Hemoglobin Trajectory Phenotypes and Neurological Outcomes in a Neurosurgical ICU Cohort

Background/Objectives: Current hemoglobin management in neurocritical care relies on static transfusion thresholds, which fail to capture the dynamic nature of hemoglobin changes during ICU care. We aimed to determine whether distinct longitudinal hemoglobin trajectory phenotypes exist among neurosurgical ICU patients and whether specific trajectory patterns independently predict unfavorable neurological outcomes. Methods: In this retrospective observational cohort study, we analyzed 8517 patients admitted to the neurosurgical ICU of a tertiary academic medical center between January 2015 and December 2024. A feature-based Gaussian mixture model was applied to trajectory-derived hemoglobin features over the first 14 ICU days to identify distinct hemoglobin trajectory phenotypes. The association between trajectory class and neurological outcomes was evaluated using propensity score matching. Results: Six distinct hemoglobin trajectory phenotypes were identified. The "Rapid Dropper" phenotype (Class 2; n = 351, 4.1%), characterized by the steepest decline velocity and highest variability, showed dramatically worse outcomes: 36.5% unfavorable neurological outcome (Glasgow Outcome Scale 1-3) versus 3.1% in all other classes combined (odds ratio [OR], 18.10; 95% confidence interval [CI], 14.08-23.27). This association persisted after propensity score matching (OR, 2.40; 95% CI, 1.77-3.26; p < 0.001). Hemorrhagic diagnoses were disproportionately concentrated in this high-risk phenotype. A combined prediction model incorporating trajectory-derived features within 72 h achieved an area under the receiver operating characteristic curve of 0.850 (95% CI, 0.829-0.870). This value reflects retrospective full-trajectory phenotyping; in a 72 h landmark analysis, early-feature prediction achieved an AUC of approximately 0.72. Conclusions: Hemoglobin trajectory phenotyping identified a high-risk "Rapid Dropper" subgroup that was significantly associated with worse short-term neurological outcomes. The rate of hemoglobin decline, rather than any single threshold, was associated with prognostic separation; prospective and external validation is required before these associations can inform transfusion strategy.