Correlation analysis and predictive model construction of metabolic syndrome, complete blood count-derived inflammatory markers, and overall burden of cerebral small vessel disease

Abstract

Background: The high burden of cerebral small vessel disease (CSVD) on neuroimaging is a significant risk factor for stroke, cognitive dysfunction, and emotional disorders. Currently, there is a lack of studies investigating the correlation between metabolic syndrome (MetS), complete blood count-derived inflammatory markers, and total CSVD burden. This study aims to evaluate the total CSVD imaging load using machine learning (ML) algorithms and to explore further the relationship between MetS, complete blood count-derived inflammatory markers, and CSVD load. Methods: We included CSVD patients from Xijing Hospital (2012–2022). Univariate and lasso regression analyses identified variables linked to CSVD neuroimaging burden. Six ML models predicted CSVD burden based on MetS and inflammatory markers. Model performance was evaluated using ROCauc, PRauc, DCA, and calibration curves. The SHAP method validated model interpretability. The best-performing model was selected to develop a web-based calculator using the Shiny package. Results: The Logistic regression model outperformed others in predicting CSVD burden. The model incorporated MetS, neutrophil-to-lymphocyte ratio (NLR), homocysteine (Hcy), age, smoking status, cystatin C (CysC), uric acid (UA), and prognostic nutritional index (PNI). Conclusion: MetS, NLR, Hcy and CSVD high load were positively correlated, and the Logistic regression model could accurately predict the total CSVD load degree.

Keywords

• Cerebral small vessel disease
• Neuroimaging burden
• Metabolic syndrome
• Complete blood count-derived inflammatory markers
• Machine learning