Accuracy and Reproducibility of Nuclear/Cytoplasmic Ratio Assessments in Urinary Cytology Specimens

Background Evaluation of the nuclear to cytoplasmic ratio is commonly used for assessment of the presence of malignancy and for grading and typing of malignant neoplasms. Despite its widespread usage, little information exists regarding the accuracy and reproducibility of non‐automated assessment. Methods Forty‐seven cells obtained from Papanicolaou stained urine cytologies were assessed by quantitative image analysis for nuclear area and cell area. The nuclear/cytoplasmic ratio was calculated. Visual estimates of the N/C ratio were made by six pathologists. Statistical analysis was performed to determine accuracy, precision, and interrater reliability. Results True N/C ratios varied from 0.02 to 0.81. 27% of cases demonstrated a true N/C ratio between 0.5 and 0.7. Quantitative estimates of N/C ratios were less precise and less accurate at high N/C ratios. The coefficient of variation was 27%. The majority of raters demonstrated decreased accuracy and precision of estimates as N/C ratio increased. Overall classification accuracy was 73%. Accuracy of classification was 53% for cases with a true N/C ratio between 0.4 and 0.8. Absolute interrater agreement was 75%. Chance corrected agreement (kappa) was 0.54. Conclusions Visual quantitation of N/C ratio showed only a fair correlation with actual N/C ratio with correlation decreasing with increasing N/C ratio. In the critical range, 0.5–0.7 N/C ratio both interobserver correlation and correlation with true N/C ratio may be insufficiently accurate for precise category assignment as used in the Paris System. Diagn. Cytopathol. 2017;45:107–112. © 2016 Wiley Periodicals, Inc.