Recommendations for reporting measures of diagnostic accuracy.

Abstract

This article serves as the third in a series that offers recommendations for optimal data reporting, specifically focusing on the statistical methods most frequently reported in Cytopathology articles. Measures of diagnostic accuracy were among the most commonly reported results in Cytopathology, with sensitivity and/or specificity reported in 32% of the reviewed articles. This article will provide a brief overview of common measures of diagnostic accuracy, including sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios, overall diagnostic accuracy, receiver operating characteristic (ROC) curve, and area under the curve (AUC). Reporting recommendations for these measures will be reviewed, including the suggestion of providing a 2 × 2 contingency table when possible, or numerator and denominator values for calculations when all values needed for a contingency table are not known, and calculation of ROC and AUC if relevant. Additionally, paired measures should be reported, including sensitivity with specificity, positive with negative predictive values, and positive with negative likelihood ratios. Calculating 95% confidence intervals (CI) for the measures is recommended, with several methods to choose from, including the Wald interval, Wilson interval, Clopper-Pearson interval, Agresti-Coull interval, and Bayesian highest posterior density (HPD) interval. Since there are various methods for CI calculations, the author encourages the reader to consult with a trained statistician to identify the most appropriate method based on the data, which should be reported in the methods section of the resulting write-up.