Age at diagnosis, stage, and MYCN amplification are the cornerstones of the risk-stratification score of neuroblastoma that enables defining patients at low- and high risk. Refinement of this stratification is needed to optimize standard treatment and to plan future clinical trials. We investigated whether two parental imprinted miRNAs (miR-487b and miR-516a-5p) may lead to a risk score with a better discrimination. Expression levels of maternal miR-487b and paternal miR-516a-5p were determined using quantitative RT-PCR both for 231 neuroblastoma tumors (derivation set) and 101 independent neuroblastoma tumors (validation set). Survival outcomes were overall survival (OS) and disease-free survival (DFS). Multivariable Cox models were developed from derivation set and their performance evaluated using Akaike's information criterion (AIC) (goodness-of-fit) and time-dependent area under curves (discrimination). The selected model was validated using internal and external validation. The prognostic model including current prognostic factors plus miR-487b, miR-516a-5p, and interaction between two miRNAs was selected. Performance of this model was better in terms of both predictive ability (smallest AIC) and discrimination power (AUC close to 0.70). This model identifies three risk groups: high (3), intermediate (2), and low (1). Hazard ratios (HR) across risk groups were HR2/1 = 6.3 (2.7-14.6), HR3/1 = 14.8 (7.2-30.2) for OS and HR2/1 = 2.8 (1.5-5.4), HR3/1 = 7.2 (3.9-13.4) for DFS. The rank between these three risk groups was maintained and validated when performing internal and external validation. Expression of maternal miR-487b and paternal miR-516a-5p improves the risk stratification. This better discrimination at diagnosis is of clinical utility both for current and future treatments of neuroblastoma patients.
Keywords: miRNA clusters; neuroblastoma; parental imprinting; prognostic model; risk score; survival outcomes.
© 2014 The Authors. Cancer Medicine published by John Wiley & Sons Ltd.