Objective To study the clinical characteristics of children with severe drowning admitted to the pediatric intensive care unit and to identify risk factors for mortality. Methods Clinical data of 49 children with severe drowning admitted between January 2015 and December 2024 were retrospectively analyzed. Multivariable logistic regression was used to determine risk factors for mortality and to construct a prediction model. Model performance for mortality risk was evaluated using the receiver operating characteristic (ROC) curve. Results Among the 49 patients, the mortality rate was 24% (12/49), and the incidence of neurological sequelae among survivors was 22% (8/37). Multivariable logistic regression analysis showed that admission Glasgow Coma Scale (GCS) score (OR=0.43, P<0.05), drowning duration (OR=1.22, P<0.05), and total prehospital time (per 10 minutes increase, OR=1.85, P<0.05) were associated with mortality. A mortality prediction model was constructed based on these three factors: logit(P)=6.26-0.85×GCS score at admission + 0.20×drowning duration (minutes) + 0.62×[total prehospital time (minutes)/10]. The area under the ROC curve was 0.912 (P<0.001), with a sensitivity of 83.3% and a specificity of 91.9%. Conclusions Severe drowning in children has a high mortality and a high rate of adverse outcomes. Admission GCS score, drowning duration, and total prehospital time are core risk factors for predicting mortality, and their combination serves as an effective indicator for early risk stratification.