【摘要】 预测模型是实现个体化用药与精准医学的关键工具。本文以《药物流行病学研究方法学指南（第2版）》为基础，系统阐述预测模型在药物疗效与安全性评价中的四大核心应用场景：治疗效应异质性识别（谁应用药）、生物标志物发现（为何有效）、用药患者疗效预测（效果如何）及药物安全性评价（是否安全）。在此基础上，进一步探讨四方面的特殊方法学考量：预测模型与因果模型的本质区别、类不平衡与罕见事件预测的处理策略、从预测到决策的三层级转化路径（内部验证、外部验证、临床效用评估），以及TRIPOD+AI与PROBAST+AI所构建的报告规范与偏倚风险评价闭环。本文旨在为药物流行病学研究者在开展预测模型研究时提供方法学指引，推动该领域从“模型开发”向“临床可用”的范式转变。更多还原

【Abstract】 Predictive models are essential methodological tools for individualized therapy and precision medicine. Based on the Guide on Methodological Standards in Pharmacoepidemiology(2 nd Edition), this paper systematically elucidates four core application scenarios of predictive models in drug effectiveness and safety evaluation: heterogeneity of treatment effects identification(who should receive the drug), biomarker discovery(why it is effective), effectiveness prediction for patients receiving drugs(how effective it is), and drug safety evaluation(whether it is safe). Furthermore, four special methodological considerations are discussed: the essential distinction between predictive and causal models, strategies for handling class imbalance and rare event prediction, the three-level translational pathway from prediction to decision-making(internal validation, external validation, and clinical utility assessment), and the closed-loop system for reporting specification and bias risk assessment formed by TRIPOD+AI and PROBAST+AI. This paper aims to provide methodological guidance for pharmacoepidemiologists and facilitate paradigm shift from "model development" to "clinical utility" in this field.更多还原