文献速递头颈与中枢神经肿瘤支持与姑息治疗

提示词引导的环境人工智能用于头颈肿瘤多学科会诊记录自动化

Prompt-directed ambient artificial intelligence for automated multidisciplinary tumor board documentation.
PubMed 文献 · Ann Oncol · 2026-06-10

该前瞻性质量改进研究在2025年于一家三级机构评估环境人工智能(AI)用于头颈肿瘤多学科会诊讨论的记录。主要干预为病例讨论中口头提示以提升记录质量,主要队列为无提示与提示引导的环境AI会诊。提示引导的环境AI较无提示者改善了肿瘤特征域评分(q=0.042)与总分(P=0.005),且未增加每患者讨论时间(P=0.158);配对比较中环境AI较人工记录改善了总分(P<0.001)与讨论域评分(q<0.001),并减少每患者行政时间(P<0.001)。

要点

  • 提示引导的环境AI可改善头颈肿瘤会诊记录质量且不增加讨论时间
  • 环境AI记录质量优于人工记录并减少行政负担
  • 内部验证队列各域评分统计学可比

核心结论

  • 头颈肿瘤会诊中口头提示引导可改善环境AI生成的记录摘要且不增加讨论时间
  • 环境AI较人工记录改善记录质量并减轻行政负担

研究设计与样本

  • 前瞻性质量改进研究,2025年于一家三级机构开展
  • 主要队列为无提示与提示引导的环境AI会诊记录
  • 配对比较子分析纳入同场次会诊的人工与环境AI记录
  • 共分析12场会诊,分四个队列:无提示环境AI、提示引导环境AI、人工记录、内部验证

主要发现

  • 主要队列中提示引导环境AI较无提示者改善肿瘤特征域评分(q=0.042)与总分(P=0.005)
  • 主要队列中每患者讨论时间无差异(P=0.158)
  • 配对比较中环境AI较人工记录改善总分(P<0.001)与讨论域评分(q<0.001)
  • 配对比较中环境AI减少每患者行政时间(P<0.001)
  • 内部验证队列各域评分统计学可比

关键数值与推荐

  • 记录质量采用改良的MODe(Metric for Observation of Decision Making) rubric评估五个域:人口学、肿瘤特征、影像学发现、病理学发现、讨论
  • 各标准按1-3分评定:1=错误/遗漏,2=部分,3=完整记录

临床怎么用

  • 在头颈肿瘤多学科会诊中可于病例讨论时使用口头提示引导环境AI记录
  • 环境AI记录可替代人工记录以提升记录质量并减少行政时间

局限性

  • 研究为单中心三级机构的前瞻性质量改进研究,样本为12场会诊,外部推广性需进一步验证

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

BACKGROUND: Tumor board documentation is challenging due to complexity of multidisciplinary input. Ambient artificial intelligence (AI) for medical dictation uses generative methods to summarize voice data. Strategic prompting can improve model output. We evaluated prompt-directed ambient AI for automated tumor board documentation and compared ambient AI with manual documentation. PATIENTS AND METHODS: We carried out a prospective quality improvement study evaluating ambient AI for documentation of multidisciplinary head and neck cancer tumor board discussions at a tertiary institution in 2025. The primary intervention was verbal prompting during case discussions to improve documentation quality. Primary cohorts were unprompted and prompt-directed ambient AI tumor boards. A paired comparator subanalysis included manual and ambient AI documentation from the same meetings. We validated our prompted ambient AI workflow with additional tumor boards. Documentation was evaluated using a modified Metric for Observation of Decision Making (MODe)-based rubric across five domains: demographics, tumor characteristics, radiology findings, pathology findings, and discussion, with criteria rated on a 1-3 scale (1 = incorrect/omitted, 2 = partial, 3 = complete documentation). RESULTS: We analyzed results of 12 tumor boards across four cohorts: unprompted ambient AI, prompt-directed ambient AI, manual documentation, and internal validation. In the primary cohorts, prompt-directed ambient AI improved tumor characteristics domain score (q = 0.042) and total score (P = 0.005) versus unprompted ambient AI, with no difference in per-patient discussion time (P = 0.158). In the paired comparator analysis, ambient AI improved total score (P < 0.001) and discussion domain score (q < 0.001) versus manual documentation, while reducing administrative time per patient (P < 0.001). The internal validation cohort showed statistically comparable scores across all domains. CONCLUSIONS: Directed prompting during head and neck tumor board improved generated summaries, without increasing discussion times. Compared with manual documentation, ambient AI improved documentation quality, while diminishing administrative burden.

来源:Ann Oncol,查看原文。版权归原作者所有。

引用

  1. Philips R, Mahajan A, Estephan L, et al. Prompt-directed ambient artificial intelligence for automated multidisciplinary tumor board documentation. Ann Oncol. 2026. PMID: 42270078. doi:10.1016/j.annonc.2026.05.709

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。