文献速递围术期管理与 ERAS胃肠外科

人工智能用于手术术前风险评估:潜在获益、障碍与未来方向

Preoperative risk assessment for surgical operations using artificial intelligence: Potential benefits, hurdles, and future directions.
PubMed 文献 · Surgery · 2026-06-09

本文为叙述性视角综述,提出人工智能在手术风险评估中的潜在价值可观但不一致。基于电子健康记录衍生的机器学习算法在大型围术期人群中能有效判别术后死亡和主要心脑血管不良事件,但吻合口漏、中转开腹等技术性并发症较难仅凭基线患者特征预测。结论认为术前表格数据对某些预测结局存在内在信息上限,关键在于更好的数据生成、严格验证、透明输出并将预测风险与验证的干预路径连接。

要点

  • 人工智能可有效判别术后死亡和主要心脑血管不良事件,但技术性并发症预测较难。
  • 术前表格数据对某些预测结局存在内在信息上限,需更好数据生成与验证。
  • 人工智能应作为围术期智能的辅助层,而非外科判断的替代。

核心结论

  • 人工智能在手术风险评估中的潜在价值可观但不一致。
  • 术前表格数据对某些预测结局代表内在上限,长期关键在于更好的数据生成、严格验证、透明预测输出及将预测风险与验证的干预路径连接。

研究设计与样本

  • 叙述性视角综述,未报告系统检索范围或样本量。

主要发现

  • 在大型围术期人群中,电子健康记录衍生的机器学习算法对术后死亡和主要心脑血管不良事件判别非常有效。
  • 技术定义的 surgical complications(如吻合口漏或中转开腹)通常较难仅凭基线患者特征预测。
  • 选择结直肠外科作为示例,因为讨论的临床事件依赖术前不可获得的信息:器官和血管灌注、组织质量、手术复杂性和术中决策。

临床怎么用

  • 外科医生应将人工智能作为围术期智能的发展层,而非判断的替代,可能改善共享决策、分诊和实时风险缓解。

局限性

  • 人工智能在手术中的潜在价值不一致。
  • 术前表格数据对某些预测结局存在内在信息上限。

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

Artificial intelligence offers the potential of examining large clinical data sets to uncover complex nonlinearities and personalized associations when it comes to surgical risk assessment. This narrative perspective review proposes that the potential value of artificial intelligence in surgery is considerable, albeit inconsistent. Across many perioperative cohorts, electronic health record-derived machine learning algorithms have been shown to discriminate very effectively for postoperative mortality and major adverse cardiovascular or cerebrovascular events in large perioperative populations, whereas technically defined surgical complications (such as anastomotic leak or conversion to laparotomy) are generally harder to predict from baseline patient characteristics alone. For these reasons, the field of colorectal surgery is selected for illustration, since the discussed clinical events have a direct impact and rely upon information not obtainable prior to surgery, namely organ and vascular perfusion, tissue quality, surgical complexity, and intraoperative decisions. The point is not that we have already reached an artificial intelligence ceiling across all types of perioperative prediction problems; rather, the conclusion is that the information available from preoperative tabular data represents an intrinsic upper limit for certain predictive outcomes. What matters in the long run is better data generation, rigorous validation, transparent predictive output, and linking predicted risks to validated intervention pathways. Surgeons should use artificial intelligence not as a substitute for judgment but as a developing layer of perioperative intelligence that may improve shared decision-making, triage, and real-time risk mitigation.

来源:Surgery,查看原文。版权归原作者所有。

引用

  1. Abdelmasseh M, Aquina CT. Preoperative risk assessment for surgical operations using artificial intelligence: Potential benefits, hurdles, and future directions. Surgery. 2026. PMID: 42385382. doi:10.1016/j.surg.2026.110374

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。