Immature Platelet Fraction as an Independent Marker of Adverse Outcomes in an Intensive Cardiovascular Care Unit

Platelet activation and thrombopoiesis are central to acute cardiovascular disease pathophysiology, yet the prognostic role of the immature platelet fraction (IPF) remains incompletely defined in heterogeneous intensive cardiovascular care unit (ICCU) populations. We conducted a single-center, observational study of 6,725 consecutive ICCU admissions with IPF measured. The primary endpoint was in-hospital all-cause mortality using multivariable logistic regression, and discrimination by receiver operating characteristic analysis. Median age was 70 years; 31.8% were women. Increasing IPF quartiles showed a stepwise rise in mortality (2.1 to 3.7%; p-trend = 0.003), acute renal failure (ARF) (2.4 to 5.1%; p < 0.001), malignant arrhythmia (0.9 to 2.2%; p = 0.002), and in-hospital cardiogenic shock (2.7 to 4.7%; p = 0.008); the regression analysis for cardiogenic shock was restricted to incident cases among non-shock admissions (n = 5,778). After multivariable adjustment, each 1 natural-log-unit increment in IPF (∼1.7× raw IPF) remained independently associated with mortality (OR 1.55, 95% CI: 1.10–2.19), ARF (OR 1.75, 1.37–2.25), malignant arrhythmia (OR 2.07, 1.44–2.98), and cardiogenic shock (OR 1.51, 1.16–1.96). Discrimination of IPF alone was modest (AUC 0.57–0.59), but incorporation into multivariable models improved performance (mortality AUC 0.78). Reclassification analysis demonstrated incremental value beyond baseline (categorical net reclassification improvement [NRI] +6.9 to +12.1%; continuous NRI 0.14–0.28). In this large, contemporary ICCU cohort, elevated IPF was independently associated with in-hospital mortality, as well as adverse outcomes. Although not a standalone discriminator, IPF is a readily available, low-cost biomarker that may enhance early mortality risk stratification when integrated with established clinical variables, supporting further evaluation in multimodal and machine learning–based prediction frameworks.