Oral Presentation Advances in Neuroblastoma Research Congress 2016

Molecular risk assessment of neuroblastoma patients eliminates the necessity of clinical prognostic markers (#116)

Carolina Rosswog 1 , Rene Schmidt 2 , André Oberthuer 1 , Dilafruz Juraeva 3 , Benedikt Brors 3 , Anne Engesser 1 , Yvonne Kahlert 1 , Ruth Volland 1 , Frank Berthold 1 , Thorsten Simon 1 , Barbara Hero 1 , Andreas Faldum 2 , Matthias Fischer 1 4 5
  1. Children’s Hospital, Department of Pediatric Oncology and Hematology, University of Cologne, Cologne, Germany
  2. Institute of Biostatistics and Clinical Research, University of Muenster, Muenster, Germany
  3. Department of Theoretical Bioinformatics, German Cancer Research Center, Heidelberg, Germany
  4. Max Planck Institute for Metabolism Research, Cologne, Germany
  5. Center for Molecular Medicine Cologne (CMMC), University of Cologne, Cologne, Germany

Purpose: Current risk stratification systems for neuroblastoma patients consider clinical, histopathological and simple genetic variables. Recent studies, however, suggested that gene expression-based predictors may add significant value for prognostic classification. We here aimed to improve neuroblastoma risk assessment by developing a risk score that integrates hazard ratios of established prognostic variables and multigene expression-based predictors.

Patients and methods: We analyzed 695 neuroblastoma patients. The entire cohort was divided into training set I (n=75) for generation of multigene predictors, training set II (n=411) for risk score development, and a validation set (n=209). Prognostic variables were selected in a multivariable Cox regression analysis based on event-free survival (EFS) using the LASSO method, followed by subsequent backward selection. Selected variables and their hazard ratios were included into a prognostic index, which was then used for development of the final risk score.

Results: The variables stage, age, MYCN status and the multigene predictors NB_th24 and NB_th44 were independent prognostic markers in the LASSO analysis. In the subsequent backward selection procedure, only the two multigene predictors were retained in the final model. Integration of the two predictors in a risk scoring system identified three patient subgroups that differed significantly in their outcome both in the training (5-year-EFS, 83.2±2.6 vs. 64.8±9.0 vs. 32.0±4.0; p<0.001) and the validation cohort (5-year-EFS, 84.9±3.4 vs. 63.6±14.5 vs. 31.0±5.4; p<0.001). Multivariable analysis of risk groups defined by the current German NB2004 stratification system and the newly developed molecular risk score revealed only the molecular risk score as independent predictor for EFS.

Conclusion: We here propose a strategy for the development of neuroblastoma risk assessment that integrates prognostic variables based on hazard ratios. The final risk score considered only two multigene predictors, supporting the notion that clinical courses of neuroblastoma are precisely reflected by the molecular properties of the tumor cells.