Next-Generation Exposure Assessment for Personalized Environmental Health: A Scoping Review of Knowledge Representation, Dynamic Modeling, and Artificial Intelligence Approaches

方法

PRISMA-ScR范围综述，检索WoS、PubMed、IEEE并分析48篇研究

发现

各方法能力互补但碎片化，极少整合，仅3篇延伸到个体预测或决策支持。

机会点

可将本体/知识图谱与ABM、LLM融合，构建个体化时空暴露与农药决策支持系统。

Background: Exposure assessment increasingly requires the integration of heterogeneous environmental, spatial, behavioral, and health information.Ontologies, knowledge graphs (KGs), agent-based modeling (ABM), large language models (LLMs), and agentic artificial intelligence (AI) provide complementary approaches for knowledge representation, data integration, dynamic simulation, information processing, and autonomous decision support.Objectives: This study aimed to map applications of these approaches in environmental health and exposure assessment and identify gaps for next-generation personalized exposure assessment.Methods: A scoping review was conducted according to PRISMA-ScR.Web of Science, PubMed, and IEEE Xplore were searched, with Google Scholar applied as a supplementary source.Forty-eight studies were analyzed by methodological approach, application purpose, eight exposure-assessment components, and extension to individual-level prediction or decision support.Results: Ontology-based approaches were identified in 26 studies (54.2%), followed by ABM in 15 (31.3%),KG in six (12.5%),LLM in four (8.3%), and agentic AI in one (2.1%).Ontologies were mainly used for semantic representation and standardization of exposure-related concepts, KGs for integrating and exploring heterogeneous exposure and health information, ABM for simulating dynamic individual exposure processes, and LLMs for extracting and reasoning from unstructured information.Across the 48 studies, pollutants (77.1%) and health outcomes (56.3%) were most frequently represented, followed by space (47.9%), whereas exposure pathways (29.2%) and management strategies (16.7%) were less frequently addressed.Integration across methodological approaches remained limited: three studies combined ontology and KG, and one combined ontology and LLM, and no study integrated ABM with ontology or KG.Only three studies extended these approaches to individual-level prediction or decision support.Conclusions: Current approaches provide complementary but fragmented capabilities.Future research should link ontology-and KG-based semantic knowledge with ABM simulation, LLM-based information processing and reasoning, and agentic AI-based monitoring and decision support, while improving semantic interoperability, validation, and representation of individual spatiotemporal and quantitative exposure information.