Machine learning-based analysis of factors influencing hospitalization costs in thyroid cancer surgery patients

Objectives: To analyze factors influencing hospitalization costs in thyroid cancer surgery patients and provide scientific evidence for optimizing medical resource allocation and reducing patient economic burden. Methods: This retrospective study included 2,498 thyroid cancer surgery patients from a tertiary hospital in Fujian Province between 2020 and 2024. Nonparametric tests were used to analyze cost composition and influencing factors. Four machine learning models (Random Forest, Support Vector Machine, Linear Regression, and XGBoost) were employed for cost prediction. Model performance was evaluated using mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination. Results: The median hospitalization cost was [Formula: see text]25,435 (interquartile range [IQR]: 22,400–28,651). Univariate analysis identified eight significant cost factors ([Formula: see text]): age group, length of stay, pathological type, surgical procedure, drug cost ratio, lymph node metastasis, ICU admission, and parathyroid transplantation. Support Vector Machine demonstrated the best performance ([Formula: see text], [Formula: see text]), followed by Random Forest ([Formula: see text], [Formula: see text]). Conclusions: Length of stay, surgical procedure type, drug cost ratio, and parathyroid transplantation are major factors influencing thyroid cancer surgery costs. Machine learning can provide valuable references for medical resource allocation, healthcare payment reform, and clinical decision-making, optimizing cost-effectiveness while ensuring medical safety.