A logistic model for thyroid lesions.

A stepwise logistic regression analysis was performed in 27 cases of papillary carcinoma thyroid (PC), 20 follicular neoplasms (FN), 30 cases of Grave's disease (GD), and 40 cases of colloid adenomatous goitre (CAG). The three most important variables in predicting PC were papillary clusters, dense cytoplasm, and intranuclear cytoplasmic inclusions, whereas the predictors of FN were high cellularity combined with a prominent acinar pattern. A few cases of GD and CAG showed a cytologic overlap with PC and FN, respectively. Regression analysis established high cellularity, fire flare appearance, and epitheloid granulomas as reliable predictors of GD, whereas abundant colloid with or without foam cells and follicles associated with colloid (FAC) were important variables in CAG.