Fig. 2

Predictive model for discontinuation. A Data from 30 patients on TFR who did not relapse and 15 patients who relapsed during TFR were included in a decision tree analysis by incorporating MCP-1 and IL-6 levels. This recursive partitioning showed that all patients were divided into groups according to the relapse condition. In the first partition, MCP-1 was considered the criterion; in a terminal node, MCP-1^hi group (> 265 pg/mL) included 25 patients, of which only 2 relapsed, while MCP-1^low group (< 265 pg/mL) was further split by incorporating IL-6 measurement. When MCP-1^low and IL-6 were combined, a relatively small percentage of patients (3/20) who did not relapse could be captured in the MCP-1^low/IL-6^hi group (> 4.5 pg/mL); on the other hand, the double low group (MCP-1^low/IL-6^low) selectively captured those patients who lost molecular response (13 out of 17 (76%)). B Confusion matrix used to define the performance of a decision tree model. To estimate the power of the model, an F1-score was calculated and a value of 0.81 was determined, suggesting a very high prediction power. C Molecular recurrence-free survival for the MCP-1^high, MCP-1^low/IL-6^high and the MCP-1^low/IL-6^low groups at the time of discontinuation. D Multivariate Cox proportional hazard analysis including the most relevant clinical and biological variables and IL-6/MCP-1 cytokines levels combined