The increasing popularity of machine learning models in real-world clinical decision support systems has underscored the need for assessing and then mitigating biases that may manifest, often spuriously, in their predictions either at the population or sub-population level. The latter often in the context of prediction fairness. These biases can be assessed in terms of calibration, performance stratification, fairness metrics, prediction interval coverages, etc., and are mainly due to poor model specification (e.g., overparameterization without regularization or loss/likelihood mismatch) or data collection issues (e.g., population misrepresentation or unmeasured confounders). Recently, we evaluated existing risk prediction models for stroke and revealed disparities between groups in model discrimination performance, with all models performing better in White compared to Black individuals. Then developed machine learning methodology tailored for more equitable risk prediction in stroke. As a use case, we leverage harmonized data from four large NIH studies, namely, ARIC, MESA, REGARDS and Framingham Offspring.