An international multi-center study to assess the clinical accuracy of the molecular subclassification of leukemia by gene expression profiling

Abstract

6522 Background: Microarray analysis can identify differentially expressed genes of pediatric and adult leukemias. Recently, the MILE (Microarray Innovations in Leukemia) study has started in 11 centers (European Leukemia Network: 7, USA: 3, Singapore: 1). MILE compares the accuracy of gene expression profiles of 16 acute and chronic leukemia subclasses, MDS, and non-leukemia as control group to routine diagnostic workup. Methods: In a pre-phase each center was trained on the identical microarray protocol (HG-U133 Plus 2.0 microarrays). 2 cell lines (MCF-7, HEPG2) and cell lysates of three leukemia patients (AML t(8;21), CML, CLL) were tested. After proficiency testing, centers started to analyze 2000 leukemia samples. Results: The pre-phase demonstrated a very high intra- and inter-laboratory comparability among centers. Using principal component analysis each sample type was clustered with no observable sub-grouping of the different centers. Replicates of leukemia samples demonstrated squared correlation coefficients between 0.930–0.997 for CML, 0.936–0.998 for CLL, and 0.940–0.999 for AML t(8;21).Here we present classification results of the first series of 607 pts. that were included prospectively in a training data set (total n=1,049) to form linear classifiers for all 153 class pairs. The average cross-validation accuracy is 89.4%. In first independent cohort (HG-U133 Plus 2.0, n=105) 89.5% classification accuracy were achieved. In a second independent cohort (n=1,094), analyzed previously in two centers on HG-U133A/B microarrays, 83.5% classification accuracy were achieved. In detail, 136 out of 139 (97.8%) chronic leukemia samples (CML or CLL) were classified fully in agreement with standard diagnostic procedures. For acute leukemia subtypes 767 out of 904 (84.8%) were classified correctly. Interestingly, an AML-like signature can be found in MDS samples with IPSS >1.5. Conclusions: This international multi-center study demonstrates a very high inter- and intralaboratory reproducibility of microarray analyses. Data will be used to design a new custom format microarray dedicated to further develop the application of gene expression profiling for diagnosis and subclassification of leukemia. No significant financial relationships to disclose.