Cardiac filling volumes versus pressures for predicting fluid responsiveness after cardiovascular surgery: the role of systolic cardiac function

The protocol was started upon arrival of the patients in the intensive care unit (ICU). Demographic characteristics were recorded, including the Acute Physiology And Chronic Health Evaluation II (APACHE-II) score and transesophageal echocardiographic findings prior to surgery. At baseline (t = 0 minute), hemodynamic measurements were performed. Heart rate (HR) and mean arterial pressure (MAP) from a radial artery were recorded at t = 0 and 90 minutes. The HR was taken from the continuously recorded electrocardiogram. The mean pulmonary artery pressure (MPAP) was measured at t = 0 and 90 minutes. Cardiac output, GEDV, CVP and PAOP were measured every 30 minutes, from t = 0 to 90 minutes. Pressures were measured with patients in the supine position after calibration, zeroing to atmospheric pressure and, for PAOP, after proper wedging, at the midchest level at end-expiration (Tramscope^®, Marquette, GE, Milwaukee, WI, USA). For the measurements of cardiac output and GEDV, the transpulmonary thermal-dye indicator dilution technique was used [1, 6]. These measurements involve a central venous injection of 15 mL of ice-cold indocyanine green in 5% glucose solution and concomitant registration of the dilution curves in the femoral artery, by a 3F catheter equipped with a thermistor (PV 2024, Pulsion Medical Systems, Munich, Germany). This catheter was inserted at the end of surgery via a 4F introducing sheath (Arrow International, Inc., Reading, PA, USA) and connected to a bedside computer (COLD Z-021, Pulsion Medical Systems, Munich, Germany. The COLD Z-021 is the precursor to the current pulse contour cardiac output (PiCCO™) technique. GEDV represents the volumes of the right and left heart at end-diastole and reflects left ventricular dimensions obtained by echocardiography in the absence of overt right ventricular distention [7, 12‐17]. The ratio between stroke volume and global end-diastolic volume/4 is defined as the global ejection fraction (GEF, normal values 25 to 35%), and is an indicator of left ventricular systolic function, provided that there is no right ventricular dysfunction [24, 25]. Reproducibility of these measurements is typically within 10% [1]. After baseline measurements were taken, fluids were given over 90 minutes on the basis of the response within predefined limits of increases in pressures (CVP or - when available - PAOP), according to a previously described protocol [23, 26, 27]. Up to 200 mL of fluid were given every 10 minutes, provided that the increase in filling pressures with the fluid loading did not exceed critical values, and this policy has been proven safe in previous studies (that is, not evoking pulmonary edema) [23, 26, 27]. The maximum amount of fluid infused was 1,800 mL. Concomitant vasoactive and sedative drug treatment and ventilatory settings remained unchanged during fluid loading. Indeed, all patients received volume-controlled mechanical ventilation and positive end-expiratory pressure (PEEP). Drainage of blood was <50 mL/hour in all patients, and no patient underwent repeated surgery for bleeding within 12 hours post surgery.

The groups to be analyzed were divided into low GEF (<20%) and near-normal GEF (≥20%). The cutoff of 20% approximately reflects a cutoff of 40% ejection fraction of the left ventricle, the lower limit of normal, as measured by echocardiography, provided that there is no right ventricular dysfunction [24, 25]. We also analyzed data according to a cutoff of 15%. Stroke volume, cardiac output and global end-diastolic volume were indexed to body surface area (BSA), giving stroke volume index (SVI, mL/m²), cardiac index (CI, L/minute/m²) and global end-diastolic volume index (GEDVI, n 680 to 800 mL/m²), respectively. Cardiac distensibility was determined as the surrogate for cardiac compliance and was calculated by GEDVI/(CVP + PAOP)/2 (mL/m²/mmHg), or GEDVI/CVP if PAOP was not available [28]. Fluid responsiveness was defined as an increase of CI or SVI ≥10% or ≥15%, in accordance with the literature [4, 9, 17], between t = 0 to 30, t = 30 to 60 and t = 60 to 90 minutes during fluid loading. For categorical data, X² and Fisher exact tests were used. Since continuous data were normally distributed (Kolmogorov-Smirnov test, P > 0.05), they were summarized by mean ± standard deviation (SD) and parametric tests were done. Paired and unpaired t-tests were used to compare data in time and between GEF groups (for Table 1). Generalized estimating equations (GEE) were used to evaluate differences in baseline and changes in variables between summated responding and non-responding fluid loading events in each GEF group (for Table 2), to evaluate their predictive and monitoring values, respectively, taking repeated measurements in the same patients into account, with the amount and type of fluid infused entered as covariates to adjust for potential confounding. Partial correlation coefficients (r), adjusted for repeated measurements by entering patient number and for type and amount of fluids as covariates were calculated. Coefficients were compared after z transformation. Receiver operating characteristic curves (ROC) plotting sensitivity against 1-specificity were constructed to evaluate the predictors of fluid responsiveness by the areas under the curve (AUC, with 95% confidence intervals) for pooled data, in the absence of accepted methods to adjust for repeated measurements, and were compared with each other. Optimum cutoff values with associated combinations of highest sensitivity and specificity were calculated (MedCalc Software, Mariakerke, Belgium). Exact two-sided P-values > 0.001 are given and considered statistically significant when <0.05. All analyses were conducted using SPSS version 15.0 (SPSS Inc, Chicago, Ill, USA).

Among the 96 fluid loading events, the proportion of responding events (increase in CI ≥10%) decreased from t = 0 to 90 minutes (P = 0.031). The amount infused was somewhat lower in non-responding than in responding events when GEF was low (<20%), (Table 2). Baseline CI was lower in responding events, regardless of GEF and cutoff percentage of fluid responsiveness. When GEF was low, baseline CVP and PAOP were lower for responding events (≥10% increase in CI) while baseline GEDVI did not differ from that in non-responding events, irrespective of the amount and type of fluids. When GEF was near-normal (≥20%), baseline GEDVI and CVP were lower for responding events (≥10% increase in CI), while baseline PAOP did not differ from that in non-responding events. Similar results were obtained for a GEF cutoff of 15% (Table S1 in Additional file 1).

For fluid responsiveness defined as an increase in CI ≥15%: only baseline PAOP and not CVP predicted fluid responsiveness in the low GEF group (Table S2 in Additional file 1). In contrast, GEDVI particularly predicted fluid responsiveness when GEF was near-normal. Changes in GEDVI paralleled CI responses in both GEF groups, while changes in CVP paralleled CI responses only in the low GEF group. Changes in PAOP particularly paralleled responses in CI when GEF was low.

For the low GEF group, baseline PAOP and CVP inversely correlated to changes in CI, irrespective of amount and type of fluids (r = -0.57 and -0.44, P = 0.008 and 0.010, respectively; Figure 1). In the near-normal GEF group, only baseline CVP inversely correlated to CI changes (r = -0.35, P = 0.009) and PAOP did not (Figure 2). Baseline GEDVI inversely correlated to changes in CI in the near-normal GEF group (r = -0.29, P = 0.03; Figure 3). Changes in CI were paralleled by changes in GEDVI (r = 0.74, P < 0.001) in the low GEF group. Changes in CI correlated to changes in both CVP and GEDVI in the near-normal GEF group (r = 0.36 and r = 0.72, P = 0.007 and <0.001, respectively). Changes in PAOP correlated better to CVP in the near-normal GEF group (r = 0.67, P < 0.001) than in the low GEF group (r = 0.21, P = 0.404).

In the near-normal GEF group, baseline GEDVI and CVP predicted fluid responsiveness (increase in both CI ≥10% and ≥15%), while in the low GEF group baseline PAOP and CVP had predictive value (Table 3). This table also shows the optimum cutoff values and associated sensitivities and specificities for fluid responsiveness. Table S3 in Additional file 1 shows identical results for cutoffs of SVI responses (0 to 90 minutes) rather than of CI responses.