Identification of key metabolic indicators associated with the comorbidity of ischemic stroke and diabetes mellitus using an optimal interpretable clinlabomics model.

Background: To identify candidate metabolic biomarkers and establish a Clinlabomics model for early precise screening of ischemic stroke (IS) with diabetes mellitus (DM) comorbidity population. Methods: A total of 2,587 IS patients were retrospectively enrolled and classified into IS-DM comorbidity and IS-only groups. A total of 16 metabolic indicators were collected, and candidate indicators were identified using univariate and multivariate logistic regression along with restricted cubic spline (RCS) analysis. The dataset was randomly split into training and test sets at a 7:3 ratio, and an additional 406 patients constituted a temporal validation set. Clinlabomics models were constructed using 11 machine learning (ML) algorithms. Model performance was evaluated using F1-score, accuracy (ACC) and area under the curve (AUC) to select the optimal algorithm, and SHapley Additive exPlanations (SHAP) analysis was performed to quantify feature contributions. Results: Multivariate logistic regression showed that 12 metabolic indicators were closely associated with IS-DM comorbidity. Notably, the triglyceride glucose (TyG) index (OR = 4.76, 95% CI: 4.01-5.64, Conclusions: This study successfully identified 9 candidate metabolic indicators of IS-DM comorbidity. The Clinlabomics model established by rpart algorithm presents excellent performance in identifying the IS-DM population.