Artificial neural network risk prediction of COPD exacerbations using urine biomarkers.

BACKGROUND

COPD exacerbations cause considerable morbidity and mortality. We sought to identify a panel of urine biomarkers that can distinguish between stable and exacerbation states and predict risk of future exacerbations.

METHODS

A retrospective discovery study was done measuring 35 biomarkers implicated in COPD pathogenesis in paired urine samples from 55 COPD subjects during stable and exacerbation states. A logistic regression model combining the 10 most discriminatory biomarkers in distinguishing between stable and exacerbation states was developed as a near-patient dipstick test with an opto-electronic reader. This biomarker panel was tested in a prospective study of 105 COPD subjects who undertook daily home urine testing over 6 months. The regression model was validated in paired samples from 26 individuals out of 105. An artificial neural network (ANN) using the urine biomarkers from 85 out of 105 subjects was developed and tested as a clinical decision tool to predict risk of an exacerbation.

RESULTS

The 10-biomarker panel (NGAL, TIMP1, CRP, fibrinogen, CC16, fMLP, TIMP2, A1AT, B2M and MMP8) was able to distinguish exacerbation stable state in the discovery study (ROC with an AUC 0.84, 95% CI 0.76-0.92; p <0.01) and validation study (AUC 0.81, 95% CI 0.70-0.92, p<0.01). The ANN model predicted an exacerbation within a 13-day window frame with an AUC 0.89 (95% CI 0.89-0.90) and identified an exacerbation median (interquartile range) 7 (5-9) days prior to clinical diagnosis.

CONCLUSION

We identified a panel of biomarkers that can distinguish between stable and exacerbation state, and using an ANN model, it can predict exacerbations before symptoms occur.