Abstract
BACKGROUND: Standard antibiotic dosing is not suitable for critically ill patients, due to altered pharmacokinetics (PK) in these patients. Knowledge of protein binding is important for optimizing antibiotic exposure because only the unbound fraction is pharmacologically active. If unbound fractions can be predicted, minimal sampling techniques and less costly methods can be routinely used. METHODS: Data from the DOLPHIN trial, a prospective randomized clinical trial that included critically ill patients, were used. Total and unbound ceftriaxone concentrations were determined using a validated UPLC-MS/MS method. A non-linear saturable binding model was made using 75% of the trough concentrations and validated on the remaining data. Our model and previously published models were tested for their performance for subtherapeutic (<1 mg/L) and high (>10 mg/L) unbound concentrations. RESULTS: In total, 113 patients were sampled [Acute Physiology And Chronic Health Evaluation version 4 (APACHE IV) score 71 (IQR 55-87), albumin 28 g/L (IQR 24-32)]. This resulted in 439 samples (trough = 224, peak = 215). Unbound fractions were significantly different between samples taken at trough and peak times [10.9% (IQR 7.9-16.4) versus 19.7% (IQR 12.9-26.6), P < 0.0001], which was not explained by concentration differences. Our model and most literature models showed good sensitivity and low specificity to determine high and subtherapeutic ceftriaxone trough concentrations using only the total ceftriaxone and albumin concentrations. CONCLUSIONS: Ceftriaxone protein binding is not concentration related in critically ill patients. Existing models show good ability to predict high concentrations, but low specificity in predicting subtherapeutic concentrations.