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

I. introduction

Chronic Obstructive Pulmonary Disease (COPD) is a condition affecting a significant portion of adults over 50 years old. Its prevalence is partly attributed to occupational exposure to risk factors such as silica dust. Early diagnosis is essential to prevent the worsening of the disease. This study aims to identify patients at higher risk of a positive COPD diagnosis using variables routinely collected in primary care.

II. libraries and databases used

III. methods

IV. computer vision methods

In addition to predictive models based on primary care variables, we employ computer vision techniques to enhance the early diagnosis of COPD by analyzing chest X-ray and CT scan images. Convolutional Neural Networks (CNNs), specifically architectures such as ResNet and EfficientNet, are trained to identify patterns consistent with COPD in lung imaging.

V. algorithm for data processing

The following algorithm was used to preprocess data before model training:

Algorithm 1

Data Preprocessing Algorithm

• Download figure
• Open in new tab

VI. expected results

The integration of machine learning models based on routine clinical data and computer vision techniques is expected to yield a robust COPD diagnostic tool. The general predictive model and the specialist model have already shown good performance (AUC of 0.847 and 0.830, respectively). For the computer vision models, we expect an AUC above 0.80, indicating reliable diagnostic support.

VII. results

The models showed good predictive performance. The general model achieved an AUC of 0.847, with a sensitivity of 0.786 and specificity of 0.765. The specialized model achieved an AUC of 0.830, with a sensitivity of 0.773 and specificity of 0.773. The main predictive variables included chronic cough, age, history of asthma, sputum production, and tobacco exposure.

VIII. discussion

The findings from this study underscore the potential of using primary care variables combined with machine learning algorithms for the early detection of COPD. The good predictive performance of both the general and specialist models indicates that routine data, such as chronic cough, age, and tobacco exposure, can effectively identify individuals at risk for COPD. Moreover, the specialist model, which focuses on participants with a history in high-risk occupations, demonstrates that occupational exposure to silica dust further enhances model performance.

Additionally, the integration of computer vision techniques, especially with convolutional neural networks (CNNs), can complement the predictive models by providing diagnostic support from chest X-rays and CT scans. While the models based on primary care variables have already shown promising results, incorporating imaging data could further improve prediction accuracy, leading to more reliable early diagnosis.

IX. conclusion

The results indicate that it is possible to predict the individual risk of a COPD diagnosis based on variables commonly collected in primary care, supplemented by computer vision techniques analyzing chest X-ray and CT scan images. This combined approach could serve as a valuable resource for early intervention in COPD patients, especially those in high-risk occupations.