Early hemoglobin trajectories predict mortality and neurological outcome after aneurysmal subarachnoid hemorrhage: a group-based trajectory modeling and machine learning study

Objective This study aimed to identify early Hb trajectories in aneurysmal subarachnoid hemorrhage (aSAH) and evaluate Hb as a dynamic biomarker for prognostic assessment. Methods We included 665 patients with aSAH. Group-based trajectory modeling classified Hb patterns during the first 7 days of intensive care unit admission. Associations with 30-day and 180-day mortality and unfavorable neurological outcomes (modified Rankin Scale score ≥ 3) were analyzed. Predictive performance was compared with static Hb measurements, and machine-learning models were developed to identify the poorest-prognosis trajectory. Results Three trajectories were identified: Class 1 ( n = 357, 53.68 %), slow decline from a lower baseline; Class 2 ( n = 214, 32.18 %), rapid decline from an intermediate baseline; and Class 3 ( n = 94, 14.14 %), slow decline from an intermediate baseline. Compared with Class 2, Class 1 and Class 3 had lower 30-day mortality ( HR 0.28, 95 % CI 0.19–0.41; HR 0.42, 95 % CI 0.24–0.75) and unfavorable neurological outcomes ( OR 0.46, 95 % CI 0.29–0.71; OR 0.13, 95 % CI 0.06–0.25). The trajectory model outperformed Day 1 Hb and Hb tertiles ( AUC 0.759 vs 0.721 and 0.720; DeLong P < 0.05). XGBoost had the highest AUC (0.832) among the machine-learning models. Conclusion Longitudinal Hb assessment may support dynamic prognostic assessment and trajectory-based risk stratification, highlighting Hb as a longitudinal biomarker in aSAH.