Abstract / Summary
Abstract Background Accurate pretherapeutic estimation of invasion depth is essential for selecting appropriate treatment for superficial esophageal squamous neoplasia (SESN). We aimed to evaluate endoscopic predictors of pathological muscularis mucosae/submucosal (MM/SM) invasion and to develop an internally validated model combining macroscopic type with intrapapillary capillary loop (IPCL) pattern. Methods This retrospective, single-center, lesion-based model-development study included 304 SESN lesions treated by endoscopic submucosal dissection at Beijing Friendship Hospital between January 2020 and October 2025. Pathological depth was dichotomized as epithelium/lamina propria mucosae (EP/LPM) versus MM/SM invasion. Logistic regression was used to evaluate predictors, and a parsimonious model combining macroscopic type and IPCL pattern was assessed by receiver operating characteristic analysis and internally validated using 1000 bootstrap resamples. Results Among 304 lesions, 235 (77.3%) were EP/LPM and 69 (22.7%) were MM/SM. In multivariable analysis, IPCL B2 pattern was independently associated with MM/SM invasion (odds ratio [OR] 6.25, 95% confidence interval [CI] 2.80-13.97; P < 0.001). Compared with 0-IIb lesions, both 0-IIa lesions (OR 8.77, 95% CI 2.65–29.01; P < 0.001) and 0-IIc lesions (OR 23.28, 95% CI 9.71–55.82; P < 0.001) had higher odds of MM/SM invasion. The combined model achieved an apparent area under the curve (AUC) of 0.915 (95% CI 0.877–0.953), with sensitivity of 84.06%, specificity of 91.91%, and accuracy of 90.13% at the Youden-optimal cut-off. Bootstrap internal validation showed minimal optimism, with an optimism-corrected AUC of 0.913, Brier score of 0.089, calibration intercept of -0.015, and calibration slope of 0.970. Conclusions Macroscopic type and IPCL pattern were complementary predictors of pathological MM/SM invasion in SESN. A simple model combining these two features showed strong discrimination and acceptable optimism-corrected performance, but external validation is required before routine clinical use. Trial registration: Not applicable.