Renal transplantation, as the most ideal renal replacement therapy for patients with end-stage chronic renal failure, has gained widespread recognition in the global field of organ transplantation. However, long-term prognosis management continues to pose significant challenges (Hariharan et al., 2021; Wolfe et al., 1999; Zhao et al., 2017). The core issue affecting the long-term survival of renal transplant recipients lies in the precise regulation of immunosuppressive therapy. The triple immunosuppressive regimen (calcineurin inhibitor [CNI], mycophenolic acid [MPA], and corticosteroids) recommended by The Transplantation Society (TTS) has become the established foundation for maintenance immunosuppression in clinical renal transplantation (KDIGO clinical practice guideline for the care of kidney transplant recipients, 2009). Especially, MPA-class agents play a pivotal role in preventing acute rejection through selective inhibition of T/B lymphocyte proliferation (Allison, 2005; Bhat et al., 2023; Behrend and Braun, 2005). Nevertheless, the clinical application of MPA-class drugs faces substantial challenges: narrow therapeutic window, significant interindividual pharmacokinetic variability, and high incidence of gastrointestinal toxicity (Bergan et al., 2021; Sobiak and Resztak, 2021). Currently, mycophenolate mofetil (MMF) and mycophenolate sodium (MPS), as two principal MPA prodrugs, exhibit distinct clinical profiles due to differences in pharmaceutical formulation despite sharing the same active metabolite (de Winter et al., 2008). The rational selection between these agents has emerged as a critical issue in optimizing immunosuppressive therapy.

Both MMF and MPS are prodrugs of MPA that require in vivo conversion to the active metabolite MPA. Their immunosuppressive effects are mediated through inhibition of purine synthesis in immune cells by blocking inosine monophosphate dehydrogenase enzyme activity (Tedesco-Silva et al., 2005). While sharing this common mechanism, their pharmacological divergence originates from distinct delivery systems (Budde et al., 2007a). MMF, an ester derivative produced as cost-effective conventional tablets, undergoes rapid hydrolysis to active MPA in the stomach and proximal small intestine with 94% bioavailability, yet exhibits significant first-pass metabolism and marked plasma concentration fluctuations (Jacqz-Aigrain et al., 2000; Tang et al., 2017). In contrast, MPS exists as a sodium salt of MPA. The enteric-coated MPS (EC-MPS) with delayed-release technology targeting the ileum’s alkaline environment, circumvents gastric acid degradation and minimizes direct mucosal irritation, thereby achieving enhanced gastrointestinal tolerability and more stable pharmacokinetic profiles (Sobiak et al., 2021; Wang et al., 2022). Clinically, MMF excels in rapid onset and cost-effectiveness, whereas EC-MPS demonstrates superior tolerability and pharmacokinetic stability. This pharmacological advantage positions EC-MPS as the preferred option for sensitive populations, including patients with diabetic gastroenteropathy, elderly recipients, and pediatric transplant cohorts, particularly given the frequent gastrointestinal complications observed with MMF in renal transplantation practice (Gabardi et al., 2003).

The area under the concentration-time curve from 0 to 12 h (AUC_0–12h) is conventionally used to evaluate MPA exposure, with an internationally accepted therapeutic target range of 30–60 mg h·L^-1 in renal transplant recipients (Chen et al., 2019; Tett et al., 2011). While phase III clinical trials have established therapeutic equivalence between EC-MPS (720 mg twice daily) and MMF (1,000 mg twice daily) in terms of efficacy and safety, significant pharmacokinetic disparities persist. EC-MPS demonstrates higher pre-dose trough concentrations, lower peak concentrations, and prolonged time to maximum concentration compared to MMF, alongside greater absorption-phase variability evidenced (Salvadori et al., 2004; Graff et al., 2016; Budde et al., 2007b). Crucially, EC-MPS exhibits interindividual variability in MPA exposure, rendering standardized dosing regimens suboptimal as evidenced by 20% of patients exceeding the therapeutic window (>60 mg h·L^-1) and 35% failing to achieve threshold exposure (<30 mg h·L^-1) under empirical dose adjustment (Budde et al., 2007b; Collins et al., 2020; Kiang and Ensom, 2018). These pharmacodynamic complexities, compounded by the narrow therapeutic index of MPA-based immunosuppressants, mandate the implementation of therapeutic drug monitoring (TDM) as a cornerstone of precision dosing strategies.

The substantial pharmacokinetic variability and definitive concentration-effect correlation of EC-MPS constitute the most compelling rationale for implementing TDM to guide AUC-based dose individualization. However, conventional TDM approaches face practical limitations, particularly the clinical infeasibility of intensive sampling protocols in transplant populations (Hougardy et al., 2016). To address this, limited sampling strategies integrating population pharmacokinetic (popPK) modeling with Bayesian forecasting have been advocated as a pragmatic solution for optimizing EC-MPS dosing regimens (Wang et al., 2022; Fromage et al., 2025). As a superior alternative to classical pharmacokinetic methods, popPK analysis enables precise quantification of inter- and intra-individual variability through sparse sampling, while facilitating identification of clinically significant covariates influencing drug exposure (Duffull et al., 2011). Nevertheless, the external validity of such models may be compromised by center-specific factors including study design heterogeneity, sample size limitations, and analytical platform discrepancies (El Hassani and Marsot, 2023). Rigorous external validation using independent multicenter datasets is therefore mandated prior to clinical implementation across diverse healthcare settings, ensuring robust model generalizability and therapeutic reliability.

Despite the development of numerous popPK models over the past 20 years to characterize the pharmacokinetics of EC-MPS, the external applicability of these models across multicenter settings remains inadequately validated. Systematic evaluation of model transferability not only addresses this knowledge gap but also facilitates identification of center-specific covariates impacting predictive performance. Furthermore, strategic selection of optimal popPK models from existing literature—rather than conducting de novo modeling—may represent a resource-efficient approach for personalized dosing guidance. To address these imperatives, this study implemented a comprehensive validation framework using independent datasets to assess the predictive capacity of published popPK models for EC-MPS within triple immunosuppressive regimens in adult renal transplant recipients.

As the two principal prodrugs of MPA, MMF and MPS demonstrate distinct clinical profiles despite sharing identical active metabolites, the difference attributable to their distinct prodrug structures and pharmaceutical formulations (Tedesco-Silva et al., 2005; Sobiak et al., 2021). To date, popPK models for MMF in renal transplant recipients have been extensively developed and externally validated (Zhang et al., 2019). In contrast, limited popPK research exists for MPS, with its external validity remaining poorly characterized. To our knowledge, this study represents the first comprehensive external validation of published MPS popPK models using an independent clinical dataset. While the validation dataset originated from a single center and had limited sample size, this investigation provides critical insights to advance research on MPS popPK characteristics.

This study evaluated the predictive performance of four published MPS popPK models using TDM data from our institutional cohort. Results demonstrated deficient predictive accuracy of existing models, with all failing to meet validation standards. Both population prediction analyses through goodness-of-fit diagnostics and prediction error testing revealed poor agreement between predicted and observed values. Individual predictions showed marginally better performance for Model 1, achieving composite precision indices (IF₂₀ > 35% and IF₃₀ > 50%), though the DV-IPRED scatter plots exhibited systematic bias and reduced overlap in high-concentration ranges due to data dispersion. In addition, the results of the VPC test based on model simulation show that the fitting results of Model 1 is relatively good, the distribution of the observed data and the simulation data is relatively close. Besides, for majority models, the observed concentration value data that fall within the confidence interval of the simulated data are relatively few, the deviation is large, and the prediction effect is poor.

The observed discrepancies between published model predictions and measured concentrations in our cohort may stem from clinical variables and methodological considerations, including interindividual pathophysiological heterogeneity, analytical variability in MPA quantification methodologies and the differential selection of pharmacokinetic modeling approaches.

Demographic characteristics, particularly race and age, emerge as significant covariates influencing MPA pharmacokinetic PK parameters, with sample size additionally affecting predictive accuracy. Variability in these key determinants across published models, including racial composition, age distribution, and cohort size, may introduce bias in popPK parameter estimation, ultimately compromising model predictive performance (Catić-Đorđević et al., 2021; Tornatore et al., 2022; Lestini et al., 2015). Comparative analysis reveals: Model 1 comprise 2,309 TDM samples from multi-regional Caucasian adults (mean age: 45 years), Model 2 comprise 232 TDM samples from American adults (mean age: 46 years), Model 3 comprise 90 TDM samples from Serbian adults (mean age: 42.97 years) and Model 4 comprise 384 TDM samples from Chinese pediatric patients (mean age: 13.3 years). Our external validation cohort comprised 186 Chinese adult TDM samples (mean age: 39.29 years). The enhanced predictive accuracy of Model 1 likely reflects its larger sample size providing greater statistical power. Notably, the complexity of race-specific metabolic variations may be the reason for the poor prediction of Model 2 and Model 3. And Model 4’s suboptimal performance may stem from fundamental PK differences between pediatric and adult populations.

Notably, the published models incorporated different analytical methods for detecting MPA blood concentration, including HPLC, HPLC/UV, and EMIT, while the external cohort dataset of our hospital adopts HPLC-MS. Variations in the accuracy and precision across these detection methodologies may introduce variability that could compromise the model’s predictive performance and stability in concentration estimation.

The selection of structural models, computational platforms, and significant covariates differs among published popPK analyses, potentially introducing variability in parameter estimation and influencing the predictive accuracy of the final population pharmacokinetic model. Research demonstrates that structural model specification directly affects derived pharmacokinetic parameters (Yu et al., 2017). Among the evaluated models, Models 1–3 implemented a two-compartment disposition model, whereas Model 4 utilized a single-compartment approximation. Given MPA’s biphasic elimination characteristics and tissue distribution profile, the two-compartment configuration appears physiologically more plausible, potentially explaining Model 4’s substantial prediction error and inflated variability estimates (de Winter et al., 2009). The enteric-coated formulation of MPS employs pH-dependent release kinetics targeting ileal absorption, thereby avoiding gastric degradation and minimizing mucosal irritation (Wang et al., 2022). This pharmacokinetic profile necessitates explicit incorporation of absorption lag-time (ALAG) in modeling (de Winter et al., 2008). The published models exhibit significant ALAG variations: Model 1 (∼2 h), Model 2 (not incorporated), Model 3 (0.21 h), and Model 4 (8.45 h). Model 1’s superior goodness-of-fit likely stems from its appropriate ALAG parameterization, which aligns with the drug’s known gastrointestinal transit dynamics.

Overall, this investigation utilized TDM data for MPA following MPS administration, collected through routine clinical monitoring protocols, thereby providing an objective representation of real-world pharmacological variability. The dataset encompassed trough, intermediate, and peak concentrations measured at standardized sampling intervals, enabling systematic evaluation of published models’ capacity to characterize pharmacokinetic profiles across critical temporal phases. The published models exhibit significant variability and unsatisfactory predictive performance, indicating that TDM remains an essential requirement for the clinical application of MPS. To advance individualized medication for MPS based on popPK, future research must prioritize the investigation of potential covariates. This will enable identification of key factors influencing MPS model predictability and facilitate the development of a popPK model suitable for patients in our hospital.