HOME HELP FEEDBACK HOW TO CITE ABSTRACTS ARCHIVE CME INFORMATION SEARCH Cancer Research Clinical Cancer Research Cancer Epidemiology Biomarkers & Prevention Molecular Cancer Therapeutics Molecular Cancer Research Cancer Prevention Research Cancer Prevention Journals Portal Cancer Reviews Online Annual Meeting Education Book Meeting Abstracts Online		QUICK SEARCH:   [advanced] Author: Keyword(s):

This Article Services Similar articles in this journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Lyons, R. Articles by Hubank, M. Search for Related Content PubMed Articles by Lyons, R. Articles by Hubank, M.

Diagnostic Technologies and Molecular and Cellular Profiling: Bioinformatics

Diagnosing pediatric leukemia: Clinical application of microarray data

Institute of Child Health, London, United Kingdom and Great Ormond Street Hospital, London, United Kingdom

Abstract

A1

Pediatric leukemias are commonly characterized by the presence of fusion genes that give rise to different subtypes of leukemia and play an important role in prognosis. Prognistically important subgroups include AML1-ETO, PML-RARA, CBFB-MYH11, MLL rearrangement (in both AML and ALL), BCR-ABL and TEL-AML1. These subtypes have unique gene expression signatures which have been used to develop diagnostic classifiers. The robustness of classifiers and their potential clinical application requires careful investigation.We have analyzed four published gene expression classifiers for diagnostic utility using three pediatric leukemia datasets (Ross et al. (2003 and 2004) and van Delft et al. 2005), totaling 392 patients. We applied a robust classification approach, using 4 different algorithms, to a training set of 268 patients and created gene lists for the leukemia subtypes listed above. Clinically important in determining the diagnosis of patients as either AML or ALL we have determined a subset of 11 genes capable of distinguishing these two classes with an accuracy of 99% in a test set of 204 samples. Application of a robust classifier of 18 genes to a test set of 128 ALL patients decreased the rate of misdiagnosis of TEL-AML1 to 1.56%, compared to 7.03% using the Ross et al. classifier and 6.25% using van Delft et al. gene list. In the AML test set (76 patients) AML1-ETO was misdiagnosed by 2.63% using van Delft et al. gene list, 1.32% by Ross et al. classifier and no patients were misdiagnosed using the genes selected by our robust classification method (52 genes*).This robust gene list selection method improves accuracy and can be applied to any microarray dataset. Assigning a probability to subtype selection allows us to attach a significance value to the certainty of correct diagnosis and so the clinician can decide whether further molecular biology tests are required.* (only 40 were used in the van Delft et al. gene list as this was the maximum number of genes available in the supplementary data)