Analysis of copd disease stages using clinical and demographic data

This study aims to predict the stages of Chronic Obstructive Pulmonary Disease (COPD) using machine learning methods, highlighting that disease progression is associated not only with the respiratory system but also with systemic factors such as cardiovascular parameters, metabolic indicators, and environmental exposures. The analysis utilized a dataset of 230 patients from a Barcelona cohort, sourced from the "COPD Analysis" collection on Kaggle, encompassing variables like age, gender, body mass index (BMI), smoking history, respiratory rate, oxygen saturation, FEV1/FVC ratio, blood pressure, heart rate, sputum, depression, and occupational exposures. The COPD stages (GOLD 1-4) were divided into binary classification groups, and various machine learning algorithms (Decision Tree, Random Forest, K-Nearest Neighbor, Gradient Boosting, Artificial Neural Networks) were tested. The highest classification performance was achieved with the K-Nearest Neighbor (KNN) algorithm for the GOLD 2-3 stage transition, yielding an AUC value of 94.9%. Performance declined in other transitions (GOLD 1-2 and 3-4), reflecting the heterogeneity in early and advanced stages. SHAP (SHapley Additive exPlanations) analysis revealed that parameters such as respiratory rate, blood pressure, sputum, and BMI play critical roles in stage prediction, while traditional metrics like FEV1 were less determinant. The findings underscore the need for individualized COPD management, advocating for the monitoring of systemic parameters and tailored treatment strategies for early intervention. This approach enhances the potential of machine learning in COPD diagnosis and prognosis, contributing to clinical practice.