Integrating Machine Learning for Early Mortality Prediction in Lung Adenosquamous Carcinoma: A Web-Based Prognostic...

Area of Science:

• Oncology
• Machine Learning in Medicine
• Cancer Research

Background:

• Lung adenosquamous carcinoma (ASC) is a rare but aggressive lung cancer subtype.
• Understanding its prognostic factors and mortality is crucial for effective treatment.
• Existing predictive models may not fully capture ASC's complex behavior.

Purpose of the Study:

• To quantify 90-day mortality in ASC patients.
• To identify significant clinical features associated with ASC outcomes.
• To develop and validate a machine learning model for predicting ASC mortality.

Main Methods:

• Retrospective analysis of 2820 ASC patients from the SEER database (2000-2018).
• Utilized logistic regression, Lasso, and XGBoost for feature selection and model development.
• Assessed model performance using AUC, KS statistic, DCA, and calibration plots; employed RCS for non-linear relationships.

Main Results:

• Identified 6 significant clinical features impacting ASC patient outcomes.
• The XGBoost model demonstrated superior predictive performance (AUC 0.97 training, 0.84 validation) over other models.
• Discovered a non-linear association between tumor size (cutoff 44 mm) and prognosis.

Conclusions:

• A novel, high-performing machine learning model for predicting 90-day mortality in ASC has been developed.
• The model, accessible via a web platform, supports personalized clinical decision-making.
• This tool can help optimize treatment strategies for lung adenosquamous carcinoma.