TY - JOUR
T1 - Endosonographic (EUS) differentiation of benign and malignant smooth muscle tumors
AU - Chak, A.
AU - Canto, M.
AU - Rosch, T.
AU - Dittler, H. J.
AU - Hawes, R.
AU - Tio, T. L.
AU - Lightdale, C. J.
AU - Boyce, H. W.
AU - Scheiman, J.
AU - Carpenter, S. L.
AU - Van Dam, J.
AU - Kochman, M.
AU - Sivak, M. V.
PY - 1996/1/1
Y1 - 1996/1/1
N2 - PRELIMINARY INVESTIGATION: A prior analysis revealed that certain EUS features [viz. cystic spaces > 4 mm (Cys), echogenic foci > 3 mm (Foc), irregular outer border (Out)] are associated with presence of malignancy in gastrointestinal smooth muscle tumors (SMT). AIM: To validate these features as predictors of malignancy and determine interobserver agreement in their interpretation. METHODS: Five blinded EUS experts (E1 · E5) reviewed videotapes of EUS examinations performed on patients prior to surgical resection of SMT. Experts graded features defined in the preliminary study (Cys, Foc, Out) and rated other features, viz., echogenicity (Ech), homogeneity (Hom), and irregular luminal border (Lum). Each expert also stated whether they thought SMT were benign or malignant. RESULTS: Twenty-five benign and 10 malignant SMT from 6 centers were studied. Accuracy for predicting malignancy for each expert was E1=56%, E2=77%, E3=69%, E4=6B%, E5=71%. Stepwise logistic regression showed that for each expert one or more of the features found to be significantly associated with malignancy in the preliminary investigation (Cys, Foc, Out) were also significantly (p<0.05) associated with malignancy. Interobserver agreement (mean kappa statistic) for Cys, Foc, and Out was 0.28, 0.39, and 0.43, respectively. If expert El (who had the lowest prediction accuracy) was excluded from the analysis then mean kappa for Cys, Foc, and Out, repectively increased to 0.39, 0.48, and 0.53. Interobserver agreement (mean kappa) for other features - Ech, Horn, and Lum was 0.12, 0.47, and 0.16, respectively. When at least two of the predictive features (Cys, Foc, and Out) were rated as being present then the sensitivity for detecting malignancy for experts El, E2, E3, E4, E5 was 100%, 80%, 90%, 90%, 80%, respectively. The likelihood of finding malignancy when all three features were rated as being absent, was 88%, 100%, 91%, 91%, for experts E2, E3, E4, E5, respectively. CONCLUSIONS: Experts are able to predict malignancy in smooth muscle tumors examined by EUS. Certain EUS features (cystic spaces, echogenic foci, and irregular outer border) are useful for the detection of malignancy in SMT. Absence of these features indicates benign disease. Agreement in interpreting these features among experts is fair to moderate.
AB - PRELIMINARY INVESTIGATION: A prior analysis revealed that certain EUS features [viz. cystic spaces > 4 mm (Cys), echogenic foci > 3 mm (Foc), irregular outer border (Out)] are associated with presence of malignancy in gastrointestinal smooth muscle tumors (SMT). AIM: To validate these features as predictors of malignancy and determine interobserver agreement in their interpretation. METHODS: Five blinded EUS experts (E1 · E5) reviewed videotapes of EUS examinations performed on patients prior to surgical resection of SMT. Experts graded features defined in the preliminary study (Cys, Foc, Out) and rated other features, viz., echogenicity (Ech), homogeneity (Hom), and irregular luminal border (Lum). Each expert also stated whether they thought SMT were benign or malignant. RESULTS: Twenty-five benign and 10 malignant SMT from 6 centers were studied. Accuracy for predicting malignancy for each expert was E1=56%, E2=77%, E3=69%, E4=6B%, E5=71%. Stepwise logistic regression showed that for each expert one or more of the features found to be significantly associated with malignancy in the preliminary investigation (Cys, Foc, Out) were also significantly (p<0.05) associated with malignancy. Interobserver agreement (mean kappa statistic) for Cys, Foc, and Out was 0.28, 0.39, and 0.43, respectively. If expert El (who had the lowest prediction accuracy) was excluded from the analysis then mean kappa for Cys, Foc, and Out, repectively increased to 0.39, 0.48, and 0.53. Interobserver agreement (mean kappa) for other features - Ech, Horn, and Lum was 0.12, 0.47, and 0.16, respectively. When at least two of the predictive features (Cys, Foc, and Out) were rated as being present then the sensitivity for detecting malignancy for experts El, E2, E3, E4, E5 was 100%, 80%, 90%, 90%, 80%, respectively. The likelihood of finding malignancy when all three features were rated as being absent, was 88%, 100%, 91%, 91%, for experts E2, E3, E4, E5, respectively. CONCLUSIONS: Experts are able to predict malignancy in smooth muscle tumors examined by EUS. Certain EUS features (cystic spaces, echogenic foci, and irregular outer border) are useful for the detection of malignancy in SMT. Absence of these features indicates benign disease. Agreement in interpreting these features among experts is fair to moderate.
UR - http://www.scopus.com/inward/record.url?scp=10544232059&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=10544232059&partnerID=8YFLogxK
U2 - 10.1016/S0016-5107(96)80503-7
DO - 10.1016/S0016-5107(96)80503-7
M3 - Article
AN - SCOPUS:10544232059
SN - 0016-5107
VL - 43
JO - Gastrointestinal endoscopy
JF - Gastrointestinal endoscopy
IS - 4
ER -