MULTIVARIATE STATISTICAL-ANALYSIS OF BILE CYTOLOGY

Cytologic features predictive of carcinoma in bile fluid were characte rized by a review of 35 cases (22 malignant and 13 benign). Thirty-two cytologic variables were subjected to a stepwise multiple regression model to separate the criteria best suited to tile benign versus malig nant groups. Six key criteria were selected as useful indicators of ma lignancy: loss of honeycomb arrangement, enlarged nuclei, loss of pola rity bloody background, flat nuclei and cell-in-cell arrangement. Thre e or more of these criteria were observed in carcinoma cases more ofte n than they were in noncancer eases (P < .02). By using these criteria , the overall sensitivity of diagnosing carcinoma by bile cytology was 86.4%, and the specificity was 76.9%.