Training and Validation of a 5-microRNA Signature to Predict Osteosarcoma Relapse

Background: Multi-kinase inhibitors have recently been found to be active in recurrent or metastatic osteosarcoma (OS), but current staging methods do not accurately predict the risk of OS recurrence. Therefore, there is an urgent need for a risk assessment method to determine an accurate treatment plan. Here we investigate combinations of candidate miRNAs as clinically applicable features to accurately predict OS relapse. Methods: Univariate Cox proportional regression analysis was performed on 624 miRNAs in gene expression profiles, and 70 miRNAs were found to be associated with disease recurrence. Then, based on a Random Forest algorithm, 5 miRNAs were screened from 70 miRNAs, and a 5-miRNA relapse prognosis classifier was constructed. In addition, the prognostic and predictive accuracy of the classifier was verified in two external independent datasets. Findings: A classifier based on 5-miRNAs was developed. It was able to classify OS patients into high- and low- disease progression risk groups, with significant differences in the disease-free survival between the two groups. In addition, 5-miRNA combined with the diagnosis of metastasis significantly improved the accuracy of prognosis of recurrence within 3 years. Furthermore, the expression levels of genes involved in endocrine and hormonal pathways were significantly correlated with 5-miRNA risk scores. Interpretation: The classifier based on 5-miRNA is a reliable prognostic and predictive tool for disease recurrence in patients with osteosarcoma, which may contribute to patient counselling and individual management. Funding: Natural Science Foundation of Chongqing. Declaration of Interest: The authors declare no conflict of interest.