Validity of an Artificial Neural Network in the Diagnosis of COPD

Background/Purpose: Neur al net wor ks anal yze a l arge amount of information and are useful in the classification of patients for the diagnosis of chronic obstructive pulmonary disease (COPD). However, its comparative performance with questionnaires for the diagnosis of COPD is unknown. The objective of the study is to evaluate the performance of a neural network against clinical questionnaires in the diagnosis of COPD. Methods: A cross-sectional study was carried out applying the clinical questionnaires and a perceptron neural network against the spirometric diagnosis of COPD. Results: A total of 1590 patients were admitted to the study, 13.5% of them were confirmed for COPD diagnosis. In the general population, average age was 67.6 years (SD = 14.0), and smoking history was 47.7% (758/1590). The questionnaire with the highest performance was the Could it be COPD with an ACOR of 0.83 (95% CI, 0.81–0.86) (p < 0.001), and the lowest performance was the LFQ with an ACOR of 0.66. (95% CI, 0.62–0.70)(p < 0.001). The ANNs showed an ACOR of 0.89 (95% CI, 0.86–0.91) (p < 0.001). Conclusion: Neural networks show a better diagnostic performance than the usual clinical questionnaires for the diagnosis of COPD.