Despite aggressive therapies, many patients with acute myeloid leukemia (AML) will ultimately die from either treatment-related toxicities or therapeutic resistance, but treatment outcomes vary considerably. Thus, the ability to accurately predict treatment-related mortality (TRM) and therapeutic resistance in individual patients would greatly facilitate clinical management as it could form the foundation for evidence-based decision- making. For example, the appropriate treatment intensity could be evaluated by weighing anticipated TRM against the benefit of more intensive therapy, while the use of standard versus investigational therapy could be evaluated by considering how likely it is that standard therapy will fail. However, a major current limitation is that no rigorous, comprehensive methods are available for this approach. Instead, treatment decisions continue to be made in an arbitrary manner and often based on single risk factors (e.g., age or performance status), although the significant shortcomings of such a strategy are well recognized. Consequently, there is a critical need to develop tools to assess an individual patient's likelihood of TRM or therapeutic resistance to curative-intent intensive therapies. The objective of this application is to develop predictive models to assess the likelihood of these outcomes in newly diagnosed AML. To this end, we have taken the lead in establishing an international alliance between four U.S. and European AML cooperative study groups and two large U.S. cancer centers. These collaborators provide large, well-annotated datasets for predictive modeling that will permit the stringent evaluation of the factors associated with TRM and therapeutic resistance, and, more importantly, the determination of the factors that help predict these outcomes for any given patient. In initial studies, we demonstrated the feasibility of our approach by empirically definin TRM and developing early models predicting TRM, which proved significantly superior to models using single factors such as age or performance status. These models have already been implemented in several research protocols that assign patients based on predicted likelihood of TRM, indicating the beginning of a shift in medical practice. Taking advantage of our large datasets, we propose the to validate the concept of TRM modeling for the treatment of adult patients with newly diagnosed AML across several independent study cohorts. We will also empirically define therapeutic resistance and develop a validated multicomponent model to predict this outcome in patients with newly diagnosed AML. Upon the successful completion of this project, it is our expectation that we will have developed validated instruments that provide the framework for personalized, informed decision- making with AML patients regarding their treatment options. Additionally, we expect that the empiric definition of therapeutic resistance developed in this project can serve as a novel, improved surrogate endpoint for use in phase 2 testing of new AML therapies. Together, our work is anticipated to have an important positive impact because it will provide novel tools that may lead to optimized outcomes for patients with AML.