Can we improve the positive predictive value of atypical glandular cells not otherwise specified?

The category of atypical glandular cells (AGC) in gynecologic cytopathology presents many well-documented diagnostic challenges, the most significant related to high interobserver variability, low specificity, and low positive predictive value. The current Bethesda System provides criteria for specific glandular categories including atypical endocervical cells not otherwise specified (AEC-NOS), AEC favor neoplastic, and atypical endometrial cells. The Bethesda System does, however, acknowledge that in some cases AGC cannot be categorized based upon cell of origin, in which case the generic term "atypical glandular cells" (AGC) may be used. We sought to determine whether further refinement of the current Bethesda System criteria for AEC-NOS might increase the positive predictive value of the general category of AGC. Fifty-three cases of AGC with documented histologic follow-up at the University of Illinois Hospital were reviewed. The cases were graded on each of the eight specific cytologic criteria recommended by the current Bethesda System for AEC-NOS using a study-developed three-tier grading system. Multiple regression analysis showed that four of the cytologic criteria in combination--nuclear enlargement, nuclear pleomorphism, increased nuclear-to-cytoplasmic (N/C) ratio, and cells occurring in sheets and strips with cell crowding and nuclear overlap--discriminated positive histologic outcome slightly better than any single criterion alone. In addition, simple logistic regression analysis showed nuclear enlargement to have a marginal independent association with positive histologic outcome (P = 0.0566). No other criterion was independently associated with outcome. Ancillary methods seem indicated to increase the positive predictive value of AGC at this time.