文献速递神经肿瘤神经重症

人工智能用于神经肿瘤症状评估与患者报告结局

Artificial intelligence for symptom assessment and patient-reported outcomes in neuro-oncology.
PubMed 文献 · J Neurooncol · 2026-10-21

本文为叙述性综述,检索2015年1月至2026年4月的文献,最终纳入53篇,总结神经肿瘤患者报告结局(PRO)及人工智能(AI)分析工具的应用、现实考量与伦理问题。作者认为AI(含机器学习与大语言模型)可建模纵向症状轨迹、从自由文本中提取症状并嵌入临床决策支持,但转化需可解释、整合电子健康记录、减轻偏倚并保护患者知情同意、隐私与自主权。

要点

  • PRO数据量与复杂度增长,传统分析方法面临挑战。
  • AI可预测症状恶化、从临床笔记提取症状并嵌入决策支持。
  • 应用需关注自主权、知情同意、数据安全、偏倚、可及性与透明度。

核心结论

  • AI在神经肿瘤PRO分析中具有建模纵向症状轨迹、提取自由文本症状和嵌入临床决策支持的潜力。
  • 成功转化需要可解释、整合电子健康记录、减轻偏倚的工具,并保护患者知情同意、隐私与自主权。

研究设计与样本

  • 叙述性综述,检索PubMed/MEDLINE、Embase和Google Scholar中2015年1月至2026年4月发表的英文文献,并补充参考文献列表回顾。
  • 初始78篇文章经筛选后纳入53篇,纳入标准为报告脑肿瘤患者PRO或症状数据,或在肿瘤学中应用计算/AI方法于PRO或症状监测。
  • 文章选择由两位资深作者(L.V.Y、A.S.)核验,保留仍为主要证据的较早期基础研究。

主要发现

  • PRO数据量和复杂性不断增加,对传统分析方法构成挑战。
  • 整合AI(包括机器学习和大型语言模型)可能有助于建模纵向症状轨迹以预测恶化、从自由文本临床笔记中提取症状,并将输出嵌入临床决策支持工具。
  • 应用时需考虑自主权、知情同意、数据安全、偏倚、可及性、透明度和问责制。

临床怎么用

  • 可探索将AI用于神经肿瘤患者PRO的纵向症状轨迹建模,以辅助预测病情恶化。
  • 可利用AI从自由文本临床笔记中提取症状信息,并嵌入临床决策支持工具。
  • 实际部署时应优先选择可解释、与电子健康记录整合且能减轻偏倚的工具,并保障患者知情同意、隐私与自主权。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

PURPOSE: Patient-reported outcomes (PROs) are critical tools for measuring how patients experience their illness and treatment. Neuropsychiatric symptoms, headaches, seizures, and neurological deficits evolve over time and impact quality of life. In this narrative review, we summarize current literature on PROs in neuro-oncology, extant artificial intelligence (AI) tools for PRO analysis, considerations for real-world application, and ethical implications. METHODS: We searched PubMed/MEDLINE, Embase, and Google Scholar for English-language articles published from January 2015 through April 2026, supplemented by reference-list reviews. Four themes were agreed upon, and authors were assigned to each theme. Authors conducted a literature review, selected articles, and synthesized the information under each of these themes. An initial 78 articles were narrowed down to 53 confirming eligibility if they reported PRO or symptom data in people with brain tumors or computational/AI methods were applied to PRO or symptom monitoring in oncology. Article selection was verified by two senior authors (L.V.Y, A.S.). Older, foundational studies were retained where they remain the primary evidence. RESULTS: The growing volume and complexity of PRO data present challenges for traditional analysis methods. Integrating AI, including machine learning (ML) and large language models (LLMs), may help model longitudinal symptom trajectories to predict deterioration, extract symptoms from free-text clinical notes, and embed outputs in clinical decision support tools. Autonomy, informed consent, data security, bias, access, transparency and accountability should be considered. CONCLUSION: Successful translation will require interpretable, electronic health record-integrated, bias-mitigating tools that allow for actionable insights and preserve patient consent, privacy, and autonomy in this highly vulnerable population. CLINICAL TRIAL NUMBER: Not applicable.

来源:J Neurooncol,查看原文。版权归原作者所有。

引用

  1. Yogendran LV, Avery C, Crooms RC, et al. Artificial intelligence for symptom assessment and patient-reported outcomes in neuro-oncology. J Neurooncol. 2026. PMID: 42905223. doi:10.1007/s11060-026-05832-9

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。