Tacrolimus is a calcineurin inhibitor largely used for the prevention of rejection in solid organ transplantation. However, it is characterized by a narrow therapeutic index and a large interindividual variability rendering its therapeutic drug monitoring mandatory. Two main markers are currently available to adjust the individual dose: the trough blood concentration (C0) largely used for practical reasons and the area under the curve (AUC) that represents overall exposure and is probably better correlated with tacrolimus efficacy than the C0.1 However, AUC is more difficult to assess than C0, especially when not using population pharmacokinetic models.2 This led us to launch in 2005 the Immunosuppressant Bayesian dose Adjustment (ISBA) website (https://abis.chu-limoges.fr) for transplant professionals, where the interdose AUC of immunosuppressants is assessed using maximum a posteriori Bayesian estimation on the basis of 3 blood concentrations and a few individual characteristics (age, posttransplantation period, drug assay). For tacrolimus in renal transplantation, the best sampling times are 0, 1, and 3 h postdosing3,4 (except for the melt dose formulation for which samples at 0, 8, and 12 h are required5), with ±50% flexibility. Actually, one of the main advantages of maximum a posteriori Bayesian estimation is that it is quite flexible regarding the actual sampling times and provides, in addition to the AUC, the modeled concentration-time curve and 1 or a range of recommended dose(s) to reach the recommended AUC targets.1,6,7 We also recently developed a machine-learning algorithm accurately estimating the interdose AUC for once-daily and twice-daily tacrolimus formulations based on only 2 samples, which should simplify AUC estimation even further.8 We previously proposed to use the AUC/C0 ratio to calculate individual C0 targets based on the desired AUC range. Actually, we found that AUC/C0 was stable during transplant follow-up contrary to AUC and C0, even in the early period after transplantation.7,9 Our hypothesis is that individualized C0 targets are better surrogates of the AUC and using them rather than statistical C0 ranges to dose-adjust patients will improve outcome. The objective of the present study was to analyze the intraindividual, time-dependent variations of AUC/C0 as compared with raw and dose-adjusted AUC and C0 in a large data set of kidney transplant patients.
MATERIALS AND METHODS
We extracted from the ISBA electronic database, authorized by the French Committee for Informatics and Liberty, CNIL (no. 1619537), fully anonymized data (data were completely deidentified in accordance with the European General Data Protection Regulation regulation) from requests posted between November 1, 2007, and the November 1, 2019, for kidney transplant recipients, whatever the drug formulation or patient age. The database includes the transplantation date, request date, patient age at request, tacrolimus dose and drug formulation (Prograf [PRO], Advagraf [ADV], and Envarsus [ENV]), the time elapsed between transplantation and the request, Bayesian estimates of the AUC and Cmax, and observed C0 values. Information about gender, ethnicity, or genetic was not available except for a small subset of patients included in the multicenter cohorts Epigren/Ephegren.10 Both adults and children were included in this analysis.
Pretreatment and Data Cleaning
A filter was applied to select patients with at least 2 tacrolimus AUC estimation requests. We multiplied AUC0-12h for twice-daily tacrolimus by 2 to be roughly comparable with the AUC0-24h for once-daily formulations (although we are aware that the morning and evening AUCs are not similar9). Some factual errors were made when the requests were filled in, leading us to apply other filters: posttransplantation time was censored at 33 y (as the earliest date of transplantation available in the database was 1986), age at 100 y, AUC at 800 µg·h/L, dose at 40 mg, and time between 2 consecutive requests at 5 y. Finally, filters were applied on AUC and C0 with the removal of values out of the 95% confidence interval (CI) of the distribution.
Variations of Metrics
Intraindividual Relative Variations
The relative variations of different metrics (C0, C0/dose, AUC, AUC/dose, AUC/C0) between 2 consecutive requests were calculated as Vn – Vn
1 (V being the value of the metric at request n or n
– 1). The CIs for these metrics were calculated (corresponding to the interindividual variability of the intraindividual variability). These values were then split depending on the time elapsed between the 2 requests or the posttransplant period (both categorized as <3, 3–12, and >12 mo) and the Pearson correlation between AUC and C0 was analyzed using scatter plots and boxplots.
Variations Between Tacrolimus Formulations
We performed similar analyses as those earlier, but split by tacrolimus formulation (PRO, ADV, or ENV). In addition, we compared AUC/dose, C0/dose, and AUC/C0 between drug formulations using 2-way analysis of variance followed by Bonferroni-corrected Tuckey posttests.
The correlation between AUC and C0 was explored graphically, entirely, and then depending on the categorized posttransplant period or whether patients were adults or not. An arbitrary acceptability interval of ±30% around the C0 values was drawn.
Finally, correlations between 2 consecutive visits or between visits n and n – 2 were investigated for AUC/C0, AUC, and C0 using scatter plots and the Pearson coefficient of correlation.
Development of an Algorithm to Estimate the Individualized C0 Target
An algorithm was developed to estimate the individual C0 target based on the AUC/C0 ratio for 2 scenarios: (1) if the patient has at least 3 prior AUC measurements, the individual C0 versus AUC regression line is drawn and the C0 value to reach any target AUC chosen is calculated directly using the linear regression equation, and (2) if it is the first AUC measurement for a given patient, the individual C0 target that allows reaching the AUC targeted is calculated using a cross-product (individual C0 × target AUC/ individual AUC). In both cases, the target AUC can be derived from one of the usual C0 targets using the population curve corresponding to the same period posttransplant (<3 mo, between 3 mo and 1 y, and >1 y) and age class (adult or pediatric). An example is provided for a virtual outlier patient.
Influence of CYP3A5 Status
A subset analysis was performed for patients also enrolled in the multicenter cohorts Epigren/Ephegren,10 for whom the CYP3A5 status was available (220 instances in 52 patients). Relative variations of C0, AUC, and AUC/C0 were investigated as function of CYP3A5 status (expressors *1/*3 or *1/*1 versus nonexpressors *3/*3) and of the time elapsed between 2 instances or the posttransplant period.
All analyses and graphs were performed in R using the Tidyverse framework.11 The code used for these analyses is provided in an HTML Rmd file (SDC, https://links.lww.com/TP/C622).
The database contained 4130 AUC estimation requests corresponding to 1369 different patients fulfilling the inclusion criteria, that is for whom we had at least 2 requests in the database. After applying the censoring rules defined earlier, the study data set contained 4084 instances from 1356 patients, and after deletion of the AUC and C0 outliers out of the 95% CI (AUC0-24h out of 80–430 mg·h/L and C0 out of 2.2–12.2 µg/L), 3827 instances from 1325 patients. This corresponded to 12.9% of the patients who had at least 1 AUC or C0 value removed. Histograms of the distribution of these variables after application of censoring rules are presented in the Rmd file (SDC, https://links.lww.com/TP/C622). The main characteristics of the study data set are presented in Table 1. The number of instances per patient ranged from 2 to 43.
TABLE 1. -
Description of the population and the subpopulation for influence of CYP3A5
Values all population
CYP3A5 subgroup analysis
|No. AUC estimation requests (“instances”)
|Date of transplantation (min; max)
||1986-11-22 ; 2019-08-03
|CYP3A5*3 nonexpressors n (%)
|CYP3A4*22 (TT) n (%)
|Age at request, y, median (IQR)
|Time posttransplantation, y, median (IQR)
|Diabetes n (%)
|Tacrolimus formulation n (%)
|Tacrolimus daily dose, mg, median (IQR)
|Observed C0, µg/L, median (IQR)
|Model estimated Cmax, µg/L, median (IQR)
|Model estimated AUC0-24h, µg·h/L, median (IQR)
|AUC0-24h dose, median (IQR)
|C0 dose, median (IQR)
|AUC C0, median (IQR)
|Time elapsed between 2 instances, y, median (IQR)
|Relative variation of AUC0-24h between 2 instances, %, median (IQR)
||–7 (–27 to 15)
||–0.04 (–0.26 to 0.14)
|Relative variation of AUC0-24h/dose between 2 instances, %, median (IQR)
||2 (–18 to 24)
|Relative variation of C0 between 2 instances, median (IQR)
||–7 (–26 to 19)
||–0.05 (–0.26 to 0.17)
|Relative variation of C0/dose between 2 instances, median (IQR)
||3 (–19 to 28)
|Relative variation of AUC0-24h/C0 between 2 instances, median (IQR)
||–1 (–13 to 12)
||–0.01 (–0.14 to 0.13)
|No. instances per patient, median (IQR)
AUC, area under the curve; C0, trough level; IQR, interquartile range; NA, not available; NS, not studied in the substudy.
Relative Variations of AUC/C0, Exposure, and Dose-adjusted Exposure Markers
The relative variations between 2 consecutive visits showed that the AUC/C0 ratio was associated with the lowest variability (Figures 1 and 2), whatever the time between 2 requests or posttransplantation period. Interestingly, the median variations of AUC/C0 do not change with the time elapsed between requests or the posttransplant period (Figure 1). The 95% CI relative fold change was (–43% to 44%) for AUC/C0, (–77% to 72%) for AUC, (–82% to 98%) for AUC/dose, (–81% to 80%) for C0, and (–94% to 117%) for C0/dose. The smoothed time plots show that there are large fluctuations of AUC or C0, and even more so of AUC/dose and C0/dose as the time posttransplantation increases, whereas the AUC/C0 ratio remains very stable, even in the early period (Figure 2).
Correlations Between AUC and C0 and Interval of Acceptability
The correlation between AUC and C0 was studied graphically at different posttransplant periods in adults or children. Considering <30% deviation from the regression line as acceptable (ie, that the same C0 range can be used in routine practice for patients within this area), we observed that quite a number of values (and patients) were out of these limits (Figure 3). Indeed, the proportion of patients out of the 30% around the C0 values were 30.5% (n = 141), 30% (n = 610), and 15% (n = 2614) in adult for the <3 mo, 3 mo to 1 y, and >1 y period, respectively, and 29.3% (n = 82), 25.5% (n = 145), and 23.4% (n = 235) in children for the <3 mo, 3 mo–1 y, and >1 y period, respectively.
Development of an Algorithm to Predict the Individualized C0 Ratio
An example is provided for a virtual, atypical patient with extremely low AUC values for standard C0 levels (of note, this patient was a real patient for whom the AUC0-12h was used instead of AUC0-24h). In the first scenario, we considered that only the first measurement was known (AUC0-24 = 104 µg·h/L and C0 = 6 µg/L at 2 y after transplantation). If a C0 target of 5 µg/L was chosen for this adult patient >1 y posttransplant, an AUC0-24h target of 185 µg·h/L could have been derived from the population regression line and equation (Figure 4). The personalized C0 was then calculated using the cross-product: measured C0 × target AUC/ measured AUC = 6 × 185/104 = 10.7 µg/L. A second measurement was drawn at 4 y posttransplant, leading to AUC/C0 = 79 µg·h/L/5.2 µg/L, which confirmed that this patient was an outlier. The second scenario could then have been used when the patient got a third value at 5.7 y posttransplant (AUC/C0 = 129 µg·h/L/8.8 µg/L). Linear regression of the individual values gave AUC = 19.4 + C0 × 12.7, which could have been used to extrapolate the individualized C0 target to reach the AUC target chosen (personalized C0 to reach 185 µg·h/L = 13.0 µg/L). This virtual case is presented in Figure 4 with 2 subsequent AUC measurements (AUC/C0 = 95 µg·h/L/6.8 µg/L and 54 µg·h/L/3.5 µg/L at 6 and 8 y posttransplant, respectively).
Influence of Tacrolimus Formulation on the Relative Fold Change in Exposure Markers
The relative variability of the different exposure markers was estimated for the different drug formulations using 858 intraindividual relative fold change values for ADV, 45 for ENV, and 2924 for PRO.
The relative fold change as a function of time between 2 instances or time posttransplantation split between the 3 drug formulations is presented in Figure S1 (SDC, https://links.lww.com/TP/C622).
No significant difference after Bonferroni correction was found between formulations for the AUC/C0 ratio (PRO and ADV, mean relative difference [95% CI], 1.37% [–0.6% to 3.4; P = 0.2489), ENV and ADV (–3.0 [–11% to 5%]; P = 0.7100), PRO and ENV (4% [–4% to 12%]; P = 0.4489). However, a significant difference was observed for AUC/dose and C0/dose between the 3 formulations (Rmd file, SDC, https://links.lww.com/TP/C622).
Correlation of Individual Metrics Between 2 Instances
The scatter plots of the different metrics between an instance and the previous or penultimate are presented in Figure 5. The correlation was best and more stable for the AUC/C0 ratio in the 2 situations (r = 0.37 and r = 0.36, respectively), as compared with C0 (r = 0.26 and r = 0.21, respectively) and AUC (r = 0.26 and 0.20, respectively).
Effect of CYP3A5 Status on Relative Variability
The description of the subpopulation concerned is provided in Table 1. There were 44 (84.6%) CYP3A5 nonexpressors, reflecting the prevalence in the population of European descent.12 The relative variations of C0, AUC, and AUC/C0 between 2 instances were not influenced by the CYP3A5 phenotype (Figure 6).
In this study of a large data set, we analyzed at the individual level the time variations (or intraindividual variability) of tacrolimus AUC/C0 ratio as compared with the usual, raw or dose-adjusted, exposure markers.
We confirmed that AUC/C0 is associated with the lowest interindividual variability of the relative fold change between 2 consecutive visits ( = intraindividual variability), whatever the time elapsed between measurements. This is in line with the results we previously obtained in a clinical trial9 and in a subset of the current data set also extracted from the ISBA website database and limited to patients treated with Advagraf.7 This is consistent with the proposal we previously made that the AUC/C0 ratio can be used to personalize the C0 target to each patient, because C0 is a surrogate of the AUC. Indeed, the correlation between AUC and C0 (Figure 3) shows that the current C0 targets, represented by a ±30% shaded area, are not accurate for all patients. We arbitrarily chose this 30% threshold to keep enough patients in the nonoutlier group. To illustrate this, for the adult groups, choosing a threshold of 20% would have led to 48.9%, 49%, and 34.5% out of the 20% interval around the regression line for <1 y, between 3 mo and 1 y, and >1 y, respectively, whereas a threshold of 30% led to 30.5%, 30%, and 15% out of the 30% interval around the regression line for <1 y, between 3 mo and 1 y, and >1 y, respectively.
This value of ±30% provides a fair estimation of the width of the usual target ranges: the target range 4–8 ng/mL represents ±33% around 6 ng/mL; 5–10 ng/mL represents ±33% around 7.5 ng/mL; and 8–12 ng/mL represents ±20% around 10 ng/mL. However, it is important to stress that this notion of an “acceptable deviation” is only based on statistics and that an individual at the lower end of a C0 range might have AUC at the upper end of the corresponding range, giving an AUC/C0 much farther apart from the population mean (which is 36.26 h). This figure also illustrates that, similar to previous reports,7 for a given AUC value, C0 varied from 3 to 4-fold between individuals, whereas for a given C0 value, AUC varied from 2 to 3-fold. This suggests that many more patients than those outside the ±30% “acceptable range” would benefit from personalized C0 targets. Increased AUC/C0 values could be because of many factors,13 including CYP3A5 expression (CYP3A5 expressors require an increased dose to reach the same trough concentration), food intake (decreasing the AUC), decreased hematocrit or serum albumin (decreasing total blood exposure), and drug-drug or drug-food interactions.
Contrary to what we expected, we could not show an effect of CYP3A5 expression on AUC/C0 values, but the number of patients with CYP3A5 status available was very low in comparison with the overall data set (3.9%, including only 8 expressors). We could not even assess the effect of the CYP3A4*22 genotype as all patients were homozygous CC carriers.
As this study shows that the individual AUC/C0 ratio is quite stable with the length of time elapsed between assessments and across different posttransplant periods, we propose that individualized C0 targets be derived from the AUC (statistical) targets and the (individual) AUC/C0 ratio averaged from at least 2 measurements. This value can then be kept as the reference until there are obvious pharmacokinetic changes (such as those induced by factors of intraindividual variation such as drug-drug interactions or bariatric surgery, for instance).
An algorithm has been developed for 2 different scenarios (at least 3 or <3 AUC estimates available) and will be implemented in the near future on our ISBA website (https://abis.chu-limoges.fr/). In the case of no prior measurement, the personalized C0 target to reach the AUC target chosen is calculated directly from the measured AUC and C0 using a cross-product. When at least 3 AUC estimates are available, the individual regression line is calculated and plotted and used to extrapolate the personalized C0 target to reach the chosen AUC target. However, this second scenario requires that the AUC and C0 are spread out over a rather wide range of values, reflecting changing tacrolimus doses, which is more likely when gathering AUC estimates from the early to stable posttransplant periods. Still, the real patient used for our virtual example exhibited C0 (between 3.5 and 6.8 µg/L) and AUC0-24 (between 108 and 258 µg·h/L) values in large ranges, although the samples were drawn between 2 and 8 y posttransplant. Interestingly, this patient’s AUC/C0 was very stable (values between 24 and 34 h). Finally, the strategy that can be recommended is that if the first AUC/C0 value is close to the corresponding population regression line, the “usual standard” C0 target can be used. If not, it is recommended to obtain a second AUC estimate to confirm an atypical behavior and avoid any artifact. As soon as a third AUC estimate is available, an individual regression line can be drawn and used to extrapolate any individualized C0 target. As mentioned earlier, the best large-scale structures with our Bayesian estimators were 0, 1, and 3 h after dose for immediate-release and prolonged-release tacrolimus (TAC), and 0, 8, and 12 h for Meltdose-TAC. This large-scale structure allows catching most types of absorption profiles. However, we never focused on patients with delayed absorption, such as patients with gastroparesis. We have made dose adjustments for such patients on our ISBA website, where we estimated very flat profiles, but in the absence of a rich pharmacokinetics (PK) profile, we could only assume that our AUC estimate was reliable.
Apart from the AUC/C0 ratio, the mean relative variations in C0, AUC, C0/dose, and AUC/dose over time do not show a linear increase with the time elapsed between instances, that is, decreasing correlation with time. Actually, up to 1 y (which corresponds to the 75th percentile of the distribution of the time difference between instances), AUC and C0 oscillate at around 5% variation, probably because of residual variability (including analytical variability and unobserved covariates such as drug-drug or food-drug interactions). However, the mean relative variation is the average of individual variations, which are subject to a very high interindividual variability. This translates into an almost doubled CI of the relative fold change for AUC or C0 than for AUC/C0. The linear time-evolution pattern of the dose-corrected metrics probably reflects the natural decrease in tacrolimus apparent clearance (Figure 2B), accompanied and followed by a dose decrease in the first 6 mo posttransplantation. The slow return to basal relative variation for instances between 6 mo to 1 y (Figure 2A) probably also reflects difficulties to achieve and maintain C0 or AUC in the target ranges.14 Basic principles of clinical pharmacokinetics may explain why the AUC/C0 ratio exhibits the lowest intraindividual variability. AUC and C0 both depend on the dose, systemic clearance, and oral bioavailability, and their ratio cancels out these factors, responsible for most of TAC variability.
For all metrics, correlation between instances was low but remained quite stable with the time elapsed between instances, the best correlation being for the AUC/C0 ratio.
Finally, the relative fold change depending on tacrolimus formulations must be interpreted carefully, given the low number of patients under ENV. However, it shows that the mean relative fold change of the AUC/C0 ratio was not significantly different between formulations, and it confirms the overall results showing that the AUC/C0 ratio is the less variable metrics between patients, whatever the formulation (Figure S1, SDC, https://links.lww.com/TP/C622).
Considering the high efficacy of the current standard of care, immunosuppressive therapy, and therapeutic drug monitoring, the clinical benefit of using the AUC/C0 ratio for individual tacrolimus dose adjustment cannot be taken for granted. However, large intraindividual variability in C0 is still associated with graft survival, acute rejection, de novo DSA appearance, chronic immunologic-mediated graft injury, and histologic lesions.15 Additionally, reduction of overimmunosuppression episodes especially in the early phase posttransplant still reduces the nephrotoxicity of immunosuppressive drugs16 and decreases the risk of posttransplantation diabetes mellitus17 and infraclinical BK virus reactivations.18 All of this knowledge suggests that there is still room for improvement and the individual C0 target strategy we propose may help toward this. It allows better reactivity than AUC-based dose adjustment and should be of particular interest in situations of large PK variability during short time periods, such as the introduction/removal of interacting drugs that modify tacrolimus oral bioavailability or clearance, hence tacrolimus AUC. Bayesian AUC estimation with our models is robust to such modifications, but the impact of drug-drug interactions on the AUC/C0 ratio has not been specifically investigated in this study.
This study has some limitations. First, the ISBA database contains no information about well-known tacrolimus PK variability factors (eg, food intake, drug-drug interactions). However, the goal of this study was only to describe the intraindividual variability of tacrolimus exposure and not to elucidate its determinants. Second, we cannot exclude that in some requests posted on our ISBA websites, a few users might have entered hypothetical data to test the system. However, as one of the selection criteria was to include patients with at least 2 visits, the probability that someone made 2 consecutive tests with the same patient ID is very low. In any case, such rarities would be diluted in the large volume of data available and should not modify the present results. Third, we made many subjective choices in data filtering or threshold selection, but we conducted a sensitivity analysis with all the AUC or C0 thresholds removed, which showed similar results despite numerical differences (Supplemental data, SDC, https://links.lww.com/TP/C622). Finally, the algorithm developed that will be implemented in the ISBA website is proposed for research purposes only until clinically validated.
In conclusion, in this study, we confirmed in a large number of kidney transplant patients on tacrolimus that the individual AUC/C0 ratio is stable over time and we propose to use it to individualize the C0 targets based on population AUC targets. An algorithm has been developed to this effect and will be implemented in the near future on our ISBA website (https://abis.chu-limoges.fr/).
The authors thank Ms Karen Poole for article editing. They thank the physicians, clinical pharmacologists, and patients for using the ISBA website.
1. Brunet M, van Gelder T, Åsberg A, et al. Therapeutic drug monitoring of tacrolimus-personalized therapy: second consensus report. Ther Drug Monit. 2019;41:261–307.
2. Woillard JB, Saint-Marcoux F, Debord J, et al. Pharmacokinetic models to assist the prescriber in choosing the best tacrolimus dose. Pharmacol Res. 2018;130:316–321.
3. Woillard JB, de Winter BCM, Kamar N, et al. Population pharmacokinetic model and Bayesian estimator for two tacrolimus formulations—twice daily Prograf and once daily Advagraf. Br J Clin Pharmacol. 2011;71:391–402.
4. Benkali K, Rostaing L, Premaud A, et al. Population pharmacokinetics and Bayesian estimation of tacrolimus exposure in renal transplant recipients on a new once-daily formulation. Clin Pharmacokinet. 2010;49:683–692.
5. Woillard JB, Debord J, Monchaud C, et al. Population pharmacokinetics and Bayesian estimators for refined dose adjustment of a new tacrolimus formulation in kidney and liver transplant patients. Clin Pharmacokinet. 2017;56:1491–1498.
6. Saint-Marcoux F, Woillard JB, Jurado C, et al. Lessons from routine dose adjustment of tacrolimus in renal transplant patients based on global exposure. Ther Drug Monit. 2013;35:322–327.
7. Marquet P, Bedu A, Monchaud C, et al. Pharmacokinetic therapeutic drug monitoring of Advagraf in more than 500 adult renal transplant patients, using an expert system online. Ther Drug Monit. 2018;40:285–291.
8. Woillard JB, Labriffe M, Debord J, et al. Tacrolimus exposure prediction using machine learning. Clin Pharmacol Ther. 2021;110:361–369.
9. Marquet P, Albano L, Woillard JB, et al. Comparative clinical trial of the variability factors of the exposure indices used for the drug monitoring of two tacrolimus formulations in kidney transplant recipients. Pharmacol Res. 2018;129:84–94.
10. Villeneuve C, Rousseau A, Rerolle JP, et al. Adherence profiles in kidney transplant patients: causes and consequences. Patient Educ Couns. 2020;103:189–198.
11. Wickham H, Averick M, Bryan J, et al. Welcome to the Tidyverse. J Open Source Soft. 2019;4:1686.
12. Kuehl P, Zhang J, Lin Y, et al. Sequence diversity in CYP3A promoters and characterization of the genetic basis of polymorphic CYP3A5 expression. Nat Genet. 2001;27:383–391.
13. Vanhove T, Annaert P, Kuypers DRJ. Clinical determinants of calcineurin inhibitor disposition: a mechanistic review. Drug Metab Rev. 2016;48:88–112.
14. Ekberg H, Mamelok RD, Pearson TC, et al. The challenge of achieving target drug concentrations in clinical trials: experience from the Symphony study. Transplantation. 2009;87:1360–1366.
15. Gonzales HM, McGillicuddy JW, Rohan V, et al. A comprehensive review of the impact of tacrolimus intrapatient variability on clinical outcomes in kidney transplantation. Am J Transplant. 2020;20:1969–1983.
16. Laskow DA, Vincenti F, Neylan JF, et al. An open-label, concentration-ranging trial of FK506 in primary kidney transplantation: a report of the United States Multicenter FK506 Kidney Transplant Group. Transplantation. 1996;62:900–905.
17. Maes BD, Kuypers D, Messiaen T, et al. Posttransplantation diabetes mellitus in FK-506-treated renal transplant recipients: analysis of incidence and risk factors. Transplantation. 2001;72:1655–1661.
18. Brennan DC, Agha I, Bohl DL, et al. Incidence of BK with tacrolimus versus cyclosporine and impact of preemptive immunosuppression reduction. Am J Transplant. 2005;5:582–594.