Construction of a prediction model for early hormone remission after prolactinoma surgery based on machine learning
10.12025/j.issn.1008-6358.2026.20260329
- VernacularTitle:基于机器学习的催乳素瘤术后激素早期缓解预测模型构建
- Author:
Yanxia DENG
1
;
Jiahui LIU
1
;
Jing XU
1
;
Weijie WANG
1
;
Kun SUN
1
;
Lianshu DING
1
;
Jian JIANG
1
Author Information
1. Department of Neurosurgery, The First People’s Hospital of Huai’an, Nanjing Medical University, Huai’an 223300, Jiangsu, China.
- Publication Type:Originalarticle
- Keywords:
prolactinoma;
transsphenoidal surgery;
hormonal remission;
prediction model;
machine learning
- From:
Chinese Journal of Clinical Medicine
2026;33(3):406-413
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore factors associated with early postoperative hormonal remission in patients with prolactinoma and to develop prediction models based on clinical and radiological features. Methods Data from 107 patients with prolactinoma who underwent transsphenoidal surgery at The First People’s Hospital of Huai’an, Nanjing Medical University between January 2020 and December 2024 was collected, included general clinical characteristics, preoperative laboratory indicators, and imaging features. Based on whether early postoperative prolactin (PRL) levels normalized, patients were divided into a remission group (n=76) and a non-remission group (n=31). Univariate logistic regression was used for preliminary evaluation of candidate variables, followed by LASSO regression for feature selection. Multiple machine learning models were constructed, including logistic regression, random forest, support vector machine, K-nearest neighbors, naive Bayes, decision tree, neural network, and gradient boosting decision tree (GBDT). All models were trained and evaluated using ten-fold cross-validation, with comprehensive assessment of model performance based on the area under the ROC curve (AUC), accuracy, sensitivity, specificity, precision, and F1 score. Results The maximum diameter of tumors in the non remission group was larger than that in the remission group (P=0.019), and the incidence of tumor stroke and preoperative PRL levels were significantly higher than those in the remission group (P＜0.001). Univariate analysis showed that sex, maximum tumor diameter, tumor stroke, and preoperative PRL levels were influencing factors for early postoperative hormone response in patients with prolactinoma (P＜0.05). The comparison results of machine learning models show that the neural network model performs the best (AUC=0.921) and has good clinical application value, followed by the GBDT model (AUC=0.893) and the support vector machine model (AUC=0.884). Other models also show certain predictive ability. Conclusions Sex, preoperative PRL levels, maximum tumor diameter, Hounsfield unit value, and tumor stroke are important factors affecting early hormone response after prolactinoma surgery. The machine learning model constructed based on the above variables has good predictive performance, and performs the best and has good clinical application value.
