This study is an open-labeled interventional controlled trial with a cross-over design where 12 healthy participants (six male and six female) underwent two infusion experiments, 3–20 weeks apart, where they received 3 mL/kg of 20% albumin by intravenous infusion during either 30 min (“fast infusion”) or 120 min (“slow infusion”).

All participants gave informed consent orally and in writing. Inclusion criteria were an age between 18 and 60 years, and absence of medical disease and medication. Exclusion criteria were pregnancy, difficulties with placement of venous cannulas, and severe allergy.

Procedure

A statistician prepared 12 envelopes for randomization of the participants to start with either the slow or the fast infusion. The envelopes were opened one day prior to the first infusion. All participants fasted from midnight before the study and throughout the study period. However, they were allowed to drink 2 dL of liquid and eat one sandwich 1.5 hours (h) prior to arriving at the research faculty at the hospital.

A venous cannula was placed in the right arm and another in the left for blood sampling and fluid infusion, respectively. The participants emptied their bladders 30 min before the study started and were then placed in a supine position until the end of the experiment. Baseline samples were withdrawn, and the participants then received the infusion. Blood was sampled on 15 occasions over a period of 6 h.

When albumin was administered over 30 min, the excreted urine was measured and sampled just before the infusion started, at 30 min after the infusion ended, and at 6 h. When albumin was administered over 120 min, the excreted urine was measured and sampled just before and at the end of the infusion, and at 6 h.

Blood and urine analyses

Whole blood was analyzed for hemoglobin (Hgb) concentration and hematocrit with a coefficient of variation (CV) of 1.0%, as given by duplicate samples at baseline, using a Cell-Dyn Sapphire instrument (Abbott Diagnostics, Abbott Park, IL). Plasma was analyzed for albumin, creatinine, sodium, and potassium on a Cobas 8000 system (Roche Diagnostics, Basel, Switzerland) at the hospital’s certified central laboratory (CVs were 2.3%, 1.9%, 0.7%, and 1%, respectively, as given by the laboratory).

The plasma colloid osmotic pressure (COP) was measured in our research laboratory on an Osmomat 050 device (Gonotec, Berlin, Germany) with a CV of 2%.

The plasma concentration of the mid-regional pro-atrial natriuretic peptide (MR-proANP) at baseline and 30 min after the infusion ended was analyzed by radioimmunoassay (Brahms MR-proANP Kryptor, Henningsdorf, Germany) with a CV of > 3.5%. The manufacturer reports a median value in healthy humans of 46 pmol/L.

Urine was analyzed for creatinine on the Cobas 8000 system with a CV of 1.9%.

Mass balance

The baseline plasma volume (PVo) was estimated from the height and weight of the participants, as suggested by Nadler *et al* [10]. The PV change to a any later Time t of the experiment (PVt) was calculated based on the hemodilution curve, with correction for sampled blood volume, as described previously [6].

The albumin mass was taken as the product of the plasma volume (PV) and the plasma albumin concentration (P-Alb). Multiplication with PV is necessary because P-Alb is diluted by the infused fluid volume and also by the oncotic-driven recruitment of extravascular fluid, which gives an unbalanced relationship between P-Alb and the intravascular albumin mass.

Capillary leakage. The capillary leakage of albumin was obtained as the change in albumin mass, with correction for the infused amount of albumin, between baseline (time 0) and a later time t [7]. The following equation was used:

Albumin leakage = Infused albumin + (PVo x P-Albo) – (PVt x P-Albt)

Half-life. The half-life of the infused albumin was obtained from the logarithm of the slope of the albumin mass, given as [(1 + PVdil) P-Albt – P-Albo] *versus* time, when an apparent first-order elimination had been established post-infusion [7]. For each experiment, the half-life of the decay of the PV expansion was estimated in the same way for the albumin mass.

Albumin and fluid kinetics

A one-compartment model was used to study the kinetics of the infused albumin mass throughout the entire experiment. In this model, albumin mixed in fluid was infused at a rate *R*o into a central body fluid space *V*c, which was then expanded to *v*c. The capillary leakage was given by the rate constant *k*b. The dependent variable was the product of the increase in measured P-Alb and the plasma dilution; the latter was given by ((Hgbo / Hgbt ) – 1) / (1 – hematocrito), where the subscript o denotes the baseline and *t* a later time. Minor correction of the plasma dilution for blood sampling was made [6].

The kinetics of the infused fluid volume was evaluated using a model with micro-constants that is developed for studies of 20% albumin [11]. This model has one infusion, one absorption route, and two elimination routes and was fitted to the plasma dilution and the urinary excretion, which served as the dependent variables.

Absorption occurred from an extravascular source, which is likely to be the interstitial fluid space, by (supposedly) oncotic forces and at a rate that was determined by a constant denoted *k*21. The interstitial fluid volume at baseline (ICFo) was assumed to contain fluid accounting for 15% of the body weight [12].

Fluid volume was eliminated by urinary excretion (*k*10) and capillary leakage (*k*b).

This “base model” was expressed by the following differential equations:

d*v*c /dt = Ro – *k*b (*v*c – *V*c) – *k*10 (*v*c – *V*c) + *k*21 ICFo

dICF/dt = – *k*21 ICFo; d*U*/dt = *k*10 (*v*c – *V*c)

where *V*c is the baseline and *v*c the expanded central volumes, and *U* the measured urinary excretion. The rate parameter *k*21 does not come into play before the infusion begins.

The fixed parameters in the albumin model (*V*c and *k*b for albumin) were estimated simultaneously for all 24 experiments using the First Order Conditional Estimation Extended Least-Squares (FOCE ELS) search routine in the Phoenix software for nonlinear mixed effects (NLME), version 8.2 (Pharsight, St. Louis, MO) and the additive model for the within-subject variability. The dependent variable was P-Alb corrected for plasma dilution.

The fixed parameters in the fluid model (*V*c, *k*10, *k*b, and *k*21) were estimated in the same way. Here, the dependent variables were the frequently measured plasma dilution and the urinary excretion measured at 1 h and 6 h.

Both base models were both refined by adding individual-specific covariates. Eleven potential covariates were examined. Age, body weight, gender, Hgbo, and urine osmolality and urine creatinine at baseline were entered once for each patient. Plasma creatinine, and MR-proANP were measured twice per experiment and applied at the point in time when measured. Plasma albumin was entered as a time-varying covariate 15 times per experiment (at the same time points as Hgb was measured).

Outcome measures

The primary outcome measure was the plasma dilution, which is an index of plasma volume expansion, following the infusions of hyper-oncotic albumin. Secondary outcome measures were the colloid osmotic pressure and how much the intravascular persistence was prolonged by oncotic-driven recruitment of fluid from of extravascular tissues.

Statistics

Power analysis prior to the study was based on the previously obtained a mean and standard deviation (SD) for the plasma volume expansion of 15.8% ± 4.9% at end of infusing 3 mL/kg of 20% albumin [11]. We aimed at identifying a difference in plasma volume expansion of 20% at the *p* < 0.05 level and with a certainty of 80%. This calculation yielded 22 experiments.

The measured variables were reported as the mean (SD) and, when appropriate, as the median and interquartile range (IQR). Differences between the two infusions were studied, depending on the distribution of the data, by the paired *t* test or Wilcoxon’s matched-pair test. *P* < 0.05 was deemed statistically significant.

Kinetic parameters were given as the best estimate and 95% confidence interval (CI). A new parameter (fixed or covariate) was accepted if its 95% CI did not include 0 and the inclusion decreased the − 2 log likelihood (-2 LL) for the model by > 3.8 points (*p* < 0.05).