Discrimination of Chronic Lymphocytic Leukemia of B-Cell Type by Computerized Analysis of 3-Color Flow Cytometric Immunophenotypes of Bone Marrow Aspirates and Peripheral Blood

Restriction of κand λ-chain expression on CD19/CD5 double-positive bone marrow and peripheral blood lymphoid cells discriminates between malignant lymphoma and reactive lymphoid cells. Although this discrimination is unproblematic for high light-chain expression, difficulties arise for low-level expression or when a few malignant cells are dispersed in a majority of normal or reactive lymphocytes. This study investigated the suitability of the κ/CD19/5, λ/CD19/5, CD45/14/20, CD4/8/3 antibody panel for automated malignancy discrimination. Normal/κ B-cell chronic lymphocytic leukemia (B-CLL)/λ B-CLL bone marrow aspirate (BMA) samples were identified with predictive values of 100.0%/100.0%/96.4%. Normal peripheral blood leukocytes (PBL) can be used as a reference for κ/λ B-CLL BMA (100.0%/92.6%/92.9%) as a simplification. PBL samples were classified with slightly lower predictive values of 95.0%/96.9%/95.0%. The unknown embedded test set of clearly recognizable κ and λ B-CLL PBL and BMA samples was identified with predictive values of 100%/100% and 90.0%/100%. Manually unclassifiable BMA and PBL samples of 5 patients were classified by the CLASSIF1 algorithm (http://www.biochem. mpg.de/valet/classif1.html) in a logically consistent way, showing the usefulness of automated classification in otherwise difficult cases. In 2 patients, PBL samples were recognized as malignant, but no distinction between κ and λ B-CLL was possible. Data pattern analysis also detects differences between κ PBL and κ BMA cells in B-CLL patients with predictive values of 98.3%/100.0% and between λ PBL and λ BMA with values of 92.6%/96.4%. Furthermore, an automated BMA quality assessment can be performed with predictive values of 98.3%/100.0% for PBL/BMA discrimination. Together, these results show the interesting potential of medical bioinformatics data mining for the automated classification of complex clinical multiparameter flow cytometry data. Lab Hematol. 2002;8:134-142