Artificial Intelligence Applied to Forced Spirometry in Primary Care.

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence in Healthcare
• Machine Learning for Diagnostics

Background:

• Early detection of obstructive lung disease is crucial for effective COPD management.
• Primary care settings offer opportunities for active case-finding in at-risk populations.
• Spirometry is a key diagnostic tool for assessing lung function.

Purpose of the Study:

• To develop an AI-based machine learning model for predicting spirometric obstructive patterns.
• To identify key predictive variables from an active case-finding program for COPD.
• To validate the model's performance in a primary care smoking population.

Main Methods:

• Utilized a dataset of 1190 smokers aged 30-80 without prior respiratory disease history.
• Employed exploratory data analysis (EDA) and mutual information analysis to select independent variables.
• Trained a gradient boosting algorithm (GBT) and validated using cross-validation.

Main Results:

• The AI model achieved high predictive power for spirometric obstructive patterns, with an Area Under the Curve near unity.
• Key predictive variable identified: FEV1_theoretical_pre values.
• Achieved high performance metrics: 93% sensitivity, 94% positive predictive value, 97% specificity, 96% negative predictive value, 95% accuracy, and 94% precision.

Conclusions:

• Machine learning models can accurately predict obstructive patterns in spirometry using FEV1_theoretical_pre values.
• The developed model demonstrates potential for early respiratory disease detection in primary care smokers.
• Further research is recommended to incorporate clinical data and integrate AI into clinical workflows.