ESMO Congress 2024 | OncologyPRO

Abstract 832P

Background

Treating pediatric acute B-cell lymphoblastic leukemia (pB-ALL) who experience central nervous system relapses (CNSR) is challenging and often results in high mortality rates. The aim of this study was to develop and validate a predictive model for CNSR in pB-ALL, allowing for early identification and assessment.

Methods

1495 patients of pB-ALL patients from the TARGET database were randomly assigned to a training cohort of 1079 patients and an internal validation cohort of 416 patients. Multi-variable COX regression analysis was applied to construct the model using the training cohort. The validation cohorts comprised of an internal validation cohort and an external validation cohort consisting of Chinese patients. The analysis was conducted using consistency index (C-index), AUROC, Kaplan-Meier curves, calibration curve, and decision curve analysis to evaluate this model.

Results

The prediction model for CNSR in pB-ALL, named CLOMB, was constructed using indicators selected from the training cohort. The model had a C-index of 0.748 and was found to be highly accurate in predicting CNSRs in both the training and validation cohorts, as determined by AUROC measures. A statistically significant difference in event-free survival was observed between pB-ALL patients divided into CNSR high-risk and low-risk groups using the CLOMB calculated risk score of 0.76 as a cut-off value. In the external validation cohort, CLOMB had an AUROC of 0.590, which was significantly higher than MRD on days 19 and days 46.

Conclusions

The newly developed predictive model, CLOMB, has shown strong power and utility in CNS relapse prediction and risk stratification, which may help modify treatment options in clinical routine.