ObjectiveThis study aimed to assess the efficacy of ligustrazine in treating cerebral ischemia-reperfusion (I/R) injury and construct a preclinical evidence framework by meta-analysis and machine learning.MethodsA systematic search was conducted for preclinical studies published in PubMed, Embase, Web of Science, and the Cochrane Library up to June 25, 2024. The inclusion criteria encompassed preclinical animal studies pertinent to the topic. Data extraction was performed independently by two individuals, Stata 17.0 software was used for quantitative analysis, R (version 4.3.3) and Python (version 3.11.4) were used for machine learning with neurological function score as the dependent variable.ResultsA total of 23 articles were included, involving 381 animals in the meta-analysis and 321 animals in the machine learning component. Ligustrazine significantly improved neurofunctional scores (NFS) [Longa criteria, SMD = −1.59, 95%CI (−2.16, −1.01), P < 0.001; mNSS criteria, SMD = −1.67, 95%CI (−2.36, −0.97), P < 0.001], cerebral infarct volume (%) [SMD = −2.56, 95%CI (−3.03, −2.09), P < 0.001], and BBB [SMD = −3.06, 95%CI (−4.53, −1.59), P < 0.001]. Furthermore, machine learning analyses, with NFS as the dependent variable, identified the time of first dose, duration, and dose as key determinants of neurofunctional improvement with ligustrazine. Notably, model interpretation suggested that greater improvements were more likely to occur when the initial administration of ligustrazine occurred within 24 h prior to (or 2.21 h post) the ischemic event, at a dosage of 23.53–34.69 mg/kg/day (or 45.71 to 75.65 mg/kg/day), and with an administration duration exceeding 71.43 h.ConclusionThe combination of meta-analysis and machine learning in this study not only confirms that ligustrazine is effective in reducing cerebral I/R injury, but also provides a framework for elucidating the preclinical intervention variables, thus offering novel insights for optimizing preclinical strategies of ligustrazine in cerebral I/R injury.
Radically different