Mitigating Hematologic Adverse Events in Patients with Myeloid Malignancies: A Novel Causal Artificial Intelligence Approach

PROJECT SUMMARY/ABSTRACT Thrombosis and bleeding are the major causes of morbidity and mortality among survivors of myeloid malignancies. Previous studies proposed clinical and laboratory variables associated with these life-threatening comorbidities, such as older age. However, these crude predictors do not capture the heterogeneity of patients’

pathology, genomics, laboratory, and clinical profiles to arrive at accurate and personalized risk assessment. In addition, conventional machine learning prediction models cannot elucidate the biological mechanisms underpinning the observed correlations, and association-based prediction models do not provide reliable treatment

suggestions. The long-term goal of this project is to reduce the hemorrhagic and thrombotic disease burden among patients with myeloid malignancies by providing personalized risk predictions that enable treatment optimization. The central hypothesis is that causal machine learning methods will empower a reliable analytical platform for

hematologic comorbidity risk prediction and mitigation. In this study, we will employ large and diverse clinical datasets from multiple clinical centers to develop integrative causal machine learning models. Specifically, we will (i) integrate laboratory, pathology, and clinical data to identify the key predictors of thrombosis for patients with myeloid

malignancies, (ii) quantify the risk of bleeding by machine learning and causal inference methods, and (iii) personalize treatment protocols to minimize risks of bleeding and thrombosis using causal machine learning. We will validate our data-driven models through rigorous external validation. Our methods are innovative because they depart

from the status quo by linking advanced causal inference methods with machine learning algorithms. Our proposed studies are significant because they will create a reliable data-driven framework for multi-modality data fusion and enable personalized treatment strategies to mitigate hematologic comorbidities. This novel causal machine

learning platform will guide individualized clinical decisions to reduce hematologic adverse events among patients with myeloid malignancies. Our approaches address the National Heart, Lung, and Blood Institute’s Compelling Question 5.CQ.10 on reducing cardiovascular morbidity and mortality in cancer survivors.