Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1956
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-20T01:09:37.924672Z
status: published
lens: g_space
sector: health
headline: Machine learning-based prediction of malignant transformation within 5 years in patients with oral leukoplakia: A retrospective cohort study
dek: Oral leukoplakia (OLK), a potentially malignant oral disorder, has 7.2 - 9.8 % transformation rate. Identifying high-risk patients within 5 years can reduce the oral cancer burden. In this retrospective study, we aimed to identify the factors associated with malignant transformation in Chinese patients with OLK and develop and evaluate machine learning (ML) models for predicting malignant transformation within 5 years. In total, 672 OLK patients (203 showing transformation; 469 with no transformation) were inclu…
gain_title: Models trained on clinicopathological features discriminated which oral leukoplakia patients would develop malignancy within 5 years, which may support individualized risk stratification.
problem_title: (none)
trace_subject: (none)
gain_reading: Models trained on clinicopathological features discriminated which oral leukoplakia patients would develop malignancy within 5 years, which may support individualized risk stratification.
gain_evidence: ML models may predict malignant transformation within 5 years in patients with OLK and may assist in individualized risk stratification.
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers analyzed records of 672 Chinese patients with oral leukoplakia, 203 of whom later developed malignancy. They trained six algorithms on 17 clinicopathological features and tested them on held-out data and on a later cohort of 73 patients.

The work suggests routine clinical variables could help flag patients at higher risk of oral cancer, so that monitoring can be focused on them. The temporal validation group was small and the findings come from one retrospective dataset, so the models are not yet ready for clinical use.
limitation: The temporal validation cohort was small and the models need testing in larger multicenter cohorts before any clinical use.
tag: Evidence-backed gain
key_points: LightGBM performed best with 17 features, with accuracy of 82.65 percent and an AUC of 0.8524 in the internal test set. | In the temporal validation cohort, LightGBM reached an AUC of 0.7959 using all 17 features. | With only five features, XGBoost reached AUCs of 0.8081 internally and 0.8092 in temporal validation. | For the specific 5-year prediction, XGBoost achieved an AUC of 0.8505 in the internal test set.
rundown: The study was a retrospective cohort of Chinese oral leukoplakia patients, a potentially malignant oral disorder with a reported transformation rate of 7.2 to 9.8 percent. The aim was to identify factors linked to malignant transformation and build prediction models.

Six methods were compared: Logistic Regression, SVM, Random Forest, GBDT, XGBoost, and LightGBM. The 672-patient dataset was split into training and internal test portions, and a separate temporal cohort had 24 transformations among 73 patients.

The authors state that the temporal validation results should be interpreted with caution because of limited sample size, and that larger multicenter cohorts are needed before clinical application.
sources:
- peer_reviewed | Oral Oncology | https://doi.org/10.1016/j.oraloncology.2026.108174 | 2026-10-17
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

1ff95fc4de5c1470fdc98034d423eef23c9465e7dff5097808c67dfc17e6b598

previous

0000000000000000000000000000000000000000000000000000000000000000