Page 113 - 《中国药房》2026年8期
P. 113
剂)由于真实病例覆盖不足,缺乏充分的情景化验证。 al. Benchmark evaluation of DeepSeek large language
此外,尽管本研究构建的“本地化LLM+知识库+RAG技 models in clinical decision-making[J]. Nat Med,2025,31
术路径”具有良好的可复制性,但由于面临医院信息系统 (8):2546-2549.
异构性、临床流程差异以及知识库本地化适配等问题,其 [13] 中华医学会糖尿病学分会 . 中国 2 型糖尿病防治指南:
在真实世界医疗场景中的落地仍面临一定挑战。需要说 2013年版[J]. 中华糖尿病杂志,2014,6(7):447-498.
明的是,这种局限性并不意味着该技术路径不可推广,而 [14] 唐密密,许世伟,葛杰,等. 胃癌患者围手术期药学监护
专 家 共 识 [J/OL]. 今 日 药 学 ,2025:1-26[2026-01-10].
是临床人工智能系统在跨机构应用中的普遍特征。
https://link.cnki.net/urlid/44.1650.R.20250110.1457.007.
综上所述,基于本地化 LLM 与结构化知识库构建
[15] OPREA A D,KESHOCK M C,O’GLASSER A Y,et al.
的药物重整模式,在胃癌术前复杂用药场景中具有较高
Preoperative management of medications for psychiatric
的准确性、一致性和临床可用性,能够提升药物重整效
diseases society for perioperative assessment and quality
率,同时降低潜在用药风险。 improvement consensus statement[J]. Mayo Clin Proc,
参考文献 2022,97(2):397-416.
[ 1 ] World Health Organization. Medication safety in poly‐ [16] ZHANG T Y,KISHORE V,WU F,et al. BERTScore:
pharmacy:technical report[EB/OL]. [2025-11-11]. https:// evaluating text generation with BERT[EB/OL]. [2025-11-
www.who.int/publications/i/item/WHO-UHC-SDS-2019.11.
11]. https://arxiv.org/pdf/1904.09675.
[ 2 ] SUNDAR R,NAKAYAMA I,MARKAR S R,et al. Gas‐ [17] LIN C Y. ROUGE:a package for automatic evaluation of
tric cancer[J]. Lancet,2025,405(10494):2087-2102. summaries[EB/OL]. [2025-11-11].https://www.microsoft.
[ 3 ] 李茁钰,刘凯,张维汉,等. 全球及中国胃癌的流行病学
com/en-us/research/publication/rouge-a-package-for-automa-
特点及趋势:2018—2022《全球癌症统计报告》解读[J]. tic-evaluation-of-summaries/.
中国普外基础与临床杂志,2024,31(10):1236-1245. [18] DING H L,XIA W J,ZHOU Y J,et al. Evaluation and
[ 4 ] LERTKOVIT S,SIRIUSSAWAKUL A,SURAARUNSUMRIT
practical application of prompt-driven ChatGPTs for EMR
P,et al. Polypharmacy in older adults undergoing major generation[J]. NPJ Digit Med,2025,8(1):77.
surgery:prevalence,association with postoperative cogni‐
tive dysfunction and potential associated anesthetic agents [19] GEORGE D,MALLERY P. IBM SPSS statistics 29 step
[J]. Front Med,2022,9:811954. by step:a simple guide and reference[M]. 18th Ed. New
[ 5 ] 马晓波,马小桐,马建清,等. 1990—2021年中国肿瘤疾 York:Routledge,2024:251-262.
病负担分析[J]. 医学新知,2025,35(8):885-891 [20] YE J C. Patient safety of perioperative medication through
[ 6 ] YAMADA Y,KOBAYASHI R,YAMAMOTO T,et al. the lens of digital health and artificial intelligence[J].
JMIR Perioper Med,2023,6:e34453.
Medication reconciliation by pharmacists for pre-
admission patients improves patient safety[J]. J Pharm [21] 中国心胸血管麻醉学会非心脏麻醉分会,中国医师协会
Health Care Sci,2024,10(1):19. 心血管内科医师分会,中国心血管健康联盟. 抗血栓药
[ 7 ] MUELLER S K,SPONSLER K C,KRIPALANI S,et al. 物围手术期管理多学科专家共识[J]. 中华医学杂志,
Hospital-based medication reconciliation practices:a syste- 2020,100(39):3058-3074.
matic review[J]. Arch Intern Med,2012,172(14):1057- [22] CASCELLA M,MONTOMOLI J,BELLINI V,et al.
1069. Evaluating the feasibility of ChatGPT in healthcare:an
[ 8 ] KENNELTY K A,CHEWNING B,WISE M,et al. Barriers analysis of multiple clinical and research scenarios[J]. J
and facilitators of medication reconciliation processes for Med Syst,2023,47(1):33.
recently discharged patients from community pharma‐ [23] CHINTAGUNTA B,KATARIYA N,AMATRIAIN X,et
cists’ perspectives[J]. Res Soc Adm Pharm,2015,11(4): al. Medically aware GPT-3 as a data generator for medical
517-530. dialogue summarization[EB/OL]. [2025-11-11]. https://
[ 9 ] HE K,MAO R,LIN Q K,et al. A survey of large lan‐ arxiv.org/pdf/2110.07356.
guage models for healthcare:from data,technology,and [24] 张微微,唐婧,冀召帅,等. 人工智能在临床药学中的应
applications to accountability and ethics[J]. Inf Fusion, 用进展[J]. 药学学报,2025,60(10):3045-3059.
2025,118:102963. [25] LI H Y,FU J F,PYTHON A. Implementing large lan‐
[10] SINGHAL K,AZIZI S,TU T,et al. Large language models guage models in health care:clinician-focused review
encode clinical knowledge[J]. Nature,2023,620(7972): with interactive guideline[J]. J Med Internet Res,2025,
172-180. 27:e71916.
[11] TEMSAH A,ALHASAN K,ALTAMIMI I,et al. Deep‐ [26] WIEST I C,FERBER D,ZHU J F,et al. Privacy-
Seek in healthcare:revealing opportunities and steering preserving large language models for structured medical
challenges of a new open-source artificial intelligence information retrieval[J]. NPJ Digit Med,2024,7(1):257.
frontier[J]. Cureus,2025,17(2):e79221. (收稿日期:2026-01-16 修回日期:2026-03-13)
[12] SANDMANN S,HEGSELMANN S,FUJARSKI M,et (编辑:舒安琴)
中国药房 2026年第37卷第8期 China Pharmacy 2026 Vol. 37 No. 8 · 1067 ·

