Home | About | Journals | Submit | Contact Us | Français |

**|**Br J Clin Pharmacol**|**v.81(4); 2016 April**|**PMC4799926

Formats

Article sections

- Abstract
- What is Already known about this Subject
- What this Study Adds
- Introduction
- Methods
- Results
- Discussion
- Competing Interests
- References

Authors

Related links

Br J Clin Pharmacol. 2016 April; 81(4): 742–752.

Published online 2016 February 5. doi: 10.1111/bcp.12851

PMCID: PMC4799926

Min Dong,^{
1
} Patrick T. McGann,^{
2
,}^{
3
} Tomoyuki Mizuno,^{
1
} Russell E. Ware,^{
2
,}^{
3
} and Alexander A. Vinks^{}^{
1
,}^{
3
}

Professor Alexander A. Vinks, PharmD, PhD, Division of Clinical Pharmacology, Cincinnati Children's Hospital Medical Center, 3333 Burnet Ave, MLC 6018, Cincinnati, OH 45229‐3039, USA.

Tel.: +1 513 636 5833Fax: +1 513 636 7247

E‐mail: gro.cmhcc@skniv.rednas,

Received 2015 July 2; Revised 2015 November 19; Accepted 2015 November 23.

Copyright © 2015 The British Pharmacological Society

Hydroxyurea has emerged as the primary disease‐modifying therapy for patients with sickle cell anaemia (SCA). The laboratory and clinical benefits of hydroxyurea are optimal at maximum tolerated dose (MTD), but the current empirical dose escalation process often takes up to 12 months. The purpose of this study was to develop a pharmacokinetic‐guided dosing strategy to reduce the time required to reach hydroxyurea MTD in children with SCA.

Pharmacokinetic (PK) data from the HUSTLE trial (NCT00305175) were used to develop a population PK model using non‐linear mixed effects modelling (nonmem 7.2). A D‐optimal sampling strategy was developed to estimate individual PK and hydroxyurea exposure (area under the concentration–time curve (AUC)). The initial AUC target was derived from HUSTLE clinical data and defined as the mean AUC at MTD.

PK profiles were best described by a one compartment with Michaelis–Menten elimination and a transit absorption model. Body weight and cystatin C were identified as significant predictors of hydroxyurea clearance. The following clinically feasible sampling times are included in a new prospective protocol: pre‐dose (baseline), 15–20 min, 50–60 min and 3 h after an initial 20 mg kg^{–1} oral dose. The mean target AUC(0,∞) for initial dose titration was 115 mg l^{–1} h.

We developed a PK model‐based individualized dosing strategy for the prospective Therapeutic Response Evaluation and Adherence Trial (TREAT, ClinicalTrials.gov NCT02286154). This approach has the potential to optimize the dose titration of hydroxyurea therapy for children with SCA, such that the clinical benefits at MTD are achieved more quickly.

- There is substantial interpatient variability in the pharmacokinetics (PK) of hydroxyurea, the primary therapeutic drug for the treatment of sickle cell anaemia (SCA).
- Currently, hydroxyurea is initiated at a relatively low dose and slowly escalated to maximum tolerated dose (MTD) to reach a goal of mild myelosuppression. Using this dose escalation strategy, MTD is reached in 6–12 months.
- The clinical and laboratory benefits of hydroxyurea are maximal at MTD, but the prolonged dose escalation process significantly delays achieving the full benefits of hydroxyurea therapy.

- A PK‐guided approach was developed to rapidly tailor individual dose for hydroxyurea in paediatric patients with SCA. This strategy will utilize a patient's own PK profile with the goal of significantly shortening the time to reach MTD.
- D‐optimal design identified that three plasma concentrations collected at 15–20 min, 50–60 min and 3 h after oral administration would effectively estimate systemic hydroxyurea exposure.
- This study, for the first time, identified cystatin C as a significant predictor for the clearance of hydroxyurea, suggesting that cystatin C might be a more sensitive marker than estimated creatinine clearance to guide dose adjustment in SCA patients.

Sickle cell anaemia (SCA) is a global epidemic disorder affecting millions of people worldwide with an estimated 300000 affected infants born annually 1. SCA is a chronic and serious medical condition with many clinical complications, including frequent severe painful vaso‐occlusive crises, stroke, chronic organ damage and premature mortality. There is a strong body of evidence demonstrating that long term daily oral hydroxyurea (hydroxycarbamide) treatment significantly improves the outcome in both adults 2, 3, 4 and children 5, 6, 7, 8. Whereas hydroxyurea was initially reserved only for patients with frequent serious clinical complications, a recent expert panel has recommended that hydroxyurea now be offered to all affected children 9 months or older, regardless of their clinical manifestations 9.

The main therapeutic effect of hydroxyurea is to increase fetal haemoglobin (HbF) levels, which have been associated with decreased frequency of painful vaso‐occlusive crises, decreased number of erythrocyte transfusions and hospitalizations and reduction in mortality 4, 5, 10. When hydroxyurea is escalated to a maximum tolerated dose (MTD, defined as mild bone marrow suppression), higher HbF levels are achieved and the clinical benefits are optimized 11, 12. The hydroxyurea MTD is highly variable in SCA patients and ranges from 14.2 to 35.5 mg kg^{–1} day^{–1} in children 13. This broad range in MTD is at least partially due to the substantial inter‐patient variability in hydroxyurea pharmacokinetics (PK) and prevents the ability to use a standard effective dose for all SCA patients. In adult patients, hydroxyurea exposure varies approximately five‐fold for a given dosing regimen 14, whereas in paediatric patients differences in exposure can vary 2–3 fold at the same mg kg^{–1} dose 13.

The traditional dosing strategy for hydroxyurea involves a titration process with all patients starting at the same empirical dose of 15–20 mg kg^{–1} day^{–1}, followed by fixed incremental increases every 4–8 weeks (by ~2.5–5 mg kg^{–1} day^{–1}) until mild myelosuppression (as defined by complete blood count and reticulocyte count) is achieved, indicating that MTD has been reached. Once the MTD is reached, patients will continue long term maintenance treatment at this dose, with adjustments as needed for weight gain or laboratory toxicities 9. This dose escalation process, however, requires frequent outpatient visits and laboratory tests and usually takes 6–12 months. In fact, in many busy clinical practices, aggressive dose escalation is not often performed. The time and effort involved in hydroxyurea dose escalation results in suboptimal dosing for many patients and limits the potential clinical benefits of hydroxyurea therapy 15. For these reasons, a personalized dose optimization process that can rapidly identify MTD for individual SCA patients is highly desirable. Ideally, an integrated pharmacokinetic–pharmacodynamic (PK/PD)‐guided dosing algorithm which takes both PK and PD variabilities into consideration would be optimal. However, until a well‐defined relationship between dose, drug exposure and outcome has been established, PK surrogate endpoints (such as AUC) could have merit for initial dose optimization.

The purpose of this study was to develop a paediatric population PK model for hydroxyurea in combination with Bayesian estimation to tailor the individual dose of hydroxyurea in paediatric SCA patients. Although a relationship between systemic exposure and its effects and/or side effects has not been demonstrated for hydroxyurea, we took the assumption that the systemic hydroxyurea drug exposure is directly correlated to clinical outcome and hypothesized that achieving an average effective AUC (derived from an earlier HUSTLE trial 13) as our initial target could have clinical benefit by potentially shortening the titration process. We also evaluated D‐optimal design to identify the most informative sampling times for Bayesian estimation and to reduce the number of observations required for robust estimation of hydroxyurea PK. These methodologies have been successfully applied by our group to clinically guide antimicrobial dosing 16, 17 and as part of concentration controlled clinical studies 18. The Bayesian approach has the potential to optimize the initial dosing of hydroxyurea such that the benefits of the MTD are achieved more quickly.

Data from the prospective HUSTLE clinical trial were used for the development of the population model 13. A total of 712 hydroxyurea plasma concentrations from 96 patients after initial treatment (day 1) were available for PK analysis. Among them, 63 patients had paired PK profiles after they had reached MTD. The original study protocol was approved by the St. Jude Children's Research Hospital institutional review boards (IRB) and all patients or parents/guardians provided written informed consent before enrolment in the study. All participating subjects with SCA were given a single oral 20 mg kg^{–1} hydroxyurea dose. Most patients were given a liquid hydroxyurea formulation to allow accurate dosing except for 10 patients early in the study, who received capsules for their day 1 dosing. Patients then followed a dose escalation protocol with periodic monitoring of toxicity until individual MTD was reached which required a minimum of 3 months of treatment to achieve a stable MTD. Most children achieved MTD between 6 and 9 months after initiating hydroxyurea therapy, based primarily on when they achieved the target levels of marrow suppression. Plasma samples were collected on the day of hydroxyurea treatment (day 1) and at MTD at the following times: pre‐dose and at 20 min, 40 min, 1, 2, 4, 6 and 8 h after oral hydroxyurea administration. Hydroxyurea concentrations were determined with a colorimetric technique as described previously 13. Demographic information and standard laboratory parameters were collected at baseline and throughout the dose escalation process. In addition, direct glomerular filtration rate (GFR) of the patients was measured by ^{99m}Tc‐DTPA plasma clearance 19. Serum creatinine was determined by a four step enzymatic method (Roche Diagnostics, Indianapolis, IN, USA) and the estimated creatinine clearance (CL_{cr}) was calculated by the Schwartz formula 20. Serum cystatin C was quantified by a turbidimetric method using latex particles which are coated with anti‐cystatin C (Roche Diagnostics, Indianapolis, IN, USA). Urine albumin measurement was performed using a turbidimetric method at Quest Diagnostics Laboratories (San Juan Capistrano, CA, USA). The percentage of fetal haemoglobin (%HbF) was measured with HPLC 13.

In order to evaluate the individual area under the concentration‐time curves (AUC) of hydroxyurea after the initial dose and at MTD, non‐compartmental PK analysis was performed using Phoenix WinNonlin software version 6.4 (Pharsight Corporation, Cary, NC, USA). AUC from the time of dosing extrapolated to infinity (AUC(0,∞)) was calculated using equation (1)

$$\text{AUC}\left(0,\infty \right)=\mathrm{A}\mathrm{U}\mathrm{C}\left(0,,,{t}_{\mathrm{last}}\right)+{C}_{\text{last}}/{\lambda}_{\mathrm{Z}}$$

(1)

where AUC(0,*t*
_{last}) is the area under the plasma concentration–time curve from time 0 to the last measurable concentration and was estimated by the linear log trapezoidal rule, *C*
_{last} is the last measured concentration and λ_{z} is the terminal rate constant which was estimated via linear regression of time *vs*. log concentration of the terminal portion of the curve. Although a clear relationship between systemic exposure and clinical outcome has not been established for hydroxyurea, it is assumed in our study that an effective AUC(0,∞) as identified in the HUSTLE trial 13 correlates with clinical outcome.

Population PK analysis was conducted using nonmem version 7.2.0 (ICON, Ellicott City, MD, USA). First order conditional estimation with interaction (FOCE‐I) was employed throughout to estimate simultaneously the typical population PK parameters, random effect of inter‐individual variability (IIV) and residual variability (RV). Model selection was based on the overall model performance including the goodness‐of‐fit plots, model stability and objective function values (OFV). The shrinkage for inter‐individual variability and residual errors was calculated for diagnostic assessment 21. Visualization of nonmem output was implemented by Xpose4 package in R (v 3.0.3.).

The IIV model was described as:

$${\mathrm{\theta}}_{\mathrm{i}}={\mathrm{\theta}}_{\mathrm{T}\mathrm{V}}\cdot \mathrm{e}\mathrm{x}\mathrm{p}\left({\eta}_{\mathrm{i}}\right)$$

(2)

where θ_{i} represents the value of the PK parameter θ for the i^{th} subject and θ_{TV} is the population mean of parameter *θ* in the structural model. The deviation of θ from the mean θ_{TV} was approximated with η_{i}, which was assumed to follow a normal distribution with a mean of 0 and a variance of ω^{2} (i.e. η_{i} ~ N[0, ω^{2}]).

The RV was described by a combined additive and proportional model, but other residual error models such as exponential, additive or proportional were also examined. The combined additive and proportional residual model was described as:

$${C}_{\mathrm{ij}}={\hat{C}}_{\mathrm{ij}}\cdot \left(1+{\epsilon}_{\mathrm{p}\mathrm{i}\mathrm{j}}\right)+{\epsilon}_{\mathrm{a}\mathrm{i}\mathrm{j}}$$

(3)

where *C*
_{ij} represents the j^{th} observed concentration in the i^{th} individual, *Ĉ*
_{ij} is the j^{th} model predicted concentration in the i^{th} individual, ε_{pij} and ε_{aij} are the proportional and additive residual random errors, respectively. ε_{pij} and ε_{aij} are assumed to be independently normally distributed with a mean of 0 and a variance of σ^{2} (i.e. ε_{pij} ~ N[0, σ_{p}
^{2}] and ε_{aij} ~ N[0, σ_{a}
^{2}]).

Patient demographics and clinical chemistry measurements were included for covariate analyses. Any covariates with more than 10% missing values were excluded from the analysis. Otherwise, the missing covariate was replaced with the median value in the population. The tested covariates were total body weight, body surface area (BSA), height, age, gender, race/ethnicity, serum creatinine, cystatin C (CysC), AST, ALT, baseline bilirubin level, blood urea nitrogen (BUN), urine albumin, estimated CL_{cr}, direct measured GFR and dose. The covariates that were not included from the analysis were white blood cell count (WBC), mean corpuscular volume (MCV), absolute neutrophil count (ANC), absolute reticulocyte count (ARC), erythropoietin (EPO), haemoglobin (Hb) and fetal haemoglobin (HbF).

The relationship between continuous covariates and the typical value of PK parameters were modelled using either normalized power or linear models as shown in equations 4 and 5, respectively.

$${\theta}_{\mathrm{T}\mathrm{V},\mathrm{i}\mathrm{j}}={\theta}_{\mathrm{T}\mathrm{V}}\cdot {{\left(\mathrm{C}\mathrm{O}{\mathrm{N}}_{\mathrm{i}\mathrm{j}}/\mathrm{C}\mathrm{O}{\mathrm{N}}_{\mathrm{r}\mathrm{e}\mathrm{f}}\right)}^{\theta}}^{\mathrm{\text{CON}}}$$

(4)

$${\theta}_{\mathrm{T}\mathrm{V},\mathrm{i}\mathrm{j}}={\mathrm{\theta}}_{\mathrm{T}\mathrm{V}}+{\theta}_{\mathrm{\text{CON}}}\cdot \left(\mathrm{C}\mathrm{O}{\mathrm{N}}_{\mathrm{i}\mathrm{j}}/\mathrm{C}\mathrm{O}{\mathrm{N}}_{\mathrm{r}\mathrm{e}\mathrm{f}}\right)$$

(5)

where θ_{TV,ij} represents the value of the PK parameter θ for the i^{th} subject on j^{th} observation, θ_{TV} is the population mean of parameter θ, CON_{ij} is the value of the continuous covariate CON in the i^{th} individual at j^{th} observation, and CON_{ref} is a reference value for the covariate and represents the median value of the population; θ_{CON} is the estimated parameters describing the magnitude of the covariate−parameter relationships.

The following equations were utilized to describe the categorical covariate–parameter relationship:

$${\theta}_{\mathrm{T}\mathrm{V},\mathrm{i}\mathrm{j}}={\theta}_{\mathrm{T}\mathrm{V}}\cdot {\theta}_{\mathrm{\text{CAT}}}{{}^{\mathrm{\text{CAT}}}}^{\mathrm{i}\mathrm{j}}$$

(6)

$${\theta}_{\mathrm{T}\mathrm{V},\mathrm{i}\mathrm{j}}={\theta}_{\mathrm{T}\mathrm{V}}+{\theta}_{\mathrm{\text{CAT}}}\cdot \mathrm{C}\mathrm{A}{\mathrm{T}}_{\mathrm{i}\mathrm{j}}$$

(7)

where CAT_{ij} is the value of the categorical covariate CAT in the i^{th} individual at j^{th} observation and θ_{CAT} is the estimated parameters describing the magnitude of the covariate−parameter relationships.

The stepwise forward addition and backward elimination approach was utilized for covariate analysis. The difference in the OFVs from nested models is assumed to be χ^{2} distributed. A drop of OFV of more than 6.63 (*P* < 0.01) in the forward selection and more than 10.83 (*P* < 0.001) in the backward elimination were used as selection criteria.

Non‐parametric bootstrap analysis was run with 1000 resampled datasets and the estimated medians and 95% confidence intervals (CIs) of parameter estimates were compared with the final model estimates. Models with more than 10% difference in the estimated medians or with 95% CI containing zero were rejected. The final model was also evaluated by prediction‐corrected visual predictive check (pcVPC) where the final model was utilized to simulate 1000 dataset and the real data observations were compared with the distribution of the simulated concentrations 22.

To visualize the effect of cystatin C, the hydroxyurea plasma concentration–time profiles were simulated with various cystatin C concentrations (0.5, 0.75, 1.25, 2.0 mg l^{–1}). A typical patient weighted 30 kg was selected and a starting dose of 20 mg kg^{–1} of hydroxyurea was given to the patients. The simulation was performed by nonmem 7.2 and was based on the final PK model. The simulated AUCs were calculated by the linear log trapezoidal rule.

To identify the optimal sampling times and number of samples required to estimate robustly individual PK parameters and total hydroxyurea exposure (AUC), a D‐optimal design analysis was performed using PopED 23. D‐optimal design implements an evaluation of the Fisher information matrix (FIM) for population PK model and provides the most informative design for getting precise parameter estimates. Based on feedback from the clinical team related to feasibility of sample collection in our paediatric patients, we constrained the design according to the following criteria: 1) a maximum of three post‐dosing plasma samples to be collected from each child and 2) the last sampling time is within 3 h after drug administration. Because of the limited D‐optimal sampling strategy, a one compartment PK model with the three primary PK parameters (absorption, clearance and volume) was used for design purposes.

Patient characteristics on day 1 and at MTD are shown in Table 1. As previous described 13, hydroxyurea plasma concentrations showed substantial inter‐individual variability especially during the absorption phase. In patients who received the standard 20 mg kg^{–1} starting dose, NCA analysis showed that the AUC(0,∞) on day 1 ranged from 40.0 to 149.2 mg l^{–1} h^{–1}. The dose at MTD varied from 14.2 to 35.5 mg kg^{–1} day^{–1}. As shown in Figure 1, the AUC(0,∞) distribution at MTD was shifted from a mean value of 91.1 to 115.7 mg l^{–1} h^{–1} compared with day 1. In our study cohort, the mean ± (SD) of AUC(0,∞) at MTD was 115.7 ± 34.0 mg l^{–1} h^{–1} (range 61.6–317.8 mg l^{–1} h^{–1}). This mean of 115 mg l^{–1} h^{–1} was chosen as the hydroxyurea AUC to be initially targeted using the PK model based approach.

The distributions of AUC(0,∞) at day 1 (A) and at the MTD (B). The red dashed lines represent 10^{th}, 50^{th} and 90^{th} percentile of the data

During the model development process, a one compartment with first order elimination was first considered. However, hydroxyurea concentrations were under‐predicted at the early elimination phase and were over‐predicted during the later elimination phase. This bias could not be removed by adding an additional peripheral compartment. A capacity‐limited elimination at therapeutic concentrations of hydroxyurea was reported previously 24. Accordingly, a one compartment with Michaelis–Menten elimination was implemented to fit the data with the apparent clearance (CL/*F*) as described by equation (8):

$$\mathrm{C}\mathrm{L}/\mathit{F}={\mathrm{V}}_{\text{max}}/\left(\mathit{C}+{\mathit{K}}_{\mathrm{m}}\right)$$

(8)

where V_{max} is the maximum elimination rate, *K*
_{m} is the Michaelis–Menten constant and is corresponded to the concentration when the elimination rate is 50% of V_{max} and *C* is the plasma concentration. Interestingly, the bias seen with the first order elimination disappeared in this model and the OFV dropped by 60 points. However, the high correlation between *K*
_{m} and V_{max} values suggested that the data did not support estimation of all the parameters and therefore *K*
_{m} was fixed to 25 mg l^{–1} based on a previous report 24. As for the absorption phase, a transit absorption model provided better flexibility and a drop of >100 points in OFV compared with when a lag time was included in the model.

Of the tested covariates, body weight (WT) was the most significant covariate for both V_{max} (ΔOFV = −169) and the apparent volume of distribution (*V*/*F*, ΔOFV = −220). The model estimated exponent of WT was 0.73 on V_{max} and was 0.97 on *V*/*F* in a power model. These values were close to the theoretical 0.75 and 1.0 for rate related parameters and for volume, respectively 25, 26. We used fixed theoretical values to scale allometrically the influence of body size on hydroxyurea PK. Of other covariates examined, cystatin C and direct GFR were found to have statistically significant impacts on elimination parameter V_{max}, with a reduction in OFV of 24.5 and 19.5, respectively (*P* < 0.001), and the inter‐individual variability was reduced from 16.6% to 14.5% by cystatin C. Only cystatin C was retained in the final model after the backward elimination step. Interestingly, compared with an OFV drop of 24.5 by cystatin C, serum creatinine (SCr) and estimated creatinine clearance (CL_{cr}) only reduced OFV by 5.05 and 4.31, respectively (*P* > 0.01).

Figure 2 shows the goodness‐of‐fit plots for the final model. Data points were symmetrically distributed around the line of identity and the regression lines revealed no obvious bias. PK parameter estimates from the final model and from bootstrap analysis are listed in Table 2. The parameters were all estimated with good precision (low % standard error of estimation) and reasonable model stability. Visual predictive check presented a good agreement between the observed and simulated concentrations (Figure 3). The final PK estimates for primary parameters were V_{max}, 490 mg h^{–1} 70 kg^{–1}; *K*
_{m}, 25 mg l^{–1} (fixed); *V*/*F*, 49.6 l 70 kg^{–1} and *K*
_{a}, 8.19 h^{−1}, for a typical patient with cystatin C level of 0.74 mg l^{–1}.

Goodness‐of‐fit plots for the final PK model. (A) Population prediction *vs.* observed concentration. (B) Individual prediction *vs.* observed concentration. (C) Conditional weighted residuals (CWRES) *vs*. population prediction. (D) Conditional **...**

Prediction‐corrected visual predictive check (pcVPC) of the final model. Open circles, observed plasma concentration; dashed red lines, lower (2.5^{th}) and upper (97.5^{th}) percentiles of the observed data; solid red line, 50^{th} percentile of the observed **...**

To visualize further the impact of cystatin C on overall hydroxyurea exposure, simulations were conducted using the final model with cystatin C concentrations of 0.5, 0.75, 1.25 and 2.0 mg l^{–1}. The first three concentrations correspond to the lowest, median and highest cystatin C concentrations reported in our study cohort. The highest cystatin C value of 2.0 mg l^{–1} represents a level associated with kidney injury. As shown in Figure 4, the elimination of hydroxyurea significantly reduced with increasing cystatin C concentration, and the predicted AUC was doubled from 72.7 to 148.6 mg l^{–1} h^{–1} as cystatin C concentrations increased from 0.5 to 2.0 mg l^{–1} in a typical paediatric patient weighted 30 kg. Compared with a normal average cystatin C concentration of 0.75 mg l^{–1}, AUC increased by 65.7% at a cystatin C concentration of 2.0 mg l^{–1} (from 89.7 to 148.6 mg l^{–1} h).

Model predicted hydroxyurea PK profiles (A) and AUCs (B) with various cystatin C concentrations in a typical paediatric patient. , Cystatin C=0.50 mg l^{−1}; , Cystatin C=0.75 mg l^{−1}; , Cystatin C=1.25 **...**

The optimal sampling analysis was performed using *a priori* hydroxyurea PK parameter estimates as determined using a pragmatic one compartment PK model with first order elimination and absorption. The PK parameter estimates applied in the optimal design are CL/*F*, 12.9 l h^{–1}; *V*/*F*, 44.5 l; *K*
_{a}, 7.11 h^{−1} and lag time 0.193 h. We found that the AUC(0,∞) estimated using parameters from this one compartment model can accurately estimate the AUC(0,∞) from the NCA analysis as shown in Figure 5 (*r*
^{2} = 0.928). Optimal design using the D‐criterion revealed that within the design constraints the optimal sampling times were 15–20 min, 50–60 min and 3 h after dose. Expected relative standard error (RSE) was below 30% for all the PK parameters and was less than 10% for hydroxyurea clearance with 20 patients under this design.

This manuscript lays out a PK‐guided dosing strategy that has great potential to individualize therapy and significantly reduce the time to reach hydroxyurea MTD for SCA patients. This strategy requires only three plasma samples to be collected at optimally designed times, 15–20 min, 50–60 min and 3 h after an initial dose, and from which drug exposure can be estimated based on prior PK information using Bayes theorem. The model will use individual PK data to produce an optimal dose with an initial target AUC(0,∞) of 115 mg l^{–1} h. A clinical software platform such as MWPharm would be optimum to implement the Bayesian estimation 27.

MTD doses were greatly variable between individuals and this variability in response to hydroxyurea has been related to both PK and PD 28. However, given that hydroxyurea exposure varied approximately five‐fold among patients as in the recent study by Paule *et al.*
14, we hypothesized that achieving an average effective AUC as our initial target could have clinical benefit by potentially shortening the titration process. We do not expect all the patients to be at their MTD at the proposed initial AUC target, as a relationship between hydroxyurea exposure and toxicity has not been clearly demonstrated yet. However, it would be useful to better control drug exposure (and PK variability) while then evaluating hydroxyurea response. By controlling systemic drug exposure, the relationship between hydroxyurea dose, systemic exposure, bone marrow suppression, effect on HbF and clinical outcomes, including safety, can be further investigated.

Currently, there is no recommended target hydroxyurea exposure for treating SCA. Actually reports of the AUCs of hydroxyurea at either initial dose or MTD dose are scant. In adult SCA patients with normal renal function (*n* = 6), the mean AUC(0,∞) after 15 mg kg^{–1} of hydroxyurea was 82.5 mg l^{–1} h 29. In another study of children and adolescents (*n* = 21) who either had MTD dose (25–35 mg kg^{–1}) or initial starting dose (15–20 mg kg^{–1}), the combined mean AUC(0,∞) was 102 mg l^{–1} h 30, but no separate mean AUC for MTD dose was reported. Our proposed 115 mg l^{–1} h is based on prospective data collected from 63 children aged 1.9–16.5 years old who achieved their empirical MTD dose. Caution should be given when applying this target AUC to patients less than 2 years old, as their renal and hepatic organ functions are still immature 31. In addition, the original MTD data were collected after carefully tailoring the individual dose over a period of months. Adjusting the dose much quicker is likely to coincide with a different relationship between dose, systemic exposure to the drug and the development of mild bone marrow suppression. Further studies are warranted to validate this target value under a much quicker dose titration protocol and in different populations, especially in very young patients.

In our study, capacity‐limited elimination was observed and was best described by the Michaelis–Menten equation. Non‐linear kinetics of hydroxyurea have been reported in several human and animal studies especially when administered with high doses 11, 24, 32, 33. For example, a PK model with parallel Michaelis–Menten metabolism and first order renal excretion was described in cancer patients 24. However, two recent studies found that linear kinetics appropriately described hydroxyurea PK in the treatment of SCA. Wiczling *et al*. found that the elimination of hydroxyurea was appropriately defined by first order kinetics with parallel renal and non‐renal clearance in a paediatric population 30. In adult SCA patients, the PK profiles of hydroxyurea were described by first order elimination with a two compartmental model 14. In our analysis, linear kinetics would lead to systemic bias where applying Michaelis–Menten kinetics eliminated the bias. However, as shown in Figure 5, a linear one compartment model also reasonably predicted the drug exposure. We believe that the non‐linear kinetics are only exhibited in a subset of the patients and are not completely distinguishable from linear PK in our data. In our study, the maximum clearance estimated by V_{max}/*K*
_{m} was 19.56 l h^{–1} 70 kg^{–1}, which is consistent with the reported CL/*F* of 11.6 l h^{–1} 70 kg^{–1} in adults 14 and 15.6 l h^{–1} 70 kg^{–1} in children 30. The estimated *V*/*F* of 49.6 l 70 kg^{–1} is also very similar with 45.3 l 70 kg^{–1} in adults 14 and 49.7 l 70 kg^{–1} in children as previously reported 30.

The goal of this study was to develop a Bayesian adaptive control strategy that can accurately estimate individual systemic exposure to help individualize hydroxyurea doses while using clinically feasible sparse sampling. Despite the fact that the developed population PK model best described the data, it requires six PK parameters to be estimated and may be difficult to implement in clinical practice. A simple one compartment model with first order elimination was used for optimal design to locate the most informative time points for AUC estimation. This one compartment model provided reasonable estimation for the total hydroxyurea exposure, especially when AUC was 80–120 mg l^{–1} h which our initially targeted AUC fell into (Figure 5). The final optimal design suggested time points for Bayesian estimation of hydroxyurea exposure were: 15–20 min, 50–60 min and 3 h after oral dose. Since most of the patients did not have samples collected at these three sampling times, we cannot perform an internal validation for our design. However, optimal design showed good prediction precision with RSE < 10% for clearance with a sample size of 20. In an earlier study, Wizcling *et al.* proposed to collect one plasma sample at 1.5 h which was internally validated using a linear regression approach (*r*
^{2} = 0.76) 30. We will prospectively evaluate our sampling strategy in a prospective hydroxyurea clinical trial, which aims to use individual PK data to tailor hydroxyurea doses. The hypothesis of this study is that this PK‐based model will result in reduced time to MTD compared to the standard empirical approach.

Our results are consistent with a previous study that identified body weight as the most significant covariate of hydroxyurea clearance and volume 30. When a power function was used to describe the relationship between body weight and PK parameters, the estimated scaling factors were 0.73 and 0.97 for V_{max} and *V*/*F*, respectively, which were very similar to the mechanistic based theoretical exponent values of 0.75 and 1. In this large study with a wide range in age and body weight, the theoretical allometric exponents can be closely identified and it is rational to apply the fixed exponent of 0.75 and 1 in our model.

Our study, for the first time, identified serum cystatin C, but not serum creatinine, as a significant predictor of hydroxyurea clearance. Currently, serum creatinine is still the most commonly used biomarker in routine clinical practice to diagnose acute kidney injury (AKI). However, its predictive performance is far from ideal, due to several drawbacks such as delayed diagnosis and due to the impact of factors unrelated to renal function. Great efforts have been made to search for alternative biomarkers for AKI diagnosis. One of the most promising and increasingly utilized biomarkers is serum cystatin C concentration, which purely relies on the filtration capacity and is more sensitive to change in kidney function. In a recent systemic review, Brou *et al.* showed that cystatin C was better correlated with drug clearance than creatinine in over 80% of 16 studied renally excreted drugs 34. However, their conclusion was drawn mostly from adult studies and the authors admitted that more evidence should be collected from children. It is believed that hydroxyurea is primarily eliminated by the kidney 35. In our study, simulation predicted hydroxyurea exposure increased by 65.7% with 2 mg l^{–1} of cystatin C compared with a normal level of cystatin C (0.75 mg l^{–1}), suggesting that cystatin C would be a better marker than creatinine clearance to guide dose adjustment for SCA patients with AKI or other renal dysfunction, and adding more evidence of the superiority of cystatin C over creatinine for predicting kidney function in children.

In conclusion, we have successfully developed a population PK model of hydroxyurea in children with SCA and identified body weight and serum cystatin C as significant predictors of hydroxyurea clearance. A Bayesian dosing strategy was proposed which will assist with the individualization of hydroxyurea dose and could help to investigate further the relationship between dose, systemic exposure, pharmacogenomics, pharmacodynamic end points including toxicity and effect on HbF. The results of the prospective TREAT study, which has already started enrolment, will test the validity of our Bayesian model and potentially shorten the time to reach hydroxyurea MTD.

All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare no support from any organization for the submitted work, no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years and no other relationships or activities that could appear to have influenced the submitted work.

The authors acknowledge funding supports from NIH grant 5T32HD069054 (MD) and from the Japan Research Foundation for Clinical Pharmacology and the Uehara Memorial Foundation (TM). The HUSTLE trial was supported by R01‐HL090941 from NHLBI (REW). The authors thank Dr. Alex Sparreboom for initial HUSTLE analyses.

Dong M., McGann P. T., Mizuno T., Ware R. E., and Vinks A. A. (2016) Development of a pharmacokinetic‐guided dose individualization strategy for hydroxyurea treatment in children with sickle cell anaemia. Br J Clin Pharmacol, 81: 742–752. doi: 10.1111/bcp.12851.

1.
Piel FB, Patil AP, Howes RE, Nyangiri OA, Gething PW, Dewi M, Temperley WH, Williams TN, Weatherall DJ, Hay SI. Global epidemiology of sickle haemoglobin in neonates: a contemporary geostatistical model‐based map and population estimates. Lancet
2013; 381: 142–51. [PubMed]

2.
Steinberg MH, McCarthy WF, Castro O, Ballas SK, Armstrong FD, Smith W, Ataga K, Swerdlow P, Kutlar A, DeCastro L, Waclawiw MA. Investigators of the Multicenter Study of Hydroxyurea in Sickle Cell A, Follow‐Up MSHP. The risks and benefits of long‐term use of hydroxyurea in sickle cell anemia: A 17.5 year follow‐up. Am J Hematol
2010; 85: 403–8. [PubMed]

3.
Charache S, Terrin ML, Moore RD, Dover GJ, Barton FB, Eckert SV, McMahon RP, Bonds DR. Effect of hydroxyurea on the frequency of painful crises in sickle cell anemia. Investigators of the Multicenter Study of Hydroxyurea in Sickle Cell Anemia. N Engl J Med
1995; 332: 1317–22. [PubMed]

4.
Voskaridou E, Christoulas D, Bilalis A, Plata E, Varvagiannis K, Stamatopoulos G, Sinopoulou K, Balassopoulou A, Loukopoulos D, Terpos E. The effect of prolonged administration of hydroxyurea on morbidity and mortality in adult patients with sickle cell syndromes: results of a 17‐year, single‐center trial (LaSHS). Blood
2010; 115: 2354–63. [PubMed]

5.
Wang WC, Ware RE, Miller ST, Iyer RV, Casella JF, Minniti CP, Rana S, Thornburg CD, Rogers ZR, Kalpatthi RV, Barredo JC, Brown RC, Sarnaik SA, Howard TH, Wynn LW, Kutlar A, Armstrong FD, Files BA, Goldsmith JC, Waclawiw MA, Huang X, Thompson BW, Investigators BABYHUG. Hydroxycarbamide in very young children with sickle‐cell anaemia: a multicentre, randomised, controlled trial (BABY HUG). Lancet
2011; 377: 1663–72. [PubMed]

6.
Lobo CL, Pinto JF, Nascimento EM, Moura PG, Cardoso GP, Hankins JS. The effect of hydroxcarbamide therapy on survival of children with sickle cell disease. Br J Haematol
2013; 161: 852–60. [PubMed]

7.
Hankins JS, Aygun B, Nottage K, Thornburg C, Smeltzer MP, Ware RE, Wang WC. From infancy to adolescence: fifteen years of continuous treatment with hydroxyurea in sickle cell anemia. Medicine
2014; 93: e215. [PubMed]

8.
Svarch E, Machin S, Nieves RM, Mancia de Reyes AG, Navarrete M, Rodriguez H. Hydroxyurea treatment in children with sickle cell anemia in Central America and the Caribbean countries. Pediatr Blood Cancer
2006; 47: 111–2. [PubMed]

9.
Yawn BP, Buchanan GR, Afenyi‐Annan AN, Ballas SK, Hassell KL, James AH, Jordan L, Lanzkron SM, Lottenberg R, Savage WJ, Tanabe PJ, Ware RE, Murad MH, Goldsmith JC, Ortiz E, Fulwood R, Horton A, John‐Sowah J. Management of sickle cell disease: summary of the 2014 evidence‐based report by expert panel members. JAMA
2014; 312: 1033–48. [PubMed]

10.
Powars DR, Weiss JN, Chan LS, Schroeder WA. Is there a threshold level of fetal hemoglobin that ameliorates morbidity in sickle cell anemia?
Blood
1984; 63: 921–6. [PubMed]

11.
Charache S, Dover GJ, Moore RD, Eckert S, Ballas SK, Koshy M, Milner PF, Orringer EP, Phillips G Jr, Platt OS, Thomas GH. Hydroxyurea: effects on hemoglobin F production in patients with sickle cell anemia. Blood
1992; 79: 2555–65. [PubMed]

12.
Ware RE, Eggleston B, Redding‐Lallinger R, Wang WC, Smith‐Whitley K, Daeschner C, Gee B, Styles LA, Helms RW, Kinney TR, Ohene‐Frempong K. Predictors of fetal hemoglobin response in children with sickle cell anemia receiving hydroxyurea therapy. Blood
2002; 99: 10–4. [PubMed]

13.
Ware RE, Despotovic JM, Mortier NA, Flanagan JM, He J, Smeltzer MP, Kimble AC, Aygun B, Wu S, Howard T, Sparreboom A. Pharmacokinetics, pharmacodynamics, and pharmacogenetics of hydroxyurea treatment for children with sickle cell anemia. Blood
2011; 118: 4985–91. [PubMed]

14.
Paule I, Sassi H, Habibi A, Pham KP, Bachir D, Galacteros F, Girard P, Hulin A, Tod M. Population pharmacokinetics and pharmacodynamics of hydroxyurea in sickle cell anemia patients, a basis for optimizing the dosing regimen. Orphanet J Rare Dis
2011; 6: 30. [PubMed]

15.
Ware RE. How I use hydroxyurea to treat young patients with sickle cell anemia. Blood
2010; 115: 5300–11. [PubMed]

16.
Hahn A, Frenck RW Jr, Zou Y, Vinks AA. Validation of a pediatric population pharmacokinetic model for vancomycin. Ther Drug Monit
2015; 37: 413–6. [PubMed]

17.
van Lent‐Evers NA, Mathot RA, Geus WP, van Hout BA, Vinks AA. Impact of goal‐oriented and model‐based clinical pharmacokinetic dosing of aminoglycosides on clinical outcome: a cost‐effectiveness analysis. Ther Drug Monit
1999; 21: 63–73. [PubMed]

18.
Scott JR, Courter JD, Saldana SN, Widemann BC, Fisher M, Weiss B, Perentesis J, Vinks AA. Population pharmacokinetics of sirolimus in pediatric patients with neurofibromatosis type 1. Ther Drug Monit
2013; 35: 332–7. [PubMed]

19.
Aygun B, Mortier NA, Smeltzer MP, Shulkin BL, Hankins JS, Ware RE. Hydroxyurea treatment decreases glomerular hyperfiltration in children with sickle cell anemia. Am J Hematol
2013; 88: 116–9. [PubMed]

20.
Schwartz GJ, Haycock GB, Edelmann CM Jr, Spitzer A. A simple estimate of glomerular filtration rate in children derived from body length and plasma creatinine. Pediatrics
1976; 58: 259–63. [PubMed]

21.
Karlsson MO, Savic RM. Diagnosing model diagnostics. Clin Pharmacol Ther
2007; 82: 17–20. [PubMed]

22.
Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Prediction‐corrected visual predictive checks for diagnosing nonlinear mixed‐effects models. AAPS J
2011; 13: 143–51. [PubMed]

23.
Nyberg J, Ueckert S, Stromberg EA, Hennig S, Karlsson MO, Hooker AC. PopED: an extended, parallelized, nonlinear mixed effects models optimal design tool. Comput Methods Programs Biomed
2012; 108: 789–805. [PubMed]

24.
Tracewell WG, Trump DL, Vaughan WP, Smith DC, Gwilt PR. Population pharmacokinetics of hydroxyurea in cancer patients. Cancer Chemother Pharmacol
1995; 35: 417–22. [PubMed]

25.
Anderson BJ, Holford NH. Mechanistic basis of using body size and maturation to predict clearance in humans. Drug Metab Pharmacokinet
2009; 24: 25–36. [PubMed]

26.
Vinks AA. Important role of population pharmacokinetic/pharmacodynamic modeling in pediatric therapeutics. J Pediatr
2011; 159: 361–3. [PubMed]

27.
Proost JH, Meijer DK. MW/Pharm, an integrated software package for drug dosage regimen calculation and therapeutic drug monitoring. Comput Biol Med
1992; 22: 155–63. [PubMed]

28.
Aarons L, Ogungbenro K. Optimal design of pharmacokinetic studies. Basic Clin Pharmacol Toxicol
2010; 106: 250–5. [PubMed]

29.
Yan JH, Ataga K, Kaul S, Olson JS, Grasela DM, Gothelf S, Kutlar A, Orringer E. The influence of renal function on hydroxyurea pharmacokinetics in adults with sickle cell disease. J Clin Pharmacol
2005; 45: 434–45. [PubMed]

30.
Wiczling P, Liem RI, Panepinto JA, Garg U, Abdel‐Rahman SM, Kearns GL, Neville KA. Population pharmacokinetics of hydroxyurea for children and adolescents with sickle cell disease. J Clin Pharmacol
2014; 54: 1016–22. [PubMed]

31.
Anderson BJ, Holford NH. Tips and traps analyzing pediatric PK data. Paediatr Anaesth
2011; 21: 222–37. [PubMed]

32.
Belt RJ, Haas CD, Kennedy J, Taylor S. Studies of hydroxyurea administered by continuous infusion: toxicity, pharmacokinetics, and cell synchronization. Cancer
1980; 46: 455–62. [PubMed]

33.
Navarra P, Del Carmine R, Ciabattoni G, D'Amato M, Ragazzoni E, Vacca M, Volpe AR, Preziosi P. Hydroxyurea: relationship between toxicity and centrally‐induced adrenal activation. Pharmacol Toxicol
1990; 67: 209–15. [PubMed]

34.
Brou NA, Jacqz‐Aigrain E, Zhao W. Cystatin C as a potential biomarker for dosing of renally excreted drugs. Br J Clin Pharmacol
2015; 80: 20–7. [PubMed]

35.
Gwilt PR, Tracewell WG. Pharmacokinetics and pharmacodynamics of hydroxyurea. Clin Pharmacokinet
1998; 34: 347–58. [PubMed]

Articles from British Journal of Clinical Pharmacology are provided here courtesy of **British Pharmacological Society**

PubMed Central Canada is a service of the Canadian Institutes of Health Research (CIHR) working in partnership with the National Research Council's national science library in cooperation with the National Center for Biotechnology Information at the U.S. National Library of Medicine(NCBI/NLM). It includes content provided to the PubMed Central International archive by participating publishers. |