Construction of risk predictive models for postoperative hyponatremia in colorectal cancer patients.

This study aimed to construct and validate a risk prediction model for postoperative hyponatremia in patients with colorectal cancer (CRC). Clinical data from 280 CRC patients hospitalized at WUZHOU RED CROSS HOSPITAL between January 2021 and December 2024 were retrospectively collected. Patients were categorized into a hyponatremia group and a normal serum sodium group according to postoperative serum sodium levels. All eligible participants were randomly allocated into a training set (70%) and an internal validation set (30%) using R software, while an additional 47 prospectively enrolled CRC patients between January and May 2025 were included as the external validation set. Independent risk factors for postoperative hyponatremia were screened, and four prediction models, including a nomogram, random forest, decision tree, and back-propagation (BP) neural network, were established and comparatively evaluated. Among the 280 enrolled patients, the overall incidence of postoperative hyponatremia was 30.71%. Significant differences were observed between the two groups in syndrome of inappropriate antidiuretic hormone secretion (SIADH) status, TNM stage, aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine (Cr), carbohydrate antigen 19-9 (CA19-9), carcinoembryonic antigen (CEA) were observed between the two groups (all P<0.05). Multivariate logistic regression identified SIADH, advanced TNM stage, elevated Cr, CA19-9 and CEA levels as independent risk factors for postoperative hyponatremia. Comparative evaluation revealed that the random forest model achieved the best predictive performance. The AUC, sensitivity and specificity were 0.985 (95% CI: 0.972-0.997), 96.6% and 92.0% in the training set, respectively; and the corresponding values were 0.845 (95% CI: 0.750-0.940), 82.1% and 83.9% in the internal validation set, respectively. The prospective external validation set further confirmed the superior predictive accuracy of the random forest model. In conclusion, the random forest-based model established in this study demonstrated favorable discrimination and generalization ability for predicting postoperative hyponatremia in CRC patients, thereby providing a useful tool for preoperative risk stratification and early targeted clinical intervention.