Abstract
Objectives: This study aims to systematically review the literature on the computational processing of the language of pain, whether generated by patients or physicians, identifying current trends and challenges. Methods: Following the PRISMA guidelines, a comprehensive literature search was conducted to select relevant studies on the computational processing of the language of pain and answer pre-defined research questions. Data extraction and synthesis were performed to categorize selected studies according to their primary purpose and outcome, patient and pain population, textual data, computational methodology, and outcome targets. Results: Physician-generated language of pain, specifically from clinical notes, was the most used data. Tasks included patient diagnosis and triaging, identification of pain mentions, treatment response prediction, biomedical entity extraction, correlation of linguistic features with clinical states, and lexico-semantic analysis of pain narratives. Only one study included previous linguistic knowledge on pain utterances in their experimental setup. Most studies targeted their outcomes for physicians, either directly as clinical tools or as indirect knowledge. The least targeted stage of clinical pain care was self-management, in which patients are most involved. The least studied dimensions of pain were affective and sociocultural. Only two studies measured how physician performance on clinical tasks improved with the inclusion of the proposed algorithm. Discussion: This study found that future research should focus on analyzing patient-generated language of pain, developing patient-centered resources for self-management and patient-empowerment, exploring affective and sociocultural aspects of pain, and measuring improvements in physician performance when aided by the proposed tools.
Abstract (translated)
研究目标:本研究旨在系统地回顾有关疼痛语言计算的相关文献,无论是由患者还是医生产生的,以识别当前的趋势和挑战。方法:遵循PRISMA指南,进行全面的文献搜索,以选择与疼痛语言计算相关的研究,并回答预先设定的研究问题。数据提取和合成是将所选研究根据其主要目的和结果、患者和痛苦人群、文本数据、计算方法以及结果目标进行分类的过程。结果:医生产生的疼痛语言,特别是从病历中提取的数据,是最常用的数据。任务包括患者诊断和分诊、疼痛提及的识别、治疗反应预测、生物医学实体提取、语言特征与临床状态的关联以及疼痛叙述的词汇-语义分析。只有1篇论文包括了他们在实验设置中之前对疼痛语句的语言知识。大多数研究将重点放在医生身上,无论是直接作为临床工具,还是作为间接知识。最少的针对性临床疼痛护理阶段是自我管理,其中患者最积极参与。最少的疼痛研究维度是情感和社会文化方面。只有2篇论文测量了医生在临床任务中表现随着所提出的算法的引入而改善。讨论:本研究发现,未来的研究应该集中于分析患者产生的疼痛语言,为自我管理和患者赋权开发基于患者的资源,探索疼痛的情感和社会文化方面,以及衡量医生在使用所提出的工具时的表现改善。
URL
https://arxiv.org/abs/2404.16226