Risk prediction models can guide prevention, monitoring, and treatment decisions in cardio-oncology, but their development requires careful attention to methodological challenges unique to cardiovascular disease arising in the context of cancer. In this state-of-the-art review, the authors use case scenarios to illustrate how prediction objectives, index dates, and time horizons must align with specific clinical decisions across the cancer trajectory. They summarize key considerations for data source selection, population definition, sample size, and variable selection and discuss challenges in incorporating treatment data, including immortal time bias and confounding by indication. The authors review regression-based, competing risk, dynamic, and machine learning approaches for model development, along with best practices for evaluating discrimination, calibration, and clinical utility. Finally, they outline principles for implementation, including workflow integration, transparency, and ongoing model updating. Together, these concepts provide a framework to support the development and adoption of rigorous, clinically meaningful risk prediction tools tailored to cardio-oncology.