Abstract
AIM: To identify preoperative risk factors for prolonged length of hospital stay (LOS) in patients undergoing surgery for primary liver cancer and to develop a predictive nomogram. METHODS: We retrospectively analyzed 702 surgical patients from a single center (2020-2023). LOS was modeled using negative binomial regression based on preoperative factors to construct a nomogram. Model performance was evaluated via internal bootstrap validation (1000 resamples), calibration plots, and decision curve analysis. Prolonged LOS was defined as >17 days (75th percentile) for a secondary logistic regression analysis. RESULTS: Four independent preoperative factors predicted longer LOS: lower serum cholinesterase, higher fibrinogen, intrahepatic cholangiocarcinoma (vs hepatocellular carcinoma), and female sex (all p<0.05). The nomogram showed moderate discriminative ability (apparent AUC ~0.67) with good calibration. The mean absolute error for LOS prediction was ~4.6 days. For predicting prolonged LOS (>17 days), the logistic model achieved an AUC of ~0.67. CONCLUSION: We developed an internally validated nomogram using routine preoperative data to estimate the risk of extended hospitalization after liver cancer surgery. This tool may help identify high-risk patients for targeted interventions, although its predictive accuracy is modest, and external validation is required before clinical application.