Abstract
Background
Positive fluid balance is associated with worse outcomes in sepsis-associated acute kidney injury (SA-AKI), but the optimal timing for active fluid removal (de-resuscitation) is unknown. We examined whether earlier loop-diuretic therapy is associated with lower mortality than later initiation.
Methods
Using MIMIC-IV v3.1, we performed a landmark-anchored cohort analysis informed by the target trial framework, with the 48-h landmark as time zero. Adults with Sepsis-3, Kidney Disease: Improving Global Outcomes (KDIGO) acute kidney injury within 48 h, a positive 48-h fluid balance and no pre-landmark continuous renal replacement therapy were classified by the timing of the first loop diuretic as early (≤24 h after the landmark) or late (>24 h to 7 days). Outcomes were 28-day (primary), in-hospital and 90-day mortality. Stabilized inverse-probability-of-treatment weighting balanced nine covariates, and odds ratios (ORs) were estimated by weighted logistic regression, with sensitivity analyses,
E
-values, subgroups and internal validation. External validation used the eICU Collaborative Research Database, with in-hospital mortality as the primary endpoint.
Results
In MIMIC-IV (1419 early, 1924 late), early de-resuscitation was associated after weighting with lower 28-day mortality (19.9% vs 25.3%; OR, 0.731; 95% CI, 0.619-0.863;
P
= .0002), in-hospital mortality (OR, 0.680; 95% CI, 0.570-0.811), and 90-day mortality (OR, 0.743; 95% CI, 0.640-0.862). In eICU (537 early, 1057 late), it was associated with lower in-hospital mortality (13.7% vs 18.3%; OR, 0.708; 95% CI, 0.541-0.928;
P
= .013) and ICU mortality (OR, 0.663; 95% CI, 0.449-0.980). The association was directionally consistent across thresholds, complete-case and continuous-exposure analyses, subgroups, and bootstrap validation.
Conclusions
In this observational analysis with external validation, earlier loop-diuretic de-resuscitation was associated with lower short-term mortality in SA-AKI with fluid accumulation. The findings are hypothesis-generating: a pragmatic randomized trial using the computable phenotype defined here is required to determine whether the association is causal.