Temporal changes in hemoglobin-to-red blood cell distribution width ratio and its association with in-hospital all-cause mortality in patients with intracerebral hemorrhage: a retrospective analysis across MIMIC-IV and eICU databases

BACKGROUND: Intracerebral hemorrhage (ICH) is associated with poor prognosis, characterized by high acute mortality and long-term disability. Real-time risk stratification using dynamic biomarkers is urgently needed to improve patient outcomes. This study aimed to explore the association between temporal changes in the hemoglobin-to-red blood cell distribution width ratio (HRR) and in-hospital all-cause mortality in patients with ICH. METHODS: We retrospectively analyzed 2,447 ICH patients from MIMIC-IV and externally validated findings in 3,142 patients from eICU. Cox regression analyzed the association between HRR (baseline and daily dynamic values) and in-hospital mortality. Restricted cubic spline (RCS) models assessed the non-linear relationship between baseline HRR and in-hospital mortality. An optimal baseline HRR cutoff for risk stratification was identified using the maximally selected log-rank statistic and subsequently externally validated. Pearson correlation analysis was applied to evaluate associations between daily HRR and in-hospital mortality, and receiver operating characteristic (ROC) curves were used to examine the time-specific predictive performance of serial HRR values. RESULTS: HRR exhibited a consistent inverse association with in-hospital all-cause mortality in both cohorts. In fully adjusted Cox models, continuous baseline HRR was associated with lower mortality risk (HR approximately 0.51 in both cohorts), and the highest HRR quartile was associated with a reduced mortality risk compared with the lowest quartile. Both survivors and non-survivors showed progressive HRR declines during 14-day hospitalization. Non-survivors had more pronounced declines (daily reduction approximately 0.017 vs. 0.014 in MIMIC-IV; P for trend difference < 0.001) and consistently lower HRR levels at all time points. Daily HRR was inversely correlated with mortality throughout hospitalization (adjusted HRs < 1.0, P < 0.05), with the strength of this inverse association increasing notably from Day 1 (r = -0.141) to Day 4 (r = -0.225). HRR demonstrated predictive performance comparable to Hb and RDW across time points. In the MIMIC-IV test set, all indicators presented a temporal upward trend within the initial few days, with AUCs reaching 0.631 (HRR), 0.594 (Hb) and 0.630 (RDW) at Day 1, and 0.663, 0.632 and 0.645 at Day 3, respectively. Calibration curves revealed poor predictive performance with overestimation of mortality risk. A baseline HRR cutoff of ≤ 0.74 identified high-risk patients, who had significantly higher mortality (25.14% vs. 12.93% in MIMIC-IV; 17.64% vs. 12.72% in eICU, all P < 0.001). CONCLUSIONS: This study preliminarily suggests that temporal changes in HRR are associated with in-hospital all-cause mortality in ICH patients. HRR may serve as a complementary dynamic prognostic marker with predictive performance comparable to Hb and RDW, though its utility is limited by poor calibration. Further prospective validation in dedicated ICH cohorts with severity and imaging data is warranted, and future studies should consider underlying diseases and inflammatory status.