Higher pre-ICU platelet count coefficient of variation was associated with in-hospital, 28-day, 90-day, and 360-day mortality in intracerebral hemorrhage patients after adjustment for clinical covariates and platelet measurement count, with the association directionally reproduced in an external Chinese cohort.
Key Findings
Results
Each 0.1-unit increase in platelet count CV was independently associated with increased in-hospital mortality in ICH patients.
OR of 1.21 (95% CI, 1.07–1.38; p = 0.003) per 0.1-unit increase in CV
Primary cohort included 2,202 patients from MIMIC-IV, of whom 294 died during the index hospitalization
Models adjusted for clinical covariates and platelet measurement count
Platelet count CV was calculated from at least three nonmissing platelet measurements obtained within 15 days before ICU admission
Results
Each 0.1-unit increase in platelet count CV was independently associated with increased 28-day mortality in ICH patients.
HR of 1.14 (95% CI, 1.04–1.26; p = 0.006) per 0.1-unit increase in CV
Analysis performed using Cox proportional hazards models
Results were consistent in fixed-three-measurement and landmark sensitivity analyses
Results
Each 0.1-unit increase in platelet count CV was independently associated with increased 90-day mortality in ICH patients.
HR of 1.11 (95% CI, 1.02–1.21; p = 0.014) per 0.1-unit increase in CV
Analysis performed using Cox proportional hazards models with multivariable adjustment
Results were consistent in sensitivity analyses
Results
Each 0.1-unit increase in platelet count CV was independently associated with increased 360-day mortality in ICH patients.
HR of 1.11 (95% CI, 1.02–1.20; p = 0.016) per 0.1-unit increase in CV
The association with mortality persisted across all time horizons examined (in-hospital, 28-day, 90-day, and 360-day)
Kaplan-Meier, restricted cubic splines, and subgroup analyses were also performed
Results
Adding platelet count CV to existing models produced modest but statistically significant improvement in patient risk reclassification.
Continuous net reclassification improvement (NRI) of 0.209 (95% CI, 0.079–0.372)
Integrated discrimination improvement (IDI) of 0.007 (95% CI, 0.001–0.019)
Reclassification metrics indicate a modest incremental prognostic value beyond established clinical covariates
Results
The association between platelet count CV and 28-day mortality was reproduced in an independent external Chinese validation cohort.
HR per 0.1-unit increase in CV of 1.141 (95% CI, 1.007–1.292; p = 0.042) after multivariable adjustment
The external cohort was an independent Chinese hospital cohort used for external validation
Authors noted that 'the effect estimate was modest' in the external cohort
The association was described as 'directionally reproduced' in the external cohort
Methods
The study used a retrospective multicenter cohort design with MIMIC-IV as the primary database and an independent Chinese cohort for external validation.
CV was calculated as the sample standard deviation divided by the mean of at least three observed, nonmissing platelet measurements
Measurements were obtained during the index hospitalization within 15 days before ICU admission
Missing measurement fields were excluded and were not treated as zero
Primary multivariable models additionally adjusted for platelet measurement count to control for monitoring frequency
Logistic regression, Cox models, restricted cubic splines, Kaplan-Meier analyses, subgroup analyses, reclassification metrics, and sensitivity analyses were performed
Conclusions
The authors concluded that prospective studies with standardized platelet-sampling schedules are needed before clinical implementation of platelet count CV as a prognostic tool.
Authors acknowledged that the effect estimate was modest in the external cohort
The call for prospective studies reflects concerns about the retrospective design and variable platelet-sampling schedules across institutions
Standardized sampling schedules were specifically identified as a prerequisite for clinical translation
What This Means
This research suggests that fluctuations in platelet counts — measured using a statistic called the coefficient of variation (CV), which captures how much platelet levels vary relative to their average — can predict the risk of death in patients hospitalized with intracerebral hemorrhage (bleeding in the brain). The study analyzed data from over 2,200 patients in the MIMIC-IV database, a large U.S. hospital records database, and found that patients whose platelet counts varied more before entering the ICU had significantly higher odds of dying in the hospital and higher risks of death at 28 days, 90 days, and one year — even after accounting for other clinical factors and how often platelets were measured.
Importantly, the finding was also tested in a separate group of patients from China, where a similar relationship between platelet count variability and 28-day mortality was observed, lending additional support to the discovery. Adding platelet CV to standard prediction models modestly improved the ability to correctly classify patients by risk level, suggesting it could add incremental value to existing prognostic tools.
This research suggests that tracking how much a patient's platelet count fluctuates — not just its absolute level — may provide useful information about their prognosis after a brain bleed. However, the authors caution that the study was retrospective (looking back at existing records) and that platelet testing was not done on a standardized schedule across patients, which could influence results. They recommend future prospective studies with consistent blood-testing protocols before this measure is used in clinical practice to guide patient care decisions.
Ge T, Qiu M, Shi T, Qiu D, Wang H, Chen P, et al.. (2026). Coefficient of variation of platelet count as a prognostic Indicator for intracerebral hemorrhage: findings from a multicenter cohort.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1868807