基于临床异构信息数据挖掘的智能化甲氨蝶呤个体化合理用药关键技术研究

甲氨蝶呤（MTX）是指南推荐的一线抗肿瘤药，因其医嘱复杂、个体差异大、毒副作用大，治疗有效性及安全性是药物管理有待解决的关键问题。我们前期研究发现目前主要基于医生经验、药动学参数或使用相关的药动学模型调整给药方案。因未考虑给药全流程和患者病生理的特异性，仅纳入较少临床变量，目前尚难实现个体化精准给药。本研究拟利用异构信息数据挖掘技术，智能地由异构数据组成的网络信息中发现和抽取信息，通过对概念及相关因素的延伸和比较，挖掘深层次关键信息。我们现已通过单中心数据挖掘发现新的高风险变量集且具有临床意义。研究拟基于多中心MTX全程治疗的异构信息，构建药物治疗知识图谱，使用数据挖掘及深度学习技术筛选影响药物治疗的综合变量集构建给药模型，进而建立与医院EMR动态交互的MTX个体化给药智能管理系统，实现全程化药物治疗方案的智能设计和调整。通过不同药物的临床数据训练实现系统泛化，可外推应用到其他高风险药物。

Methotrexate (MTX) is a first-line anti-tumor agent recommended by several guidelines. Due to its complicated orders, large individual variability and severe adverse drug reactions, the effectiveness and safety are key issues in MTX administration. Our previous results indicated that current individualized regimen adjustments were based on physicians’ experience, part of pharmacokinetic parameters and population pharmacokinetic models. It is worth noting that all the existing methods above merely including a small number of variables without considering the whole process of complex administration of MTX and the sophisticated physiological specificity of cancer or pregnant patients. Heterogeneous information mining refers to the process of using data mining technology to automatically discover and extract knowledge from network information composed of heterogeneous data, and to find out the deep-level knowledge required by users through the extension and comparison of concepts and related factors. We have found a new set of high-risk variables through single center data mining which has great clinical significance. Hence, this study will draw heterogeneous data from expanded multi-centers, constructing the knowledge graph of MTX and choosing all the significant variables by screening comprehensive parameters during the whole administration in the aid of data mining and deep learning algorithms, finally obtaining individualized MTX administration model and establishing an intelligent decision system that can achieve pharmacist-electronic medical record interoperability, precisely adjust the administration and recommend the plan of next circle. The system can be extrapolated to other drugs by training clinical data of different drugs.