Background Valproic acid is an antiepileptic that has a broad antiepileptic spectrum with high protein binding. Due to the large variability in protein binding, it is recommended that malnourished patients receiving valproic acid therapy are supervised via therapeutic drug monitoring, but is not always technically possible or possible because of cost.

Purpose Correlate serum valproic acid concentrations with unbound valproic acid concentrations.

Material and methods A retrospective and observational study including critically ill and malnourished patients treated with valproic acid was conducted. Data pairs collected total valproic acid and unbound valproic acid determined at the same time. Dose of valproic acid (mg), weight (kg), age, sex, serum protein (g/dL), serum albumin (g/dL), serum creatinine (g/dL), serum urea (g/dL), serum total bilirubin (g/dL) and serum glutamate pyruvate transaminase (g/dL) were collected from electronic clinical records. Statistical analysis was performed with NONMEM, fitting to the Langmuir equation (Ct=((Bm*Cf)/(Kd+Cf)+Cf); where Ct is total valproic acid concentration, Cf is unbound valproic acid concentration, Bm is maximum concentration of valproic acid binding site on the serum protein and Kd is the dissociation constant between serum protein and valproic acid). All parameters were provided with interindividual variability. Visual predictive check (<5% of the observations must fall outside the range of 95% prediction) and bootstrap were performed to assess the predictive ability of the final model and ensure the validity of the method, respectively.

Results 17 malnourished adults were included (0.86 men/women). The final model took into account linearly the addition of albumin on Kd (slope=m). Final model parameters were Bm=47.3 mg/L (95% CI 35.0 to 76.7), Kd=127*1/mg (95% CI 57.5 to 313) and m=-13.6 (95% CI -29.9 to -79.1). Visual predictive check and bootstrap confirmed the intern validity of the final model (3.57% of the observations were excluded in the 95% CI and the calculated parameters for the model were within the 95% CI and the means were below 8.5%).

Conclusion This correlation model provided an estimation of unbounded valproic acid in critically ill and malnourished patients, saving money and time on determination. A study with more patients would give the model more robustness.

No conflict of interest.