Interpretation of diagnostic cytology with likelihood ratios.

Abstract

Sensitivity, specificity, and predictive values are commonly used to report the performance characteristics of a diagnostic test. With tests that by their nature have more than two classes of results, calculation of sensitivity and specificity demands the combination of categories into two classes, leading to a loss of diagnostic information. Fine-needle aspiration cytology of the breast is usually reported using four categories: definitely malignant, suspicious, benign, and unsatisfactory. We used likelihood ratios to show that fine-needle aspiration cytology of the breast appreciates each of these four categories fully. This approach is preferred above the usual one, which combines "definitely malignant" and "suspicious" categories as abnormal and often discards unsatisfactory aspirations from calculation. Knowledge of the prior probability of breast cancer and the likelihood ratio for a particular class of results of fine-needle aspiration cytology of the breast allows for easy calculation of the posttest probability of the disease. Application of this method should lead to more rational support from fine-needle aspiration cytology for clinical decision making.