Association between dynamic estimated plasma volume trajectories and acute kidney injury after coronary artery bypass grafting.

BACKGROUND: Cardiac surgery-associated acute kidney injury (CSA-AKI) is driven partly by venous congestion, yet no single bedside measure fully captures volume status: central venous pressure (CVP) reflects venous pressure, not plasma-volume expansion. Whether dynamic estimated plasma volume status (ePVS), a non-invasive marker derived from routine hemoglobin and hematocrit, conveys AKI-relevant information independent of and complementary to CVP is unknown. METHODS: In 4,818 adult patients with coronary artery bypass grafting (CABG) from Medical Information Mart for Intensive Care IV, version 3.0 (MIMIC-IV v3.0), serial ePVS over the first 48 h after admission to the intensive care unit (ICU) was modeled by latent-class linear mixed modeling to identify dynamic phenotypes. The primary outcome was Kidney Disease: Improving Global Outcomes (KDIGO) acute kidney injury (AKI) stage ≥ 2 within 72 h. Independence from CVP was tested by the ePVS-CVP correlation, progressive adjustment for admission CVP, CVP-stratified analysis with a trajectory by CVP interaction term, and parallel multiple-mediator analysis (24-h ΔHgb, peak CVP, net fluid balance). Forward-time landmark analyses and incremental-value metrics [Δ area under the curve (ΔAUC), net reclassification improvement (NRI), integrated discrimination improvement (IDI), and decision-curve analysis] were also performed. RESULTS: Four phenotypes emerged: Low-stable (58.5%), Gradually-increasing (6.9%), Rapid-decline (8.7%), and High-stable (25.9%). Admission ePVS and CVP were essentially uncorrelated (Spearman r = 0.04), indicating non-redundant axes. Only the Gradually-increasing phenotype was independently associated with AKI (Model 3 OR 1.47, 95% CI 1.14 to 1.89), and this was barely changed by adjustment for admission CVP (from 1.58 in Model 1 to 1.49 in Model 2); peak CVP mediated only 9.9% of the total effect, the majority not being attributable to any single measured pathway. Forward-time landmarking amplified the association (OR 2.10 at 24 h; 2.26 at 48 h), arguing against reverse causation. The trajectory-by-CVP interaction was not significant (p = 0.20); the phenotype effect was numerically larger at low than high admission CVP (OR 1.76 versus 1.20), an exploratory pattern. Adding trajectory class to a static model produced modest gains (NRI 9.9%, IDI 0.4%, ΔAUC 0.003). CONCLUSION: Four distinct ePVS trajectories within 48 h after CABG carry differential AKI risk. A Gradually-increasing ePVS trajectory after CABG signifies AKI risk that is largely independent of and complementary to CVP, providing a non-invasive volume axis not captured by pressure-based monitoring.