A Network Science Approach to Conflicts of Interest: Metrics, Policies, and Communication Design

该项目的主要目的是开发新的衡量标准和机制，以评估 生物医学研究企业中的利益冲突风险。COI风险的创新 评价和沟通将从将COI概念化转变为一个问题 个体研究人员将COI理解为网络现象。紧迫的问题 COI及其灌输的偏见不是来自个人，而是来自COI的聚合 研究人员和资助者的网络。研究团队将利用机器学习和高效率 性能计算，以(1)跟踪COI在生物医学决策网络中的循环，以及(2) 评估某些冲突网络配置文件预测患者伤害风险增加的程度。这个 团队将测试候选COI指标与美国食品和药物管理局的不良事件 药品安全数据上报系统。以这些数据为基础，本项目还将(3)开发和 评估旨在防范不正当行为的COI偏向风险沟通策略 目前披露做法的结果。这项研究的结果将为新的基于证据的 关于生物医学研究中COI政策的建议以及提高效率的建议 披露惯例。为了实现这些目标，该项目将利用机器学习来识别系统 针对特定药物产品的生物医学研究企业内的COI网络。研究小组将 然后评估网络指标在多大程度上预测不良事件发生率和严重性的相对增加 对于已识别的产品。

The primary purpose of this project is to develop new metrics and mechanisms for the evaluation of conflicts of interest (COI) risks in the biomedical research enterprise. The innovation in COI risk evaluation and communication will draw on a shift away from the conceptualization of COI as a problem of individual researchers toward an understanding of COI as network phenomena. The pressing problems of COI and the bias it inculcates stem not from individuals but from the aggregation of COI across networks of researchers and funders. The research team will leverage machine-learning and high performance computing to (1) track the circulation of COI within biomedical decision networks, and (2) evaluate the extent to which certain conflict network profiles predict increased risks of patient harm. The team will test candidate COI metrics with U.S. Food and Drug Administration's Adverse Events Reporting System drug safety data. Using these data as a foundation, this project will also (3) develop and evaluate strategies for communicating COI risks of bias designed to safeguard against the perverse outcomes in current disclosure practices. The results of this research will underwrite novel evidence-based recommendations for COI policies in biomedical research as well as recommendations for more effective disclosure practices. To achieve these aims, this project will leverage machine-learning to identify systemic COI networks within the biomedical research enterprises for specific drug products. The research team will then evaluate to what extent network metrics predict relative increases in adverse event rates and severity for identified products.