Perioperative predictors of biochemical remission in acromegaly: a predictive modeling study

Biochemical remission after transsphenoidal resection for acromegaly remains challenging, and early prognostication is limited. This study applies statistical and predictive modeling approaches to identify perioperative predictors of remission.

We retrospectively analyzed 86 patients with acromegaly who underwent endoscopic endonasal transsphenoidal resection of GH-secreting primary pituitary adenomas between January 2018 and January 2025. Preoperative demographic, biochemical, radiologic, and histopathologic variables were evaluated. Patients were randomly split into training and test sets. Missing data were addressed using median and KNN imputation. Five machine learning models (GBM, RF, GLMNET, KNN, and Nnet) were trained using exhaustive feature subset selection and evaluated using AUROC and accuracy.

Of 86 patients, 62 (72.1%) achieved biochemical remission. Remission was associated with older age (p = 0.016), round tumor shape (p < 0.0001), gross total resection (p < 0.0001), lower Knosp grade (p < 0.0001), and dense CAM 5.2 staining (p = 0.037). The best-performing model was RF using four features (gender, tumor shape, extent of resection, CAM 5.2), achieving an accuracy of 0.8235 and AUROC of 0.8542 on the test set.

Predictive modeling may help estimate biochemical remission after surgery for acromegaly. This exploratory perioperative framework may support postoperative risk stratification and patient counseling while complementing standard endocrine follow-up. Further external validation in larger cohorts is warranted.