国内外急性肾损伤预测研究进展

目的：梳理和归纳国内外急性肾损伤预测研究进展，为后续研究提供借鉴。方法：通过关键词检索、参考文献回溯等方法从PubMed、中国知网等数据库检索文献，经人工筛选后获得84篇文献，从研究主题、研究方法和研究数据3方面对其进行述评。结果：预测标志物挖掘、发病风险预测、预后预测是目前急性肾损伤预测研究的3类核心主题。相关性分析、传统回归分析、机器学习与深度学习共同构成急性肾损伤预测研究方法体系，并且深度学习更受重视。真实世界电子病历数据正逐步成为急性肾损伤预测研究的主要数据类型。结论：未来急性肾损伤预测研究可围绕新型标志物挖掘、基于多模态数据预测、医学领域知识和数据融合驱动、连续动态预测等关键点展开。

Advances in foreign and domestic prediction of acute kidney injury

Objective To provide reference for the future research on acute kidney injury by stratifying and summarizing the advances in foreign and domestic prediction of acute kidney injury. Methods The papers on prediction of acute kidney injury were retrieved from PubMed and CNKI by searching the key words and retrospective references, from which 84 were selected by manual screening and their research topics, research methods and research data were reviewed. Results Mining the prediction markers, predicting the disease onset risk, and predicting the disease outcome were the key topics on prediction of acute kidney injury at present. Correlation analysis, traditional regression analysis, machine learning and deep learning were the component parts of the research method system for predicting acute kidney injury with importance attached to deep learning. The real world electronic medical record data were gradually becoming the major data type for predicting acute kidney injury. Conclusion The future research on prediction of acute kidney injury can be carried out by focusing on mining the novel markers of acute kidney injury, predicting multimodal data-based acute kidney injury, driving the integrated medical domain knowledge and data of acute kidney injury, and continuously and dynamically predicting acute kidney injury.