Furosemide initiation during norepinephrine infusion and proximal hemodynamic destabilization: a target trial emulation in MIMIC-IV with external validation in eICU-CRD

Technical Summary: Furosemide Initiation During Norepinephrine Infusion and Proximal Hemodynamic Destabilization

Problem Statement
Intravenous furosemide is frequently administered to intensive care unit (ICU) patients receiving norepinephrine to manage fluid overload. However, the proximal hemodynamic consequences of initiating diuresis while patients are under active vasopressor support remain inadequately characterized. While prior research has focused on long-term efficacy and mortality, there is a lack of rigorous evidence regarding the immediate hemodynamic safety of this practice. Specifically, it is unclear whether the reduction in effective intravascular volume caused by diuretics compromises the dose–response coupling required to maintain perfusion in vasopressor-dependent states, potentially triggering immediate hemodynamic destabilization. Previous studies, such as Bandak et al., have yielded conflicting or null results, often due to methodological limitations including a lack of target trial emulation, absence of event-based outcomes capable of detecting clinically meaningful escalation, and failure to characterize treatment effect heterogeneity.

Methodology
This study employed a target trial emulation framework using retrospective data from two large, independent critical care databases: the MIMIC-IV (derivation cohort, 2008–2022) and the eICU-CRD (validation cohort, 2014–2015).

• Study Population: Adults (≥18 years) with a cumulative norepinephrine infusion duration of at least 120 minutes. Patients with end-stage kidney disease, early renal replacement therapy, or prior furosemide administration before norepinephrine initiation were excluded.
• Treatment Strategies:
  • Treated: Initiation of intravenous furosemide during an active norepinephrine infusion.
  • Control: No furosemide administration while on norepinephrine.
  • Time Zero: Defined as the first furosemide dose for the treated arm and a randomly sampled "on-norepinephrine" hour for the control arm to ensure temporal alignment and avoid immortal time bias.
• Primary Outcome: A 24-hour composite indicator of hemodynamic destabilization, defined as either (a) sustained norepinephrine dose escalation (≥1.3 times the index dose for ≥120 minutes) or (b) initiation of a new vasoactive agent.
• Secondary Outcomes: Progression of Acute Kidney Injury (AKI) within 7 days (KDIGO criteria) and all-cause mortality within 28 days (or in-hospital mortality in eICU).
• Statistical Analysis:
  • Causal Inference: The primary analysis utilized 1:2 propensity-score matching (PSM) with a caliper of 0.2 standard deviations. Sensitivity analyses included exact-plus-propensity matching (controlling for time-related covariates) and stabilized inverse probability of treatment weighting (IPTW).
  • Heterogeneity: Causal forests (4,000 trees) were used to estimate conditional average treatment effects (CATEs) and identify patient subgroups. A depth-2 policy tree was fitted to translate these effects into interpretable decision rules.
  • Robustness: E-values were calculated to quantify the strength of unmeasured confounding required to explain the observed associations.
  • Validation: The primary findings were externally validated in the eICU-CRD using a parallel analytic pipeline.

Key Results

• Primary Outcome (Hemodynamic Destabilization): In the MIMIC-IV derivation cohort (n=6,759; 3,836 matched), furosemide initiation was associated with a significantly higher risk of the primary composite outcome (Odds Ratio [OR] 2.03, 95% CI 1.77–2.34). This effect was driven principally by sustained norepinephrine dose escalation (OR 2.66, 95% CI 2.27–3.12) rather than the initiation of new vasopressors (OR 1.15).
• Causal Estimates: The doubly robust average treatment effect (ATE) corresponded to an absolute risk difference of 0.161 (95% CI 0.135–0.188), implying a Number Needed to Harm (NNH) of approximately 6.
• Heterogeneity: Causal forest analysis revealed that all 6,759 patient-level CATE estimates were positive, indicating no subgroup derived a net hemodynamic benefit. The current norepinephrine dose at the index time was the most significant driver of treatment effect heterogeneity.
• Secondary Outcomes:
  • AKI: The primary PSM analysis showed no association with AKI progression (OR 1.03). However, the exact-plus-propensity sensitivity analysis showed a positive association (OR 1.25, 95% CI 1.10–1.43).
  • Mortality: No significant association was found with 28-day all-cause mortality in MIMIC-IV (OR 1.05) or in-hospital mortality in eICU-CRD (OR 1.10).
• External Validation: The eICU-CRD validation cohort (n=7,137; 2,307 matched) directionally confirmed the primary findings. Furosemide initiation was associated with the primary composite (OR 1.71, 95% CI 1.40–2.09) and norepinephrine dose escalation (OR 1.62, 95% CI 1.28–2.06). The eICU cohort also showed a significant association with AKI progression (OR 2.01).
• Robustness: E-values for the primary composite ranged from 2.52 to 2.69, and up to 3.75 for norepinephrine dose escalation, suggesting that a substantial unmeasured confounder would be required to fully explain the observed associations.

Significance and Claims
The authors claim that this study provides the first rigorous, causally interpretable evidence that initiating furosemide during active norepinephrine infusion is associated with proximal hemodynamic destabilization, characterized primarily by the need to escalate vasopressor doses. The study highlights a "proximal-distal dissociation": while the immediate hemodynamic cost is substantial and consistent across patient phenotypes, this perturbation does not appear to translate into increased short-term mortality, likely because clinicians detect and correct the acute instability in real time.

The paper emphasizes that no patient subgroup was identified that would derive a net hemodynamic benefit from this intervention; the treatment effect was uniformly positive (harmful) across the cohort. The findings suggest that the timing of diuretic administration relative to the trajectory of vasopressor dependence is a critical clinical consideration. The authors conclude that these results support careful consideration of timing in vasopressor-dependent patients and call for prospective evaluation to confirm these causal inferences. The study does not claim to have established a definitive clinical rule but rather provides a hypothesis-generating framework for understanding the risks of concurrent diuresis and vasopressor support.