Heping Wang

Herbal prescription recommendation (HPR) in traditional Chinese medicine (TCM) is a challenging clinical decision-support task that requires fusing heterogeneous patient information, domain knowledge, and institution-specific clinical constraints. Although recent methods leveraging deep learning and large language models (LLMs) have advanced this task, their limited use of these complementary sources may constrain prescription accuracy and weaken clinical safety assurance. To address these limitations, we propose ReaRx, a novel deep learning model for personalized and safe prescription recommendation. ReaRx comprises a prescription reasoning module and a prescription refinement module, which together emulate the clinical workflow of TCM physicians. The prescription reasoning module fuses heterogeneous clinical information to enhance patient representation and generate foundational prescriptions. For the prescription refinement module, we introduce an expert-guided herbal prescription refinement framework, named EGHPR, which incorporates institution-specific guidelines into a multi-step decision-making process within a policy-based reinforcement learning framework to iteratively refine the generated prescriptions. Experimental results under five-fold cross-validation on the TCM-GCD dataset demonstrate that ReaRx achieves state-of-the-art performance. Ablation studies, parameter analysis, and case studies further validate the effectiveness and practicality of the proposed modules. EGHPR can also be seamlessly integrated into existing LLM-based prescription recommendation models, improving prescription accuracy without retraining or updating the underlying models. Overall, ReaRx closely aligns with the clinical prescription process of TCM physicians, while EGHPR further enhances the accuracy, clinical relevance, and safety of generated prescriptions, demonstrating practical value for personalized and safe prescription recommendation.

Backgrounds: Colorectal cancer (CRC) is one of the common malignant tumors, with a gradually increasing incidence. Due to late detection and poor sensitivity to chemotherapy, it has become a difficult problem in tumor prevention and treatment at present. Exploring or discovering new combinations is a significant strategy for the treatment of CRC. Compound kushen injection (CKI) is a traditional Chinese medicine injection extracted from Sophora flavescens Ait. and Smilax glabra Roxb., which is widely used in the comprehensive treatment of CRC in China. This systematic review is aimed to ascertain the clinical efficacy and safety of CKI combined with chemotherapy in the treatment of advanced CRC based on available data. On this basis, the specific application of CKI in combination with chemotherapy in clinical practice is further discussed. Methods: PubMed, Web of Science, the Cochrane Library, EMBASE, China National Knowledge Infrastructure, Chinese Scientific Journals Database, Wanfang Database, Chinese Biomedicine Database Searches, the Chinese Clinical Trial Registry, and ClinicalTrials.gov were searched systematically, from inception to April 20, 2024. We adopted the ROB2 tool to assess quality of the included trials, Stata 16 for data analysis, and evaluated the publication bias with the funnel plot and Egger’s test. The quality of the evidence was justified according to GRADE. We also used trial sequential analysis (TSA) to calculate the final required sample size in this meta-analysis and to verify whether the results present a reliable conclusion. The protocol for this systematic review was registered on PROSPERO (CRD42022380106) and has been published. Results: Sixteen trials that examined 1378 patients were included in this study. Meta-analysis revealed that compared with chemotherapy, objective response rate (ORR, RR = 1.30, 95% CI: 1.18-1.44), disease control rate (DCR, RR = 1.08, 95% CI: 1.03-1.13), and KPS score improvement rate were improved (RR = 1.18, 95% CI: 1.07-1.31) by the combination of CKI and chemotherapy in patients with advanced CRC. Additionally, CKI combined with chemotherapy was associated with lower adverse reactions such as leukopenia (RR = 0.74, 95% CI: 0.62-0.87), thrombocytopenia (RR = 0.68, 95% CI: 0.49-0.94), gastrointestinal reactions (RR = 0.72, 95% CI: 0.55-0.94), and liver damage (RR = 0.48, 95% CI: 0.30-0.79), higher CD4⁺ ratio (MD = 9.70, 95% CI:8.73-10.68) and CD4⁺/CD8⁺ ratio (MD = 0.25, 95% CI: 0.22-0.28), and lower CD8⁺ T cell ratio (MD = −5.25, 95% CI: −5.94 to −4.56). Subgroup analysis demonstrated that ORR and DCR in patients with advanced CRC were improved when CKI combined with FOLFOX and 5Fu + L-OHP. Both 15 and 20 ml/day of CKI combined with FOLFOX provided a significant effect in ORR. Moreover, ORR was improved when the accumulated CKI dose reached 280 ml per course and 420 ml in total. 7 days/course as well as 14 days/course of CKI combined with FOLFOX were effective durations in ORR. As for DCR, 7 days/course of CKI combined with FOLFOX could improve efficacy. Furthermore, CKI + FOLFOX may be useful in ORR and DCR for at least 4 cycles of combination therapies. The TSA showed that firm results in ORR and DCR were established and additional trials were unlikely to change the results. Conclusion: CKI combined with chemotherapy provides a statistically significant and clinically important effect in the improvement of ORR, DCR, performance status, ADR reduction, and immune function in patients with CRC. However, more rigorously designed, large-scale, and multi-center RCTs are needed in the future.

Introduction. Brain metastases (BMs) are common in non‐small‐cell lung cancer (NSCLC), which leads to a poor prognosis. As the two most effective strategies available, the use of combination of epidermal growth factor receptor tyrosine kinase inhibitors (EGFR‐TKIs) and radiotherapy (RT) is still controversial. This protocol proposes a methodology for carrying out a systematic review and meta‐analysis that is aimed at (1) focusing on the efficacy and safety role of EGFR‐TKIs combined with RT for BMs from NSCLC and (2) displaying the difference in efficacy of EGFR‐TKIs owing to the sites and number of BMs, different types of RT, EGFR mutation status, and the subtypes of EGFR mutations by subgroup analysis. Methods and Analysis. Electronic databases including PubMed, Embase, CENTRAL, Web of Science, CBM, CNKI, Wanfang database, and VIP database will be searched from their inception until May 2022. Only randomized controlled trials evaluating the clinical efficacy and safety of EGFR‐TKIs combined with RT on BMs of NSCLC will be included. Two reviewers will select the articles, assess the risk of bias, and extract data independently and in duplicate. The RoB 2 tool will be used to assess the quality of included studies. The meta‐analysis of data synthesis will be performed with Stata 16. Publication bias will be assessed with the funnel plot method and the Egger test. Quality of the evidence will be evaluated by the GRADE system. Discussion. The approval of an ethical committee is not required. All the included trials will comply with the current ethical standards and the Declaration of Helsinki. Given the ongoing controversies regarding the optimal sequencing of the available and expanding treatment options for EGFR‐TKIs in NSCLC with BMs, a synthesis of available, high‐quality clinical research evidence is essential to advance our understanding in the treatment of this complex and common disease. This systematic review will evaluate available evidence, will try to provide optimized advice in the applications of EGFR‐TKIs, and will be published in a high‐quality journal. This study is registered with PROSPERO registration number CRD42021291509.

Background. Kanglaite injection (KLTi) has shown good clinical efficacy in the treatment of pancreatic ductal adenocarcinoma (PDAC). While previous studies have demonstrated the antitumor effects of the oil compounds in KLTi, it is unclear whether the unsaponifiable matter (USM) also has antitumor effects. This study used network pharmacology, molecular docking, and database verification methods to investigate the molecular biological mechanisms of USM. Methods. Compounds of USM were obtained from GC‐MS, and targets from DrugBank. Next, the GEO database was searched for differentially expressed genes in cancerous tissues and healthy tissues of PDAC to identify targets. Subsequently, the protein‐protein interaction of USM and PDAC targets was constructed by BisoGenet to extract candidate genes. The candidate genes were enriched using GO and KEGG by Metascape, and the gene‐pathway network was constructed to screen the key genes. Molecular docking and molecular dynamic simulations of core compound targets were finally performed and to explore the diagnostic, survival, and prognosis value of targets. Results. A total of 10 active compounds and 36 drug targets were screened for USM, 919 genes associated with PDAC, and 139 USM candidate genes against PDAC were excavated. The enrichment predicted USM by acting on RELA, NFKB1, IKBKG, JUN, MAPK1, TP53, and AKT1. Molecular docking and dynamic simulations confirmed the screened core targets had good affinity and stability with the corresponding compounds. In diagnostic ROC validation, the above targets have certain accuracy for diagnosing PDAC, and the combined diagnosis is more advantageous. As the most diagnostic value of RELA, it is equally significant in predicting disease‐specific survival and progression‐free interval. Conclusions. USM in KLTi plays an anti‐PDAC role by intervening in the cell cycle, inducing apoptosis, and downregulating the NF‐κB, MAPK, and PI3K‐Akt pathways. It might participate in the pancreatic cancer pathway, and core target groups have diagnostic, survival, and prognosis value biomarker significance.