Background: Diabetes insipidus (DI) is a common complication after endoscopic transsphenoidal surgery (TSS) for pituitary adenoma (PA), which affects the quality of life in patients. Therefore, there is a need to develop prediction models of postoperative DI specifically for patients who undergo endoscopic TSS. This study establishes and validates prediction models of DI after endoscopic TSS for patients with PA using machine learning algorithms.
Methods: We retrospectively collected information about patients with PA who underwent endoscopic TSS in otorhinolaryngology and neurosurgery departments between January 2018 and December 2020. The patients were randomly split into a training set (70%) and a test set (30%). The 4 machine learning algorithms (logistic regression, random forest, support vector machine, and decision tree) were used to establish the prediction models. Area under the receiver operating characteristic curves were calculated to compare the performance of the models.
Results: A total of 232 patients were included, and 78 patients (33.6%) developed transient DI after surgery. Data were randomly divided into a training set (n = 162) and a test set (n = 70) for development and validation of the model, respectively. The area under the receiver operating characteristic curve was highest in the random forest model (0.815) and lowest in the logistic regression model (0.601). Invasion of pituitary stalk was the most important feature for model performance, closely followed by macroadenomas, size classification of PA, tumor texture, and Hardy-Wilson suprasellar grade.
Conclusions: Machine learning algorithms identify preoperative features of importance and reliably predict DI after endoscopic TSS for patients with PA. Such a prediction model may enable clinicians to develop individualized treatment strategy and follow-up management.
Keywords: Diabetes insipidus; Machine learning; Pituitary adenoma; Prediction model; Transsphenoidal surgery.
Copyright © 2023 Elsevier Inc. All rights reserved.