Abstract
Multivariate analysis classification of chronic lymphocytic leukemia (CLL) and lymphoma (non-CLL) disorders is investigated in 299 patients by an extended panel of surface markers, and compared with Matutes classical scoring proposal. Diagnosis was based on clinical features, cell morphology, node or bone marrow histology, and immunological scoring system. Results are obtained on directly labeled tumoral cells by flow cytometry gating. Patients included 154 CLL, 2 Richter transformation, and 143 lymphoma (26 follicular, 49 lymphocytic, 18 other low-grade, 7 Waldenström macroglobulinemia, 13 mantel, 11 diffuse large-cell, 6 Burkitt, 4 marginal zone-cell, 5 hairy-cell leukemia, 2 MALT, 1 prolymphocytic leukemia, 1 SLVL). For CD43, FMC7, CD23, CD5, CD79b (% stained cells) and CD20, CD22 surface antigen intensities Chi-Square values indicate very high probability of correct classification (varing from 621 to 94.9; p<0.0000). If, alternatively, % of CD22, CD20, CD19 and intensities of CD79b, CD5, CD19, CD43, CD23 and kappa/lamba chains are employed, Chi-Square yields values of lower significance (varing from 65 to 0.1; p<0.0000 to 0.6573). Using classical panel scoring with CD79b, 82.4 % of patients were correctly classified, compared to 84.5% after replacing CD79b by CD22 intensity. If CD43 is added, correct classification increased to 89.6% and 88.1% of patients, respectively; this improvement is due to better allocation of CLL. In discriminant analysis 91.3% of patients are correctly classified with the panel including CD79b, and 90.9% with CD22 intensity. CD43 enhances the allocation of either one to 94.3%. Using our previous discriminant analysis with CD79b (Jung G, et al. Br J Haematol.2003; 120:496–499), this blind analysis correctly classified the population in 87.1%, compared to 91.3% with the new one. By adding CD43, it moved from 92.4% up to 94.3%. In order to find the optimal combination of the selected best markers, a stepwise probit discrimination was performed. Using CD43 and FMC7 yields a correct classification of 90.3%; after addition of CD5, CD79b, CD23, and CD22 intensity, efficiency increased to 94.6%. Further added markers don’t improve classification. Efficiency of this panel was further confirmed by hierarchical cluster and principal components analysis. Cluster analysis with squared Euclidian distances separated CLL from non-CLL patients with low overlaps: 86.6% of cases are correctly identified. Separated points in the plot representing patients with CLL and non-CLL, obtained by principal components analysis of surface markers, confirm the high predictive potential of this panel. The same analysis of surface marker positions for non-CLL suggests use of: % of CD79b, FMC7, and CD22 intensity, and for CLL: % of CD5, CD23, CD43. So, the addition of CD43 improves as well the discriminant function as the scoring system. Our selected panel of best markers is useful in distinguishing CLL from non-CLL and offers a better distinction by discriminant analysis. Furthermore quantitative expression of each marker and its predictive value improve diagnosis and classification.
The Use of Artificial Neural Networks and a General Discriminant Analysis for Predicting Culling Reasons in Holstein-Friesian Cows Based on First-Lactation Performance Records