Using vascular biomarkers to assess heart failure event risk in hospitalized patients with and without AKI

Background: Patients with AKI experience higher rates of heart failure (HF). This study seeks to identify criteria to assess the risk of heart failure post-hospitalization, with a special focus on AKI patients. We hypothesized that the combined use of 9 vascular biomarkers would predict future heart failure events after AKI. Using a study of 1497 hospitalized patients with and without AKI, we found that these 9 vascular biomarkers successfully stratified patients into different risk groups for HF, and were able to improve prediction of HF when added to routine clinical variables. Methods: Using the ASSESS-AKI cohort, we performed an unsupervised spectral cluster analysis with 9 plasma biomarkers measured at 3 months post-hospitalization [Angiopoietin (angpt)-1, angpt-2, vascular endothelial growth factor (VEGF)-A, VEGF-C, VEGF-d, VEGF receptor 1 (R1), solubleTie-2 (sTie-2), placental growth factor (PlGF), and basic fibroblast growth factor (bFGF)] in 1,497 patients, half of whom had AKI. We used a Cox regression analysis to evaluate the associations between the clusters and HF. Models were adjusted for demographics, cardiovascular disease risk factors, medications, ICU status, lung disease, sepsis, clinical center, and 3-month post-discharge serum creatinine and proteinuria. We calculated change in the area under the curve (AUC) for the prediction of HF or death at 3 years by adding the biomarkers to a clinical model selected by a penalized regression with LASSO. We also calculated a net reclassification index for the addition of the biomarkers to the clinical model. Results: Three biomarker-derived clusters were identified: Cluster 1 [n = 302, Vascular Injury (Injury) Phenotype] had higher levels of injury markers, whereas Cluster 2 [n = 728, Vascular Repair (Repair) Phenotype] had higher levels of repair markers. Cluster 3 (n = 467) had lower levels of all markers (Dormant Phenotype). Across the entire cohort, those with the Injury Phenotype had twofold higher risk of a HF event compared to the Repair Phenotype [aHR 2.24 (95% CI: 1.57–3.19)] and noted in both participants with AKI [aHR 2.12 (95% CI: 1.35–3.34)] and without AKI [aHR 2.94 (95%CI: 1.57–5.50)]. The Dormant Phenotype was associated with higher risk of HF events only in participants without AKI. The AUC for the prediction of HF event or death at 3 years by the biomarkers was 0.76 (95% CI: 0.73–0.80), 0.77 (95% CI: 0.73–0.80) for the clinical model, and 0.80 (95% CI: 0.77–0.83) for the combined model. The addition of the biomarkers significantly improved reclassification of HF event or death. Conclusions: Vascular biomarkers can be used to derive phenotypes capable of stratifying future risk of HF events in recently hospitalized patients with or without AKI.