New predictors of malignant adenopathy by endoscopic ultrasound (EUS)

Commonly used EUS criteria (size, roundness, echogenicity, homogeneity) for discriminating benign from malignant adenopathy are only 70-80% accurate. We prospectively evaluated EUS findings to determine optimal predictors. Methods: 32 patients undergoing EUS evaluation (Olympus GFUM20) of cancers of the pancreas (II), esophagus (11), stomach (4), rectum (3), bile duct (1), lung (1), or primary adenopathy (1), all with lymph nodes identified on EUS, underwent surgical (27) or needle aspiration (5) biopsy. EUS data were prospectively collected and assessed by a single observer: EUS TNM stage (standard criteria), tumor size, largest node size, number of nodes (#nodes), proximity (cm) of nodes to tumor, and on a 5 point subjective scale (1-5): roundness, echogenicity, homogeneity (homo), and likelihood of metastasis (likely). Associations were assessed by Chi square, Student T, and Fisher tests. Results: 17 patients had malignant and 15 had benign nodes. Significant associations with malignant adenopathy: #nodes (4.12±0.47 vs. 2.73±0.48 (mean ±SEM), p<0.05), the presence of multiple nodes (more than 1)(p<0.01), proximity (1.07±0.22 vs. 2.1210.37 cm, p<0.017), nodes within 1.3 cm of the tumor (p< 0.04), homo>3 (on a 5 pt scale) (p<0.04), and likely>3 (p<0.01). Size, T-stage roundness and echogenicity were not significantly associated. Sens Spec PPV NPV Ace. P-value N-stage 100% 40% 65% 100% 72% <0.01 Multiple 94% 47% 67% 88% 72% <0.01 Likely>3 94% 47% 67% 88% 72% <0.01 Combined 65% 93% 92% 70% 78% <0.001 Model 100% 77% 85% 100% 90% <0.001 EUS N-stage by endosonographer assessment of common criteria (roundness, echogenicity, homogeneity) was 72% accurate; the presence of Multiple nodes or Likely>3 was similar to N-stage (Table). The Combined presence of multiple nodes, nodes within 1.3cm, and homo>3 had the best specificity and positive predictive value (>90%). A logistic regression Model using the presence of multiple nodes, proximity, homogeneity and likelihood had the best accuracy. Conclusions: 1. In clinical situations requiring a highly sensitive test (e.g., if neoadjuvant therapy may be given), the presence of multiple nodes, or the endosonographer's estimate of likelihood should be used. 2. In situations requiring a highly specific test (e.g., pancreatic cancer staging where node positivity may contraindicate surgery), the combined criteria should be used. 3. Regression modeling may improve accuracy. 4. Larger studies with multiple observers are needed.