Cyclosporine (CsA), a classic calcineurin inhibitor, has long served as a cornerstone of immunosuppressive therapy. It is widely applied to prevent organ transplant rejection and graft-versus-host disease after bone marrow transplantation, and also treats multiple autoimmune diseases such as nephrotic syndrome and psoriasis^[1,2]. Although tacrolimus has emerged as an alternative in recent years, CsA remains irreplaceable in clinical practice. Metabolized mainly by CYP3A4 in the liver and intestines and transported by P-glycoprotein, CsA has a low bioavailability of only 27%^[3,4].

Cyclosporine has an extremely narrow therapeutic window and exhibits significant interindividual variability in pharmacokinetics. Factors such as age, body weight, liver and kidney function, and concomitant medications can all influence its plasma concentrations. Inappropriate dosing can easily lead to inadequate therapeutic effects or drug toxicity; therefore, therapeutic drug monitoring (TDM) has become an essential tool in clinical practice. Traditional monitoring often uses pre-dose trough concentrations as an indicator; however, this metric fails to accurately reflect overall drug exposure. Post-dose concentrations at 2 hours have been demonstrated to be a more ideal alternative indicator. With the growing prevalence of precision medicine, model-guided precision dosing utilizing population pharmacokinetic models and Bayesian prediction techniques can fully leverage the value of TDM data to achieve individualized cyclosporine therapy. With this objective in mind, this study aimed to construct a population pharmacokinetic model for cyclosporine specifically tailored to patients in Uruguay.

A total of 53 patients treated with CsA were enrolled and divided into two groups. The model-building group included 37 patients with at least one complete pharmacokinetic profile, generating 621 steady-state CsA blood concentration samples. The other 16 patients formed the validation group with 81 samples to assess the model’s predictive performance. Researchers collected comprehensive clinical information including gender, age, body weight, medication history, comedications, post-transplant duration and renal function. Whole blood samples collected at different time points after drug administration were tested for CsA concentration via chemiluminescent microparticle immunoassays.

Cl_i is the individual clearance of CsA, Cl_pop is the population clearance determined in the PK model, ClCrea is the individual value of creatinine clearance, ClCrea_pop is the mean value for creatinine clearance (98.62 mL/min), and β_ClCrea represents how this covariate impacts on the parameter.

Covariate analysis proved that creatinine clearance was the only significant covariate affecting CsA clearance. In the initial prediction, MPPE values of 1-hour and 2-hour post-dose concentrations were below 50%. After Bayesian forecasting with accumulated monitoring data, the prediction accuracy was greatly improved, and the predictive error of trough concentration decreased from 98% to 27%.

Figure 1 Diagnostic evaluations of the final model for CsA. (a) Plots of observed concentrations versus population and individual predictions; (b) NPDE versus time and individual predicted concentrations; (c) Prediction-corrected visual predictive check (pcVPC).

Prediction-corrected visual predictive check and other diagnostic plots verified that the final model fitted the observed data well and could accurately describe the in-vivo metabolic characteristics of CsA. With the accumulation of monitoring data, the predictive error decreased continuously and the stability was significantly enhanced.

Figure 2 Boxplots of relative individual predictive error (rIPE) at different monitoring occasions. Occasion 1 represents prediction without prior data; Occasion 2 and 3 represent prediction with previous monitoring data.