Machine Learning and Statistical Analysis of Biochemical Remission Predictors in Somatotroph Adenoma Resections
Journal of Neurological Surgery Part B: Skull Base · 01.02.2024 · 85 (S 01 S 01) · S1-S398
DOI: 10.1055/s-0044-1780160 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Neurological Surgery Part B: Skull Base, erschienen am 01.02.2024. Verfasst von Robert C. Osorio und 12 weiteren Autor:innen.
Abstract (Originalsprache)
Introduction
There is persistent debate in the literature surrounding the true predictors of biochemical remission after surgical resection of somatotroph adenomas. A multimodal analysis of a large number of patients is needed to better understand which patients may be at higher or lower risk for remission failure after surgery.
Methods
A retrospective review was performed on patients undergoing transsphenoidal surgery for somatotroph adenomas at a single high-volume academic pituitary center from 2012 to 2019. Biochemical remission was defined as normalization of age and sex-adjusted serum IGF-1 levels at least six months after surgery. Covariates examined included patient sex, age at surgery, tumor size and cavernous sinus invasion, number of prior pituitary surgeries, preoperative serum growth hormone (GH) and IGF-1, extent of resection, postoperative tumor size, and immediate postoperative GH and IGF-1. Tumor size was defined both as the largest of 3-axis measurements and as a 3D volumetric recording from pre and postoperative imaging studies. Preoperative variables’ effects on remission status were examined by Pearson’s chi-squared tests, Wilcoxon rank sum tests, and Fisher’s exact tests, where appropriate. Pre- and postoperative variables were included in a Random Forest machine learning model to assess for their importance in determining remission status. Preoperative variables found to be significant remission predictors on statistical testing and important in the Random Forest model were subsequently assessed via receiver operating characteristic (ROC) analysis to determine numeric thresholds that optimally predicted preoperative likelihood of remission success or failure.
Results
A total of 97 patients were identified with somatotroph adenomas who underwent transsphenoidal resection, with 78 patients (80%) achieving biochemical remission. Statistical testing found that patients who failed remission were more likely to have larger tumors (1.9 vs. 1.6 cm by largest axis, p = 0.031, and 3.61 vs. 2.57 cm 3 by 3D volume, p = 0.009) that invaded the cavernous sinus more frequently (68% vs. 27%, p < 0.001), and have higher preoperative GH (24.3 vs. 18.7 ng/mL, p = 0.041) and IGF-1 (878 vs. 670 ng/mL, p = 0.045). A random forest machine learning model with 10,000 iterations found 3D tumor volume (mean decrease in model accuracy 2.3) and preoperative IGF-1 (1.85 decrease) to be important preoperative predictors in determining remission status. ROC analysis revealed that among patients with biochemical remission, 83.3% had preoperative GH levels below 8.21 ng/mL (AUC 0.695, p = 0.012), 94.7% had preoperative 3D tumor volume less than 1.51 cm 3 (AUC 0.696, p = 0.001; [ Fig. 1 ]), and 100% had preoperative IGF-1 less than 718.5 ng/mL (AUC=0.742, p = 0.001).
Conclusions
Important preoperative predictors of postoperative remission for somatotroph adenoma resections include serum HGH, serum IGF-1, cavernous sinus invasion, and tumor size. 95% of patients who achieve post-op remission have pre-op 3D tumor volumes less than 1.51 cm 3 . Fig. 1 Results of a random forest machine learning model to predict biochemical remission. Variables are scored by mean decrease in model accuracy when the variable is removed from the model. A higher score indicates increased importance. Fig. 2 Receiver Operator Characteristic analysis of preoperative 3D tumor volume as a predictor of postoperative biochemical remission failure. Publication History Article published online: 05 February 2024 © 2024. Thieme. All rights reserved. Georg Thieme Verlag KG Rüdigerstraße 14, 70469 Stuttgart, Germany
Zitieren
APA
Osorio, R. C., Kabir, A., Haddad, A. F., Badani, A., Khela, H., Saha, A., Peeran, Z., Theodosopoulos, P. V., Kunwar, S., II, J. G., El-Sayed, I. H., Jr, L. S. B., & Aghi, M. K. (2024). Machine Learning and Statistical Analysis of Biochemical Remission Predictors in Somatotroph Adenoma Resections. Journal of Neurological Surgery Part B: Skull Base, 85(S 01 S 01), S1-S398. https://doi.org/10.1055/s-0044-1780160
Vancouver
Osorio RC, Kabir A, Haddad AF, Badani A, Khela H, Saha A, et al. Machine Learning and Statistical Analysis of Biochemical Remission Predictors in Somatotroph Adenoma Resections. Journal of Neurological Surgery Part B: Skull Base. 2024;85(S 01 S 01):S1-S398. doi:10.1055/s-0044-1780160
BibTeX
@article{Osorio2024,
author = {Osorio, Robert C. and Kabir, Aymen and Haddad, Alexander F. and Badani, Aarav and Khela, Harmon and Saha, Atul and Peeran, Zain and Theodosopoulos, Philip V. and Kunwar, Sandeep and II, Jose Gurrola and El-Sayed, Ivan H. and Jr, Lewis S. Blevins and Aghi, Manish K.},
title = {Machine Learning and Statistical Analysis of Biochemical Remission Predictors in Somatotroph Adenoma Resections},
journal = {Journal of Neurological Surgery Part B: Skull Base},
year = {2024},
volume = {85},
number = {S 01 S 01},
pages = {S1--S398},
doi = {10.1055/s-0044-1780160}
}
EUVIMED