Effect of dexamethasone pretreatment using deep learning on the surgical effect of patients with gastrointestinal tumors

Abstract

To explore the application efficacy and significance of deep learning in anesthesia management for gastrointestinal tumors (GITs) surgery, 80 elderly patients with GITs who underwent surgical intervention at our institution between January and September 2021 were enrolled. According to the preoperative anesthesia management methodology, patients were rolled into a control (Ctrl) group (using 10 mg dexamethasone 1–2 hours before surgery) and an experimental (Exp) group (using a deep learning-based anesthesia monitoring system on the basis of the Ctrl group), with 40 cases in each group. A comprehensive comparative analysis was performed between the two cohorts, encompassing postoperative cognitive evaluations, Montreal Cognitive Assessment (MoCA) scores, gastrointestinal functionality, serum biomarkers (including interleukin (IL)-6, C-reactive protein (CRP), and cortisol levels), length of hospitalization, incidence of complications, and other pertinent metrics. The findings demonstrated that anesthesia monitoring facilitated by deep learning algorithms effectively assessed the anesthesia state of patients. Compared to the Ctrl group, patients in the Exp group showed significant differences in cognitive assessments (word recall, number connection, number coding) (P<0.05). Additionally, the Exp group exhibited a notably increased MoCA score (25.3±2.4), significantly shorter time to first flatus postoperatively (35.8±13.7 hours), markedly reduced postoperative pain scores, significantly shortened time to tolerate a liquid diet postoperatively (19.6±5.2 hours), accelerated recovery of serum-related indicators, and a significantly decreased mean length of hospital stay (11.4±3.2 days) compared to the Ctrl group. In summary, administering dexamethasone under the anesthesia management of GITs surgery based on gradient boosting decision tree (GBDT) and pharmacokinetics pharmacodynamics (PKPD) models can promote patient recovery, reduce the incidence of postoperative cognitive impairment (POCD), and improve patient prognosis.