An International Multi-Center Study To Define the Application of Microarrays in the Diagnosis and Subclassification of Leukemia (MILE Study): Interim Analysis Based on 1,889 Patients Achieves 95.4% Prediction Accuracy.

We have demonstrated that microarrays can identify differentially expressed genes associated with distinct clinical and prognostically relevant classes of pediatric and adult leukemias. In 2005, the MILE (Microarray Innovations in LEukemia) study research program was initiated within 11 centers: 7 from the European Leukemia Network (WP13), 3 from the US, and 1 in Singapore. In this study, the clinical accuracy of gene expression profiles (GEP) was compared to current routine diagnostic workup. As a first step, microarray analysis protocols and procedures were successfully standardized. All participating laboratories demonstrated high proficiency in this pre-phase. Here, we present data on MILE stage I where n=1,889 samples were profiled between 10/2005 and 7/2006 (HG-U133 Plus 2.0, Affymetrix). In 98.24%, the generated GEP did pass our strict quality acceptance criteria. GEP from this study were then combined with previous data obtained by the research groups from Munich (Haferlach) and Memphis (Downing) to generate a dataset of n=2,916 samples that includes 16 subclasses of acute and chronic leukemias (mature B-ALL with t(8;14), pro-B-ALL with t(11q23)/MLL, c-ALL/pre-B-ALL with t(9;22), T-ALL, ALL with t(12;21), ALL with t(1;19), ALL with hyperdiploid karyotype, c-ALL/Pre-B-ALL without t(9;22), AML with t(8;21), AML with t(15;17), AML with inv(16)/t(16;16), AML with t(11q23)/MLL, AML with normal karyotype or other abnormalities, AML with complex aberrant karyotype, CML, CLL), MDS, as well as non-leukemia and healthy bone marrows as a control group. A linear discriminant classification algorithm was created and predicted these 18 classes with a 91.4% accuracy (30-fold CV). Miscalls were mostly observed in the distinction between MDS and AML with normal karyotype. As a consequence, a classification model trained on n=2,647 samples representing 17 classes and excluding MDS reached a prediction accuracy of 95.4%. In particular, the classes Pro-B-ALL with t(11q23)/MLL, T-ALL, ALL with t(1;19), AML with t(8;21), AML with t(15;17), AML with inv(16)/t(16;16), CLL, CML, as well as non-leukemias were classified with a ≥97.2% sensitivity. The specificity was ≥98.4% for all 17 classes, respectively. Based on these results a customized microarray was designed using 1,449 probe sets. In conclusion, this international multi-center research study demonstrates a very high accuracy for leukemia classification by GEP. This effort led to an innovative customized microarray designed for a routine diagnostic application. Starting in 9/2006 additional n=2,000 samples will be prospectively analyzed with the custom research AmpliChip Leukemia Test (MILE Stage II).