LNM

LNM predicts lymph node metastasis and survival outcomes for patients with soft tissue sarcoma (STS) and kidney cancer using statistical and machine learning models.

Key Features:

• Predictive models for soft tissue sarcoma (STS): Utilizes data from the Surveillance, Epidemiology, and End Results (SEER) database and an external medical institution dataset and applies univariate and multivariate logistic regression to identify independent risk factors and construct diagnostic nomograms validated by ROC, calibration curves, and decision curve analysis (DCA).
• Prognostic nomograms for OS and CSS: Builds prognostic nomograms for overall survival (OS) and cancer-specific survival (CSS) in STS and assesses prognostic accuracy using area under the curve (AUC) metrics and calibration consistency.
• Machine learning models for kidney cancer: Trains machine learning algorithms on SEER data with emphasis on XGBoost (XGB) to predict lymph node metastasis and reports model discrimination by AUC.
• Feature selection and predictor identification: Applies least absolute shrinkage and selection operator (LASSO) alongside univariate and multivariate logistic regression to identify independent predictors of LNM.
• Model validation and clinical thresholding: Validates models using 10-fold cross-validation, ROC/AUC, calibration curves, DCA, probability density functions (PDFs), and clinical utility curves (CUCs) to evaluate performance and identify clinical utility thresholds.

Scientific Applications:

• Early detection and prognosis: Facilitates identification of lymph node metastasis in STS and kidney cancer and prediction of survival outcomes to inform prognosis and treatment planning.
• Research and development: Supports investigation of factors influencing metastasis and provides a methodological framework for developing comparable predictive models across cancer types.

Methodology:

Uses SEER and an external institution dataset; applies univariate and multivariate logistic regression, LASSO, Cox regression, XGBoost, nomogram construction, 10-fold cross-validation, ROC/AUC, calibration curves, decision curve analysis, probability density functions, and clinical utility curves.