Abstract: Objective To analyze the influencing factors of postoperative facial paralysis in patients with parotid tumors by using machine learning technique. Methods The data of patients with parotid tumors in the medical record system of Guigang People's Hospital from January 2013 to December 2023 were collected retrospectively. LASSO regression was used to screen the potential risk factors of postoperative facial paralysis, followed by an disordered multi⁃classification Logistic regression analysis to determine the risk factors of postoperative facial paralysis. Results 54 of 395 patients with parotid tumors experienced facial paralysis, including 40 cases of transient facial paralysis and 14 cases of permanent facial paralysis. Disordered multi⁃classification LASSO regression identified 26 risk factors of postoperative facial paralysis in patients with parotid tumors under the optimal lambda value, which were included in the disordered multi⁃classification Logistic regression analysis. The results showed that tumor location, tumor nature, preoperative mass tenderness, smoking, glutamyl transferase, chloride concentration, magnesium concentration, and partial thromboplastin time were the influencing factors for transient facial paralysis (all P<0.05); tumor nature tumor volume, age, and body mass index were influencing factors for permanent facial paralysis (all P<0.05). Conclusions The key influencing factors of postoperative facial paralysis in patients with parotid tumors are identified by using the machine learning technique, helping the early identification of high⁃risk groups for postoperative facial paralysis and providing a scientific basis for the prevention and prediction of facial paralysis.

Key words: Parotid tumor, Postoperative facial paralysis, Risk factors, Machine learning