Predicting Chronic Obstructive Pulmonary Disease (COPD) Diagnosis Using Primary Care Variables and Machine Learning Algorithms

Chronic Obstructive Pulmonary Disease (COPD) affects many adults over the age of 50. Part of its incidence in the population is attributed to work and exposure to risk factors such as silica dust, and anticipating the diagnosis can prevent its worsening. This study aims to identify patients at higher risk of having a positive COPD diagnosis using variables routinely collected in primary care. A total of 120,294 participants from the UK Biobank database, recruited between 2006 and 2010, were analyzed. Of these, 1,837 (1.5%) had a positive COPD diagnosis. A total of 20 variables, including demographic data, laboratory tests, habits, and symptoms, were selected to build predictive models of COPD using five machine learning algorithms (artificial neural networks, extra trees, random forests, catboost, and extreme gradient boosting). Additionally, a subset of 7,628 participants with a history in the construction and mining industries was selected to train a specialized model. Among them, 248 (3.25%) had a positive diagnosis. Data were randomly divided, with 70% allocated for training the models and 30% for performance testing. Both models showed good predictive performance. The general model achieved an AUC of 0.847, sensitivity of 0.786, and specificity of 0.765. In the specialist model, an AUC of 0.830, sensitivity of 0.773, and specificity of 0.773 were obtained. The five main predictive variables were chronic cough, age, history of asthma, sputum production, and tobacco exposure. The results demonstrate that it is possible to predict the individual risk of COPD diagnosis using variables commonly collected in primary care.