An artificial neural network predicted transsphenoidal endoscopic pituitary adenomectomy duration with a test-set mean absolute error of about 38 minutes

The study built an artificial neural network to predict operative duration for transsphenoidal endoscopic pituitary adenomectomy and reported a test-set MAE of 0.636 hours (about 38 minutes). Where surgical duration prediction has often relied on traditional methods, this work offers a preoperative-variable-driven machine learning model with error expressed explicitly in hours and minutes. Based on retrospective data from 100 patients and 22 preoperative variables, with R², MAE, and RMSE reported; however, the loaded text is an incomplete version and does not give the R² or RMSE values or the details of the train-test split.

Within clinically acceptable error ranges, 63.3% of the ANN model's predictions fell within ±30 minutes and 76.7% within ±45 minutes. Beyond statistical error, the study introduces two clinical accuracy thresholds, ±30 and ±45 minutes, tying model performance directly to the time tolerances used in nursing and operating room scheduling. Clinical accuracy is reported as percentages derived from test-set predictions; the text does not state the size of that test set.

The artificial neural network outperformed the random forest model in predictive performance. The study compares two algorithms on the same data and variable set, providing a direct head-to-head basis for choosing a modeling approach. The text states that the ANN outperformed random forest but does not report the random forest's specific error values, so the magnitude of the gap cannot be quantified from the loaded text.

Tumor recurrence, distance from the third ventricle floor, and the cinch sign were identified as key predictors of operative duration. The study moves predictor discussion from general clinical experience toward specific preoperative variables, offering leads for later variable selection and model interpretation. The key predictors come from the model analysis; the text does not report importance rankings or statistical significance for each factor.

The model is intended for preoperative duration estimation in transsphenoidal endoscopic pituitary adenomectomy, in perioperative management settings where nursing staffing and operating room time blocks must be planned, with nursing managers and scheduling staff as users. Its value lies in converting preoperative variables into a quantifiable hour-level duration prediction and defining clinical acceptability through the ±30-minute and ±45-minute thresholds, thereby providing objective support for scheduling. The loaded text is an incomplete version containing only the aim, background, methods, results, and conclusion in summary form, without figures or full numerical detail, so the conclusions described here should be read as an interpretation of that summary.

The loaded text is an incomplete version and lacks the specific R² and RMSE values, the train-test split, the random forest's comparative error, and the importance rankings of the key predictors, so the model's overall goodness of fit and the size of the gap between the two models remain open questions. In addition, the single-center retrospective data spanning 100 patients from 2016 to 2025 may coincide with changes in surgical technique and workflow over that period, so model stability on new data or at other institutions awaits prospective validation. Clinical accuracy is presented as test-set percentages, but the test-set sample size is not stated in the text, so confidence intervals for those proportions cannot be assessed.