Mortality Risk in Cerebral Infarction Tied to Three Routine ICU Variables
A new cohort study built from the MIMIC-IV database points to a simple bedside message for clinicians caring for patients with cerebral infarction in the intensive care unit: mortality risk tracked closely with the anion gap, platelet count, and glucose concentration. In a sample of 109 eligible patients admitted between 2008 and 2019, those three variables emerged as prognostic markers after sequential univariate screening, least absolute shrinkage and selection operator regression, and multivariate Cox modeling.
The findings are clinically familiar in a way that matters. Metabolic acidosis, thrombocytopenia, and hyperglycemia often travel with severe neurologic illness, systemic stress, and poor physiologic reserve. Here, they were not treated as background noise; they were carried into a risk model that separated patients into high- and low-risk groups with visibly different survival probabilities.
A Small ICU Cohort, But a Clear Signal
The study population came exclusively from intensive care admissions, which makes the question highly relevant to acute stroke management. Patients with cerebral infarction who arrive in the ICU are not the typical ward stroke population; they often carry more organ dysfunction, more hemodynamic instability, and a narrower margin for recovery. A risk tool built in that setting has to compete with rapid bedside decision-making, where clinicians need markers that are already available and immediately interpretable.
The model’s appeal rests on its ingredients. Anion gap is a blunt but useful marker of unmeasured acids and disturbed metabolic balance. In neurologically injured patients, rising anion gap may reflect tissue hypoperfusion, lactic acidosis, renal dysfunction, or the physiologic fallout of critical illness. Each of those pathways can worsen cerebral injury indirectly by compromising systemic stability and oxygen delivery.
Platelet count moved in the opposite direction, with lower counts associated with greater mortality risk. Platelets are not merely clotting cells in this context. They sit at the intersection of thrombosis, inflammation, endothelial injury, and microvascular flow. In a patient already suffering an ischemic brain event, a depleted platelet reserve may mark more severe illness or impaired hemostatic balance, both of which can complicate recovery.
One sentence captures the bedside consequence: deranged metabolism, impaired hemostasis, and stress hyperglycemia clustered with worse survival.
Why Glucose and Acid-Base Status Matter After Brain Infarction
Glucose concentration also carried prognostic weight. Hyperglycemia is a frequent companion of acute stroke, and its presence may reflect the catecholamine surge of critical illness, pre-existing dysglycemia, or both. Excess glucose can aggravate ischemic injury by promoting oxidative stress, worsening lactate accumulation, and increasing blood-brain barrier disruption. In that setting, glucose is not just a number on the chemistry panel; it is a marker of a metabolic environment that may amplify neuronal damage.
The study identified anion gap with a hazard ratio of 1.1040, platelet count with a hazard ratio of 0.9965, and glucose concentration with a hazard ratio of 1.0050. Those effect sizes should be read in the context of a multivariable survival model rather than as isolated bedside thresholds. Even so, the direction of effect is biologically coherent and consistent with the kind of physiologic distress that drives poor outcomes in critically ill stroke patients.
Model Performance and the Limits of a 109-Patient Dataset
After the prognostic variables were selected, the investigators constructed a risk model that classified patients into high- and low-risk cohorts. High-risk patients had lower survival probabilities, and receiver operating characteristic analysis showed areas under the curve above 0.7 for the model’s performance. In practical terms, that suggests the model discriminated better than chance and may have some value as a prognostic support tool.
Still, the dataset was small. Just 109 patients remained after exclusion criteria were applied, and that leaves the model vulnerable to overfitting even when machine-learning selection methods are used. Internal performance can look encouraging while external performance falls apart when the same model is tested in a different hospital system, a different ICU case mix, or a broader stroke population. The study does not resolve that gap.
Even with those constraints, the work has an advantage that many prognostic models lack: its variables are ordinary. Anion gap, platelet count, and glucose are already measured in routine critical care. No specialized biomarker platform, imaging postprocessing suite, or rare molecular assay is required. If the model holds up elsewhere, it could be folded into early triage or mortality counseling without adding procedural burden.
Bedside Use Still Depends on Validation
For neurologists, intensivists, and stroke teams, the practical implication is straightforward. A patient with cerebral infarction who presents with elevated anion gap, reduced platelet count, and elevated glucose should not be viewed as carrying risk from stroke severity alone; the metabolic and hematologic context matters. Those variables may identify a subgroup with higher short-term vulnerability, where closer monitoring, tighter physiologic management, and earlier escalation of supportive care could be justified.
The study was supported by the Chengdu Science and Technology Program under grant agreement 2024-YF05-01785-SN, and the authors declared no competing interests. Ethics approval was listed as not applicable in this database-based analysis.
What remains unresolved is external validation. A broader multicenter cohort, preferably with prospective assessment and calibration testing, is needed before this model can be treated as more than a promising retrospective signal. Without that step, the predictors are best understood as plausible markers of risk rather than a finished clinical decision tool.