Enhancing deep learning models for predicting smoking Status using clinical data in patients with chronic obstructive...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Pulmonology

Background:

• Chronic obstructive pulmonary disease (COPD) is a major health concern, often complicated by persistent smoking.
• Predicting smoking persistence in COPD patients is crucial for effective intervention.
• Current prediction models may not fully capture the complexity of smoking behavior in this population.

Purpose of the Study:

• To develop and evaluate deep learning models for improved prediction of persistent smoking in COPD patients.
• To integrate behavioral, psychosocial, and clinical data for enhanced predictive accuracy.
• To identify key predictors of persistent smoking for targeted cessation interventions.

Main Methods:

• Developed and assessed three deep learning models and one machine learning model.
• Utilized clinical, behavioral, and psychosocial data from 350 COPD patients.
• Employed data preprocessing, hyperparameter optimization (Optuna), and cross-validation; used SHAP for interpretability.

Main Results:

• A Residual Neural Network achieved the highest performance with a macro F1 score of 0.87.
• Key predictors identified include professional advice to quit, employment status, sputum symptoms, perceived stress, health check-ups, and health literacy.
• Shapley Additive Explanations (SHAPs) provided insights into feature importance.

Conclusions:

• Integrating behavioral and psychosocial data significantly improves the prediction of persistent smoking in COPD.
• Multidimensional data enhance the identification of high-risk individuals for smoking cessation.
• Findings support the development of targeted cessation strategies tailored to COPD patient needs.