Key points are not available for this paper at this time.
In childhood acute lymphoblastic leukemia, univariate analyses have identified more than thirty clinical and laboratory prognostic factors. Multivariate analysis is used to sort out the most independent and predictive of these factors, but the methods used are mathematically and conceptually complex. In this report, we depict the interactions between multiple variables graphically using 3-dimensional data displays. This approach helps conceptualize the multivariate process and provides an alternative method. We applied the method to the 2,987 children in the CCG-160 series of studies, and specifically to the "high-risk" subgroups: initial white cell count (WBC) greater than 50,000/microliter; white count greater than 50,000/microliter in conjunction with a mediastinal mass; age at diagnosis less than 1 year; and, central nervous system (CNS) leukemia at diagnosis. Analyzed in this fashion, T-cell immunophenotype, the presence of a large mediastinal mass, CNS leukemia, and the lymphomatous pattern lose their prognostic value when outcome is stratified by age and WBC. The 3-dimensional method confirms the mathematical analyses and provides graphic evidence for the conclusions.
Bleyer et al. (1986) studied this question.