2026 Volume 28 Issue 1 Published: 28 January 2026
  

  • Select all
    |
  • Interpretation of the International Expert Consensus on Hospital Intelligent Pharmacy
    Jiang Wenshuo, Li Cao, Shi Weizhong, Zhao Zhigang
    2026, 28(1): 2-9. https://doi.org/10.3760/cma.j.cn114015-20250728-00396
    Abstract ( ) PDF ( )
    To enhance the understanding of shared issues related to hospital intelligent pharmacy (HIP), experts from various fields at home and abroad collaboratively developed the International Expert Consensus on Hospital Intelligent Pharmacy through questionnaire-based surveys and the Delphi consensus method. To facilitate a better understanding of this consensus by pharmaceutical professionals in domestic medical institutions, this article summarized and elucidated the key points of the consensus. On the one hand, the consensus clarified the definition and connotation of HIP and emphasized its significance in intelligent hospital. On the other hand, a system covering important functional modules was constructed, including intel- ligent drug supply chain management, drug dispensing, prescription review, pharmacovigilance, medication therapy management, therapeutic drug monitoring, telepharmacy services, pharmacy administration, science popularization, and clinical trials. In addition, the consensus emphasized the importance of professional talent cultivation and academic carrier construction, and proposed to provide support for continuous innovation and practical promotion in this field through specialized talent cultivation and high-quality academic platforms.
  • Seizing the opportunity of artificial intelligence development to promote the quality, efficiency, and innovation in pharmacovigilance
    Cui Xiaohui, Wang Tianlin
    2026, 28(1): 10-14. https://doi.org/10.3760/cma.j.cn114015-20251107-00577
    Abstract ( ) PDF ( )
    Pharmacovigilance is the core component of ensuring the safety of drugs throughout their entire lifecycle, undertaking the important mission of monitoring, identifying, evaluating and controlling adverse drug events, and safeguarding public health. In 2019, the newly revised Drug Administration Law of the People′s Republic of China officially established the pharmacovigilance system as one of the basic systems for drug management in China. In 2021, Good Pharmacovigilance Practice further clarifies regulatory requirements. However, existing information system of pharmacovigilance still faces bottlenecks such as a surge in adverse reaction reports, difficulties in integrating multisource data, and passive and lagging monitoring modes. The breakthrough development of artificial intelligence (AI) technology provides an important opportunity to overcome these transformation challenges in pharmacovigilance. In this paper, the technological innovation pathways of AI-empowered pharmacovigilance, including multisource and multimodal data integration, avoidance of violation risks, and reduction of technical application thresholds are elaborated; the application value of AI in pharmacovigilance scenarios such as automated report generation, signal screening, audit verification, and risk assessment is analyzed; the practical challenges that AI faces in terms of data quality, model efficiency, human-machine collaboration, cost control, and ethical standards are explored, and at last, a vision for the future development direction of the integration of AI technology and pharmacovigilance from 5 dimensions are proposed, including data governance, technological optimization, capacity building, cost control, and ethical norms.
  • Construction and application of an agent for personalized medication recommendation by integrating large language models with population pharmacokinetic models
    Guo Heng, Li Zhe, Liu Yi, Wang Weina, Li Xingang
    2026, 28(1): 15-21. https://doi.org/10.3760/cma.j.cn114015-20250418-00208
    Abstract ( ) PDF ( )
    Objective To develop an agent integrating a large language model (LLM) with population pharmacokinetic (PPK) models for personalized medication recommendation via natural language interaction.　Methods Using the Dify workflow as the framework and employing DeepSeek-R1 as the LLM, an intelligent agent system was constructed by integrating functional modules including retrievalaugmented generation (RAG), question classification, parameter extraction, and result integration and output. The system was connected to a local knowledge base integrating clinical guidelines and pharmacokinetic research literature, as well as a PPK model application developed in Python. RAG technology was utilized to enhance response accuracy and knowledge timeliness. A structured process enabled full automation from user natural language queries to parameter extraction, model selection, PPK calculation, and result output.　Results A personalized medication recommendation agent integrating a local knowledge base and PPK models was successfully developed. This agent could respond correctly to clinical medication-related queries via natural language interaction. Its core functionalities included the following: answering pharmaceutical consultation questions based on the local knowledge base and providing reference sources; completing individualized dose recommendations, simulation calculation of drug concentration, and plotting of drug concentration-time curves based on PPK models and individual patient characteristics (e.g., age, body weight, pathological status); accurately estimating individual patient pharmacokinetic parameters using Bayesian feedback with sparse measured drug concentration data. Validated in clinical scenarios such as treatment of neonatal Candida albicans infection and individualized tacrolimus therapy in pediatric post-liver transplantation patients, the agent generated precise medication recommendations, parameter results, and drug concentration-time curves, effectively supporting clinical individualized medication decision-making.　Conclusion The intelligent agent constructed in this study effectively integrates LLM with PPK models and achieves personalized medication recommendation through natural language interaction. Dose recommendations based on the PPK model provide quantitative evidence for clinical medication, helping to reduce unnecessary drug exposure and lower the risk of adverse drug reactions and harmful drug interactions. Additionally, this study demonstrates the feasibility of a cooperative framework combining a front-end LLM application with a back-end professional model in the field of pharmaceutical services.
  • Preliminary development and evaluation of a multi-modal agent for medication education in tuberculosis patients
    Nai Jingxue, Liu Shuo, Xu Rui, Yan Chenxia, Cheng Kai
    2026, 28(1): 22-29. https://doi.org/10.3760/cma.j.cn114015-20250711-00368
    Abstract ( ) PDF ( )
    Objective To preliminarily develop a medication educational agent based on multi- modal interactive and solve the problem of information understanding obstacles in medication education for tuberculosis patients.　Methods Based on the Dify 1.4.1 platform, an agent workflow was constructed using large language model (LLM) and retrieval-augmented generation. A total of 50 tuberculosis patients, who discharged from Beijing Chest Hospital, Capital Medical University from March to September 2025, were selected. The structured information was collected and input into the agent to evaluate its intrinsic performance, including structured information extraction capability (precision, recall, and micro-averaged F1-value), text generation quality ［bilingual evaluation understudy (BLEU) 4 and series of indicators of recall-oriented understudy for gisting evaluation (ROUGE)］, as well as stability (interaction success rate), efficiency (response time and average material generation time), compliance (compliance rate), and user experience (satisfaction scores). A total of 38 medical professionals (physicians and pharmacists), who worked in the hospital from February to March 2025, were selected as survey subjects. Using the Wenjuanxing platform, the accuracy, comprehensiveness, readability, humanistic care, and personalization of medication education materials from 3 sources were evaluated, including the hospital′s current standardized medication guidance template (material 1), medication education materials directly generated by a generalpurpose LLM (material 2) and materials generated by the agent developed in this study (material 3). And the application effectiveness of the agent was assessed via the survey results.　Results The evaluation result of the agent using structured information of 50 tuberculosis patients showed that the precision was 95%, the recall was 92%, and micro-averaged F1-value was 0.93. The agent was scored (18.61±4.06), (38.60±5.93), (22.40±5.13), and (29.42±6.81) points in BLEU-4, ROUGE-1, ROUGE-2, and ROUGE-L, respectively. The interaction success rate was 96% (48/50), with an average response time of  (3.1±0.6) s and a material generation time of (27.4±1.5) s, the compliance rate was 100% (50/50), and the patient satisfaction score for the agent generated text was (84.5±5.5) points. The survey results of 38 medical professionals showed in dimensions of readability, humanistic care and personalization, the scores of material 3 were better than materials 1 and 2, and the differences were statistically significant (all P<0.016 7). The score of material 3 in the comprehensiveness dimension was better than the material 1 (P=0.003). In these medical professionals, 71.1% (27/38) were satisfied with material 3.　Conclusions A multi-modal interactive agent for medication education in tuberculosis patients is successfully developed. The multiple performance indicators of this agent have good feasibility and reliability, and have certain advantages in readability, humanistic care, and personalization. By providing multi-modal and layered outputs, this agent offers a novel paradigm for medication education in tuberculosis patients.
  • Knowledge structure and research hotspot of artificial intelligence in drug-induced diseases: a visual bibliometric analysis
    Liang Shuo, Wang Xue, Ren Zhao, Sun Li Chaoyue, Su Su, Liu Hua, He Aini, Song Haiqing, Yan Suying
    2026, 28(1): 30-38. https://doi.org/10.3760/cma.j.cn114015-20250612-00316
    Abstract ( ) PDF ( )
    Objective To understand the knowledge structure of artificial intelligence (AI) technology applications in the field of drug-induced diseases (DID), and to reveal the research hotspots and development trends.　Methods Relevant literature on drug safety research with AI technologies was systematically retrieved from PubMed, Embase, Web of Science, SinoMed, CNKI, and Wanfang databases (up to December 2024). The bibliometric online analysis platform, VOSviewer 1.6.18 and CiteSpace 6.4.R2 were used to analyze and visualize the indicators such as publication volume, author countries/institutions, publishing journals, highly cited literature, and keywords.　Results A total of 1 857 articles were included, comprising 1 433 English and 424 Chinese publications. Annual publication volumes for both English and Chinese literature showed continuous growth. Authors from China contributed the most English publications (462 articles), followed by those from the United States (401 articles); also, collaboration between the 2 countries was the most frequent. About the author institutions, Harvard University had the largest volume of publications (243 articles), followed by Central South University (112 articles). The English journal with the most publications in this field was Frontiers in Pharmacology (94 articles), while the primary Chinese journal was China Pharmacy (22 articles). The top 10 articles ranked by total cited frequency were all in English, focusing on the use of machine learning algorithms and deep learning models for early warning and data mining for DID. Keyword clustering revealed 11 clusters for English literature and 10 clusters for Chinese literature. In the English clusters, chemical and drug induced liver injury was the largest cluster (started in 2015 and included 87 keywords), Bayes theorem was the earliest cluster (started in 2009 and included 34 keywords), and random forest was the latest cluster (started in 2019 and included 42 keywords). In the Chinese clusters, data mining was the largest cluster (started in 2016 and included 38 keywords), and signal mining was the latest cluster (started in 2023 and included 18 keywords).　Conclusions The application of AI technologies in DID researches is rapidly expanding. Research hotspots have evolved from machine learning to deep learning and large language models, with the liver injury, kidney injury, and cardiac injury induced by drugs as the key diseases. Future researches in this field will primarily focus on 2 aspects, further integration of advanced technologies and tools for the first, and expansion of data scale and diversification of data sources for the second, to promote the development of this field towards proactivity, precision, and intelligence.
  • Drug-induced fingerprint loss: a study based on the US FDA Adverse Event Reporting System database and published case reports
    Shen Jie, Deng Lin, Xie Qiaojin, Rao Mei
    2026, 28(1): 39-44. https://doi.org/10.3760/cma.j.cn114015-20250423-00219
    Abstract ( ) PDF ( )
    Objective To mine the potential risk drugs associated with fingerprint loss and provide reference for clinical safe drug use.　Methods Adverse event (AE) reports with the preferred term  “fingerprint loss” were collected by searching US Food and Drug Administration Adverse Event Reporting System (FAERS) database from the first quarter of 2004 to the first quarter of 2025. Reporting odds ratio (ROR) and proportional reporting ratio (PRR) methods were used to mine the AE risk signals. Drugs meeting the criteria of ≥3 reports, 95% confidence interval lower limit of ROR>1, PRR≥2, and χ2≥4 were consi- dered to have a potential risk signal for fingerprint loss. The identified drugs were classified according to the Anatomical Therapeutic Chemical (ATC) classification system. In addition, relevant literature databases were searched to collect case reports of drug-induced fingerprint loss, and descriptive analysis were conducted on the reported clinical characteristics and medications involved.　Results A total of 134 AE reports of drug-induced fingerprint loss were collected within the limited time period, and 87 reports were detected as risk signals by ROR and PRR methods; it involved 7 drugs, all of which were classified as antineoplastic agents according to the ATC system. Ranked in descending order of signal strength, the 7 drugs were capecitabine (ROR=267.47), tucatinib (ROR=146.54), ribociclib (ROR=24.61), fluorouracil (ROR=21.41), pal- bociclib (ROR=16.99), trastuzumab (ROR=12.52), and bevacizumab (ROR=10.47). Among them, only the label of capecitabine had records of fingerprint loss. A total of 13 case reports of drug-induced fingerprint loss were retrieved, involving 14 patients; related drugs included capecitabine in 9 cases, fluorouracil, osimertinib/anlotinib, doxorubicin, paclitaxel, and venlafaxine in 1 case each.　Conclusions The main risk drugs for fingerprint loss are antineoplastic drugs, among which capecitabine has the strongest risk signal. Other risk drugs include osimertinib/anlotinib, doxorubicin, paclitaxel, and venlafaxine.
  • Efficacy and safety of ensifentrine in the treatment of chronic obstructive pulmonary disease: a meta-analysis
    Liu Yanyan, Ma Yin, Tang Yanfen, Wang Yu
    2026, 28(1): 45-53. https://doi.org/10.3760/cma.j.cn114015-20250215-00070
    Abstract ( ) PDF ( ) Supplementary files
    Objective To systematically evaluate the efficacy and safety of ensifentrine in the treatment of chronic obstructive pulmonary disease (COPD).　Methods Randomized controlled trials (RCTs) of ensifentrine in the treatment of COPD were collected by searching related databases at home and abroad (up to February 11, 2025). On the basis of conventional treatment for COPD, patients in in the trial group were given ensifentrine, while those were given placebo additionally in the control group. The efficacy outcomes included forced expiratory volume in 1 second (FEV1), sores of Evaluating Respiratory Symptoms in COPD (E-RS-COPD) and St.George′s Respiratory Questionnaire (SGRQ), transition dyspnea index (TDI), and daily rescue medication use (RMU). The safety outcomes included the incidences of adverse events (AEs) and acute exacerbation of COPD during treatment. Quality of methodology was evaluated using bias risk assessment tool of Cochrane collaboration networks. Meta-analysis was performed using RevMan 5.3 software. The effect sizes of measurement data were mean difference (MD) and its 95% confidence interval (CI), while the effect sizes of counting data were odds ratio (OR) or relative risk (RR) and its 95%CI.　Results A total of 4 RCTs and 2 367 patients were entered in the analysis, including 1 629 patients in the trial group (divided into 5 dose groups of 0.357, 0.75, 1.5, 3, and 6 mg) and 738 in the control group. The meta-analysis results showed that compared with the control group, the peak FEV1(MD=139.03 ml, 95%CI: 121.54-156.52 ml), area under the curve of FEV1 from 0 to 12 h (MD=91.59 ml, 95%CI: 67.39- 115.79 ml), and FEV1 morning trough values (MD=43.44 ml, 95%CI: 18.88-68.00 ml) in the trial group were significantly increased at the end of the observation compared to baseline, while the scores of E-RS-COPD (MD=-1.25 points, 95%CI: -1.60- -0.91 points) and SGRQ (MD=-2.45 points， 95%CI: -3.42- -1.47 points) were significantly decreased compared to baseline, and the differences were statistically (all P<0.001). In addition, the TDI of the 1.5 mg, 3 mg, and 6 mg subgroups in the trial group was improved more significantly respectively compared to those in the control group, the average daily RMU in the 6 mg subgroup was significantly reduced compared to that in the control group, and the differences were statistically significant (all P<0.05). There was no significant difference in the incidence of AEs ［5.6% (91/1 629) vs. 5.0% (37/738), RR= 1.00, 95%CI: 0.69 1.45］, incidence of serious AEs ［4.2% (69/1 629) vs. 5.0% (37/738), RR=1.01, 95%CI: 0.69 1.49］, or incidence of AEs leading to drug discontinuation ［5.9% (96/1 629) vs. 7.2% (53/738), RR=0.92, 95%CI: 0.67-1.28］ during the study period between the trial group and the control group (all P>0.05).　The acute exacerbation rate of COPD in the trial group was significantly lower than that in the control group［7.1%(115/1 629) vs. 13.4%(99/738), RR=0.61, 95%CI: 0.48-0.79, P<0.001］. Conclusion Ensifentrine can significantly improve the lung function and dyspnea in COPD patients, and reduce the risk of acute exacer- bation, with a good safety profile.
  • Application and prospects of prompt engineering in large language models for clinical pharmacy
    Xin Hao, Zheng Jianling, Cui Xiaohui, Tang Ying, Liu Dong, Liu Zongtao, Li Guanghui, Cao Jianhua
    2026, 28(1): 54-61. https://doi.org/10.3760/cma.j.cn114015-20250605-00306
    Abstract ( ) PDF ( )
    Large language models have developed rapidly in the field of healthcare in recent years, particularly in clinical pharmacy, where they demonstrated broad application prospects for knowledge retrieval, information integration, risk assessment, and decision support. The prompt engineering played an important role in ensuring the accuracy, interpretability, and clinical applicability of output content. Drawing upon the current practical experience in clinical pharmacy, this article systematically expounds the fundamental concepts, working principles, and main types of prompt engineering, explores its major application scenarios, optimization strategies, and common challenges, and further discusses future development directions, aiming to provide pharmacists and relevant researchers with systematic references and practical guidance.
  • Acute generalized exanthematous pustulosis induced by belimumab
    Yang Xiaowen, Wuriliga, Wu Guiying
    2026, 28(1): 62-64. https://doi.org/10.3760/cma.j.cn114015-20250207-00060
    Abstract ( ) PDF ( )
    A 56-year-old male patient received an intravenous infusion of 720 mg belimumab once every 2 weeks for systemic lupus erythematosus. On the second day of medication, the patient developed small red papules on the skin of the chest and abdomen, which gradually spread to the head, face, behind the ears, earlobes, neck, back, and groin, manifesting as extensively and symmetrically distributed edematous papules. The lesions on the back skin merged into large patches, with target-like lesions visible locally. The pathological results of the patient′s chest skin biopsy showed a large number of pustules in and below the stratum corneum, dilated and congested blood vessels in the superficial and middle layers of the dermis, and infiltration of surrounding lymphocytes and neutrophils, which was consistent with the diagnosis of acute generalized exanthematous pustulosis. It was considered to be induced by belimumab. Belimumab was discontinued immediately and antiallergic treatments with methylprednisolone and loratadine, combined with topical application of fluticasone propionate cream and fusidic acid cream. After 1 week of treatments, the rash was improved significantly, and after 1 month, the rash basically disappeared, leaving behind pigmentation and desquamation.
  • Fulminant type 1 diabetes mellitus caused by serplulimab
    Hou Leping, Zhang Dan
    2026, 28(1): 65-67. https://doi.org/10.3760/cma.j.cn114015-20250109-00020
    Abstract ( ) PDF ( )
    A 60-year-old male patient with penile squamous cell carcinoma received postopera- tive treatment with serplulimab (200 mg intravenous infusion on day 1, every 21 days as one cycle). Pre- treatment fasting blood glucose was 5.6 mmol/L. Four days after administration of the 6th cycle, the patient developed symptoms of dry mouth, polydipsia, and polyuria, followed by nausea and vomiting 3 days later. Laboratory tests showed random blood glucose 31.9 mmol/L, β-hydroxybutyrate 6.9 mmol/L, blood pH 7.21, base excess -15.4 mmol/L, bicarbonate concentration 10.9 mmol/L, serum potassium 4.59 mmol/L, blood lactate 1.7 mmol/L, urine glucose (++++), and urine ketones (++). Diabetic ketoacidosis was diagnosed and considered to be induced by serplulimab. The drug was therefore discontinued, and supportive treatments including fluid replacement, potassium supplementation, and continuous insulin infusion for glycemic control were initiated. The next day, fasting blood glucose was 15.1 mmol/L and glycated hemoglobin 8.1%. One day later, the insulin pump was stopped and replaced with a regimen of insulin lispro, insulin glargine, and acarbose. The following day, both fasting and 1 hour/2 hour postprandial C-peptide levels were<0.01 nmol/L, and diabetes-related autoantibodies were all negative. Two days later, the fasting blood glucose still elevated at 12.7 mmol/L, which was considered as fulminant type 1 diabetes with poor glycemic control, and metformin was added. Two days afterwards, urine ketones negatire, the fasting blood glucose decreased to 6.0 mmol/L. At a 40 day follow-up, the glycated hemoglobin was 7.7%.
  • Acute urinary retention induced by doxofylline in an elderly patient
    Xiao Sa, Lei Luwen
    2026, 28(1): 67-69. https://doi.org/10.3760/cma.j.cn114015-20250224-00092
    Abstract ( ) PDF ( )
    An 82-year-old male patient with chronic obstructive pulmonary disease was treated with doxofylline injection (0.3 g intravenous infusion, once daily) due to worsened condition of recurrent cough and dyspnea. After 3 days of treatment, the patient′s dyspnea was significantly improved, but he suddenly developed urgent and frequent micturition, and dysuria. Color Doppler ultrasound examination of urgent urinary system and residual urine volume measurement revealed prostate enlargement and a residual urine volume of 281 ml. Acute urinary retention was considered and suspected to be caused by doxofylline. The drug was immediately discontinued, and catheterization was performed, along with oral administration of tamsulosin and finasteride. After 3 days of comprehensive treatments, the patient′s symptoms of urgent and frequent micturition and dysuria were improved, tamsulosin and finasteride were discontinued. At one-month of follow-up, the patient′s urination returned to normal, with no recurrence of urinary retention.
Office
Journal
Download
More...
Link
More...