Work

NGAL and traditional biomarkers in cervical cancer: a comparative evaluation using random forest

Year: 2026

Type: article

Abstract: Introduction: Tumor stage and biological aggressiveness have a significant impact on the prognosis of cervical cancer, which is still a serious global health concern. Methods: This work assesses an exploratory, leakage-free Random Forest model and describes longitudinal NGAL and traditional biomarker patterns in 100 individuals with cervical cancer. Since there was no independent radiologic, pathological, or defined response endpoint in the database, the machine-learning result was specifically characterized as an NGAL-decrease phenotype (post-treatment NGAL < pre-treatment NGAL), not clinical treatment response. All NGAL-derived variables and all post-treatment variables were removed from the predictor set in order to prevent target leakage. Results: The mean accuracy using stratified 5-fold cross validation was 43.0% (SD 7.6%) and the mean balanced accuracy was 42.1%, suggesting that there was no discriminative performance that was useful. Greater histological grades and more advanced broad-stage groups had descriptively greater NGAL concentrations. Discussion: These results do not prove clinical predictive usefulness; rather, they are hypothesis-generating.

Source: Frontiers in Oncology

DOI: 10.3389/fonc.2026.1933456