A prediction score model of difficult endoscopic submucosal dissection for early esophageal cancer and precancerous lesions

Background To establish a prediction model to triage difficult esophageal endoscopic submucosal dissection (ESD) in order to reduce complications.Methods This retrospective study enrolled 742 patients who underwent ESD for early esophageal cancer and precancerous lesions. Difficult ESDs were defined as those meeting any of the following criteria: (1) long procedure time (≥ 90 min), (2) occurrence of intraoperative adverse events (muscularis propria injury, perforation, and massive hemorrhage), or (3) piecemeal resection. Logistic regression analysis was performed to identify the predictors of technical difficulty. Then we developed a score prediction model according to the Framingham system for the first time and presented a convenient table containing the risk probability for each patient. In addition, the score model was validated in the validation cohort.Results The risk factors for difficult ESDs were as follows: lesion total length 4–6 cm (1 point), 7–9 cm (2 points) or ≥10 cm (3 points); lesion ≥ 2/3 circumference (2 points); preoperative invasion depth extending to submucosa (2 points). Areas under ROC curves for the score model in the derivation and validation sets were 0.768 (95% CI, 0.726–0.811) and 0.719 (95% CI, 0.646–0.793), respectively. In addition, the percentage of difficult ESDs in easy (score = 0), intermediate (score = 1–2), difficult (score = 3–4), and very difficult (score ≥5) categories were 14.8%, 36.1%, 62.6%, and 87.5% in the derivation cohort and 14.8%, 33.7%, 56.3%, and 68.8% in the internal validation cohort.Conclusions We successfully established a prediction model to assess the difficulty of esophageal ESD and it can provide a scientific basis for clinical management to reduce complications.