Machine Learning-Based Model to Predict Long-Term Tumor Control and Additional Interventions following Pituitary Surgery

Objective: Pituitary adenomas (PAs) are common benign tumors causing complex symptoms due to their mass or excessive hormone production. PAs are curable by surgery for a long period of time; however, tumors with identical histopathology may respond differently, necessitating retreatment in the event of a lesion or endocrinological recurrence. Accurately predicting patients at high risk of recurrence from preoperative factors may assist in patient management from a long-term perspective. This study aimed to identify the potential factors predicting long-term outcomes of endonasal transsphenoidal surgery (ETS) for PAs using trained multiple tree-based machine learning (ML) algorithms.

Methods: The authors reviewed the patients who underwent ETS for PAs between 2013 and 2023. Data on patients’ baseline characteristics, intervention details, histopathology, surgical outcomes, and neurological and endocrine functions were collected. The primary outcome was the intervention-free survival (IFS) rate, and the therapeutic outcomes were labeled as “under control” or “treatment failure,” depending on whether additional therapeutic interventions were required. The secondary outcomes included overall survival (OS) and neurological and endocrine outcomes. The decision tree and random forest classifiers were trained and tested to predict post-ETS outcomes based on unseen data using an 80/20 split.

Results: Data on 884 ETSs for 700 patients with a median follow-up period of 60 months were extracted, including 360 (52%) nonfunctioning, 98 (14%) growth-hormone-secreting (GH), 150 (22%) adrenocorticotropic hormone (ACTH)-secreting, 41 (6%) prolactin-secreting, 24 (3%) pluripotent hormone-secreting, and 27 (4%) other PAs. In the entire cohort, 195 patients (28%) required additional interventions. Consequently, IFS rates following ETS alone were 76% at 5 years and 62% at 10 years in the entire cohort ([Fig. 1]).

Fig. 1

Multivariable Cox proportional hazard analysis demonstrated that gross total resection (GTR) (hazard ratio [HR], 3.38; 95% confidence interval [CI], 2.32–4.93; p = 0.001), Knosp-Steiner (KS) grades 0–2 (HR, 2.27; 95% CI, 1.56–3.29; p = 0.001) and older age at primary surgery (HR, 1.02; 95% CI, 1.01–1.03; p = 0.001) were significantly associated with better IFS. In the decision tree analysis, when patients with GTR were younger than 52 years and had a KS grade 4, they had a 70% risk of additional intervention. When patients without GTR had KS grades 3 to 4 and were younger than 45 years, they had a 79% risk of additional intervention; in contrast, those older than 50 years with GH-, ACTH-, or pluripotent-secreting PAs showed an 86% risk of additional intervention, respectively (accuracy 79%; [Fig. 2]).

Fig. 2

Random forest analysis (n = 500 trees) revealed that GTR (mean minimal depth = 1.23), KS grade (1.3), patient age (1.43), tumor size (1.71), and tumor type (1.73) were the five most significant predictors of long-term tumor control and intervention needs ([Fig. 3]).

Fig. 3

OS rate following ETS was 97% at 10 years in the entire cohort ([Fig. 4]).

Fig. 4

Conclusions: Key factors enhancing better IFS included GTR, KS grades 0–2, and older age at primary surgery. Notably, younger patients with GTR and a higher KS grade had up to a 70% recurrence risk. The top predictors for long-term recurrence encompassed GTR, KS grade, patient age, tumor size, and type.

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
05 February 2024