Objectives: Non-small cell lung cancer (NSCLC) is the leading cause of cancer-related mortality. Development of an early diagnosis method may improve survivals. We aimed to develop a new diagnostic model for NSCLC using serum biomarkers.
Methods: We set up a patient group diagnosed with NSCLC (n = 122) and a healthy control group (n = 225). Thirty serum analytes were selected on the basis of previous studies and a literature search. An antibody-bead array of 30 markers was constructed using the Luminex bead array platform (Luminex Inc, Austin, Tex) and was analyzed. Each marker was ranked by importance using the random forest method and then selected. Using selected markers, multivariate classification algorithms were constructed and were validated by application to independent validation cohort of 21 NSCLC and 28 control subjects.
Results: There was no difference in demographics between patients and the control population except for age (64.8 ± 10.0 for patients vs 53.0 ± 7.6 years for the control group). Among the 30 serum proteins, 23 showed a difference between the 2 groups (12 increased and 11 decreased in the patient group). We found the highest accuracy of multivariate classification algorithms when using the 5 highest-ranked biomarkers (A1AT, CYFRA 21-1, IGF-1, RANTES, AFP). When we applied the algorithms on a validation cohort, each method recognized the patients from the controls with high accuracy (89.8% with random forest, 91.8% with support vector machine, 88.2% with linear discriminant analysis, and 90.5% with logistic regression).
Conclusions: We confirmed that a new diagnostic method using 5 serum biomarkers profiling constructed by multivariate classification algorithms could distinguish NSCLC from healthy controls with high accuracy.
Copyright © 2012 The American Association for Thoracic Surgery. Published by Mosby, Inc. All rights reserved.