Critical Care Intensivist Cooper University Hospital, United States
Introduction: Positive fluid balance in critically ill patients has become widely recognized as potentially deleterious, especially in sepsis. However, fluid administration remains a cornerstone of therapy across most diagnoses. While prior scholarship has focused predominantly on sepsis, the effect of fluid balance on mortality risk for other major critical illnesses remains less well delineated.
Methods: We conducted a retrospective analysis using the EICU dataset, extracting a cohort of patients admitted for common critical diagnoses excluding sepsis. Missing data was imputed using iterative imputation with predictive mean matching. Features were scaled using robust scaling. Fluid balance was the treatment and Mortality was the outcome. We developed a causal inference methodology using the DoWhy Python library, implementing propensity score weighting to estimate the heterogeneous treatment effect of positive fluid balance on in-hospital mortality for each admission diagnosis.
Results: Our models revealed significant heterogeneity in the causal relationship between fluid balance and mortality. Congestive heart failure patients demonstrated the highest mortality risk from positive fluid balance (ATE 0.062), while post-CABG patients derived benefit (ATE -0.015). Stroke, diabetic ketoacidosis, and arrhythmia admissions showed more modest or neutral effects.
Conclusions: These findings elucidate meaningful variability in the causal effects of fluid resuscitation across diseases. As fluid administration diffuses into clinical practice for increasingly diverse conditions, these results highlight the need for evidence-based, diagnosis-specific fluid management grounded in causal methodology. Further prospective confirmation is warranted before translating these conclusions to practice.
