A Population Pharmacokinetic Model and Dosing Algorithm to Guide the Tacrolimus Starting and Follow-Up Dose in Living and Deceased Donor Kidney Transplant Recipients

Tacrolimus treatment after kidney transplantation is complicated by its narrow therapeutic range and its large inter- and intra-individual variability. The current dosing strategy in most transplant centers consists of a standard, bodyweight-based dose followed by therapeutic drug monitoring (TDM). However, bodyweight is a poor predictor for an individual’s dose requirement [1‐3]. Consequently, both over- and under-exposure to tacrolimus occur frequently in the critical early post-transplantation phase, which puts patients at an increased risk for toxicity and rejection, respectively. Furthermore, this may also lead to more TDM requests as more dose adjustments are required, which paradoxically increases intra-individual variation.

Theoretically, population pharmacokinetic (popPK) models, incorporating demographic and clinical patient characteristics that are associated with tacrolimus pharmacokinetics, can better predict an individual’s dose requirement than bodyweight alone. Therefore, multiple research groups have developed popPK models and dosing equations that incorporated such factors [2, 4‐24]. However, most of these popPK models were developed with data from one center, and therefore cannot be used in a different population or setting without validation.

Recently, we prospectively tested a tacrolimus starting dose algorithm (developed by our group) [16]. The algorithm successfully predicted the tacrolimus starting dose in adult recipients of a living donor kidney [25]. Using the algorithm, at first steady state (3 days after transplantation), 58% of the patients had a tacrolimus pre-dose concentration within the target concentration range, whereas this was as low as 37.4% in kidney transplant recipients receiving a standard bodyweight-based tacrolimus starting dose [25, 26]. Although these results indicate that algorithm-based tacrolimus dosing has the potential to minimize under- and over-exposure to this critical dose drug, there remains a considerable amount of unexplained variability in tacrolimus’ pharmacokinetics. Earlier dosing algorithms were often based on datasets limited to 50–300 patients. To identify factors with a smaller effect on tacrolimus pharmacokinetics, which can explain this unexplained variability, and to estimate effects with more precision, high statistical power and large datasets are required. Furthermore, our dosing algorithm was developed for recipients of a living donor kidney and therefore might be less suitable for patients who receive a deceased donor kidney transplant. Deceased donor kidneys constitute a large portion of all kidney transplants and since these patients are at higher risk of allograft loss, they might benefit even more from algorithm-based tacrolimus dosing compared with patients transplanted with a living donor kidney [27, 28].

The aim of this study was: (i) to identify factors that affect tacrolimus pharmacokinetics and can explain its inter- and intra-patient variability; and (ii) develop a popPK model that can be used to guide tacrolimus dosing in both living and deceased donor kidney transplant recipients.

In this large, international, multicenter study, a popPK model for tacrolimus was developed using data of 1180 living and deceased donor kidney transplant recipients. Factors that were significantly associated with tacrolimus pharmacokinetics and were included in the final model were: age (in years), serum creatinine (in µmol/L), CYP3A4 (*22 allele carriers versus non-carriers) and CYP3A5 genotype (*3/*3 versus *1/*3 versus *1/*1), height (in cm), and hematocrit. With these covariates, the popPK model explained 35% of the inter-individual variability in tacrolimus CL/F.

In previous pharmacokinetic studies, various covariates were found to be associated with tacrolimus, including age, albumin, ASAT, bodyweight, BSA, CYP3A4 and CYP3A5 genotype, serum creatinine, hematocrit, LBW, post-operative day, phase angle, and sex [23].

The present study confirms the well-known relationship between a patient’s CYP3A4 and CYP3A5 genotype and their tacrolimus dose requirement [5, 23, 35]. In contrast to previous models, the present model can better differentiate the effect on tacrolimus CL/F for both homozygous and heterozygous CYP3A5*1 allele carriers, which allows a more tailored dosing approach. CYP3A5 expressers with the CYPA5*1/*3 genotype required a 1.7-times higher tacrolimus dose, and patients with the CYP3A5*1/*1 genotype required a 2-times higher tacrolimus dose, than CYP3A5 non-expressers (having the CYP3A5*3/*3 genotype). Carriers of the CYP3A4*22 allele require 0.8-times the tacrolimus dose compared with non-carriers.

Lower age, serum creatinine, and hematocrit were associated with higher tacrolimus CL/F [16]. The association between age and tacrolimus CL/F (θ = − 0.309) is in line with previous studies (θ = − 0.43 to − 0.50, or dose × 1.24, if recipient age is 18–34 years, or dose − 0.205, if recipient is age ≥ 65 years) [5, 16, 23, 36, 37]. This association may be explained by age-dependent drug metabolism, but changes in body composition may also have a role in the altered pharmacokinetics [36, 38, 39]. Although multiple studies found an association between serum creatinine concentrations and tacrolimus CL/F [16, 40, 41], the explanation for the relationship between serum creatinine and tacrolimus CL/F remains unclear. Less than 1% of tacrolimus undergoes renal clearance, and therefore, serum creatinine concentrations are not likely to be directly correlated with renal clearance of tacrolimus. A possible explanation might be that creatinine is a surrogate for a patient’s health status, which in turn affects a patient’s body composition, and consequently tacrolimus pharmacokinetics [36]. The negative relationship between hematocrit and tacrolimus CL/F can be explained by the fact that tacrolimus is highly bound to erythrocytes. With lower hematocrit values, and thus lower erythrocyte counts, the unbound tacrolimus fraction in blood is larger. As this unbound fraction is available for clearance, this can explain the relationship between hematocrit and clearance.

The relationship between body composition parameters and tacrolimus pharmacokinetics differs greatly between studies, and it is yet unclear which parameter best predicts an individual’s dose requirement. In previous models fat mass, LBW, weight, and BSA have been associated with both CL/F and V1/F [16, 23, 42, 43], whereas in the current model, height was best correlated with tacrolimus CL/F. A previous study showed that obese patients are more likely to be overdosed, suggesting bodyweight is not the ideal body composition measure to guide tacrolimus dosing [1]. As height relates to the ideal bodyweight, this might be a more appropriate predictor for a patient’s volume of distribution and fraction of tacrolimus available for clearance. In a previous study, in which we measured the body composition of kidney transplant recipients using bio-impedance spectroscopy (instead of estimating the body composition parameters based on bodyweight and height), we found that the phase angle correlates best with tacrolimus pharmacokinetics [36]. The phase angle relates to body cell mass, membrane integrity, and hydration status and may therefore better correlate with a patient’s volume of distribution, clearance, and consequently, a patient’s dose requirement. However, this measure is not routinely measured in clinical practice and therefore not always available for dose prediction.

As compared with the population pharmacokinetic model developed by Andrews et al. [16], we did not find new covariates associated with the pharmacokinetics of tacrolimus despite the large dataset. Both models include the CYP3A4 and CYP3A5 genotypes, hematocrit, serum creatinine, and age as covariate factors associated with tacrolimus pharmacokinetics. However, an advantage of the present model is that it can differentiate between homozygous and heterozygous CYP3A5*1 allele carriers, which allows for a more tailored tacrolimus dosing approach. Another important advantage of the present model is that it was developed in kidney transplant recipients who received a kidney from either a living or a deceased donor. This makes model-based tacrolimus dosing available for a much larger group of patients, as in many centers the proportion of deceased donor kidney transplantations is > 50% of all transplantations. Also, the inclusion of multiple centers in this dataset reduces the risk of overfitting and increases the generalizability of the model. Moreover, the proportion of explained inter-individual variability on CL/F is larger in the present study compared with the study of Andrews et al. (35.0% versus 30.4% for the full model, and 33.5% versus 27.5% for the starting dose model). The total inter-individual variability for both the base model and the final model was larger in the present study, compared with the models of Andrews et al. (51.3% versus 46.3% for the base model, and 41.4% versus 38.6% for the final model). This might be explained by the fact that more patients from different populations were included in the present analysis. In conclusion, the present model has several advantages over the older model. However, before implementing a model in clinical practice we believe a model should be tested prospectively.

The pcVPCs of the full model show a slight underprediction of peak concentrations. This may be caused by the relatively large number of pre-dose concentrations. To minimize this effect, it was evaluated whether estimating K_a, V1, and V2, which are parameters that are best estimated with dense sampling, using only data from patients with at least one available concentration-versus-time curve, improved the model fit. Moreover, the pcVPCs stratified for the different centers showed a deviating fit for the data from Barcelona. These differences may be attributed to differences in clinical practice (such as differences in dosing regimens, target concentrations, or analysis methods) or, more likely, unknown population characteristics affecting tacrolimus pharmacokinetics (such as body composition, diet, or the gut microbiome).

One of the strengths of this study is the large number of included patients and tacrolimus concentrations (both pre-dose concentrations and samples throughout the dosing interval). This increases the statistical power and, theoretically, makes it possible to detect smaller associations between patient characteristics and tacrolimus pharmacokinetics, as well as provide more precise/reliable estimates. However, in this study we did not find more factors associated with the pharmacokinetics of tacrolimus than that were already known from previous models. An explanation could be that other covariates that are thought to be associated with tacrolimus pharmacokinetics [for example, the gut microbiome, drug interactions, a patient’s body composition [36, 44‐46]] were not included in the dataset used in this study, as these were not measured during routine clinical care, could not be reliably extracted from the electronic patient files, or were not included in the previous pharmacokinetic studies on which the present analysis was based. Therefore, it is recommended for future studies in kidney transplant recipients to consider these covariates and perform appropriate measurements.

Another strength of this study is the inclusion of kidney transplant recipients of both living and deceased donors, whereas in our previous model only kidney transplant recipients from living donors were included. Prior to this, it was not clear whether the type of kidney donor affects tacrolimus pharmacokinetics. As deceased donor kidneys constitute a large part of all kidney transplants, it is important to include these patients in pharmacokinetic studies. Theoretically, tacrolimus CL/F may be different in deceased donor kidney transplant recipients, since they have an increased risk of a lower kidney function, and serum creatinine has been associated with tacrolimus CL/F [16, 40, 41]. In this study, donor type was not significantly associated with tacrolimus CL/F. Moreover, kidney transplant recipients from different transplant centers in Europe were included, which makes it possible to extrapolate the results to other centers. This is important as a dosing algorithm may not fit another population than the one for which it was developed, which may be caused by underlying differences in factors that affect the pharmacokinetics of tacrolimus but are not yet identified and incorporated in the model [47]. Consequently, models should be extensively validated and prospectively tested in a specific population before using the model in clinical practice.

The main limitation of the study is its retrospective and observational design. Some of the data were collected as part of pharmacokinetic studies, whereas some of the data were retrospectively collected from electronic patient files. In the latter, the exact times of the tacrolimus dose ingestions and consequently, the exact time after dose of the tacrolimus concentrations were unknown. Therefore, we assumed that patients followed the instructions of their transplant physicians and that they took their recommended tacrolimus dose at the recommended time. To limit the effect of possible deviations from this recommendation, we checked the data for outliers before and during modeling. If adherence problems or mistakes in tacrolimus dosing were noted in the electronic patient files, the tacrolimus pre-dose concentration was corrected or excluded from the dataset. For the densely sampled curves the dose ingestions and time after dose was registered more strictly. The retrospective design also resulted in missing data. Some data was missing at random, as some factors were not collected as part of specific pharmacokinetic studies. This missingness is therefore related to a specific study population, but not to the outcome. The other missing data can be considered missing completely at random, as not all factors were measured as part of routine clinical practice, and, to the best of our knowledge, no systematic differences exist between participants with missing data and those with complete data. During pharmacokinetic modeling, we used accepted methods to deal with missing data, and covariate estimates were based on available data. Therefore, we believe the missing data will not introduce bias in the estimates.

In this large international multicenter study, a popPK model with its accompanying starting dose algorithm was developed using data of 1180 kidney transplant recipients who received kidneys from both living and deceased donors. Higher age, serum creatinine, and hematocrit, as well as lower height were associated with lower tacrolimus clearance. Tacrolimus clearance was 1.7–2 times higher for CYP3A5*1 allele carriers as compared with CYP3A5 non-expressers (*3/*3), and 0.8 times lower for CYP3A4*22 carriers as compared with CYP3A4*1/*1 individuals. Together, these covariates explained 35% of the variability in clearance.