Clinical significance of intraoperative dissection time and machine learning-based prediction in laparoscopic distal gastrectomy for gastric cancer.

BACKGROUND: Establishing a clinically meaningful cutoff value for intraoperative dissection time and developing predictive models for this metric among patients with distal gastric cancer. METHODS: This retrospective multicenter study included 849 patients with distal gastric cancer who underwent laparoscopic distal gastrectomy between November 2020 and May 2025 at three tertiary centers. Using postoperative complications as the status variable, determine the optimal cutoff value for intraoperative dissection time. Patients were subsequently classified into short intraoperative dissection time (SIDT) and long intraoperative dissection time (LIDT) groups. Six machine learning predictive models were developed and validated. RESULTS: An intraoperative dissection time of 133 min was identified as the optimal cutoff value. Compared with patients in the SIDT group, patients in the LIDT group had a higher incidence of postoperative complications and worse overall survival and recurrence-free survival. Although the LIDT group showed worse OS and RFS in the unadjusted analyses, intraoperative dissection time was not independently associated with prognosis in the multivariable Analysis. The LIDT group was also characterized by a higher body mass index, larger tumors, a more advanced clinical stage, and a higher proportion of patients who received neoadjuvant chemotherapy. Among the evaluated models, logistic regression achieved the highest area under the curve (AUC) in the test cohort (0.770), whereas random forest achieved the highest AUC in the validation cohort (0.759). CONCLUSIONS: For patients undergoing laparoscopic radical gastrectomy for distal gastric cancer, an intraoperative dissection time of 133 min was identified as a critical threshold, beyond which there was a higher incidence of complications. Although there were differences in prognosis, intraoperative dissection time was not an independent risk factor. Among the evaluated models, the random forest model showed the most stable predictive performance and may offer additional value for the preoperative evaluation of patients at risk of prolonged intraoperative dissection time.