Machine Learning–Based Prediction of Early Visual Field Nonrecovery After Endoscopic Surgery for Nonfunctioning Pituitary Adenomas With Clinically Accessible Biomarkers: A Dual‐Center Study

Objective To identify preoperative and perioperative predictors of early postoperative visual field (VF) nonrecovery after endoscopic endonasal transsphenoidal surgery (EETS) for nonfunctioning pituitary adenomas (NFPAs) and to develop a practical prediction model. Methods This dual‐center retrospective study included 368 patients with histologically confirmed NFPAs and preoperative bilateral VF impairment who underwent EETS between July 2023 and July 2025. The Kunming cohort ( n = 268) served as the training set, and the Zhuji cohort ( n = 100) served as the validation set. Early postoperative VF recovery was defined as an improvement in mean deviation of at least 3 dB in either eye on postoperative Day 7. Candidate predictors were screened using least absolute shrinkage and selection operator regression and multivariable logistic regression. A nomogram was constructed and evaluated by discrimination, calibration, and decision curve analysis. Results Early postoperative VF recovery occurred in 123 of 268 patients (45.90%) in the training cohort and 44 of 100 patients (44.00%) in the validation cohort. Age, vertical tumor diameter, and fasting glucose were independently associated with VF nonrecovery in both cohorts. In the training cohort, the odds ratios were 1.029 for age ( p = 0.007), 1.099 for vertical diameter ( p < 0.001), and 1.274 for fasting glucose ( p = 0.036); in the validation cohort, the corresponding odds ratios were 1.047 ( p = 0.012), 1.135 ( p < 0.001), and 1.617 ( p = 0.021). The nomogram demonstrated good discrimination in the training cohort (C‐index 0.747, 95% CI 0.691–0.803) and validation cohort (C‐index 0.753, 95% CI 0.663–0.843), with satisfactory calibration. Conclusions Older age, greater vertical tumor diameter, and higher fasting glucose were associated with early postoperative VF nonrecovery after EETS for NFPA. This study provides a simple, routinely available risk stratification tool based on these clinically accessible variables. After further prospective validation and refinement, this model could potentially aid in perioperative counseling and expectation management.