{
  "abstract": "目的: 诊断证据不足与边界模糊影响临床判断和科研标签可靠性，而规范诊断所需的访谈成本高、普及难度大，因此可借助常规留存的住院病历文本开展辅助诊断复核。本研究旨在评估常用的几种通用大语言模型(large language model，LLM)基于精神科入院病历开展诊断复核时与临床参考诊断的一致性及多轮输出稳定性，分析其诊断信息完整度、诊断复杂度与模型首选诊断(Top-1)一致性的关系，并探索高频Top-1不一致病例的特征，为后续人工复核提供参考。\n\n方法: 采用回顾性诊断一致性评估设计，纳入125例精神科住院病历，临床参考诊断分属精神分裂症谱系、双相情感障碍和抑郁障碍3个《精神障碍诊断与统计手册(第五版-修订版)》(Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision，DSM-5-TR)诊断大类。模型仅接收经标准化和去标识化处理的入院阶段病历文本。采用GPT-5.5、Gemini 3.5 Flash和Qwen 3.7 Max分别对全部病例的诊断信息完整度和诊断复杂度进行3轮独立评估。首先以45例分层样本对10个候选模型进行初步筛选，随后选取Qwen 3.7 Max、GPT-5.5、Gemini 3.5 Flash、Qwen 3.6 35B-A3B、DeepSeek-R1-Distill-Qwen-32B和Gemma 4 26B-A4B共6个模型进行正式评估。每个模型对125例病例独立运行3轮，共形成2 250次诊断判断。主要结局为Top-1一致率，次要结局包括Top-2一致率和跨轮次一致性。采用多因素线性概率模型分析Top-1一致率的相关因素，并对高频Top-1不一致的病例进行探索性回顾分析。\n\n结果: 汇总6个模型的3轮运行结果后，总体Top-1一致率为93.0%(2 092/2 250，95% CI 89.0%~96.4%)，Top-2一致率为95.6%(2 151/2 250)，各模型跨轮次一致性均超过92.0%。按每个模型3轮共375次诊断判断汇总计算，在线模型中Qwen 3.7 Max的Top-1一致率最高，为94.4%；本地部署模型中Qwen 3.6 35B-A3B的Top-1一致率最高，为93.1%。在基于全部6个模型和3轮运行结果的多因素分析中，调整诊断信息完整度、模型、诊断大类和运行轮次后，诊断复杂度每增加1个标准差，Top-1一致率平均降低3.11个百分点(95% CI -5.66~-0.57，P=0.017)；诊断信息完整度与Top-1一致率呈正向关联趋势，但未达到统计学意义(P=0.075)。共识别7例高频Top-1不一致病例。探索性回顾初步显示，其诊断分歧可能涉及心境症状与精神病性症状的时序关系、横断面表现与纵向病程的整合，以及关键鉴别信息的缺失或不明确。\n\n结论: 在精神分裂症谱系、双相情感障碍和抑郁障碍3个诊断大类层面，通用LLM与临床参考诊断具有较高的一致性，多轮输出也较为稳定，较高的诊断复杂度与较低的Top-1一致率相关。高频Top-1不一致病例可作为人工复核的探索性提示，但不能直接解释为模型诊断错误或临床误诊。这一提示的实际价值仍需通过独立专家评价和外部样本验证。",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": "research article",
  "arxiv": null,
  "author_total": 6,
  "authors": [
    {
      "name": "升 徐",
      "orcid": "0009-0000-7923-0869"
    },
    {
      "name": "孜妍 林"
    },
    {
      "name": "涵俪 汪"
    },
    {
      "name": "宁 袁"
    },
    {
      "name": "超 陈",
      "orcid": "0000-0002-5114-2282"
    },
    {
      "name": "春宇 刘",
      "orcid": "0000-0002-5986-4415"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/48df2ba7e8eqs.bib",
  "cited_by": 0,
  "doi": "10.11817/j.issn.1672-7347.2026.260431",
  "fulltext_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13710913/",
  "id": "48df2ba7e8eqs",
  "issns": [
    "1672-7347"
  ],
  "landing_url": "https://doi.org/10.11817/j.issn.1672-7347.2026.260431",
  "language": "zh",
  "licence": "CC-BY-NC-ND-4.0",
  "licence_label": "CC BY-NC-ND 4.0",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by-nc-nd/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "heading",
      "Topics on mental and Psychological Health in the Era of Digital Intelligence"
    ],
    [
      "openalex",
      "2718|Artificial Intelligence in Healthcare and Education"
    ]
  ],
  "openalex": "W7223884396",
  "pdf_url": null,
  "pmcid": "PMC13710913",
  "pmid": "42935190",
  "published": "2026-08-28",
  "publisher": "Central South University",
  "ris": "https://research.draconitic.com/paper/48df2ba7e8eqs.ris",
  "sources": [
    {
      "first_seen": "2026-10-31T12:21:47Z",
      "native_id": "PMC13710913",
      "source": "pmc",
      "url": "https://doi.org/10.11817/j.issn.1672-7347.2026.260431"
    }
  ],
  "subjects": [
    {
      "field": "medicine",
      "origin": "openalex:2718|Artificial Intelligence in Healthcare and Education",
      "subfield": "epidemiology-public-health"
    }
  ],
  "title": "大语言模型诊断精神障碍的一致性及相关因素分析",
  "title_text": "大语言模型诊断精神障碍的一致性及相关因素分析",
  "type": "peer-reviewed",
  "updated_at": "2026-10-31T12:21:47Z",
  "url": "https://research.draconitic.com/paper/48df2ba7e8eqs",
  "venue": "Journal of Central South University Medical Sciences",
  "versions": []
}