Eradicating health disparity is a new focus for precision medicine research. Identifying patient subgroups is an effective approach to customized treatments for maximizing efficiency in precision medicine. Some features may be important risk factors for specific patient subgroups but not necessarily for others, resulting in a potential divergence in treatments designed for a given population. In this paper, we propose a tree-based method, called Subgroup Detection Tree (SDT), to detect patient subgroups with personalized risk factors. SDT differs from conventional CART in the splitting criterion that prioritizes the potential risk factors. Subgroups are automatically formed as leaf nodes in the tree growing procedure. We applied SDT to analyze a clinical hypertension (HTN) dataset, investigating significant risk factors for hypertensive heart disease in African-American patients, and uncovered significant correlations between vitamin D and selected subgroups of patients. Further, SDT is enhanced with ensemble learning to reduce the variance of prediction tasks.