Construction and practice of application model for localized large language model in preoperative medication reconciliation for gastric cancer
- VernacularTitle:本地化大语言模型在胃癌术前药物重整中的应用模式构建与实践
- Author:
Yuxuan ZHU
1
;
Jizhong ZHANG
1
;
Yuhao SUN
1
;
Jiayu WEN
1
;
Xin LIU
2
;
Jifu WEI
2
;
Lingli HUANG
2
Author Information
1. Dept. of Pharmacy，Jiangsu Cancer Hospital/Jiangsu Institute of Cancer Research/Nanjing Medical University Affiliated Cancer Hospital/Jiangsu Key Laboratory of Innovative Cancer Diagnosis and Therapeutics，Nanjing 210009，China;School of Basic Medicine and Clinical Pharmacy，China Pharmaceutical University，Nanjing 211198，China
2. Dept. of Pharmacy，Jiangsu Cancer Hospital/Jiangsu Institute of Cancer Research/Nanjing Medical University Affiliated Cancer Hospital/Jiangsu Key Laboratory of Innovative Cancer Diagnosis and Therapeutics，Nanjing 210009，China
- Publication Type:Journal Article
- Keywords:
medication reconciliation;
artificial intelligence;
large language model;
gastric cancer;
preoperative medication
- From:
China Pharmacy
2026;37(8):1062-1067
- CountryChina
- Language:Chinese
-
Abstract:
OBJECTIVE To construct a preoperative medication reconciliation model assisted by a localized large language model （LLM） for gastric cancer and evaluate its clinical efficacy. METHODS A total of 249 gastric cancer patients with a history of continuous medication before admission in the Gastric Surgery Department of Jiangsu Cancer Hospital were retrospectively enrolled. Patients were divided into training set （154 cases） and validation set （95 cases） based on the order of time. Based on guidelines， drug package inserts， and other evidence， a standardized medication reconcili ation process and a structured knowledge base were constructed. DeepSeek-V3 LLM was deployed privately in the hospital， combined with retrieval-augmented generation technology， to achieve automated integration of medication information， risk screening， and generation of personalized recommendations. The quality of LLM-generated recommendations was evaluated using automatic metrics （BERT Score and ROUGE-1， 2， L） and manual scoring [seven-dimensional index （7DI） ] . Spearman correlation analysis was performed to explore the correlation between automatic scores and manual scores. Cronbach’s α coefficient was used to test the internal consistency of manual scoring results. The time consumed by manual and LLM-assisted medication reconciliation was compared across tasks of different difficulty levels （simple， moderate， and high）. RESULTS A structured knowledge base covering 8 major drug categories was finally established， covering common and high-risk preoperative medication scenarios and providing structured retrieval support for the LLM. For automatic evaluation， the precision， recall， and F1-score of BERT Score were 0.783±0.033， 0.811±0.038， and 0.796±0.028， respectively. The F1-scores of ROUGE-1， ROUGE-2 and ROUGE-L were 0.566±0.067， 0.338±0.076 and 0.468±0.082， respectively. The 7DI scores from three manual raters ranged from 32.06 to 33.45. The F1-score of automatic scoring was significantly positively correlated with the 7DI score of manual scoring （maximum coefficient of determination＝0.611， P ＜0.001）， and the internal consistency of manual scoring was good （Cronbach’s α ＝ 0.876）. In terms of efficiency， LLM-assisted medication reconciliation reduced time consumption by more than 90% compared with manual reconciliation in the simple， moderate， and high-difficulty groups （ P ＜0.001）. CONCLUSIONS The medication reconciliation model constructed based on a localized LLM and structured knowledge base shows high accuracy， consistency， and clinical applicability in complex preoperative medication scenarios for gastric cancer. It can improve the efficiency of medication reconciliation and reduce potential medication risks.
