Abstract
BACKGROUND: This study was designed to establish a biomarker risk model for predicting bone metastasis in stage III non-small cell lung cancer (NSCLC). METHODS: The model consists of 105 cases of stage III NSCLC, who were treated and followed up. The patients were divided into bone metastasis group (n = 45) and non-bone metastasis group (other visceral metastasis and those without recurrence) (n = 60). Tissue microarrays were constructed for immunohistochemical study of 10 molecular markers associated with bone metastasis, based on which a model was established via logistic regression analysis for predicting the risk of bone metastases. The model was prospectively validated in another 40 patients with stage III NSCLC. RESULTS: The molecular model for predicting bone metastasis was logit (P) = - 2.538 + 2.808 CXCR4 +1.629 BSP +0.846 OPN-2.939 BMP4. ROC test showed that when P ≥ 0.408, the sensitivity was up to 71% and specificity of 70%. Model validation in the 40 cases in clinical trial (NCT 01124253) demonstrated that the prediction sensitivity of the model was 85.7%, specificity 66.7%, Kappa: 0.618, with a high degree of consistency. CONCLUSION: The molecular model combining CXCR4, BSP, OPN and BMP4 could help predict the risk of bone metastasis in stage IIIa and IIIb resected NSCLC.