Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1766
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-18T01:09:44.932238Z
status: published
lens: trace
sector: health
headline: Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer Using Routinely Collected Clinical Data: Multicenter Retrospective Machine Learning Model Development and External Validation Study
dek: Background Preoperative nodal assessment in colorectal cancer (CRC) remains imperfect, and uncertainty regarding pathological lymph node metastasis (LNM) may persist despite conventional imaging. Machine learning using routinely available preoperative information may provide complementary patient-level risk estimates without requiring radiomic feature extraction or molecular testing. Objective This study aimed to develop and externally validate machine learning models for preoperative estimation of pathologicall…
gain_title: A random forest using six routine preoperative variables estimated lymph node metastasis risk in colorectal cancer and added predictive information beyond clinical tumor stage alone.
problem_title: External discrimination was only moderate, and the model is research-stage and cannot replace imaging, pathology, or clinical judgment before prospective validation.
trace_subject: Machine learning prediction of lymph node metastasis in colorectal cancer
gain_reading: A random forest using six routine preoperative variables estimated lymph node metastasis risk in colorectal cancer and added predictive information beyond clinical tumor stage alone.
gain_evidence: provided predictive information beyond cT alone
problem_reading: External discrimination was only moderate, and the model is research-stage and cannot replace imaging, pathology, or clinical judgment before prospective validation.
problem_evidence: The model showed moderate external discrimination | should not replace imaging, pathological evaluation, or clinical judgment
quick_read: Researchers built machine learning models to estimate, before surgery, whether colorectal cancer patients have pathologically confirmed lymph node metastasis. They used 2725 patients from two hospitals, training on one center's data and testing the selected random forest on 1651 patients from a second center.

The work matters because preoperative nodal assessment remains imperfect even with conventional imaging, and a model built on routine data could offer complementary patient-level risk estimates. However, performance was moderate and the retrospective design means it is not ready for clinical use until prospective testing in broader settings.
limitation: The model is intended only for research-stage adjunctive risk stratification, showed moderate external discrimination, and needs prospective validation in broader populations and clinical settings before implementation.
tag: Dual reading
key_points: Six predictors were retained: BMI, preoperative carcinoembryonic antigen, primary tumor site, clinical tumor stage, histological type, and tumor differentiation. | Seven algorithms were compared, and random forest was chosen based on internal test performance. | In external validation, sensitivity was 0.677, specificity 0.693, positive predictive value 0.569, and negative predictive value 0.781. | The six-predictor model discriminated significantly better than clinical tumor stage alone, though that stage contributed most in SHAP analysis.
rundown: The study included newly diagnosed colorectal cancer patients undergoing direct radical surgery without preoperative anticancer treatment. The internal cohort of 1074 patients was split into 753 for training and 321 for internal testing.

Predictors came from routine clinical, laboratory, tumor-marker, imaging-based, and endoscopic biopsy data and were selected with univariable and multivariable logistic regression. The approach avoids radiomic feature extraction and molecular testing.

Evaluation covered discrimination, threshold-dependent metrics, calibration, and decision curve analysis, with SHAP used for interpretation. External accuracy was 0.687, the Brier score was 0.190, and the calibration slope was 1.061.
sources:
- peer_reviewed | JMIR Cancer | https://doi.org/10.2196/101162 | 2026-10-07
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

487ade3367157275c52e05dbaf592de6ce5425dbc51973d31098580b7c3b4951

previous

0000000000000000000000000000000000000000000000000000000000000000