Purpose: Predictive models for medical applications have shown promise, yet their clinical adoption remains limited due to concerns about prediction uncertainty, lack of interpretability, and limited reproducibility. Goal of this study was to design a deployable ML development framework that operationalizes trustworthy AI for clinical prediction. Furthermore, the study aimed to demonstrate applicability and validity of the framework in real world clinical practice by using it to build a patient no-show predictor in collaboration with a large pediatric hospital in the United States. Methods and Data: This study is guided by Design Science Research Methodology (DSRM) that is well-suited for research where the primary contribution is an artifact intended for real-world use. An ambulatory clinic patient no-show prediction application was chosen as the framework validation use case and was constructed using five years of electronic health record data (7.9 million appointments) from the collaborating hospital. Result: We designed and evaluated a novel socio-technical artifact—a trustworthy ML design framework (UML-Med) for clinical prediction—that addresses the trust gap in medical ML deployment. The resulting framework artifact (UML-Med) comprised of integrated uncertainly quantification module, multi-level explainability layer (local and global), reproducibility pipeline, structured evaluation governance, and an uncertainty-aware model selection method involving competing ML models based on performance-uncertainty tradeoff – not accuracy alone. All code and data are made publicly available to ensure full reproducibility. The UML-Med framework contributes as a generalizable development pipeline and a reusable blueprint for delivering trustworthy clinical prediction applications.