Plasma glutamic acid predicts myelosuppression and mortality in septic patients using machine learning

Myelosuppression is a common secondary manifestation of sepsis and is associated with increased morbidity and mortality. Recent evidence suggests that amino acid metabolism, particularly that of glutamic acid, may influence hematopoietic function and inflammatory responses¹. Using a machine learning technique, we aimed to evaluate whether plasma amino acid levels at hospital admission are associated with myelosuppression and 30-day mortality in septic patients, with a focus on glutamic acid as a potential early biomarker. We conducted a prospective observational study on 390 adult patients hospitalized with sepsis at the University Hospital of Trieste. Patients were stratified into two groups on the basis of peripheral blood counts: myelosuppressed and non-myelosuppressed. Plasma levels of 13 amino acids were quantified at admission via high-performance liquid chromatography. Inflammatory markers, clinical outcomes, and 30-day mortality were recorded. Predictive models for myelosuppression were developed via Random Forest and interpreted via SHapley Additive exPlanations (SHAP) analysis. While nutritional intake and longitudinal amino acid trends were not available, we accounted for major comorbidities in the analysis. Myelosuppression was identified in 152 patients (39%) and was associated with significantly higher levels of inflammatory markers (e.g., C-reactive protein (CRP) and procalcitonin), an increased incidence of bacteremia, and increased 30-day mortality (20% vs. 12%, p = 0.03). Patients experiencing myelosuppression had lower median glutamic acid concentrations (134 vs. 154 µmol/L, p = 0.05). Among the 44 clinical and metabolic features, glutamic acid was the strongest predictor according to SHAP analysis. Furthermore, glutamic acid concentrations less than 134 µmol/L were associated with reduced survival according to Kaplan Meier analysis (p = 0.013). Reduced plasma glutamic acid levels at admission are independently associated with myelosuppression and worse prognosis in septic patients. Although causality cannot be established and nutritional/temporal factors were not captured, glutamic acid may represent an early biomarker for risk stratification and deserves further investigation in longitudinal and interventional studies.