利用真实世界数据预测脆弱性的人工智能进展：范围综述。

Advances of artificial intelligence in predicting frailty using real-world data: A scoping review.

机构信息

Department of Health Outcomes and Biomedical Informatics, University of Florida, Gainesville, FL 32611, United States.

出版信息

Ageing Res Rev. 2024 Nov;101:102529. doi: 10.1016/j.arr.2024.102529. Epub 2024 Oct 5.

DOI:10.1016/j.arr.2024.102529

Abstract

BACKGROUND

Frailty assessment is imperative for tailoring healthcare interventions for older adults, but its implementation remains challenging due to the effort and time needed. The advances of artificial intelligence (AI) and natural language processing (NLP) present a novel opportunity to harness real-world data (RWD) including electronic health records, administrative claims, and other routinely collected medical records for frailty assessments.

METHODS

We followed the PRISMA-ScR guideline and searched Embase, Web of Science, and PubMed databases for articles that predict frailty using AI through RWD from inception until October 2023. We synthesized and analyzed the selected publications according to their field of application, methodologies employed, validation processes, outcomes achieved, and their respective limitations and strengths.

RESULTS

A total of 23 publications were selected from the initial search (N=2067) and bibliography. The approaches to frailty prediction using RWD and AI were categorized into two groups based on the type of data utilized: 1) AI models using structured data and 2) NLP techniques applied to unstructured clinical notes. We found that AI models achieved moderate to high predictive performance in predicting frailty. However, to demonstrate their clinical utility, these models require further validation using external data and a comprehensive assessment of their impact on patients' health outcomes. Additionally, the application of NLP in frailty prediction is still in its early stages. Great potential exists to enhance frailty prediction by integrating structured data and clinical notes.

CONCLUSION

The combination of AI and RWD presents significant opportunities for advancing frailty assessment. To maximize the advantages of these technological advances, future research is needed to rigorously address the challenges associated with the validation of AI models and innovative data integration.

摘要

背景

对老年人进行医疗保健干预的量身定制至关重要，但由于需要付出努力和时间，其实施仍然具有挑战性。人工智能 (AI) 和自然语言处理 (NLP) 的进步为利用包括电子健康记录、行政索赔和其他常规收集的医疗记录在内的真实世界数据 (RWD) 进行脆弱性评估提供了新的机会。

方法

我们遵循 PRISMA-ScR 指南，从开始到 2023 年 10 月，在 Embase、Web of Science 和 PubMed 数据库中搜索使用 AI 通过 RWD 预测脆弱性的文章。我们根据应用领域、使用的方法、验证过程、取得的结果以及各自的局限性和优势对选定的出版物进行综合和分析。

结果

从最初的搜索（N=2067）和参考文献中总共选择了 23 篇出版物。使用 RWD 和 AI 进行脆弱性预测的方法分为两组，基于使用的数据类型：1）使用结构化数据的 AI 模型，2）应用于非结构化临床记录的 NLP 技术。我们发现，AI 模型在预测脆弱性方面取得了中等至高的预测性能。然而，为了证明它们的临床实用性，这些模型需要使用外部数据进一步验证，并全面评估它们对患者健康结果的影响。此外，NLP 在脆弱性预测中的应用仍处于早期阶段。通过整合结构化数据和临床记录，可以极大地提高脆弱性预测的能力。