| Popis: |
Introduction: Lung cancer is a leading cause of cancer mortality. Exhaled-breath analysis of volatile organic compounds (VOC’s) may detect lung cancer at an early stage, possibly leading to better outcomes. Artificial neural networks (ANN) have proven to be accurate tools to diagnose lung cancer by distinguishing between breath profiles of healthy and sick individuals. Adding readily available clinical information to the ANN may improve the diagnostic accuracy. Methods: Subjects with non-small cell lung cancer (NSCLC) and healthy controls breathed into the Aeonose™ (The eNose Company, Zutphen, Netherlands). Diagnostic accuracy, presented as Area under the Curve (AUC) was studied in a prospective, multi-center study in 282 individuals of whom 140 had confirmed NSCLC. We compared a 6-element breath-vector (Aeonose-only) with a 10-element vector (Aeonose plus age, pack years, COPD-presence, and gender). Results: Confirmed NSCLC patients (67.1 (9.0) years; 57.6% male) were compared with non-NSCLC controls (62.1 (7.1) years; 40.4% male). The AUC based on the Aeonose-only classification was 0.75 (95% CI: 0.69-0.81). Adding age, number of pack years, presence of COPD, and gender to the ANN resulted in an improved performance with an AUC of 0.84 (95% CI: 0.79-0.88). By choosing an appropriate threshold value in the ROC-diagram of the multivariate model, we observed a sensitivity of 92.9%, a specificity of 53.3%, and a positive and negative predictive value of 66.3% and 88.4%, respectively. Conclusion: The diagnostic accuracy to predict presence or absence of lung cancer can be improved significantly by adding readily available clinical information to the ANN. |
