Abstract
OBJECTIVES: Significant practice variation exists in the amount of resuscitative IV fluid given to patients with sepsis. Current research suggests equipoise between a tightly restrictive or more liberal strategy but data is lacking on a wider range of resuscitation practices. We sought to examine the relationship between a wide range of fluid resuscitation practices and sepsis mortality and then identify the primary driver of this practice variation. DESIGN: Retrospective analysis of the Premier Healthcare Database. SETTING: Six hundred twelve U.S. hospitals. PATIENTS: Patients with sepsis and septic shock admitted from the emergency department to the ICU from January 1, 2016, to December 31, 2019. INTERVENTIONS: The volume of resuscitative IV fluid administered before the end of hospital day- 1 and mortality. MEASUREMENTS AND MAIN RESULTS: In total, 190,682 patients with sepsis and septic shock were included in the analysis. Based upon patient characteristics and illness severity, we predicted that physicians should prescribe patients with sepsis a narrow mean range of IV fluid (95% range, 3.6-4.5 L). Instead, we observed wide variation in the mean IV fluids administered (95% range, 1.7-7.4 L). After splitting the patients into five groups based upon attending physician practice, we observed patients in the moderate group (4.0 L; interquartile range [IQR], 2.4-5.1 L) experienced a 2.5% reduction in risk-adjusted mortality compared with either the very low (1.6 L; IQR, 1.0-2.5 L) or very high (6.1 L; IQR, 4.0-9.0 L) fluid groups p < 0.01). An analysis of within- and between-hospital IV fluid resuscitation practices showed that physician variation within hospitals instead of practice differences between hospitals accounts for the observed variation. CONCLUSIONS: Individual physician practice drives excess variation in the amount of IV fluid given to patients with sepsis. A moderate approach to IV fluid resuscitation is associated with decreased sepsis mortality and should be tested in future randomized controlled trials.