AML is one of the most common and lethal hematopoietic malignancies. Currently, there is a need to improve the predictive nature of available prognostic biomarkers and to develop more precise methods for risk- stratifying AML patients. In keeping with these needs, we will determine whether biomarker assays can be significantly improved by examining these biomarkers in subpopulations of AML blasts. Furthermore, we will develop risk-assessment models using a combination of biomarker results and other prognostic factors that will more accurately predict clinical outcomes. Specific Aim 1. Determine if measuring biomarkers in enriched populations of AML blasts significantly improves their predictive accuracy. Diagnostic samples will be obtained from pediatric (N = 250) and adult (N = 192) patients with AML. Previously recognized genomic (e.g., FLT3, etc.) and transcript(e.g., BAALC, etc.) biomarkers will be examined in 4 cell populations from these diagnostic samples: mononuclear cells (MNCs), total AML blasts, more differentiated AML blasts, and less differentiated AML blasts. Multivariate analyses, incorporating the effects of other prognostic factors, will be used to identify independent associations with clinical outcomes. Comparative analyses of performance characteristics will be used to determine if measuring the biomarker in enriched populations of AML blasts, including the less differentiated AML blasts, significantly improves the predictive accuracy of the biomarker. Specific Aim 2. Develop and validate novel risk-assessment models for predicting clinical outcomes for AML patients. Using the data generated from Specific Aim 1, we will develop risk-assessment models for the following clinical outcomes: complete response (CR), resistant disease (RD), relapse-free survival (RFS), and overall survival (OS). These models will combine the results from the most predictive biomarkers, whether in enriched populations of AML blasts or not, with other risk factors (age, cytogenetics, performance status, etc.). We will initially develop these models independently for each outcome and the two populations of patients (i.e., children and adult, separately). However, we will also examine whether a comprehensive risk- assessment model can be developed, which predicts for multiple clinical outcomes (CR, RD, RFS, and OS) across both pediatric and adult populations. We will then validate these predictive models in similarly treated populations of pediatric (N = 250) and adult (N = 191) patients.