Endosonographic differentiation of benign and malignant stromal cell tumors.

Background: Endosonography (EUS) is a valuable technique for diagnosing gastrointestinal stromal cell tumors. However, EUS features that are predictive of malignancy in these tumors have not been defined.

Methods: Videotapes and photographs of EUS examinations performed prior to surgical resection of 35 stromal cell tumors (9 malignant) were blindly reviewed by a single examiner. EUS features associated with malignancy were determined. Interobserver agreement in interpreting these features was then measured among a panel of five expert endosonographers who judged EUS videotapes of 35 resected stromal cell tumors (10 malignant).

Results: Stepwise logistic regression analysis demonstrated that tumor size (diameter > 4 cm), irregular extraluminal border, echogenic foci, and cystic spaces were independently associated with malignancy in stromal cell tumors (p < 0.05). Interobserver agreement for irregular extraluminal border, echogenic foci, and cystic spaces, as measured by mean kappa statistic, was 0.43, 0.39, and 0.28, respectively. For the five experts, the sensitivity for detecting malignancy ranged between 80% to 100% when at least two of the three features were judged to be present. The likelihood of finding malignancy ranged between 0% to 11% for the experts when all three features were judged absent.

Conclusions: Tumor size and certain EUS features are useful for predicting malignancy in stromal cell tumors. Absence of these features indicates benign disease. Agreement among experts in interpreting these EUS features is fair to moderate.