The fluid challenge (FC) response is usually evaluated as binary, which may be inadequate to describe the complex interactions between heart function and vascular tone response after fluid administration. We applied a clustering approach to assess the different phenotypes of cardiovascular responses to FC administration, considering the associations of all the baseline variables potentially influencing pressure and flow response to a FC. Secondarily, we evaluated the reliability of baseline hemodynamic variables in discriminating fluid responsiveness, which is considered the standard approach at the bedside. Five merged datasets from elective surgical patients receiving a FC dose ≥4 mL/kg, infused over 10 minutes. In a principal component approach, hierarchical clustering was used to define hemodynamic phenotypes of response to FC administration. Hierarchical cluster analysis with Ward linkage was carried out to define similar patient groups using the Gower distance for the mixed combination of continuous and categorical variables. No a priori criteria of fluid responsiveness were applied. The area (AUC) under the pre-FC variables' receiver operating characteristic curves (ROC) was also built to predict fluid responsiveness, defined as SVI ≥10% after FC. We analyzed 223 patients. The cluster analysis identified three hemodynamic clusters of patients: cluster 1 (98 patients, 44.0%) showed an average increase of mean arterial pressure (MAP) and Stroke Volume Index (SVI) of 17.3% (11.9-23.1) and 13.1% (0.5-23.4) at the end of FC, respectively. These patients showed baseline flow and pressure variables slightly below physiological ranges, with high pulse pressure variation (PPV). Cluster 2 (68 patients, 30.5%) showed no increase of MAP and SVI at the end of FC. These patients showed baseline flow and pressure variables within physiological ranges, with low hear rate (HR) and PPV. Cluster 3 (57 patients, 25.5%) showed no MAP increase and an SVI increase of 13.1 (2.1-19.6). These patients showed baseline pressure variables within physiological ranges, low flow variables associated to high HR and PPV. The pulse pressure variation (PPV) showed an AUC of 0.82 (0.03), with a grey zone ranging from 6% to 12%, including 86 (38.5%) patients. Clustering analysis identified three hemodynamic clusters with different response phenotypes to FC. This promising approach may enhance the ability to detect fluid responsiveness at the bedside, by considering the specific association of parameters and not the presence of a single one, such as the PPV. In fact, in our cohort the reliability of the PPV was limited, showing high sensibility and specificity only above 12% and below 6%, respectively, and a grey zone including 38.5% of patients.
Read full abstract