Rethinking Autoantibody Signature Panels for Cancer Diagnosis.

INTRODUCTION: Most pulmonary nodules found on imaging studies are indeterminate, but because of the concern for lung cancer, all patients require further evaluation with resultant radiation risk, significant cost, and delays in diagnosis. We hypothesized that a diagnostic blood test based on detection of autoantibodies against cancer antigens would be able to distinguish a benign nodule from lung cancer. METHODS: We identified a panel of 25 serum autoantibodies associated with NSCLC and constructed a protein microarray containing the autoantigens. We tested the microarray with human sera (from 125 patients with NSCLC and 125 matched controls with a benign nodule) and attempted to develop a classification algorithm that would separate the two groups. RESULTS: In the training data set the logistic regression c-index statistic was 0.691; in the validation data set, the model predicting the score generated from the training set model had a c-index of 0.490. The relationship between the score and outcome (final diagnosis) was not statistically significant (p = 0.460). CONCLUSIONS: When the current panel of antigens and assay format was used, classification algorithms based on levels of autoantibodies to cancer antigens did not prove to have statistically significant value for predicting the presence of cancer. We suggest that there are inherent biological limitations to this approach.

Duke Scholars

Author Edward F. Patz Jr. Radiology, Cardiothoracic Imaging

Author James Emmett Herndon II Biostatistics & Bioinformatics, Division of Biostatistics