GENERATIVE ARTIFICIAL INTELLIGENCE IN DIGITAL HEALTH: CLINICAL TRANSLATION, SAFETY GOVERNANCE, REGULATORY SCIENCE, AND IMPLEMENTATION ROADMAP

Md Gulzar Ahmad*, Hasibullah Khairi, Ravi Dhakad, Prem Shankar, Md Faiyaz, Alisha Singh

Background: Generative artificial intelligence (AI), including large language models and multimodal systems, is increasingly evaluated in digital health for clinical documentation, information synthesis, decision support, diagnostic assistance, and workflow optimization. Despite rapid technical progress, safe clinical translation remains limited by unresolved challenges related to clinical validity, real-world performance, workflow integration, safety monitoring, regulatory readiness, ethical governance, and accountability. Objective: This narrative review proposes a translational regulatory-science framework for the responsible integration of generative AI into digital health systems by connecting clinical use cases, generative AI-specific safety risks, task-risk stratification, global regulatory expectations, ethical governance, multimodal accountability, lifecycle monitoring, and de-implementation readiness. Review Approach: A structured literature review was conducted of peer-reviewed studies, clinical AI reporting guidelines, implementation science literature, and official regulatory or governance documents published between 2000 and 2026. The review prioritized high-impact biomedical, digital health, regulatory science, and AI governance sources, including guidance from the World Health Organization, the U.S. Food and Drug Administration, European Union authorities, Health Canada, and the International Medical Device Regulators Forum. Main Synthesis: Generative AI can support digital health only when deployed through a lifecycle-based system that integrates intended-use definition, clinical validation, workflow assessment, human oversight, equity evaluation, privacy and cybersecurity safeguards, post-deployment monitoring, controlled model updating, and withdrawal pathways when safety or clinical value is not sustained. Framework Contribution: The proposed framework aligns clinical application categories with task-risk profiles, maps generative AI-specific risks to validation and governance requirements, and links regulatory expectations with accountability structures involving developers, healthcare organizations, clinicians, regulators, and patients. Conclusion: Trustworthy implementation of generative AI in digital health requires technical robustness, ethical oversight, regulatory compliance, real-world monitoring, and accountable governance, rather than relying on model capabilities alone.