Objective:

To identify metabolic biomarkers and establish a Clinlabomics model for screening ischemic stroke (IS) with diabetes mellitus (DM) comorbidity.

Approach:

• Study Design: Retrospective enrollment of 2,587 IS patients classified into IS-DM comorbidity and IS-only groups.
• Data Collection: Collection of 16 metabolic indicators and identification of candidate indicators using univariate and multivariate logistic regression and restricted cubic spline (RCS) analysis.
• Model Construction: Dataset split into training and test sets (7:3 ratio) with an additional 406 patients for temporal validation. Clinlabomics models constructed using 11 machine learning algorithms.
• Model Evaluation: Model performance evaluated using F1-score, accuracy (ACC), and area under the curve (AUC). SHAP analysis performed to quantify feature contributions.

Key Findings:

• 12 metabolic indicators were closely associated with IS-DM comorbidity.
• TyG index (OR = 4.76) and AIP (OR = 3.60) significantly associated with increased risk of IS-DM comorbidity.
• RCS revealed non-linear associations between 9 indicators and comorbidity.
• The rpart algorithm model achieved the best performance with ACC of 0.885, F1-score of 0.818, and AUC of 0.910.
• 9 candidate metabolic biomarkers identified: TyG, UA, TC, LDL-C, AIP, AC, CRI-I, CRI-II, and HbA1c/HDL-C.

Interpretation:

The Clinlabomics model established by the rpart algorithm demonstrates strong performance in identifying the IS-DM population.

Conclusion:

This study identified 9 candidate metabolic indicators of IS-DM comorbidity.