ABSTRACT Health disparities widely exist in the United States. African Americans and Hispanic populations have higher probabilities of cancer incidence and poor outcomes. There are treatments and policies designed trying to improve health outcomes and reduce disparities. Research methods are needed that help monitor the trend of health disparities, evaluate efficiencies of interventions, identify risk factors that contribute to the start and the change in health disparities, and quantify the effects of identified risk factors. In this study, we propose to develop modern analytical methods that address these requests in research. We propose single- and multi-level moderation analysis methods that adopt recently developed data mining and machine learning techniques in the general mediation analysis. Using these methods, researchers will be able to differentiate and make inferences on effects of risk factors on the observed health disparities from both individual and environmental levels. In addition, we will be able to identify changes of health disparities over time and over employment of new policies. An R package will be developed for the implementation of the proposed novel methods. We will demonstrate the method in 1) identifying the trend of racial disparity in using the Oncotype diagnosis among eligible breast cancer patients, and identifying and making inferences on effects of risk factors that contribute to the change of racial disparity over time; and 2) Investigating the interactive effects of individual factors and social/physical environments on explaining racial disparities in breast cancer outcomes in Louisiana.