Development and Validation of Novel Deep-Learning Models Using Multiple Data Types for Lung Cancer Survival

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning in Medicine

Background:

• Lack of a comprehensive lung cancer survival prediction model integrating multiple data types and novel machine learning algorithms.
• Need for identifying critical factors influencing non-small cell lung cancer (NSCLC) patient survival.

Purpose of the Study:

• To develop and validate a robust lung cancer survival prediction model using diverse data and machine learning.
• To identify key prognostic factors for non-small cell lung cancer survival.

Main Methods:

• Retrospective cohort study of 3714 NSCLC patients (2008-2018) from Taipei Medical University and Taiwan Cancer Registry.
• Utilized demographics, comorbidities, medications, lab results, and gene tests as input variables.
• Applied nine machine learning algorithms, with performance evaluated using Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• The artificial neural network (ANN) model achieved the best performance (AUC=0.89) when integrating all data types.
• Key predictors identified include cancer stage, tumor size, age at diagnosis, smoking/drinking status, EGFR gene mutation, and BMI.
• The ANN model demonstrated high predictive accuracy (Accuracy=0.82, Precision=0.91, Recall=0.75, F1-score=0.65).

Conclusions:

• Integrating diverse data types significantly enhances the predictive performance of lung cancer survival models.
• The developed ANN model offers a promising tool for predicting NSCLC patient survival and understanding prognostic factors.
• This study addresses a gap in the literature by providing a well-validated, multi-modal predictive model for lung cancer survival.