As standardized in our ICU, epinephrine was used as the first-line vasopressor. Epinephrine (diluted to 1:10 in 0.9% saline) was started intravenously using a programmable syringe pump (Pilote IEC, Fresenius-Vial, Bressins, France) at a rate of 0.15 μg/kg/min. The infusion rate was then adjusted to obtain a mean arterial pressure between 65 and 75 mmHg. Of note, continuous epinephrine infusion is required in septic shock patients to maintain arterial pressure until the patient's hemodynamic status improves, generally over several hours or days; thus duration of perfusion or total cumulative dose cannot be predicted. In this study, neither intravenous hydrocortisone nor recombinant human activated protein C were used, and epinephrine was the exclusive catecholamine used.

An initial blood sample (C₀) was drawn within the 15 minutes preceding epinephrine infusion. A second blood sample (C₁) was drawn when the cumulative epinephrine dose adjusted to body weight (BW) reached a threshold fixed arbitrarily at 0.15 mg/kg, provided the epinephrine infusion rate remained steady and fluid loading was not used in the preceding 15 minutes. In the case of modification of the epinephrine infusion rate or fluid loading in the 15 minutes preceding the threshold dose, C₁ was delayed until a 15-minute period of stability for the infusion rate was obtained. The 15-minute steady-state interval was chosen according to epinephrine plasmatic half-life in healthy subjects 12. The epinephrine infusion rate (mg/hr) at C₁ time was recorded. Blood drawn on C₀ was used to assess plasma levels of endogenous adrenal axis hormones: epinephrine, norepinephrine, renin, aldosterone, and cortisol. Blood drawn on C₁ allowed for the measurement of epinephrine and norepinephrine plasma levels.

Blood assigned to catecholamine assays was sampled in EDTA-tubes and immediately centrifuged at 3000 g for five minutes. The plasma samples were then immediately stored at -80°C before assay. Blood assigned to renin and aldosterone assays was also sampled in EDTA-tubes and centrifuged at 3000 g for five minutes. The plasma was then separated and stored at -20°C. Blood assigned for cortisol assays was allowed to clot at room temperature for 30 minutes, then centrifuged at 3000 g for five minutes. Samples were stored at -20°C.

Epinephrine and norepinephrine concentrations were measured in plasma using high-pressure liquid chromatography (HPLC) with coulometric detection 1314. The limit of quantification (defined by a variability between measurements of <10%) for HPLC was 0.10 nmol/L. The epinephrine concentration measured on C₁ was the sum of endogenous and exogenous epinephrine, as the two compounds are strictly identical with regard to chromatographic detection. Plasma aldosterone was measured in duplicate by RIA using a commercial kit from the Diagnostic Products Corporation (Los Angeles, CA, USA). Renin and cortisol concentrations were measured on a fully automated chemiluminescence analyzer (LIAISON^® Analyzer, DiaSorin S.p.A, Salluggia-Vercelli, Italy). Plasma renin concentration was measured with a Direct Renin assay. This two-site immunometric assay was calibrated according to World Health Organization reference material (National Institute for Biological Standards and Control, code 58/356). Normal ranges in the supine position for plasma renin, aldosterone, and cortisol concentrations were 10 to 25 mU/L, 80 to 400 pmol/L, and 330 to 500 nmol/L, respectively.

Baseline demographic data included sex, age, BW, new simplified acute physiology score (SAPS) II at study inclusion 15, ICU length of stay before inclusion and cause of septic shock. The volume of fluid administered for resuscitation of shock before C₁ was recorded. Invasive blood pressure and heart rate were collected at C₀ and C₁.

Data were analyzed using the nonlinear mixed effect modeling software program NONMEM version VI driven by Wings for Nonmem (WfN, Free Software Foundation, Boston, MA, USA) 16. The FOCE method was used. This method allows for the estimation of both the pharmacokinetic and statistic parameters of the model, that is, the elimination clearance is estimated along with its corresponding between-subject variability (BSV). Any residual variability, including measurement errors, can also be estimated. Epinephrine pharmacokinetics was ascribed to a one-compartment open model with first order elimination. Parameters for the model were plasma clearance (CL), volume of distribution (V), and baseline rate of epinephrine infusion (R0). The R0 parameter allowed us to take into account the baseline epinephrine concentration at C₀. BSVs were assumed to be exponential and their variances and standard deviations (SD) were denoted as ω²CL and ωCL, respectively. Covariances were also estimated. When a full covariance matrix could not be estimated, the following algorithm was applied: ω²CL was always retained; and if the correlation between terms was low, it was fixed at 0.

Proportional, additive or mixed error models were investigated to describe any residual variability. The main covariates of interest in the population were sex, age, BW, SAPS II, volume of liquid infused, and plasma hormone concentrations. Parameter estimates were standardized for a mean standard covariate using an allometric model: P_i = P_STD × (BW_i/BW_STD)^θBW where P_STD and θ_BW are the standard parameter value for a patient with the standard BW value and the power parameter estimate for BW, and P_i and BW_i are the parameter and BW of the ith individual. Graphical evaluation of goodness-of-fit was mainly assessed by observed vs. predicted concentrations and weighted residuals vs. time and/or weighted residuals vs. predicted concentrations. The final population model was also ascertained by normalized prediction distribution error metrics 17. The stability of the model and accuracy of the parameters were assessed by a bootstrap method implemented in Wings for Nonmem 18 and diagnostic graphics and distribution statistics were obtained using R for Nonmem 18 via the R program 19.

The sample size was calculated based on data from a previous study where the lower limit of the 95% confidence interval of the correlation coefficient r² between epinephrine perfusion rate and its plasma concentration was 0.4. To detect such a correlation (we used a β risk of 20% and an α risk of 5%), 28 patients or more were required. Results are expressed as numbers (%), means ± SD, or medians (range) for data not normally distributed. Analyses were conducted with SPSS 11.5 software (SPSS Inc., Chicago, IL, USA). Wilcoxon and Mann-Whitney tests were applied for comparison of relevant variables. Shapiro-Wilks test was used for the assessment of normality. We considered a difference to be significant when the α risk was < 5% (P < 0.05).

Thirty-eight consecutive patients satisfying the entry criteria were recruited and had their plasma epinephrine concentrations analyzed. We were able to sample C₁ at the exact threshold dose in all of these patients. Characteristics and demographic data of the enrolled patients are summarized in Table 1. Three C₁ plasma tubes assigned for epinephrine and norepinephrine blood concentrations were lost due to a technical problem during handling.

Epinephrine infusion significantly increased arterial blood pressure and heart rate from C₀ to C₁, and was associated with a significant decrease in plasma norepinephrine concentration (Table 2). The median fluid volume administered for shock resuscitation until C₁ was 4650 ml (range 500 to 11,400). In all patients, 90% of this volume had been administered before C₀.

Thirty-eight patients and 73 plasma epinephrine concentrations were available for pharmacokinetics evaluation. The median delay from the start of epinephrine infusion to C₁ was 415 minutes (range 90 to 1260). Despite the fact that the drug dosage was arbitrarily normalized on BW, the infusion rates varied widely at C₁ due to the significant variations in patients' requirements to achieve hemodynamic goals: there was a greater than 100-fold difference between the lowest and highest rates (Table 2). A one-compartment model adequately described the data. In a first step, only BSV for CL (ωCL) and an additive component for the residual variability could be estimated. The use of a one-compartment model with Michaelis-Menten (saturable) elimination did not improve fit and could not provide reliable Vmax and Km estimates. However, the accuracy of the residual variability parameter was very poor. Thus, in a second step, the residual variability parameters were fixed as follows: 10% and 0.1 nmol/L for the proportional and additive components, according to the assay quantification as stated in Methods. In this manner, the BSV for CL and R0 could be accurately estimated. The parameter estimates of this basic model were CL, 108 L/hr (ωCL = 0.44), V, 9.1 L, and R0, 43.5 nmol/hr (ωR0 = 1.21). The corresponding half-life was 3.5 minutes. The accuracy of estimates varied from 6% for CL to 36% for V. Only BW and SAPS II at ICU admission significantly influenced epinephrine CL, reducing the objective function value by 20 units and ωCL to 0.33. The final relationship for epinephrine CL was thus: CL_i (L/hr) = 127 × (BW/70)^0.60 × (SAPS II/50)^-0.67 where 127 L/hr is the typical CL for an individual weighing 70 kg with a SAPS II of 50. This relationship shows that CL increases with BW and decreases with SAPS II. Using CL, the prediction of the epinephrine plateau concentration at the steady-state infusion rate was: C_plateau (nmol/L) = (rate of infusion + R0)/(127 × (BW/70)^0.60 × (SAPS II/50)^-0.67). Baseline norepinephrine, aldosterone, renin, and cortisol blood concentrations as well as volume of liquid administered for shock resuscitation had no impact on CL. Table 3 summarizes the final population pharmacokinetics estimates including the bootstrap verification. Figure 1 depicts the predicted versus observed concentrations and Figure 2 shows the corresponding normalized prediction distribution error test for this data.