Predictive modeling of arginine vasopressin deficiency after transsphenoidal pituitary adenoma resection by using multiple machine learning algorithms

• Published in: Scientific reports 2024-09, Vol.14 (1), p.22210-13, Article 22210
• Main Authors: Chen, Yuyang, Zhong, Jiansheng, Li, Haixiang, Lin, Kunzhe, Wei, Liangfeng, Wang, Shousen
• Format: Article
• Language: English
• Subjects: 692/308; 692/4028; 692/617; 692/698; Adenoma; Adenoma - pathology; Adenoma - surgery; Adrenocorticotropic hormone; Adult; Aged; Algorithms; Arginine Vasopressin - blood; Arginine Vasopressin - deficiency; Arginine Vasopressin - metabolism; Argipressin; Cortisol; Decision Trees; Diabetes insipidus; Diaphragm; Female; Hormones; Humanities and Social Sciences; Humans; Learning algorithms; Machine Learning; Male; Middle Aged; multidisciplinary; Pituitary; Pituitary adenoma; Pituitary Neoplasms - pathology; Pituitary Neoplasms - surgery; Postoperative Complications - etiology; Predictive model; Retrospective Studies; Science; Science (multidisciplinary); Surgery; Transsphenoidal surgery; Tumors; Vasopressin
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-024-72486-w

This study aimed to predict arginine vasopressin deficiency (AVP-D) following transsphenoidal pituitary adenoma surgery using machine learning algorithms. We reviewed 452 cases from December 2013 to December 2023, analyzing clinical and imaging data. Key predictors of AVP-D included sex, tumor height, preoperative and postoperative changes in sellar diaphragm height and pituitary stalk length, preoperative ACTH levels, changes in ACTH levels, and preoperative cortisol levels. Six machine learning algorithms were tested: logistic regression (LR), support vector classification (SVC), random forest (RF), decision tree (DT), k-nearest neighbors (KNN), and extreme gradient boosting (XGBoost). After cross-validation and parameter optimization, the random forest model demonstrated the highest performance, with an accuracy (ACC) of 0.882 and an AUC of 0.96. The decision tree model followed, achieving an accuracy of 0.843 and an AUC of 0.95. Other models showed lower performance: LR had an ACC of 0.522 and an AUC of 0.54; SVC had an ACC of 0.647 and an AUC of 0.67; KNN achieved an ACC of 0.64 and an AUC of 0.70; and XGBoost had an ACC of 0.794 and an AUC of 0.91. The study found that a shorter preoperative pituitary stalk length, significant intraoperative stretching, and lower preoperative ACTH and cortisol levels were associated with a higher likelihood of developing AVP-D post-surgery.