137P - Event-free survival prediction using lncRNAs in pediatric B-cell acute lymphoblastic leukemia

Background

Despite outcomes in pediatric B-cell acute lymphoblastic leukemia (B-ALL) haing improved greatly in the past decades, 10-15% of patients will still experience an event after initial complete remission. In this context, identification of new outcome predictors and therapeutic targets is still needed. For that purpose, here we studied long non-coding RNAs (lncRNAs) as potential novel biomarkers and outcome predictors in pediatric B-ALL.

Methods

Total RNA, extracted from tumor samples at diagnosis of 50 patients from three different Spanish hospitals (development cohort, DC) and 72 samples from CHU Sainte-Justine hospital in Montreal, Canada (validation cohort, VC), was sequenced on NovaSeq 6000 System (Illumina), with a mean depth of ≈180 million paired-reads. Reads were aligned with STAR and quantified with featureCounts using lncRNAKB annotation (hg38). Univariate Cox Proportional Hazard Models (UVC) were used to identify significant genes (p < .01 & HR > 1) for five-year Event-Free Survival (EFS) in each cohort. Then, different EFS prediction models were adjusted from overlapping genes in the DC and validated in the VC. The best model was selected based on overall performance, assessed using metrics such as AUC, concordance and scaled Brier score. Finally, patients were grouped in very high-risk (R1, highest 10% predicted risk samples), high-risk (R2, next 20% highest predicted risk samples), or standard-risk (R3, lowest 70% predicted risk samples) groups to perform Kaplan-Meier (KM) survival curves.

Results

UVC resulted in 769 and 1686 significant genes for DC and VC, respectively, from which 47 were common (42 lncRNAs). Starting from those 47 genes, a model with 19 genes (16 lncRNAs) was selected. In the DC 100% of R1 patients, 20% from R2, and none of the patients from R3 group reported an event during first five years of follow-up. In the VC, 87.5% of patients from R1 had an event, 28.6% from R2, and just one patient (2%) from R3 group; being both analyses significant (p <00001) in KM analyses.

Conclusions

Our EFS prediction model is able to significantly discriminate risk groups predicting an event during the first five years of follow-up from diagnosis, being a potential prognostic predictor tool in the near precision oncology future.

Editorial acknowledgement

Clinical trial identification

Legal entity responsible for the study

The authors.

Funding

Eusko Jaurlaritza (Basque Government).

Disclosure

All authors have declared no conflicts of interest.