Abstract
BACKGROUND: Computed tomography (CT) scanning has emerged as an effective means of early detection for lung cancer. Despite marked improvement over earlier methodologies, the low level of specificity demonstrated by CT scanning has limited its clinical implementation as a screening tool. A minimally-invasive biomarker-based test that could further characterize CT-positive patients based on risk of malignancy would greatly enhance its clinical efficacy. METHODS: We performed an analysis of 81 serum proteins in 92 patients diagnosed with lung cancer and 172 CT-screened control individuals. We utilize a series of bioinformatics algorithms including Metropolis-Monte Carlo, artificial neural networks, Naïve Bayes, and additive logistic regression to identify multimarker panels capable of discriminating cases from controls with high levels of sensitivity and specificity in distinct training and independent validation sets. RESULTS: A three-biomarker panel comprised of MIF, prolactin, and thrombospondin identified using the Metropolis-Monte Carlo algorithm provided the best classification with a %Sensitivity/Specificity/Accuracy of 74/90/86 in the training set and 70/93/82 in the validation set. This panel was effective in the classification of control individuals demonstrating suspicious pulmonary nodules and stage I lung cancer patients. CONCLUSIONS: The selected serum biomarker panel demonstrated a high diagnostic utility in the current study and performance characteristics which compare favorably with previous reports. Further advancements may lead to the development of a diagnostic tool useful as an adjunct to CT-scanning.