Prediction of minimal residual disease in pediatric ALL: Comparison and evaluation of different mathematical methods applied to data from the ALL-BFM 2000 trial

A Torge; M Zimmermann; A Möricke; R Köhler; J Alten; CR Bartram; M Schrappe; M Stanulla

doi:10.1055/s-0034-1374828

Klinische Pädiatrie, Table of Contents

Prediction of minimal residual disease in pediatric ALL: Comparison and evaluation of different mathematical methods applied to data from the ALL-BFM 2000 trial

A Torge ¹^,², M Zimmermann ¹, A Möricke ², R Köhler ³, J Alten ², CR Bartram ³, M Schrappe ², M Stanulla ¹

¹Pediatric Hematology and Oncology, Hannover Medical School, Hannover, Germany
²Department of Pediatrics, University Hospital Schleswig-Holstein, Kiel Germany
³Department of Human Genetics, University of Heidelberg, Heidelberg, Germany

Abstract

In pediatric acute lymphoblastic leukemia (ALL), the most important prognostic factor for the risk of relapse is the evaluation of minimal residual disease (MRD) at defined timepoints. The results of MRD analyses are not available at diagnosis to guide selection of the optimal treatment intensity already early on.

In trial ALL BFM 2000, multi-level patient data were collected and are now used to develop models to predict MRD.

Different methods of model developing like neural networks, decision trees, genetic algorithms and support vector mashines are compared focusing on their potential and their accuracy in predicting MRD.

Involving gene expression data might help to improve the predictions, but due to the big amount of additional information, leads soon to overfitting. Therefore the genes considered in the analysis have to be chosen carefully.