Machine learning (ML) models show promise of assisting physicians with better prognostic predictions, outperforming simple statistical models and conventional one-size-fits-all risk scores. However, there are a number of possible barriers to its clinical adoption. 1. ML model can be difficult to interpret because it can capture non-linear, complex patterns that are hard for human interpretation. 2. Physicians might not be confident in its adoption if they cannot confidently assess the adoptability of the ML model to their specific patient situation. For example, one might question if the training data lacks African representation when adopting for an African patient. 3. ML prediction might not sufficiently capture clinical interests, rendering its results little values in realistic clinical settings. In this paper, we explore how these three barriers can be overcome by 1, provide explanations to ML prediction, 2, present relevant clinical data and 3, develop a Decision Support System that allows physicians to meaningfully interact with ML model for personalized clinic cases. A survey is developed to assess how our solutions affect the confidence with which human subjects (including physicians, researchers, and medical students) utilize ML prognosis for their potential clinical use.