Abstract: The rapid evolution of digital healthcare systems has induced a transformative shift in the acquisition, analysis, and application of medical data, yielding notable enhancements in patient care. Among emerging technologies, federated learning (FL) has surfaced as a promising method, facilitating collaborative model training across distributed healthcare institutions while safeguarding data privacy. FL empowers the training of machine learning (ML) models on decentralized data dispersed across a spectrum of Internet of Health Things (IoHT) devices, encompassing smartphones, wearables (e.g., fitness trackers, smartwatches), and implantable healthcare devices such as pacemakers. Significantly, FL assures the privacy and security of these devices and raw data throughout the learning process. However, integrating FL into contemporary digital healthcare systems raises challenges and risks that warrant meticulous consideration to ensure the ethical and secure utilization of sensitive patient information. Accordingly, this study comprehensively explores the multifaceted challenges, problems, and risks of FL within digital healthcare systems. We underscore potential solutions and outline future directions for mitigating these challenges and risks effectively. The insights presented here serve as invaluable guidance for researchers, students, and diverse stakeholders navigating the intricate landscape of FL in digital healthcare systems, with a steadfast commitment to upholding ethical principles and security standards.