Standardized model for diagnosing cervical carcinoma in situ based on the cytologic signs

An attempt was made to optimize the recognition of cervical intraepithelial carcinoma in situ (CIS) by the adoption of a standard set of rules derived from a case-control study of 50 histologically confirmed incidence cases of CIS and a simple random sample of 100 controls from the same mass screening population examined between 1977 and 1984. One cytotechnologist screened the slides for all occurrences of a standard set of classic cytopathologic signs. Chromatin clumping, dyskeratosis, anisonucleosis and an increased nuclear-cytoplasmic ratio in cells from the deep layer and chromatin clumping and anisonucleosis in cells from the superficial layer showed an association with the diagnosis of CIS, in decreasing order of magnitude. These variables make up a binary discriminant model; adding vacuolization, anisocytosis, prominent nucleoli and multinucleation to this model did not appear to improve its discriminatory power. The discrimination function accommodates any a priori judgment regarding the probability and utility of diagnostic outcomes. Cytologists normally manipulate the probabilities and utilities intuitively; the model requires the translation of intuitive judgments into quantitative estimates. For example, in our cytology laboratory, the CIS pick-up rate is approximately 1 per 1,000 smears; a 1:1,000 negative utility ratio of a false positive versus a false negative is thus used in our model. Given this, our model has a sensitivity of 86% and a specificity of 96% upon reclassification, and its predictive power (true positives/false positives) is 21.5. This standardized cytodiagnosis model is relevant to any local diagnosis-specific probability and utility estimates.