随着新药研发加速和临床用药复杂性提升，药物不良反应（ADR）的预测与监管面临日益严峻的挑战。传统药物警戒体系主要依赖自发报告系统和统计学方法，存在报告滞后、漏报以及难以处理高维、非线性复杂数据等问题。近年来，人工智能（AI）技术的快速发展，尤其是机器学习（ML）、深度学习（DL）及多模态数据融合方法的突破，为突破传统范式局限、实现ADR的早期识别与动态评估提供了新的技术路径。系统综述AI在ADR预测中的方法体系，包括ML、DL及多模态数据融合等技术，并探讨其在药品全生命周期监管中的应用场景与价值。同时，分析当前AI在监管应用中面临的可解释性、数据质量、伦理合规等挑战，并展望从“被动响应”向“主动预测”监管范式转型的未来方向，旨在为构建智能、高效、可信的药品监管新范式提供理论与技术参考。

With the acceleration of new drug research and development and the increasing complexity of clinical medication, the prediction and supervision of adverse drug reactions (ADR) are facing increasing challenges. Traditional pharmacovigilance systems mainly rely on spontaneous reporting systems and statistical methods, which suffer from problems such as reporting lag, underreporting, and difficulty in handling high-dimensional, nonlinear and complex data. In recent years, the rapid development of artificial intelligence (AI) technology, especially the breakthrough of machine learning (ML), deep learning (DL) and multimodal data fusion methods, provides a new technical path for breaking through the limitations of traditional paradigms and realizing the early identification and dynamic evaluation of ADR. This paper systematically reviewed the method system of artificial intelligence in ADR prediction, including ML, DL and multimodal data fusion, and discussed its application scenarios and value in the supervision of the whole life cycle of drugs. At the same time, this paper analyzes the current challenges of interpretability, data quality, ethical compliance and other regulatory applications of AI, and looks forward to the future direction of the transformation from “passive response” to “active prediction” regulatory paradigm. This study aims to provide theoretical and technical reference for building a new paradigm of intelligent, efficient and credible drug regulation.

R951