overlooked mistakes. This article aims to guide clinicians to avoid common pitfalls in ROC analysis.