Determinants of fluid use and the association between volume of fluid used and effect of balanced solutions on mortality in critically ill patients: a secondary analysis of the BaSICS trial.

PURPOSE: Fluid use could modulate the effect of balanced solutions (BS) on outcome of intensive care unit (ICU) patients. It is uncertain whether fluid use practices are driven more by patient features or local practices. It is also unclear whether a "dose-response" for the potential benefits of balanced solutions exists.

METHODS: The secondary analysis of the Balanced Solution in Intensive Care Study (BaSICS) compared 0.9% saline versus Plasma-Lyte 148® (BS) for fluid therapy in the ICU. The relative contribution of patient features and enrolling site (the random effect) on the volume of fluid used up to day 3 after admission was assessed using different methods, including a Bayesian regression, a frequentist mixed model, and a random forest, all adjusted for relevant patient confounders. Subsequently, a variety of methods were used to assess whether volume of fluid used modulated the effect of BS on 90-day mortality, including a traditional subgroup analysis for patients that remained alive and in the ICU up to 3 days, a Bayesian network accounting for competing risks, and an analysis based on site practices.

RESULTS: 10,505 patients were analyzed. Median fluid use in the BS arm and in the 0.9% saline arm were 2500 mL and 2488 mL, respectively. The random effect in the Bayesian regression explained 0.32 (95% credible intervals (CrI) 0.24-0.41) of all model variance (0.33, 95% credible intervals from 0.32-0.35). Frequentist and random forest models produced similar results. In the analysis including only patients alive and in the ICU at 3 days, there was a strong suggestion of interaction between fluid use and the effect of BS, driven mostly by a lower mortality with BS compared to 0.9% saline as fluid use increased for patients with sepsis. These results were consistent in the Bayesian network analysis and in an analysis based on site practices, where septic patients enrolled to BS at high fluid use sites had a lower mortality (absolute risk reduction of - 0.13 [95% credible interval - 0.27 to - 0.01]; 0.98 probability of benefit).

CONCLUSION: Baseline patient characteristics collected in the BaSICS trial explain less of the variance of fluid use during the first 3 days than the enrolling site. Volume of fluid used and the effects of BS appear to interact, mostly in the sepsis subgroup where there was a strong association between fluid use after enrollment and the effect of BS on 90-day mortality.