艾昆纬-值得信赖的AI-ML用于患者分析和研究（英）

White PaperTrustworthy AI/ML for Patient Analytics and ResearchAI-secure, privacy-first, with continuous monitoring and oversightTable of contentsResponsible innovation in patient analytics and research 2Heightened care in AI/ML 2Adopting a principled, AI-secure approach to AI/ML 2Shifting baseline for AI-Secure AI/ML 3Bridging AI and data protection with federated modeling 3Collection limitation and data minimization 3Use limitation and purpose specification 4Security safeguards 4Accountability and oversight 4Openness and transparency 4Federated learning for AI/ML 5Understanding the data journey for federated learning 6Source ingestion: pseudonymization and segregation 6Horizontal federated learning: generating synthetic trends 6Vertical federated learning: AI-secure AI/ML 7Safe outputs 7Beyond de-identification: managing reconstruction risk 8Leveraging synthetic trends for AI/ML 8Managing reconstruction risk 9AI governance and privacy operations (AI PrivOps): an integrated governance function 10Continuous monitoring of AI PrivOPs metrics 10Oversight without exposure 11Human-in-the-loop for accountability 11Ethics Board for patient analytics and research 12Conclusion 13Acknowledgment 13 iqvia.com | 1This paper outlines our approach to Artificial Intelligence (AI) and Machine Learning (ML) that withstands today’s threat landscape and serves as a blueprint for sustainable innovation. It’s how we raise the bar for defensible AI in healthcare applications and beyond, representing a shift from traditional data practices to AI-security as a design feature.2 | Trustworthy AI/ML for Patient Analytics and ResearchResponsible innovation in patient analytics and researchLife sciences are being transformed by Artificial Intelligence (AI) and Machine Learning (ML). But with that transformation comes a critical question: how do we unlock value from sensitive health data without undermining trust, transparency, or control? Traditional safeguards are no longer enough in an era of AI/ML, where subtle patterns can be used — or misused — in unanticipated ways. The stakes are especially high in healthcare, where data utility must be balanced with rigorous protection.Robust de-identification methods, which remove identifying elements, can be used but the industry lacks widespread adoption of standardized practices. This absence of fixed standards provides space to explore forward-looking approaches, especially in light of emerging AI/ML threats that will need to be addressed. As AI/ML and other developing technologies reshape the landscape, more sophisticated strategies are needed to balance AI/ML and data protection with responsible use.Heightened care in AI/MLThis whitepaper introduces a novel privacy-first and AI-secure architecture for defensible AI developed by IQVIA. In response to AI and data protection concerns, the platform combines synthetic data abstractions, federated learning, and integrated AI Governance and Privacy Operations (AI PrivOps) mo