Abstract: OBJECTIVE To construct the risk prediction model for postoperative infections in elderly patients with early gastric cancer undergoing endoscopic submucosal dissection (ESD) with the use of the least absolute shrinkage and selection operator (LASSO) regression so as to provide scientific bases for prevention of postoperative infections. METHODS Totally 545 elderly patients with early gastric cancer who underwent ESD in the Affiliated Hospital of Traditional Chinese Medicine of Southwest Medical University from May 2022 to May 2025 were recruited as the research subjects. The major variables were screened by LASSO regression; the intrinsic connections among the various factors were analyzed through Bayesian network, and the efficacy of the model was evaluated. RESULTS Totally 12 major variables for the postoperative infections in the patients undergoing ESD were screened out by LASSO regression, namely diabetes mellitus history, smoking history, intraoperative bleeding, intraoperative perforation, postoperative delayed perforation, operation duration, American Society of Anesthesiologists (ASA) classification, and preoperative albumin (ALB), lymphocyte count (Lym), C-reactive protein (CRP), hemoglobin (Hb), and prealbumin (PA). The Bayesian network model revealed that the preoperative Hb, CRP, intraoperative perforation and postoperative delayed perforation were directly correlated with the postoperative infections; the diabetes mellitus history and smoking history indirectly affected the emergence of postoperative infections by influencing the preoperative Hb, Lym and PA; ASA classification was indirectly associated with the preoperative ALB, PA, and preoperative CRP, and it affected the occurrence of postoperative infections through preoperative CRP; the operation duration was indirectly associated with the intraoperative hemorrhage and preoperative ALB, which then resulted in the postoperative infections. The accurate rates of the model were 90.97% and 89.68% before and after the Bootstrap validation, respectively, indicating that the Bayesian network risk prediction model had favorable performance (P＜0.001). CONCLUSION The LASSO variable selection model that is constructed targeting the influencing factors for the postoperative infections in the elderly patients with early gastric cancer conforms to the clinical and theoretical laws and has high clinical value in risk prediction of infections.