ORIGINAL RESEARCH article

Front. Pharmacol.

Sec. Ethnopharmacology

Association of Adjunctive Traditional Chinese Medicine Injections with 90-Day Readmission and Mortality After Ischemic Stroke: A Real-World Evidence Study Incorporating Machine Learning

  • 1. Shanxi University of Chinese Medicine, Taiyuan, China

  • 2. Shanxi Digital Government Construction and Operation Co., Ltd., Taiyuan, China

The final, formatted version of the article will be published soon.

Abstract

Background: Readmission and mortality remain important contributors to the long-term disease burden of ischemic stroke. Traditional Chinese Medicine Injections (TCMIs) are widely used as adjunctive therapies in China; however, robust real-world evidence regarding their association with clinical outcomes and healthcare utilization remains limited. Methods: We conducted a retrospective cohort study using a municipal medical insurance claims database from Shanxi Province, China. Adult patients hospitalized for ischemic stroke (ICD-10: I63) from January 2023 to September 2024 were classified into TCMI users and non-users according to exposure during the index hospitalization. Propensity score matching (1:1) was performed to reduce baseline imbalances between groups. Cox proportional hazards models and Fine–Gray competing risk models were used to evaluate 90-day clinical outcomes, accounting for competing mortality where appropriate. Generalized linear models with appropriate distribution and link functions were applied for healthcare cost analysis. Machine-learning models were developed to identify important predictors of 90-day outcomes. Results: After 1:1 propensity score matching, 2,308 patient pairs were included in the analysis. Adjunctive TCMI use was associated with a reduced risk of the 90-day composite outcome of stroke-specific readmission or all-cause mortality (HR=0.69, 95% CI: 0.55–0.88) and stroke-specific readmission under competing-risk analysis (subdistribution HR [sHR]=0.75, 95% CI: 0.57–0.98). TCMI use was also associated with reduced index hospitalization costs (adjusted cost ratio=0.87, 95% CI: 0.84–0.91). After Holm correction, none of the TCMI categories retained statistical significance. Salvia miltiorrhiza-based injections showed nominal statistical significance before correction, whereas Ginkgo biloba-based injections were not significant either before or after correction. Among machine-learning models, LightGBM demonstrated the highest discriminative performance (AUC=0.804), with hospitalization duration, medication burden, age, and TCMI exposure identified as important predictive features. Conclusions: In this real-world study, adjunctive TCMI use was associated with a reduced risk of the 90-day composite outcome and lower index hospitalization costs after propensity score matching. The association with stroke-specific readmission should be interpreted cautiously, as it was not statistically significant in the unmatched competing-risk analysis. Outcome prediction involved multiple clinical and healthcare-related factors. These findings should be interpreted as observational and hypothesis-generating and warrant confirmation in prospective studies.

Summary

Keywords

ischemic stroke, machine learning, Mortality, Propensity score matching, Readmission, Real-world evidence, Traditional Chinese medicine injections

Received

09 March 2026

Accepted

01 October 2026

Copyright

© 2026 Zhi, Cui, Xue, YUAN, Zhu, Cheng, Meng and Yan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Juanjuan Yan

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Share article

Article metrics