The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2415
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-07T12:02:08.961542Z
status: published
lens: g_space
sector: health
headline: Multiple Machine Learning Algorithms for Predicting Postoperative Residual Tumor in Pituitary Neuroendocrine Tumors
dek: Aims To develop and validate an interpretable machine learning (ML) model for predicting radiologically confirmed postoperative residual tumor (RT) in pituitary neuroendocrine tumors (PitNETs). Methods We retrospectively analyzed 584 surgically treated patients with PitNETs from a single center, using a random 7:3 training-test split and internal cross-validation. The primary endpoint was radiologically confirmed postoperative RT, assessed using MRI obtained within 48 h after surgery and longitudinal follow-up M…
gain_title: In a single-center study of 584 pituitary tumor patients, a random forest model predicted residual tumor after surgery with an AUC of 0.945 in internal testing, versus 0.715 for logistic regression.
problem_title: (none)
trace_subject: (none)
gain_reading: In a single-center study of 584 pituitary tumor patients, a random forest model predicted residual tumor after surgery with an AUC of 0.945 in internal testing, versus 0.715 for logistic regression.
gain_evidence: the random forest model achieved high discrimination (AUC = 0.945; accuracy = 0.938) | logistic regression yielded a test‐set AUC of 0.715
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers at Xiangya Hospital of Central South University built machine learning models to predict residual tumor after surgery for pituitary neuroendocrine tumors. They analyzed 584 patients treated between 2017 and 2023, of whom 72 (12.3%) had radiologically confirmed residual tumor. A random forest model reached an AUC of 0.945 and accuracy of 0.938 in the internal test set.

The model uses eight routinely available inputs, including tumor diameters, age, Knosp grade, carotid artery invasion and sphenoid sinus pneumatization. It was released as a web tool that shows each patient's predicted probability and a SHAP chart explaining which factors drive it. Larger sagittal and coronal diameters pushed predicted risk up most.

The results come from one hospital using a random split of its own patients, so performance elsewhere is unknown. The authors say external and temporal validation, with calibration checks, is needed before clinical use, and that postoperative imaging is still required to confirm residual tumor.
limitation: Evaluation used only a single-center random split and internal cross-validation, so generalizability to other centers and future patients is unproven and internal performance may be overstated.
tag: Evidence-backed gain
key_points: Residual tumor occurred in 72 of 584 patients (12.3%); the final label used MRI within 48 hours plus follow-up MRI. | Nine algorithms were compared; random forest had the highest test AUC (0.945) but was not statistically different from LightGBM (p = 0.108), SVM (p = 0.726) or XGBoost (p = 0.063). | LightGBM had the lowest Brier score (0.047) versus 0.064 for random forest, so random forest probabilities are less well calibrated. | A logistic regression comparator scored an AUC of 0.715 and a Brier score of 0.105 in the test set. | For random forest, recall for residual tumor was 92.3% and precision was 70.6%, so about three in ten positive predictions were false alarms. | Agreement between 48-hour MRI and follow-up classification was 96.4% (kappa 0.837); 21 patients (3.6%) were reclassified. | Random forest AUC was 0.850 in functional tumors and 0.927 in nonfunctional tumors; it was 0.725 in giant adenomas and 0.717 in non-giant adenomas.
rundown: Patients were all adults treated by the same neurosurgeon, with diagnoses confirmed under the 2022 WHO classification. Of 617 candidates, 5 were excluded for no surgery and 28 for missing data. Two senior neurosurgeons blinded to predictors read the MRI scans, with a third consulted when needed.

Variables were narrowed from 50 candidates using LASSO regression to eight predictors. Patients were split 7:3 into training and test groups, with oversampling applied only to training data. Interobserver agreement was high, with kappa 0.825 for retrosellar invasion, 0.960 for carotid artery invasion and 0.994 for Knosp grade.

Patients with residual tumor had larger tumors (maximum diameter 40.2 vs. 25.9 mm) and more retrosellar invasion (31.9% vs. 6.05%). In SHAP analysis, sagittal diameter ranked highest, followed by coronal diameter, age and Knosp grade. In one high-risk case the predicted probability rose from a 0.271 baseline to 0.941.

The authors said the model should not be read as better than existing scoring systems, because no direct comparison was made. Subgroup estimates were imprecise, with only 7 to 19 residual tumor cases per stratum, and some molecular marker data were incomplete.
sources:
- peer_reviewed | CNS Neuroscience & Therapeutics | https://doi.org/10.1002/cns.71244 | 2026-11-01
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

34845b6ef3d35c72994dbd62ff0fe450b019f7a7a73503d0848403d7def7e9ba

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000