This exploratory retrospective cohort study analyzed adult intensive care unit (ICU) patients discharged alive using the publicly available MIMIC‑IV database (version 3.1). The analysis included patients with available pre-hospital discharge hemoglobin measurements. The primary analytic sample comprised 20,233 ICU survivors; the reported median age was 67 years. Data cover ICU admissions at Beth Israel Deaconess Medical Center spanning the database timeframe cited by the source.
The study evaluated pre-discharge red blood cell (RBC) laboratory indices: hemoglobin, red cell distribution width (RDW), mean corpuscular volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), and RBC count. Anemia was classified using the World Health Organization (WHO) definition; exact hemoglobin thresholds used to define anemia were those referenced by WHO as reported in the source. All RBC indices used were the last available values prior to hospital discharge as documented in the data source.
Investigators first examined hemoglobin and anemia in relation to one-year mortality. They then standardized the RBC indices and applied Gaussian mixture modeling to identify patient clusters of correlated hematologic traits. Clustering and model development were performed in a derivation cohort (70% of patients) with validation in a held-out cohort (30%). Associations between anemia, hemoglobin quartiles, cluster membership, and one-year mortality were assessed using multivariable-adjusted Cox proportional hazards models. Survival differences were tested using log-rank statistics.
Using the WHO definition, 17,247 of 20,233 ICU survivors (85.2%) were classified as anemic at hospital discharge. This high prevalence indicates that post-critical illness anemia was common in the study sample when measured immediately prior to hospital discharge.
Overall one-year mortality for the cohort was reported as 20%. Mortality varied across hemoglobin quartiles, with the highest mortality in the lowest hemoglobin quartile (28.8%) and the lowest mortality in the highest quartile (12.5%); the difference across quartiles met statistical significance (log-rank P < 0.005). In adjusted Cox models, the presence of anemia at discharge was associated with increased one-year mortality (adjusted hazard ratio 1.42, P < 0.001). The authors also report a graded relationship: lower hemoglobin values were associated with progressively higher adjusted risk of death at one year.
Gaussian mixture modeling of the standardized RBC indices identified six distinct hematologic clusters. Cluster discovery was performed in the 70% derivation sample and the cluster structure was evaluated in the 30% validation sample (held-out cohort). The six clusters reflected differing combinations of the RBC indices, integrating information from RDW, MCV, hemoglobin, MCH, MCHC, and RBC count.
One-year mortality across the six clusters ranged from 10.1% to 31.4% (log-rank P < 0.005). The highest-risk cluster was characterized primarily by elevated RDW and MCV and showed a twofold higher adjusted one-year mortality risk compared with the lowest-risk cluster (adjusted HR 2.14; 95% CI, 1.81–2.53). The source reports this as a robust signal across derivation and validation samples, though full cluster composition details beyond the prominence of RDW and MCV are provided in the source materials.
The authors conclude that both anemia and hematologic clusters derived from routine RBC indices measured at hospital discharge are associated with one-year mortality after critical illness. They frame these results as hypothesis-generating and suggest that standard RBC indices may serve as candidate biomarkers for post-discharge risk stratification. The authors further propose that the biological processes reflected by these indices—such as inflammation, nutritional status, and hematopoietic function—warrant further study as potential therapeutic targets to improve long-term outcomes.
The study used the de-identified MIMIC‑IV research database; the authors state appropriate ethical approvals and oversight consistent with use of that dataset. All data produced in the study are available on request from the authors, and the underlying MIMIC‑IV data source is publicly accessible. The source frames the analysis as exploratory and hypothesis-generating. Specific limitations beyond the exploratory, retrospective design (for example, residual confounding, missingness patterns, or unmeasured covariates) are not detailed in the abstract; further limitations and sensitivity analyses, if present, are reported in the full preprint supplementary material linked by the source.
Conclusion
In this large retrospective cohort of ICU survivors, discharge anemia and distinct hematologic phenotype clusters—especially a cluster marked by elevated RDW and MCV—were independently associated with higher one-year mortality. The findings support additional research to validate these associations, to assess potential mechanisms linking RBC indices to post-discharge outcomes, and to evaluate whether interventions targeting the underlying processes could modify long-term risk.