Prognostic Value of Admission D-dimer Levels and Total Bleeding Volume in Aneurysmal Subarachnoid Hemorrhage: A Retrospective Cohort Study with Machine Learning-Based Modeling.

BACKGROUND: Plasma D-dimer levels are independently associated with poor prognosis following aneurysmal subarachnoid hemorrhage (aSAH). However, the underlying mechanisms contributing to early D-dimer elevation remain unclear. This study aimed to evaluate the association between admission D-dimer levels and total bleeding volume (TBV) and to further explore their combined predictive power for functional outcomes using interpretable machine learning approaches. METHODS: We analyzed data from 473 patients with aSAH enrolled in the retrospective PROSAH-MPC cohort, including clinical, radiological, and laboratory parameters. Patients were stratified by D-dimer quartiles. Multivariable logistic regression was employed to assess associations between D-dimer, TBV, and 12-month modified Rankin scale (mRS) outcomes. Boruta algorithm was applied for feature selection, identifying D-dimer, TBV, age, Hunt-Hess grade, and modified Fisher score (mFS) as top predictors. Subsequently, seven machine learning models were developed for outcome prediction. SHapley Additive exPlanations (SHAP) analysis was used to interpret the contribution of these features. RESULTS: Unfavorable outcomes occurred in 125 patients (26.4%). Elevated D-dimer levels were significantly associated with unfavorable outcomes (adjusted OR 1.08; 95% CI 1.02-1.16). A significant interaction between TBV and D-dimer was observed in relation to functional outcome (P for interaction = 0.032). Combined D-dimer + TBV + Hunt-Hess yielded the highest area under the curve (AUC; 0.867; 95% CI 0.764-0.856), outperforming D-dimer (0.735), TBV (0.783), and Hunt-Hess (0.833). Among machine learning models, XGBoost achieved the highest discriminative performance (AUC = 0.904), with SHAP confirming D-dimer and TBV as major contributors. CONCLUSIONS: Admission D-dimer levels are a key biomarker associated with long-term functional outcomes after aSAH, and their prognostic value is significantly influenced by total bleeding volume. Integrating machine learning and SHAP interpretability enhances our understanding of these relationships and may inform early risk stratification in clinical practice.