Objective This study aims to extract medical events from the ancient Chinese medical book"Treatise on Febrile Diseases"and explore their internal connections.By constructing an event evolution graph,this study visualizes the progression of diseases related to the three Yang and three Yin,provides new ideas for the digitization of ancient Chinese medical literature,and offers more intuitive learning and reference material for modern clinical practice and education in Traditional Chinese Medicine(TCM).Methods Taking the classic TCM literature"Treatise on Febrile Diseases"as the research subject,we initially used a combination of the BERT model and LSTM-CRF model to identify medical events and their argument constituents in the ancient text.Then,an improved SpERT model was employed to identify multi-event relationships.Finally,we constructed an event evolution graph of"Treatise on Febrile Diseases"with medical events as nodes and event relationships as edges,which represents the internal connections among medical events.Results The models mentioned above achieved a precision rate of 0.768,a recall rate of 0.761,and an F1 score of 0.772 for identifying medical events and their argument constituents.Additionally,achieving a precision rate of 0.736,a recall rate of 0.682,and an F1 score of 0.687 for recognizing complex event relationships.Through the above model,the text of Treatises of Febrile Diseases was extracted,and finally the theory graph was constructed by Neo4j,which contained 3518 medical events and 5294 event relationships.Conclusion The event evolution graph organizes medical events in a cohesive manner,facilitating understanding of the relationships among diseases,patterns,treatments,prescriptions,and outcomes.Therefore,it provides a multidimensional approach for learning and guiding clinical practice in TCM.

OBJECTIVE Aiming at the current vacancy of large language models(LLMs)in TCM evaluation,a TCM benchmark dataset is designed and constructed to comprehensively and objectively evaluate the mastery and reasoning performance of LLMs in TCM knowledge,providing scientific and reliable basis for optimizing the performance of LLMs in the field of TCM.METHODS This benchmark includes 29 506 questions across 13 subjects,with data collected from standardized TCM exams and textbooks.Three gen-eral-purpose models(GPT-3.5,ChatGLM3,Baichuan)and five Chinese medical models(PULSE,BenTsao,HuatuoGPT2,Bian-Que2,ShenNong)were evaluated with answer prediction and answer reasoning tasks.The evaluation results were quantitatively as-sessed using metrics including accuracy,F1 score,BLEU,and Rouge.RESULTS For the answer prediction task,Baichuan had the highest accuracy of 36.07%in single-choice questions,while ChatGLM3 achieved the highest accuracy of 18.96%and F1 score of 76.31%in multiple-choice questions.For the answer reasoning experiment,Baichuan scored highest on BLEU-1 with 24.71,while ChatGLM3 achieved the highest Rouge-1 score of 44.64.CONCLUSION In this study,general LLMs performed slightly better than Chinese medical LLMs.Meanwhile,all models'accuracy on choice questions remained below 60%,reflecting the significant challen-ges and room for improvement that LLMs still face in the field of TCM.

OBJECTIVE Aiming at the current vacancy of large language models(LLMs)in TCM evaluation,a TCM benchmark dataset is designed and constructed to comprehensively and objectively evaluate the mastery and reasoning performance of LLMs in TCM knowledge,providing scientific and reliable basis for optimizing the performance of LLMs in the field of TCM.METHODS This benchmark includes 29 506 questions across 13 subjects,with data collected from standardized TCM exams and textbooks.Three gen-eral-purpose models(GPT-3.5,ChatGLM3,Baichuan)and five Chinese medical models(PULSE,BenTsao,HuatuoGPT2,Bian-Que2,ShenNong)were evaluated with answer prediction and answer reasoning tasks.The evaluation results were quantitatively as-sessed using metrics including accuracy,F1 score,BLEU,and Rouge.RESULTS For the answer prediction task,Baichuan had the highest accuracy of 36.07%in single-choice questions,while ChatGLM3 achieved the highest accuracy of 18.96%and F1 score of 76.31%in multiple-choice questions.For the answer reasoning experiment,Baichuan scored highest on BLEU-1 with 24.71,while ChatGLM3 achieved the highest Rouge-1 score of 44.64.CONCLUSION In this study,general LLMs performed slightly better than Chinese medical LLMs.Meanwhile,all models'accuracy on choice questions remained below 60%,reflecting the significant challen-ges and room for improvement that LLMs still face in the field of TCM.

Objective This study aims to extract medical events from the ancient Chinese medical book"Treatise on Febrile Diseases"and explore their internal connections.By constructing an event evolution graph,this study visualizes the progression of diseases related to the three Yang and three Yin,provides new ideas for the digitization of ancient Chinese medical literature,and offers more intuitive learning and reference material for modern clinical practice and education in Traditional Chinese Medicine(TCM).Methods Taking the classic TCM literature"Treatise on Febrile Diseases"as the research subject,we initially used a combination of the BERT model and LSTM-CRF model to identify medical events and their argument constituents in the ancient text.Then,an improved SpERT model was employed to identify multi-event relationships.Finally,we constructed an event evolution graph of"Treatise on Febrile Diseases"with medical events as nodes and event relationships as edges,which represents the internal connections among medical events.Results The models mentioned above achieved a precision rate of 0.768,a recall rate of 0.761,and an F1 score of 0.772 for identifying medical events and their argument constituents.Additionally,achieving a precision rate of 0.736,a recall rate of 0.682,and an F1 score of 0.687 for recognizing complex event relationships.Through the above model,the text of Treatises of Febrile Diseases was extracted,and finally the theory graph was constructed by Neo4j,which contained 3518 medical events and 5294 event relationships.Conclusion The event evolution graph organizes medical events in a cohesive manner,facilitating understanding of the relationships among diseases,patterns,treatments,prescriptions,and outcomes.Therefore,it provides a multidimensional approach for learning and guiding clinical practice in TCM.

Objective:In general, patients with seropositive rheumatoid arthritis (RA) are considered to show an aggressive disease course. However, the relationship between the two subgroups in disease severity is controversial. Our study is aimed to compare the clinical characteristics and prognosis of double-seropositive and seronegative RA in China through a real-world large scale study.Methods:RA patients who met the 1987 American College of Rheumatology (ACR) classification criteria or the 2010 ACR/European Anti-Rheumatism Alliance RA classification criteria, and who attended the 10 hospitals across the country from September 2015 to January 2020, were enrolled. According to the serological status, patients were divided into 4 subgroups [rheumatoid factor (RF)(-) anti-cyclic citrullinated peptide (CCP) antibody (-), RF(+), RF(+) anti-CCP antibody(+), anti-CCP antibody(+)] and compared the disease characteristics and treatment response. One-way analysis of variance was used for measurement data that conformed to normal distribution, Kruskal-Wallis H test was used for measurement data that did not conform to normal distribution; paired t test was used for comparison before and after treatment within the group if the data was normally distributed else paired rank sum test was used; χ2 test was used for count data. Results:① A total of 2 461 patients were included, including 1 813 RF(+) anti-CCP antibody(+) patients (73.67%), 129 RF(+) patients (5.24%), 245 RF(-) anti-CCP antibody(-) patients (9.96%), 74 anti-CCP antibody(+) patients (11.13%). ② Regardless of the CCP status, RF(+) patients had an early age of onset [RF(-) anti-CCP antibody(-) (51±14) years old, anti-CCP antibody(+) (50±15) years old, RF(+) anti-CCP antibody(+) (48±14) years old, RF(+)(48±13) years old, F=3.003, P=0.029], longer disease duration [RF(-) anti-CCP antibody(-) 50 (20, 126) months, anti-CCP antibody(+) 60(24, 150) months, RF(+) anti-CCP antibody(+) 89(35, 179) months, RF(+) 83(25, 160) months, H=22.001, P<0.01], more joint swelling counts (SJC) [RF(-) anti-CCP antibody(-) 2(0, 6), Anti-CCP antibody(+) 2(0, 5), RF(+) anti-CCP antibody(+) 2(0, 7), RF(+) 2(0, 6), H=8.939, P=0.03] and tender joint counts (TJC) [RF(-) anti-CCP antibody(-) 3(0, 8), anti-CCP antibody(+) 2(0, 6), RF(+) anti-CCP antibody(+) 3(1, 9), RF(+) 2(0, 8), H=11.341, P=0.01] and the morning stiff time was longer [RF(-) anti-CCP antibody(-) 30(0, 60) min, anti-CCP antibody(+) 20(0, 60) min, RF(+) anti-CCP antibody(+) 30(10, 60) min, RF(+) 30(10, 60) min, H=13.32, P<0.01]; ESR [RF(-) anti-CCP antibody(-) 17(9, 38) mm/1 h, anti-CCP antibody(+) 20(10, 35) mm/1 h, RF(+) anti-CCP antibody(+) 26(14, 45) mm/1 h, RF(+) 28(14, 50) mm/1 h, H=37.084, P<0.01] and CRP [RF(-) anti-CCP antibody(-) 2.3 (0.8, 15.9) mm/L, Anti-CCP antibody(+) 2.7(0.7, 12.1) mm/L, RF(+) anti-CCP antibody(+) 5.2(1.3, 17.2) mm/L, RF (+) 5.2(0.9, 16.2) mm/L, H=22.141, P<0.01] of the RF(+)patients were significantly higher than RF(-) patients, and RF(+) patients had higher disease severity(DAS28-ESR) [RF(-) anti-CCP antibody(-) (4.0±1.8), anti-CCP antibody(+) (3.8±1.6), RF(+) anti-CCP antibody(+) (4.3±1.8), RF(+) (4.1±1.7), F=7.269, P<0.01]. ③ The RF(+) anti-CCP antibody(+) patients were divided into 4 subgroups, and it was found that RF-H anti-CCP antibody-L patients had higher disease severity [RF-H anti-CCP antibody-H 4.3(2.9, 5.6), RF-L anti-CCP antibody-L 4.5(3.0, 5.7), RF-H anti-CCP antibody-L 4.9(3.1, 6.2), RF-L anti-CCP antibody-H 2.8(1.8, 3.9), H=20.374, P<0.01]. ④ After 3-month follow up, the clinical characteristics of the four groups were improved, but there was no significant difference in the improvement of the four groups, indicating that the RF and anti-CCP antibody status did not affect the remission within 3 months. Conclusion:Among RA patients, the disease activity of RA patients is closely related to RF and the RF(+) patients have more severe disease than RF(-) patients. Patients with higher RF titer also have more severe disease than that of patients with low RF titer. After 3 months of medication treatment, the antibody status does not affect the disease remission rate.