Prediction of Lung Cancer Metastasis Using Machine Learning Models Based on Clinical Laboratory Data.

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Lymph node (N) or distant metastasis in lung cancer significantly worsens prognosis.
• Current diagnostic methods like CT scans often require combination with other tests for effective metastasis assessment, limiting clinical utility.
• Predicting N involvement and skip N metastasis is crucial for accurate lung cancer staging and treatment planning.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting lymph node invasion (N) and skip N metastasis (M) in lung cancer.
• To identify key clinical laboratory parameters that serve as predictors for N and M metastasis.
• To leverage diverse clinical data for enhanced accuracy in lung cancer metastasis assessment.

Main Methods:

• Regression analysis was performed on histopathologically diagnosed lung cancer cases.
• Univariate analysis and LASSO regression were used to identify significant predictors from clinical laboratory data.
• Four ML algorithms were employed to build predictive models, with logistic regression identified as optimal.

Main Results:

• Analysis of 1629 cases revealed significant differences in numerous parameters between N and M groups.
• LASSO regression identified 13 key factors for N metastasis and 12 for M metastasis.
• The optimal logistic regression models achieved high predictive performance with AUC values of 0.888 for N and 0.875 for M metastasis.

Conclusions:

• Machine learning algorithms utilizing clinical laboratory data show significant potential for predicting N involvement and skip N metastasis in lung cancer.
• The identified key predictors offer insights into the biological mechanisms underlying lung cancer metastasis.
• These ML-driven predictive models could enhance clinical decision-making for lung cancer patients.