ALL-125: PHi-RACE: PGIMER In-House Rapid and Cost-Effective Classifier for the Rapid Detection of BCR-ABLI-Like Acute Lymphoblastic Leukemia in Indian Patients

Context: For the detection of BCR-ABL1-like ALL cases, two methodologies, specifically gene expression profiling (GEP) or next-generation targeted sequencing (NGS) with TaqMan based low-density (TLDA) card, are being used. NGS is very costly, and TLDA is not commercially available in India. We have built a binary logistic 'PHi-RACE’ classifier for rapid detection in Indian patients. Objective: BCR-ABL1-like ALL signature identification, followed by the creation of a PHi-RACE classifier using machine learning approach. Design: This study was conducted at PGIMER, a tertiary care center in India, for a period of 5 years (2017–2021). Flow cytometric immunophenotyping (FCM-IP) and multiplex RT-PCR were performed on 629 B-ALL cases to detect the positivity of BCR-ABL1 chimeric fusion transcripts. Further, 12 BCR-ABL1-positive cases were subjected to transcriptome profiling using Affymetrix microarray (U133 Plus 2.0 Array). A total of 536 B-ALL cases were subjected to GEP of 8 selected genes, followed by PHi-RACE classifier generation and validation (n=93). Patients: We examined 629 freshly diagnosed and treatment-naive B-ALL cases with complete laboratory workup. Results: Multiplex RT-PCR assay revealed BCR-ABL1 transcripts in 17.17% (108/629) of cases. Global transcriptome profiling of 12 BCR-ABL1 RNA transcripts revealed a total of 1574 differential expressed (DE) genes. DE genes were further filtered, and 45 genes with 10- to 86-fold change were identified. Based upon regression coefficient values, 8 best classifier genes were selected using penalized logistic regression. Out of 536 examined B-ALL cases, we identified 26.67% (143/536) BCR-ABL1-like ALLs using hierarchical clustering and principal component analysis. BCR-ABL1-like ALL cases were significantly older at presentation (p=0.036) and had male preponderance (p=0.047) compared to BCR-ABL1-negative ALL cases. Lastly, we built a PHi-RACE classifier (cut-off = 0.28, sensitivity = 95.2%, specificity= 83.7%, AUC= 0.927) using glm function in R, validated with 93 BCR-ABL1-negative ALL cases. Conclusions: We developed and validated a PHi-RACE classifier for the first time in a developing country. This predictive classifier is rapid, cost-effective [USD 42.00 (INR 3,000.00)/sample], with short turnaround time (4 hours), including testing and interpretation. This classifier is advantageous over other approaches for the prompt detection at baseline to start tailored treatment regimes in BCR-ABL1-like ALL cases.