Abstract Background The efficacy of granulocyte-colony stimulating factor (G-CSF) treatment for acute-on-chronic liver failure (ACLF) remains controversial. The aim of this study, which is a secondary analysis of the GRAFT study (NCT02669680), was to identify potential prognostic biomarkers in ACLF and to find markers that could be used to predict response to G-CSF treatment. Methods Blood samples from 79 patients randomized in the GRAFT study to receive G-CSF ( n = 40) or standard medical therapy ( n = 39) were collected at different timepoints. Samples were analyzed for cells, cytokines, extracellular particles, cell-free DNA, and functional properties. An exploratory approach was taken, whereby the measured variables were subjected to univariate and multivariate Cox analyses, and the patients were stratified into clusters by unsupervised hierarchical clustering. Results Patients with ACLF had increased plasma levels of pro-inflammatory cytokines and increased absolute levels of specific cell populations when compared to healthy controls. ROC curve analysis suggested that plasma VEGF-A levels at baseline (timepoint) are a prognostic factor of transplant-free survival (AUC: 0.70; 95%CI 0.56–0.84). Hierarchical clustering of circulating immune cell populations at baseline appeared to define a patient subset within the analyzed G-CSF-treated patients with improved median transplant-free survival (102 days vs 16 days; p = 0.002). In our study, a higher percentage of CD39 + lymphocytes serves as an independent predictor of unfavorable outcomes following G-CSF treatment (HR: 1.05, 95%CI: 1.01–1.09, p = 0.008). Conclusion VEGF-A appears to be a suitable biomarker for predicting the prognosis of patients with ACLF. With the help of hierarchical cell clusters, ACLF patients who could benefit from G-CSF therapy could be identified and selected prior to treatment. CD39 expression on lymphocytes seems suitable to allow stratification. Further testing and validation are necessary and could help to adapt treatment options for each individual patient. Graphical abstract
Splith et al. (2026) studied this question.