Machine learning prediction of thrombolysis efficacy using hs-CRP and inflammatory markers in stroke.
The aim of this study was to investigate the relationship between serum ultrasensitive C-reactive protein (hs-CRP) levels and stroke incidence and to assess its potential role in decision-making for thrombolytic therapy in stroke. Given that hs-CRP is a well-recognized marker of inflammation, its association with cardiovascular disease makes this area of research significant. Using NHANES data (3144 participants, including 239 stroke cases), we analyzed associations between inflammatory markers and stroke via bivariate correlation, restricted cubic spline regression, and neural network modeling. Model performance was evaluated by receiver operating characteristic curves and confusion matrices. Stroke patients exhibited significantly higher hs-CRP (3.1 mg/L) and ferritin (133 μg/L) but lower granulocyte-to-lymphocyte ratio (GLR) (P <.05). Hs-CRP >2 mg/L markedly increased stroke risk (HR >1.0, P <.05). Correlation analysis confirmed hs-CRP (R = 0.050) and ferritin (R = 0.052) as positive risk factors, while GLR was protective (r=-0.054). The neural network model prioritized ferritin (importance = 0.257) and hs-CRP (0.246), achieving 88.7% accuracy. Additional risk factors included depression (OR = 2.12), smoking, and advanced age. Elevated hs-CRP (≥2 mg/L) and ferritin are robust predictors of stroke risk and may indicate adverse outcomes post-thrombolysis. Combined with GLR, these markers support personalized thrombolytic therapy decisions. Further validation in prospective cohorts is warranted.