Integrating transcriptomic profiling and machine learning: A clinically actionable prognostic model for infant acute myeloid leukemia.

Infant acute myeloid leukemia (AML), particularly in those under 3 years of age, presents poor prognostic outcomes and distinct biological characteristics that require age-specific risk assessment. This study, utilizing data from four pediatric AML (pAML) trials conducted by the Children's Oncology Group, aimed to develop a simple RNA expression-based prognostic model to refine risk stratification for infant AML. Expression data from 213 infant AML patients were analyzed using machine-learning algorithms to develop the infant-prognostic-score (IPSscore), or IPSgroup when categorized. To validate the stability of the model, internal validation was conducted on a set of 127 cases, and external validation was performed using a separate set of 63 patients from a different ethnic background. Furthermore, we compared its prognostic prediction capability with that of other AML models and explored its potential clinical decision-making value for infant AML patients. The IPSgroup independently and specifically predicted outcomes in infant AML, outperforming several previously published RNA expression-based models. Infant patients categorized into the high-risk group based on IPSgroup may benefit from hematopoietic stem cell transplantation (HSCT), while those in the low-risk group are not suitable for HSCT. Additionally, when combined with the current pAML stratification system used in clinical trials, the IPSgroup enabled re-stratification of 43% of infant AML patients into more accurate risk groups, highlighting the advantage of incorporating gene expression analysis into clinical decision-making. Infant AML demonstrates significant heterogeneity at clinical, molecular, and prognostic levels. The newly proposed model surpasses existing AML stratifications, offering a valuable tool for clinical decision-making and treatment strategies.