Population pharmacokinetics of weekly docetaxel in patients with advanced cancer

Abstract

Aims

Previous pharmacokinetic studies of the 3-weekly regimen (100 mg m⁻² every 3 weeks) of docetaxel have shown that docetaxel clearance is affected by liver function, body surface area, age, serum α₁-acid glycoprotein and cytochrome P450 3A4 (CYP3A4) activity. However, the pharmacokinetics of a weekly docetaxel (40 mg m⁻² week⁻¹) schedule are not well characterized. The aims of this study were (a) to investigate the pharmacokinetics of docetaxel (40 mg m⁻² week⁻¹) using sparse concentration-time data collected from patients with advanced cancer and (b) to utilize a population pharmacokinetic approach to identify patient covariates that significantly influence the clearance of docetaxel when administered according to this regimen.

Methods

A two-compartment pharmacokinetic model was used to describe the docetaxel concentration–time data from 54 patients with advanced cancer. The mean population and individual posterior Bayesian estimates of docetaxel clearance were estimated using P-PHARM. The relationships between docetaxel clearance and 21 covariates were investigated. This included estimates of CYP3A4 function in each patient using the erythromycin breath test (1/t_max). Significant covariates were included into the final population pharmacokinetic model. Pharmacokinetic models were validated using a data splitting approach with a dataset consisting of 16 patients.

Results

Significant relationships were found between docetaxel clearance and 1/t_max (erythromycin breath test parameter) and several of the liver function enzymes and CL was best described by the equation; CL = 21.51 + 217 (1/t_max) − 0.13 (ALT). This final population pharmacokinetic model provided both precise and unbiased predictions of docetaxel concentrations in a validation group of patients and an estimate of the population mean (95% confidence interval) clearance of docetaxel was 30.13 l h⁻¹ (12.54, 46.04 l h⁻¹) with an intersubject variability 30%.

Conclusions

A population pharmacokinetic model has been developed and validated for weekly docetaxel (40 mg m⁻²) in patients with advanced cancer. These results indicate that CYP3A4 activity and hepatic function have an impact on the pharmacokinetics of docetaxel when administered weekly.

Introduction

Docetaxel (Taxotere®) is a semisynthetic member of the taxane family with significant activity in solid malignacies, including breast, nonsmall cell lung, prostate, ovarian and head and neck cancers [1]. The most frequently used schedule of docetaxel is 100 mg m⁻² administered every 3 weeks. However, in recent years, a more dose-intensive weekly regimen using doses ranging from 35 to 40 mg m⁻² week⁻¹, has been introduced. This regimen is well tolerated in the advanced cancer population and shows similar if not superior response rates to the 3-weekly schedule [2–6]. The profile of toxicity is different from that of the 3-weekly regimen with a reduced incidence of neutropaenia and mucositis. However, both regimens show wide variability in both efficacy and toxicity.

Large scale pharmacokinetic/pharmacodynamic studies have previously been reported for the 3-weekly regimen of docetaxel [7–9]. These used population modelling approaches, which allow for the evaluation of interindividual variability in the pharmacokinetics and pharmacodynamics in large numbers of patients. Similar pharmacokinetic/pharmacodynamic studies have not to date been conducted with weekly docetaxel.

Using data from several phase II studies of 3-weekly docetaxel (100 mg m⁻² every 3 weeks), Bruno et al. estimated the total body clearance (CL) to be 36.3 l h⁻¹ with a 95% confidence interval of 17.5 l h⁻¹ to 59.3 l h⁻¹[8]. Inter-patient variability in CL of a similar magnitude has been observed in other pharmacokinetic studies of docetaxel [10–13] and attempts have been made to identify the factors responsible. In particular, Bruno et al. demonstrated total CL to be related to serum α₁-acid glycoprotein concentration, hepatic function tests, body surface area and age. The inclusion of these factors as covariates in a population pharmacokinetic model reduced the interpatient variability of CL values by 34% to a final interpatient coefficient of variation of 33%. Of these factors, α₁-acid glycoprotein and hepatic function were found to be most important. The influence of α₁-acid glycoprotein on CL can be rationalized in terms of protein binding as this protein accounts for high but variable binding of docetaxel in plasma [14]. Protein binding is a particularly important factor in determining CL for those drugs that have low to moderate hepatic extraction fractions. The clearance of such drugs is also influenced by the intrinsic clearance in the liver. In humans, docetaxel is metabolized by CYP3A4 [15] and correlations between CL and CYP3A4 activity have been sought [13, 16]. One of the most widely used phenotyping assays for CYP3A4 activity is the erythromycin breath test [17]. Hirth et al. found that baseline erythromycin breath test results (as determined by the percentage of ¹⁴C exhaled h⁻¹) were significantly correlated with docetaxel CL [13]. The erythromycin breath test was more predictive of docetaxel CL than the covariates identified by Bruno et al. including α₁-acid glycoprotein, alanine transferase (ALT) and alkaline phosphatase (ALP). The importance of CYP3A4 activity was confirmed by Yamamoto et al. using an alternative phenotyping method, namely the 24 h urinary recovery of 6β-hydroxycortisol after exogenous cortisol administration, although α₁-acid glycoprotein, AST and age were also identified as important covariates [16].

The pharmacokinetics and pharmacodynamics of weekly docetaxel have not been extensively investigated and most of the data are only preliminary [18–21]. However, given that the pharmacokinetics of 3-weekly docetaxel are dose-proportional [22], it is likely that the same covariates would also be of importance with weekly docetaxel. The aim of this study was to investigate the pharmacokinetics of weekly docetaxel in patients with advanced cancer and to examine the role of patient covariates on the CL of docetaxel using a population pharmacokinetic approach.

Methods

Patients

Patients with advanced cancer (metastatic or locally advanced, histologically confirmed malignancy) and who were suitable for palliative chemotherapy were enrolled into the trial at the Sydney Cancer Centre, Sydney, Australia. Patients had to be > 18 years of age and have an Eastern Cooperative Oncology Group performance status ≤ 3. Patients remained eligible if they had comorbid disease or were concurrently taking medications that were known to induce or inhibit CYP3A4 activity; however, the dose, frequency and duration of treatment with these drugs were noted. Patients were excluded if they had received chemotherapy less than 1 month prior to the study or were allergic to erythromycin. Written informed consent was obtained from all patients, and the Ethics Committee of the Central Sydney Area Health Service approved the study.

Clinical study design

The study was carried out over 2 days with an initial consultation visit on day 1 and a study and treatment visit on day 2. Routine evaluation was conducted on day 1 and included a complete medical history and collection of blood samples for the determination of baseline haematological counts, acute phase reactant proteins (C-reactive protein and α₁-acid glycoprotein) and serum biochemistry. These samples were processed by the Departments of Haematology, Immunology and Biochemistry of the Royal Prince Alfred Hospital, Sydney, Australia, respectively.

Additionally, baseline liver function test results were categorized into various hepatic indices as previously described by Bruno et al. [8]. Briefly, patients were classified as HEP1 when serum transaminase concentrations were> 1.5 times the upper limit of normal (82.5 U l⁻¹), HEP2 when serum ALP concentrations> 2 times the upper limit of normal (300 U l⁻¹) and finally HEP12 when there were concomitant elevations of both serum transaminases and ALP (i.e. positive for both HEP 1 and HEP 2).

On day 2, the modified erythromycin breath test was performed as described previously [23]. Briefly, 4 µCi of ¹⁴C-erythromycin (N-methyl-¹⁴C, 55 mCi mmol⁻¹, NEN Life Science Products Inc, Boston, MA) was injected intravenously and breath samples were collected into gas-tight balloons (Pytest®, Ballard Medical Products, Utah) 5, 10, 15, 20, 25, 30 and 40 min later. Breath samples were processed by bubbling the collected gas through a capture solution consisting of hyamine hydroxide 10X (Packard, Sydney, NSW, Australia) in 50 : 50 methanol : ethanol v : v to which a trace of phenolphthalein was added. After the addition of scintillant (Ultima Gold®, Packard, Sydney, NSW, Australia) and counting, the data were expressed in terms of % of dose exhaled min⁻¹ at each time point by assuming a CO₂ output of 5 mmol min⁻¹ m⁻²[24]. Parameters of the erythromycin breath test, C_{20 min} and 1/t_max, were calculated [23]. After the erythromycin breath test, all patients received 40 mg m⁻² week⁻¹ docetaxel (Taxotere®, Aventis Pharma, Australia) administered as a 2-h intravenous infusion.

The first 10 patients received standard oral steroid prophylaxis consisting of 8 mg dexamethasone (Aventis Pharma, Australia) twice daily the day before, on the day and the day after docetaxel treatment together with 5 mg tropisetron orally (Navoban®, Novartis, Australia) on the day of treatment. However, the steroid prophylaxis was reduced to a single intravenous dose of 4 mg dexamethasone (Faulding Company Ltd, Mulgrave, Australia) just prior to the commencement of treatment for the remaining patients because of problematic Cushingoid symptoms.

Pharmacokinetic blood collection and sampling

Blood samples (10 ml) were collected into heparinized collecting tubes (Becton Dickinson, North Ryde, Australia) by venepuncture in the contralateral arm prior to the docetaxel infusion and then from one or more of three sampling time windows (0–1 h, 1–4 h or 4.5–24 h) after initiation of the infusion. Plasma was harvested by centrifugation for 10 min at 4 °C and 2000 g and transferred into polypropylene vials, which were stored at −70 °C until analyzed by high performance liquid chromatography with UV detection.

Analysis of docetaxel plasma concentrations

Determination of plasma docetaxel concentrations was performed by HPLC using a slight modification of the method of Garg & Ackland [25]. Briefly, patient plasma samples (1 ml), standards (9, 37, 185, 556 and 1700 ng ml⁻¹), and plasma quality controls (25 and 250 ng ml⁻¹) were loaded onto previously conditioned (1 ml methanol and 1 ml water) CN solid phase extraction cartridges (1 cc, Waters, Rydalmere, Australia) after spiking with 2 µg of the internal standard, paclitaxel.

The columns were rinsed sequentially with 1 ml water, 1 ml 80 : 20 water : methanol and compounds eluted with 1 ml methanol into 1.5 ml Eppendorf tubes. The eluent was dried under a stream of nitrogen at room temperature and the residue reconstituted into 50 µl of mobile phase (55 : 45 v : v, 20 mm potassium phosphate, pH 3.0 : acetonitrile).

The reconstituted samples were vortex-mixed, centrifuged and transferred to HPLC insert vials. Samples were injected (10–20 µl) onto a Luna Phenyl-Hexyl column (3 µm particle size, 2 × 150mm, Phenomenex, Pennant Hills, Australia) using a cooled (15 °C) autoinjector and mobile phase flow rate of 0.2 ml min⁻¹. The LC eluent was monitored at 227 nm over a 25-min run. Concentrations were determined by the area ratio method with a log-log transformed standard curve. All standard curves had r² > 0.99 (mean = 0.998, n = 20) and the QC samples, which were each run in duplicate had mean accuracy and imprecision (CV%) of 95.4% (9.2%) and 93% (6.3%), respectively. Back-calculated concentrations of standards from the standard curves yielded average accuracy and imprecision of 99% and 8.2%, respectively. At the lower limit of quantification (9 ng ml⁻¹), the average accuracy and total imprecision were 105% and 12.8%, respectively.

Population pharmacokinetic modelling of docetaxel

A population pharmacokinetic approach was used to analyze plasma concentration-time data (P-PHARM Version 1.5, Innaphase, Champs-Sur-Marne, France). This approach implements nonlinear mixed effects modelling using a parametric expectation-maximization algorithm [26]. Posterior Bayesian estimates of individual pharmacokinetic parameters were generated. The docetaxel pharmacokinetic parameter of primary interest was CL. A full list of the 21 covariates investigated is included in the table of patient characteristics (Table 1).

Table 1.

Characteristics of the patients in the total dataset.

Patient characteristics (n = 54)	n	Mean	Median	Range	Covariate symbol
Demographics
Gender (Male: female)	31 : 23				GENDER
Age			63	40–83	AGE
Weight (kg)		70	70	44–108	WEIGHT
Body surface area (m2)		1.77	1.74	1.14–2.27	BSA
ECOG performance status					PS
0	6
1	25
2	19
3	4
Primary cancer site
NSCLC	19
Breast	10
Head and neck	11
Prostate	4
Other	10
Metastatic disease (present : absent)	46 : 8
Liver metastasis (present)	10				HEPMETS
Previous chemotherapy
0	19
1	17
2	9
>2	9
Biochemical markers
Albumin (g l−1)		36.5	37	18–44	ALB
Pre-albumin (mg l−1)		0.22	0.22	0.05–0.49	PREALB
Bilirubin (µmol l−1)		8.4	7	3–31	BILI
Alkaline phosphatase (U l−1)		194	104	46–864	ALP
Aspartate amino transferase (U l−1)		34	22	8–173	AST
Alanine amino transferase (U l−1)		32	23	3–212	ALT
Gamma glutamyl aminotransferase (U l−1)		210	43	1–2468	GGT
Hepatic function indices
HEP1 (yes)	4				HEP1
HEP2 (yes)*	7				HEP2
HEP12 (yes)*	3				HEP12
Erythromycin breath test parameters
1/tmax (min−1)		0.06	0.05	0.02–0.15	1/TMAX
C20 min (% dose)		0.05	0.04	0.002–0.1	C20MIN
Inflammatory markers
α1-acid glycoprotein (g l−1)		1.44	1.25	0.5–3.18	AAGP
C-reactive protein (mg l−1)		34	13	1–291	CRP
Prognostic inflammatory and nutritional index		31.1	1.68	0.06–1028	PINI

ECOG: Eastern Co-operative Oncology Group; NSCLC: Non-small cell lung cancer, HEP 1: AST or ALT elevated 1.5× the upper limit of normal (>82.5 U l⁻¹), HEP2: ALP elevated 2× upper limit of normal (>300 U l⁻¹), HEP12: both positive for HEP1 and HEP2, other abbreviations see text.

^*Three individuals had bony metastases in conjunction with elevation in ALP. All three are included in HEP result, but only one is included in HEP12.

Population pharmacokinetic modelling strategy

The population pharmacokinetic modelling strategy was based on an approach employed by Nath et al. [27] and consisted of four steps; [Step 1] an initial population pharmacokinetic modelling of the data and generation of individual posterior Bayesian estimates of CL without considering potentially explanatory covariates; [Step 2] an exploratory analysis of the relationships between patient covariates and the individual posterior Bayesian estimates of CL; [Step 3] influential covariates subsequently included into the population pharmacokinetic model and re-estimation of CL values and [Step 4] reparameterization of the model to allow estimation of the population mean (and variance) of intercepts and coefficients of the regression equation describing CL and covariates.

Development of the basic population pharmacokinetic model without covariates (Step 1)

A basic population pharmacokinetic model for docetaxel was developed using the Model Development Dataset and starting estimates were derived from a previous study [28]. A number of pharmacokinetic and pharmacostatistical models were investigated. These varied in the following: the number of compartments (two vs three), description of the intersubject variability of pharmacokinetic parameters (log vs log-normal), and description of the residual error variance (heteroscedastic vs homoscedastic). The most appropriate model was selected based on the residual error estimate (sigma), estimated interindividual variability in CL, Akaike Information Criterion, Maximum Likelihood estimate, correlations between predicted and actual concentrations, and residuals with a mean value not significantly different from zero and normal distribution [27].

The population mean and variance of docetaxel clearance was determined from the model and utilizing empirical Bayesian estimation, individual posterior Bayesian estimates of CL were produced.

Exploratory analysis of relationships between covariates and estimated CL values (Step 2)

Identification of significant relationships between CL and continuous patient covariates were initially investigated using linear regression analysis [29]. The Mann–Whitney U-test and the Kruskal–Wallis test were used to test whether there were significant differences in CL when individuals were grouped as categorical covariates (e.g. gender and hepatic function indices). Inter-relationships between covariates were investigated using Spearman's correlation analysis. All significant covariates from univariate analysis were then examined using multiple linear regression with forward stepwise inclusion and deletion of covariates with the threshold F statistic of 10 (P-PHARM version 1.5).

Development of a covariate population pharmacokinetic model (Step 3)

Significant covariates from the multiple linear regression analysis were included to form the covariate population pharmacokinetic model. The P-PHARM software-generated values for AIC, sigma, Maximum Likelihood and intersubject variability of CL values were used to assess whether the covariate population pharmacokinetic model was better than the basic model.

Model reparameterization (Step 4)

Once influential covariates were investigated the final two-compartment model was reparameterized in P-PHARM to allow estimation of the population mean and variance of the intercept and coefficients of the regression equation that described the relationship between docetaxel CL and influential covariates.

Validation of the population pharmacokinetic models for docetaxel

Docetaxel concentrations in the model validation dataset were predicted using posterior empirical Bayesian estimation and the parameters of the population pharmacokinetic model derived from the Model Development Dataset. The bias (mean prediction error) and precision (root mean squared prediction error) of the docetaxel concentrations predictions were investigated as previously described [30].

Results

Patient characteristics and data collection

Fifty-six patients with advanced cancer were recruited into the study, between July 2000 and November 2001. Two patients (3.7%) were ineligible for pharmacokinetic analysis because they were not sampled in the first course of docetaxel therapy. The median dose administered was 69 mg with a range from 55 to 80 mg. The total data cohort (n = 145 concentration-time observations) consisted of pharmacokinetic and covariate data from 54 patients receiving weekly docetaxel. All patients had a baseline pharmacokinetic sample obtained and a median of two additional samples per patient (range: 1–4 samples per patient). Five samples were obtained during the docetaxel infusion, 46 samples between the end of infusion and 1 h after the end of infusion, 28 samples between 1 and 4 h after the end of infusion and 12 samples after 4 h postinfusion. The patients were randomly split in a ratio of 2 : 1 [31]. As a result, 38 data sets comprising 103 observations were retained for the development of the model (referred to as the Model Development Dataset) and the remainder of patients (n = 16 with 42 observations) were used for the validation of the model and estimation of model parameters (referred to as the Model Validation Dataset). There were no significant differences in the patient characteristics between the three datasets (P > 0.05, anova).

The patient characteristics are shown in Table 1. The median age was 63 years; 58% were male, and most had performance status of 0 or 1 (57.4%). The most common tumour types were nonsmall cell lung cancer (35.2%), followed by head and neck (20.4%), breast (18.5%) and prostate cancer (7.4%). Other cancers included gastric, ovarian, hepatocellular cancers and patients with adenocarcinoma of unknown primary. Most patients had metastatic disease (85%) and had received more than one prior chemotherapy regimen (64%).

In addition to age, gender and Eastern Co-operative Oncology Group performance status, 18 other covariates were explored and these are listed in Table 1. Weight and body surface area ranged from 44 to 108 kg and 1.14–2.27 m², respectively. Four patients (7.4%) were categorized as positive for HEP 1, seven patients (13%) were positive for HEP 2 (although three of these patients had bone metastases) while three were positive for HEP 12. The range of erythromycin breath test parameters, 1/t_max and C_{20 min}, ranged from 0.02 to 0.15 min⁻¹ and 0.002–0.1% dose exhaled min⁻¹, respectively.

Selection of the pharmacokinetic and pharmacostatistical model

A two-compartment pharmacokinetic model with first order elimination provided the best description of the docetaxel concentration-time data compared with the three-compartment model. The two-compartment model using the Model Development Dataset had the lowest AIC (−0.195) and residual error value (0.079), and highest maximum likelihood value (20.89) than any three compartmental model. Additionally, the predicted and observed concentrations were in close agreement, the pattern of residuals was favourable and the intersubject variability of the pharmacokinetic parameters was minimal. The residual error model selected was 1/y heteroscedastic and a log-normal distribution best described the interpatient variability in CL. Thus, the mean population estimate (95% confidence interval) for CL using the two-compartment basic population pharmacokinetic model was 28.42 l h⁻¹ (14.78, 41.67 l h⁻¹, Table 2). The interindividual variability (CV percentage) in docetaxel CL was 43%.

Table 2.

Population pharmacokinetics of docetaxel

	Basic model		Covariate model
Parameter	Mean	95% CI*	Mean	95% CI*
Clearance (l h−1)†	28.42	14.78–41.67	30.13	12.54–46.04
Volume of distribution (l)†	7.91	7.51–8.21	7.90	7.38–8.45
k12 (h−1)†	1.16	0.43–1.62	1.13	0.50–1.66
k21 (h−1)†	0.15	0.10–0.23	0.15	0.12–0.23
Prediction error (µg ml−1)#	0.0013	−0.0935–0.0960	0.0017	−0.1012–0.1046
Root squared prediction error (µg ml−1)#	0.0476	−0.0720–0.1670	0.0515	−0.0796–0.1822

^*95% CI: lower and upper 95% confidence intervals;

^†derived from the individual posterior Bayesian estimates of the total dataset;

^#derived from individual posterior Bayesian estimates of the validation database.

Inter-relationships in the covariates

There were significant correlations between several covariates investigated in this study. 1/t_max was found to significantly correlate with measures of inflammation and nutrition, including CRP, pre-albumin and AAGP (P < 0.001, Spearman's correlation with Bonferroni correction). There was a lack of correlation between the 1/t_max parameter and markers of liver function. C_{20 min} parameter was significantly correlated with the 1/t_max parameter (P = 0.0002) and pre-albumin concentrations (P = 0.002). All markers of liver function (i.e. bilirubin, ALP, ALT, AST, and GGT) were highly interrelated (P < 0.002). However, there was no significant correlation observed between ALT and bilirubin concentrations. In addition, no significant correlations with age and any of the other covariates were observed. BSA was only significantly correlated with an individual's weight (Spearman's rho = 0.9, P < 0.0001).

Exploratory analysis

As individual covariates, many of the liver function enzymes were shown to significantly correlate with the estimates of docetaxel CL (r² = 0.22, P = 0.0002; r² = 0.21, P = 0.0005, r² = 0.19; P = 0.001 and r² = 0.12, P = 0.006, for AST, ALT, ALP and GGT, respectively). The HEP indices did not improve the relationships between individual liver enzymes with CL. Docetaxel CL correlated significantly with either measure from the EBT (r² = 0.19, P = 0.0005 and r² = 0.15, P = 0.003, for C_{20 min} and 1/t_max, respectively). Natural log transformation of erythromycin breath test parameters also did not improve the relationship with these parameters and CL. Docetaxel CL was also correlated with C-reactive protein and α₁-acid glycoprotein (r² = 0.09, P = 0.017 and r² = 0.09, P = 0.014, respectively). Other covariates including age, body surface area, weight, Eastern Co-operative Oncology Group performance status and serum bilirubin did not correlate significantly with docetaxel CL.

Using step-wise multiple linear regression, only a combination of ALT and 1/t_max significantly improved on the univariate regressions (r² = 0.36, P = 0.00001).

The covariate population pharmacokinetic model for docetaxel

After inclusion of the most predictive covariate combination (ALT and 1/t_max) into the population pharmacokinetic model, the population mean estimates of CL (95% confidence intervals) were 30.13 l h⁻¹ (14.78, 41.67 l h⁻¹, Table 2). The interindividual variability (% CV) in clearance estimates was 31.7%. The volume of distribution had a population mean (95% confidence intervals) of 7.9 l (7.51, 8.21 l). The regression equation describing the relationship between CL and influential covariates used in the reparameterized pharmacokinetic model (Step 4) is shown in Equation 1:

$$\text{CL}={\theta }_1+{\theta }_2(1/t_{\text{max}})-{\theta }_3(\text{ALT})$$ (2)

where the regression coefficients θ_1,θ_2, and θ₃ have a mean value (± SD) of 21.51 (± 2.9), 217 (± 8.7) and 0.13 (± 0.2), respectively.

The final population pharmacokinetic model, using the total dataset, incorporating covariates had a lower residual error (0.052 vs 0.066) and AIC (−0.353 vs −0.290) values and higher maximum likelihood results (40.15 vs 34.41) compared with the basic population pharmacokinetic model. The covariate population pharmacokinetic model explained 30% of the interpatient variability observed in CL values. Population fitted data for the covariate population pharmacokinetic model is shown in Figure 1. For patients with elevated transaminases alone (HEP 1), CL was decreased by 30% (range: 40% increase to 82% decrease in CL, Figure 2), while patients with increased ALP alone (HEP 2) had on average a 34% reduction in CL (range: 11% increase to 82% decrease in CL). Three individuals with elevations in both transaminases and ALP (HEP 12) had approximately 39% lower CL values than the rest of the patient group (range: 20–58% decrease in CL).

Figure 1.

Docetaxel population pharmacokinetic profile in 54 patients. The line denotes the predicted profile calculated for the mean disposition parameters

Figure 2.

Docetaxel pharmacokinetic profile in a representative patient with elevated transaminase levels (closed circles) and a patient with normal transaminase levels (open circles). Lines denote model predictions after Bayesian estimation from the covariate population pharmacokinetic model

Performance and validation of the basic and covariate population pharmacokinetic models for docetaxel

The population pharmacokinetic models (with and without covariates) provided close predictions of the observed docetaxel concentration data. The residuals for both population pharmacokinetic models were normally distributed, as indicated by a Kolmogorov-Smirnov test, and the means were not significantly different from zero. The bias and precision of the model-estimated concentrations of docetaxel using the Model Validation Dataset indicated that the imprecision was low while the mean prediction error (bias) was not significantly different from zero (Table 2). Figure 3 shows the scatterplot of the posterior Bayesian predicted docetaxel concentrations in the total dataset from the covariate population pharmacokinetic model. The slope of the regression line was 0.93 and the regression coefficient (r²) was 0.94 indicating a close agreement between the observed and posterior predicted docetaxel concentrations.

Figure 3.

Scatterplot of observed and posterior Bayesian predicted docetaxel concentrations in the total dataset from the covariate population pharmacokinetic model. The solid line denotes the linear regression between the observed and predicted docetaxel concentrations and the dashed line denotes the line of identity

Discussion

In this study of patients with advanced cancer we have investigated the population pharmacokinetics of docetaxel administered at 40 mg m⁻² weekly. This was successfully achieved using sparse data collection (typically two samples per patient), population pharmacokinetic modelling and posterior empirical Bayesian estimation for individual values of CL. At this dose and with this modelling approach, the data were compatible with a two-compartment model rather than the three-compartment model utilized to describe the pharmacokinetics of higher doses of docetaxel [8, 13, 22]. One of the possible reasons for this finding is that the terminal phase is not as readily characterized at lower doses, because assay sensitivity becomes limiting. This is probably why early phase I studies found that the pharmacokinetics of docetaxel at doses greater than 85 mg m⁻² exhibited tri-phasic elimination and those under this dose were better described using two-compartment pharmacokinetic model [22]. A second factor is that the discrimination of multiple phases of elimination requires more intensive sampling, which was not possible in our setting.

Using the basic population pharmacokinetic model, we observed large interpatient variability in docetaxel CL with a mean of 28.42 l h⁻¹ (95% CI 14.78, 41.67 l h⁻¹). Similar variability in clearance has been observed in other population pharmacokinetic studies of docetaxel [8, 10, 13]. The slightly lower mean CL value found in our study may partly be due to the more advanced nature of the cancers involved. In another study, we found that drug metabolism, as determined by the erythromycin breath test, was reduced in advanced cancer patients with an acute phase reaction [32]. Markers of acute-phase reaction are strong prognostic indicators of poor physical functioning, extent of disease, and survival in advanced cancer [33]. In this study, approximately 40% of patients had a performance status value = 2.

Exploratory univariate analyses with the estimates of clearance showed relationships with liver function enzymes (ALP, AST, ALT and GGT), CYP3A4 function (C_{20 min} and 1/t_max) and with acute phase reactants (α₁-acid glycoprotein and C-reactive protein). Investigation of the inter-relationships between covariates in this study revealed that both markers of CYP3A4 activity, C_{20 min} and 1/t_max were interrelated. However, only the 1/t_max parameter was found to be highly correlated with various measures of inflammation. It is well documented that CYP3A4 activity and mediated-clearance is reduced in the presence of various inflammatory stimuli [34, 35]. In contrast to CYP3A4 activity, the liver function enzymes showed no correlation with markers of inflammation or either erythromycin breath test parameter. However, all liver function enzymes showed a high degree of correlation.

When step-wise multiple linear regression analysis was employed, the most influential covariate effects on docetaxel clearance were those of the erythromycin breath test parameter, 1/t_max, and serum ALT concentrations. In addition, the lack of correlation between liver function enzymes and CYP3A4 suggests that the factors are independently influencing the pharmacokinetics of docetaxel. When these two covariates were subsequently included, the resulting covariate population pharmacokinetic model performed better according to standard criteria although its predictive performance was somewhat similar to that of the basic model.

The relationship between CYP3A4 and docetaxel CL confirms the findings of other pharmacokinetic studies of docetaxel [13, 16]. The variability in pathophysiology, disease extent and performance status observed in the patients of this study is representative of patients with advanced cancer and despite the lower correlation between the docetaxel CL and CYP3A4 activity and liver function in comparison with other studies, the similarities in both the relationships between influential covariates and docetaxel clearance as well as the population values for docetaxel clearance in this study support that the methodology employed was valid and may be of benefit for the individualization of weekly docetaxel. One main difference between our study and that conducted by Hirth et al. [13] was the parameter derived from the erythromycin breath test. Various erythromycin breath test parameters have been used to describe hepatic CYP3A4 activity in vivo. We recently investigated the various advantages and disadvantages of these [36] following our finding that the 1/t_max parameter was most predictive of erythromycin clearance in advanced cancer patients, whereas the C_{20 min} and various other parameters failed to correlate [23].

Additionally, similar relationships between liver function enzymes and docetaxel clearance have been documented in prior pharmacokinetic studies [8, 9, 11]. The incorporation of liver function enzymes in the calculation of docetaxel clearance suggests that factors other than metabolic capability of the liver (e.g. liver blood flow, hepatocellular uptake or biliary excretion) may be important for the disposition of docetaxel and may be identified using the various liver function tests. This may also explain why a combination of covariates including a hepatic function enzyme and a measure of CYP3A4 activity were most predictive of docetaxel clearance.

Furthermore, factors influencing the concentration of unbound drug available for liver uptake and metabolism could be important for the pharmacokinetics of docetaxel. Docetaxel is highly bound to α₁-acid glycoprotein [14] and this may reduce docetaxel CL. Indeed, some pharmacokinetic studies have revealed that docetaxel clearance is inversely related to α₁-acid glycoprotein concentrations [8], although the effect may be weak [13]. Although the concentration of α₁-acid glycoprotein may influence the free-fraction of docetaxel, it is likely that a direct measure of the docetaxel protein-binding would be more predictive of total clearance. Unfortunately, this was not technically feasible in our study. Alternatively, α₁-acid glycoprotein being an acute phase reactant may act as a surrogate marker of inflammatory status, which is known to correlate with reduced CYP3A4 activity [32]. The failure to retain α₁-acid glycoprotein in our population pharmacokinetic model may simply be due to the presence of a more direct measure of hepatic CYP3A4 activity.

In conclusion, we have described the population pharmacokinetics of weekly docetaxel (40 mg m⁻²) in advanced cancer patients. The findings confirmed results from prior pharmacokinetic studies of the 3-weekly dose of docetaxel (100 mg m⁻²) [8, 13, 16] indicating that the clearance of docetaxel (40 mg m^{−2 −1} week) is influenced by both CYP3A4 activity and global hepatic function. Further studies investigating the relationships between docetaxel exposure and toxicity and/or efficacy are now needed to identify a safe and effective target AUC for this dose schedule. Currently, we are evaluating these relationships in advanced cancer patients with the identified population pharmacokinetic model.