Machine learning-based preoperative predictive analytics in transsphenoidal surgery for Cushing´s disease: can we predict the future?
EANS Academy. Staartjes V. 09/27/19; 276011; EP04068
Mr. Victor Staartjes
Mr. Victor Staartjes

Access to this content is reserved for EANS members and attendees of this event. Click here to become an EANS member and gain your access to the full content of the EANS Academy

Discussion Forum (0)
Rate & Comment (0)
Background: Although postoperative morbidity and mortality have become relatively low in transnasal transsphenoidal surgery (TSS) for pituitary adenomas, long-term prognosis remains difficult. Especially in Cushing's disease, where residual tumor and hormonal endpoints - both hard to predict - have been demonstrably linked to residual morbidity and mortality, the ability to pre- and perioperatively predict such outcomes with high reliability would be valuable in patient counseling. We aimed to apply machine learning algorithms to predict four relevant endpoints after surgery for Cushing's disease.
Methods: Endpoints were biochemical remission, GTR, recurrence, and achievement of normosecretion. A range of demographic and disease-specific inputs were collected, including tumor size, Hardy, Wilson, and Knosp grading, tumor size, preoperative hormonal function, prior treatments, histology, and immediate postoperative hormonal status. Factors univariately associated with functional impairment were identified. For machine learning-based modelling, data was randomly split into two sets in a 80%/20% ratio for bootstrapped training and testing, respectively. We trialed seven algorithms and tuned for AUC.
Results: We included 151 patients, of which 146 (97%) achieved biochemical remission, 107 (71%) achieved normosecretion, 137 (91%) underwent GTR, and 26 (17%) experienced recurrence. The four final models were based on support vector machines, random forests, and neural networks. At internal validation, bochemical remission, normosecretion, and GTR were predicted with AUCs of 0.68-1.0, accuracy of 74.2%-100%, and Brier scores of 0.035-0.177. However, recurrence was only predictable with AUC of 0.576, accuracy of 74.2%, and Brier score of 0.250. No reliable predictors of endocrinological outcome and recurrence with relevant effect size were identified using traditional interferential statistics.
Conclusions: We trained and internally validated robust machine learning models that identify clinically relevant endpoints in pituitary surgery. Machine learning algorithms may predict outcomes and adverse events that were previously near unpredictable, thus enabling safer and improved patient care and better patient counseling.
Code of conduct/disclaimer available in General Terms & Conditions
Anonymous User Privacy Preferences

Strictly Necessary Cookies (Always Active)

MULTILEARNING platforms and tools hereinafter referred as “MLG SOFTWARE” are provided to you as pure educational platforms/services requiring cookies to operate. In the case of the MLG SOFTWARE, cookies are essential for the Platform to function properly for the provision of education. If these cookies are disabled, a large subset of the functionality provided by the Platform will either be unavailable or cease to work as expected. The MLG SOFTWARE do not capture non-essential activities such as menu items and listings you click on or pages viewed.

Performance Cookies

Performance cookies are used to analyse how visitors use a website in order to provide a better user experience.

Google Analytics is used for user behavior tracking/reporting. Google Analytics works in parallel and independently from MLG’s features. Google Analytics relies on cookies and these cookies can be used by Google to track users across different platforms/services.

Save Settings