How low is too low? Use of cluster analysis to define low levels of mercaptopurine metabolites.

BACKGROUND: To group patients receiving treatment for acute lymphoblastic leukemia (ALL) according to their oral mercaptopurine (6-MP) metabolite levels and to establish cut-off points to screen for potential poor clinical outcome. PROCEDURE: Methodological study using 6-MP metabolite levels from 48 adolescent ALL patients enrolled in a multicenter adherence study. Cluster analysis was the primary analytical technique. We used two validation methods (a split-sampling and a simulation technique) for validating the results.
RESULTS: Four clusters were retained in our initial analysis using our first group of patients (n = 27). Three clusters (labeled 1, 2, and 4) exhibited the expected negative correlation between the two metabolites, that is, "high" values of one were associated with "low" values of the other. One cluster (labeled 3) had "low" levels for both TGN and MMP. Five of the 27 adolescents had their 6-MP "held" during the study. Post-hoc examination of the results revealed that all five grouped in Cluster 3 during the time that their medications were stopped, but grouped in other clusters at other times. The median ANC was highest in Cluster 3, consistent with low therapeutic drug levels. Parameters were reproducible with both validation methods. Values below the respective 75th centile for both TGN and MMP in Cluster 3 for the complete sample (n = 48) are suggested as representing a potentially higher risk for relapse.
CONCLUSIONS: This study provides an objective method for identifying patients at risk for treatment failure due to suboptimal 6-MP therapy; the clinical significance of this approach should be examined in future studies.