1 Introduction
Tacrolimus is the most commonly used immunosuppressant to prevent acute rejection following renal transplantation [
1‐
4]. Because of its huge medical impact, tacrolimus was chosen by scientists as one of the five molecules to take to a remote island [
5]. Nonetheless, prolonged use of tacrolimus leads to substantial toxicity, including increased rates of infection, post-transplant diabetes mellitus, nephrotoxicity, neurotoxicity, hypertension, and gastrointestinal disturbances [
6‐
9]. These adverse events contribute to the limited long-term patient and kidney allograft survival and patient non-adherence [
10,
11]. Adverse events seem to be related to higher tacrolimus concentrations, whereas rejection rates seem to be related to lower concentrations [
12,
13]. It is thus important to reach the tacrolimus target concentration as soon as possible to limit the risk of rejection and reduce toxicity [
12,
14].
Tacrolimus is a critical dose drug with a narrow therapeutic index and large intra- and interpatient variability, for which therapeutic drug monitoring (TDM) is routinely performed [
13]. Many factors, including age [
15,
16], bodyweight [
17‐
19], cytochrome P450 (CYP) 3A genotype [
17,
19], drug–drug interactions [
20,
21], ethnicity [
22,
23], and hematocrit [
17,
19,
24,
25] influence the pharmacokinetics of tacrolimus. Contrary to adults, most published pediatric population pharmacokinetic (PK) models have included either bodyweight or age as a significant covariate influencing clearance (CL) [
21,
26]. To reach the target range, children aged younger than 5 years require higher weight-normalized tacrolimus doses than older children [
15]. Currently, in clinical practice and at first steady state, only 30% of patients are within the target range. Two thirds of children have a concentration outside the target range, 63.5% having subtherapeutic concentrations, and 6.5% have supratherapeutic concentrations [
17]. In daily practice, the starting dose is often based solely on bodyweight, subsequent doses are adjusted using TDM, which limits the time a patient is exposed to concentrations outside the target range, but it can still take up to 3 weeks before target concentrations are reached [
17].
The use of a population PK model may help in predicting an individual’s tacrolimus exposure and can be applied before the start of therapy. Recently, our group developed a dosing algorithm to predict the right tacrolimus starting dose in pediatric renal transplant recipients [
17]. In this model, the starting dose is based on bodyweight, CYP3A5 genotype, and donor status (living vs. deceased). The model was extensively validated, both internally (bootstrap analysis, visual predictive check [VPC] and normalized prediction distribution errors) and externally (VPC) in an independent cohort consisting of 23 pediatric renal transplant recipients.
Here, we report the results of a prospective clinical trial in pediatric renal transplant recipients in which the tacrolimus starting dose was based on this dosing algorithm [
17]. The aim of this trial was to determine if basing the starting dose of tacrolimus on the validated dosing algorithm leads to a higher proportion of patients reaching the tacrolimus target pre-dose concentration (
C0) range (10–15 ng/mL) at day 3 after transplantation. The number of children we planned to include was 28 and an interim analysis was planned after the inclusion of 16 children. The interim analysis demonstrated that the algorithm did not adequately predict the tacrolimus exposure and therefore the trial was stopped. Subsequently, a new and improved dosing algorithm was developed in a cohort in which the total number of included children was doubled compared to the original cohort.
4 Discussion
In this prospective trial, 31% (five out of 16) of the children had a tacrolimus C0 within the target range on day 3 following transplantation when prescribed a tacrolimus starting dose based on the original dosing algorithm. Two children (12.5%) had a markedly subtherapeutic tacrolimus C0 (< 5 ng/mL) on day 3, and 3 (19%) had a markedly supratherapeutic (> 20 ng/mL) tacrolimus C0. The original algorithm performed worse than anticipated and was comparable to the weight-based prescription, and therefore the trial was ended prematurely.
Contrary to what we expected, only 31% of patients had a tacrolimus
C0 within the target range. When the tacrolimus starting dose is only based on bodyweight in children, 30% is on target on day 3 [
17]. For the model to be a meaningful addition to standard bodyweight-based dosing, our estimate was that at least 55% would have to be on target. Based on our results, it seems that basing the starting dose on the original dosing algorithm does not increase the percentage of patients on target and does not reduce the number of extreme high and low tacrolimus
C0 compared with standard bodyweight-based dosing.
There is compelling evidence that CYP3A5 expressers require a 1.5- to 2-fold higher tacrolimus dose than non-expressers [
15,
17‐
19,
30‐
36]. A randomized clinical trial of age- and genotype-guided tacrolimus dosing in children concluded that CYP3A5 genotype-guided dosing stratified by age resulted in earlier attainment of therapeutic tacrolimus concentrations and fewer out-of-range concentrations [
37]. Clinical outcomes were not studied. Two large clinical trials in adults studied whether basing the tacrolimus dose on CYP3A5 would lead to more patients within the target
C0 range. Both studies concluded that optimization of the initial tacrolimus dose using CYP3A5 genetic testing does not improve clinical outcomes when TDM is performed [
38,
39]. As the variability in CL is not solely based on CYP3A5, basing the starting dose on a dosing algorithm including clinical, genetic, and demographic factors seemed the sensible next step.
On day 3 following transplantation, five patients were on target. As they were all CYP3A5 non-expressers who received a kidney from a living donor, they were all prescribed a dosage of approximately 0.3 mg/kg/day. This is the same dose as the standard bodyweight-based dose according to the package leaflet [
40]. These children would have been on target regardless if they participated in this trial.
Six children had a subtherapeutic tacrolimus C0 on day 3. All of these children received a dose between 0.29 and 0.54 mg/kg/day, which is equal to or higher than the standard bodyweight-based dose of 0.3 mg/kg. This suggests that if they had not participated in the trial and received a standard bodyweight-based dose, these children would also all have had a subtherapeutic tacrolimus exposure. Of these children, one was a CYP3A5 expresser and one received a kidney from a deceased donor. The doses were calculated correctly. It seems other factors not included in the original algorithm increased tacrolimus CL in these patients.
In three patients, the tacrolimus dose was reduced on days 1–2 following transplantation owing to a high tacrolimus
C0. If these doses had not been reduced, these patients would have likely had toxic tacrolimus
C0 on day 3. These patients all received a kidney from a deceased donor and were CYP3A5 expressers. It seems that the original dosing algorithm overestimates the CL of tacrolimus in this group, and therefore overestimates the required tacrolimus dose. To our knowledge, no other publication has found a relationship between donor type and tacrolimus CL. As tacrolimus undergoes hepatic metabolism, a higher tacrolimus CL in kidneys from a deceased donor seems highly unlikely. All patients received the same immunosuppressive protocol, no patients experienced delayed graft function. Dialysis prior to transplantation and the number of human leukocyte antigen mismatches were not significantly associated with the tacrolimus CL. The higher tacrolimus CL in kidneys from a deceased donor is probably caused by other unknown parameters that could not be tested as covariates and therefore cannot be corrected for. The cohort in which the model was developed [
17] consisted of 46 children of whom only two were
CYP3A5 expressers and received a kidney from a deceased donor. It seems that there was insufficient power in the cohort in which the model was developed to determine adequate tacrolimus exposure predictions in this specific subgroup.
After improving the original dosing algorithm, CYP3A5 expressers required a 1.4-fold higher tacrolimus dose than CYP3A5 non-expressers in the improved dosing algorithm. This is in line with previous research mentioned above. As approximately 70–80% of tacrolimus is distributed in erythrocytes, low hematocrit reduces the whole-blood concentration of tacrolimus [
41]. In this study, we concluded that patients with a higher hematocrit had a lower CL/
F. Previous research substantiates these findings [
17,
19,
31,
35,
36,
42‐
45]. Hematocrit did not influence the starting dose and was therefore not included in the dosing algorithm. Patients with higher serum creatinine levels had a decreased CL/
F. Tacrolimus undergoes hepatic elimination and almost no renal elimination, thus the explanation for this observation remains unclear. Some studies have reported a correlation between creatinine and tacrolimus CL [
31,
46,
47], whereas others found no such effect [
48‐
50]. Four decades previously, Sheiner et al. concluded that forecasting a concentration based on covariates does not improve accuracy and precision as much as one previous concentration [
51]. However, when predicting the optimal starting dose, no previous concentrations are available. In this first exploratory study, we chose to focus on the starting dose with a historic cohort. For a new study, it will be interesting to adjust the subsequent doses using the dosing algorithm in combination with the previous concentration rather than just TDM.
The main difference between the improved PK model and the original model used to determine the tacrolimus starting dose in this trial is that donor status was no longer a significant covariate on CL. It remains unclear why a recipient of a kidney from a deceased donor would have a higher tacrolimus CL. The other difference between the two models is that allometric scaling was coded with an estimated exponent on CL and CYP3A5 expressers should receive a 1.46-fold higher dose in the improved algorithm compared with 1.82 in the original. Simulations of the improved model showed better description of the data compared with the previously published model, as shown in the VPC (Fig. S3 of the ESM). It will be interesting to see if the improved model is able to adequately predict the tacrolimus exposure when used in clinical practice. We are currently planning a new prospective clinical trial using this new and improved algorithm.
The main strength of this study is that this is the first attempt at predicting the optimal starting dose of tacrolimus in children in clinical practice with a dosing algorithm. Many PK models that have been published in the literature were developed retrospectively but were not tested in prospective studies. This study has demonstrated that even though on paper the algorithm was validated extensively and performed well, in clinical practice, it was simply inadequate. As the old proverb says, the proof of the pudding is in the eating. A second strong point is that because of the chosen methodology, a limited sample size was sufficient to answer the research question. The final strength of this study is that an improved starting dose model was developed and designed for clinicians, making it easy to use the dosing algorithm in clinical practice.
The main limitation of this study is that in the PK model building cohort two different analytical techniques were used: immunoassay and liquid chromatography–tandem mass spectrometry. However, to solve this issue, this difference was built into the residual error model. We chose not to exclude the immunoassay concentrations as they were included in the original model. Furthermore, the relatively large proportion of Caucasian patients in our center is a limitation as this may not reflect pediatric transplant populations worldwide.