Population Pharmacokinetics and Pharmacodynamics of Paclitaxel and Carboplatin in Ovarian Cancer Patients: A Study by the European Organization for Research and Treatment of Cancer-Pharmacology and Molecular Mechanisms Group and New Drug Development Group

Ovarian cancer is still the most common cause of death from gynecologic cancer in the Western world. The combination of paclitaxel and carboplatin is seen as standard treatment for advanced ovarian cancer based mainly on three studies showing that paclitaxel plus carboplatin is equally effective to paclitaxel plus cisplatin with a superior safety profile and response rates of roughly two thirds in patients with measurable disease (1–3). Most patients with advanced ovarian cancer achieve a clinical complete remission following cytoreductive surgery and chemotherapy with paclitaxel and carboplatin. The same combination has also proven to be the most effective treatment regimen for late relapses of ovarian cancer (i.e., in patients who progress ≥6 months after completion of upfront therapy). Small-to-moderate gains in progression-free and overall survival were achieved with the paclitaxel/carboplatin combination therapy as opposed to carboplatin monotherapy in patients with late (≥6 months) tumor relapses (4). Accordingly, the combination of a 3-h infusion of paclitaxel (175 mg/m²) followed by carboplatin area under the concentration-time curve (AUC) 5 mg/mL*min has been used as the control arm in clinical studies of first-line (5) and second-line treatment of platinum-sensitive or late relapses of ovarian cancer (6).

Single-agent paclitaxel showed overall response rates of 28% in platinum-pretreated patients with advanced ovarian cancer (7). Paclitaxel is known for its nonlinear pharmacokinetics (8–11), and population pharmacokinetic models have mostly implemented saturable elimination and distribution to peripheral compartments (12–14). Carboplatin has less nonhematologic toxicity than cisplatin, and most randomized trials have reported comparable activity between cisplatin and carboplatin in chemotherapy-naive patients with advanced ovarian cancer (15).

The exposure toxicity relationship of paclitaxel has been described by threshold models, whereas the time above a paclitaxel plasma concentration of between 0.05 and 0.2 μmol/L was correlated with hematologic toxicity (8, 12, 16–18). Two of these trials studied single-agent paclitaxel in patients with advanced ovarian cancer (8, 16). The study of Gianni et al. (8) studied 30 patients with relapsed ovarian or breast cancer and found a correlation between neutropenia and the duration that paclitaxel plasma concentration was ≥0.05 μmol/L, a relationship that was well described by a sigmoid maximum response (E_max) model. In the study of Huizing et al. (16), the duration of paclitaxel plasma concentration ≥0.1 μmol/L (t_{C > 0.1}) was correlated with neutropenia and leucopenia, suggesting that myelosuppression could be predicted by measuring the duration of paclitaxel plasma concentration ≥0.1 μmol/L. The combination of paclitaxel and carboplatin has been studied for its pharmacokinetic-pharmacodynamic relationship in patients with advanced non–small cell lung cancer, where a paclitaxel plasma concentration >0.1 μmol/L for more than 15 h was correlated with improved overall survival (19). In ovarian cancer, limited data suggest the absence of a pharmacokinetic interaction between paclitaxel and carboplatin (20), but there is a well-known pharmacodynamic interaction resulting in a platelet-sparing effect of paclitaxel in combination with carboplatin as first-line treatment (21, 22). However, studies implementing combined pharmacokinetic-pharmacodynamic assessment of the drug combination in an extended group of ovarian cancer patients have not been done to our knowledge.

Besides threshold models as used by Gianni et al. (8) and Huizing et al. (17, 19), more physiology-based modeling approaches have been adapted recently. Friberg et al. (23) described a semimechanistic (pharmacokinetic-pharmacodynamic) model of paclitaxel-induced myelosuppression by using nonlinear mixed-effects modeling. The semimechanistic modeling approach typically uses drug-related and system (patient)-related variables and is capable to explain and predict both the degree and duration of hematologic toxicity. Such more physiology-based models are often preferred because they generally are more predictive, with variables that refer to actual processes and conditions (23). The present study was initiated to build a population pharmacokinetic-pharmacodynamic model of paclitaxel and carboplatin in ovarian cancer patients and to quantitatively assess the relationship between pharmacokinetic, hematologic toxicity, and treatment response.

Patient population, blood sampling, and bioanalysis. Twelve study centers participated in the trial (Appendix). All centers had approval from the local Medical Ethics Committee. Written informed consent was obtained from all patients. One hundred thirty-nine female patients with ovarian cancer were recruited between 1998 and 2001. Inclusion criteria for study participation included histologically or cytologically verified epithelial ovarian cancer, patients with microscopic local, locally advanced or metastatic disease, performance status (WHO scale) ≤2, life expectancy ≥3 months, absolute neutrophil count (ANC) ≥1,500/μL, platelets ≥100,000 μL, age ≥18, creatinine clearance ≥40 mL/min, serum bilirubin ≤26 μmol/L (1.5 mg/dL), and alanine aminotransferase and aspartate aminotransferase ≤2× the upper limit of normal (unless related to liver metastases, then aspartate aminotransferase and alanine aminotransferase may be 5× the upper limit of normal), and a minimum of 4 weeks (8 weeks in case of extensive prior radiotherapy) must have elapsed between the end of the prior radiotherapy and entry into the protocol. Patients may be premenopausal or postmenopausal. Patients may (but do not necessarily) have unidimensionally or bidimensionally measurable lesions. Patients must be able and willing to undergo pharmacokinetic analysis. Patients are excluded in case of prior chemotherapy for adjuvant or advanced disease; symptomatic brain and/or leptomeningeal disease; previous or current malignancies at other sites (with exception of adequately treated in situ carcinoma of the cervix uteri and basal or squamous cell carcinoma of the skin); active uncontrolled infection; women with childbearing potential not practicing adequate contraception, pregnancy, or lactation; and active clinically significant and/or treatment requiring cardiac arrhythmias and/or congestive heart failure. It was anticipated to include 150 patients in this trial. However, this was not based on a formal sample size calculation because several factors were uncertain a priori (e.g., the number of patients with evaluable disease). Furthermore, the number of patients was based on a reasonable estimate of the patient accrual per center in the study period. All patients received Cremophor EL–formulated paclitaxel (175 mg/m²) i.v. over 3 h immediately followed by a 30-min infusion of carboplatin AUC 5 mg/mL*min, with the initial dose calculated according to the method described by Calvert et al. (24), in which glomerular filtration rate was calculated using the Cockroft-Gault formula. This dose remained fixed for all cycles, unless toxicity necessitated any dose reduction. Standard premedication was administered, including 20 mg dexamethasone orally at 12 and 6 h before paclitaxel, 2 mg clemastine i.v., and a H₂ receptor antagonist 30 min before paclitaxel, to prevent hypersensitivity reactions. Antiemetics were used according to local guidelines (metoclopramide, ondansetron, or granisetron). Cycles were repeated at three-weekly intervals, and routine weekly hematologic assessments were done, including hemoglobin, leucocyte count, ANC, lymphocyte count, and thrombocyte count (PLT). Pharmacokinetic analysis was done during the first cycle of paclitaxel and carboplatin treatment. The samples for paclitaxel analysis were collected in heparinized tubes. The blood samples for paclitaxel were taken before the start of the infusion, at 1.5 h after the start, at the end of the infusion, and at 45 min and 2, 6, 9, 12, and 24 h after the end of the infusion. Whole blood was centrifuged immediately after withdrawal for 5 min (3,000 rpm, 4°C), and the plasma fraction was stored at −20°C until analysis. The plasma concentrations of paclitaxel were determined by a validated isocratic high-performance liquid chromatographic method with solid-phase extraction as the sample pretreatment procedure as described previously (17). The quantitation range of the high-performance liquid chromatographic method was 10 to 10,000 ng/mL (11.7-11,710 nmol/L). Within-day imprecision for paclitaxel was between 1.2% and 8.5% and accuracy was >95%. Bioanalysis of paclitaxel was done in European Organization for Research and Treatment of Cancer-Pharmacology and Molecular Mechanisms centers (see Appendix). The blood samples for carboplatin were taken before the start of the infusion, at the end of the infusion, and at 2, 5, and 9 h after the start of the infusion. Whole blood samples were collected in 5 mL heparin-containing tubes and centrifuged immediately after withdrawal for 5 min (3,000 rpm, 4°C). Plasma ultrafiltrate (pUF) was obtained by centrifuging the plasma fraction through a 30-kDa cutoff ultrafiltrate filter (Centriplus YM-30, Millipore Corp.) for 15 min (3,000 rpm, 20°C). The preparation of pUF from plasma was done immediately after blood collection to prevent the decrease of free carboplatin levels due to ex vivo binding of carboplatin to plasma proteins and erythrocytes. All samples were stored at −20°C until analysis. Carboplatin analysis was also done in European Organization for Research and Treatment of Cancer-Pharmacology and Molecular Mechanisms centers (see Appendix) according to a previously described atomic absorption spectroscopy assay (25). Bioanalysis of paclitaxel and carboplatin took place after extensive cross-validation of all involved centers. This was a two-step process. At first, all participating centers had to send their validation report of the assay to the coordinator of the bioanalytic steering committee (J.H.B.), who reviewed the reports according to international standards (19). Second, when the report of a site was considered appropriate, five blinded samples containing two blanks and three samples with spiked concentrations of carboplatin and paclitaxel were sent to the participating bioanalytic centers. The spiked samples contained concentrations in the lower quartile, in the middle, and in the upper quartile of the calibration curve of the respective assays. For paclitaxel, the concentrations were in the range of 20 to 10,000 μg/L (23.4-11,710 nmol/L) and, for carboplatin, in the range of 556 to 5,568 μg/L (1.5-15 μmol/L). The centers had to achieve the minimal acceptance levels for successful validation of their assay, which means that the deviations from the nominal value had to be <20%.

Population pharmacokinetic model. Population pharmacokinetic analysis of the concentration-time data of paclitaxel and carboplatin was done using the nonlinear mixed-effect modeling program (NONMEM) version V (double precision, level 1.1; refs. 26, 27). NONMEM uses a maximum likelihood criterion to simultaneously estimate population values of fixed-effects variables (e.g., drug clearance) and values of the random-effects variables (e.g., interindividual, interoccasion, and residual variability).

Concentration-time data of paclitaxel were described by applying a three-compartment model as described previously (28). The following pharmacokinetic variables were estimated for paclitaxel: volume of the central (V₁ in L) and second peripheral compartment (V₂ in L), maximal elimination rate (VM_EL in μmol/h), Michaelis-Menten elimination constant (KM_EL in μmol/L), maximal transport rate to the first peripheral compartment (VM_TR in μmol/h), total plasma concentration of paclitaxel at half VM_TR (KM_TR in μmol/L), and intercompartmental clearance between the central and second peripheral compartment (Q in L/h). Concentration-time data of carboplatin were described by a two-compartment model with linear distribution and elimination as described previously (29). The following pharmacokinetic variables were estimated for carboplatin: volume of the central (V₁ in L) and peripheral compartment (V₂ in L), drug clearance (CL in mL/min), and intercompartmental clearance between the central and peripheral compartment (Q in L/h).

Interindividual variability was estimated using a proportional error model. For paclitaxel, two subpopulations for intraindividual or residual variability were modeled as described previously (28). NONMEM estimates of the individual pharmacokinetic variables for paclitaxel and carboplatin were subsequently used as input variables for the pharmacokinetic-pharmacodynamic model.

Because bioanalysis was done at the participating centers, albeit after a thorough cross-validation procedure, careful consideration was given at possible differences in pharmacokinetic variables (including variability) between the data from the different study centers. This was done by visual inspection and by a jackknife procedure, in which data from the different centers were deleted to assess the effect of that center on the variable estimates.

Semiphysiologic modeling of hematologic toxicity. A semiphysiologic model as used in this study to describe drug-related neutrophil and thrombocyte toxicity was introduced by Friberg et al. (30). The model comprised a compartment representing the proliferating cells linked to a compartment representing the systemic circulation through three transit compartments, mimicking precursor cell maturation within the bone marrow. The chain of transit compartments allows the description of the time delay between drug exposure, impaired cell proliferation, or cell killing and the resulting effect on circulating blood cells. The transition rate constant (k_tr) between the compartments was supposed to be first order and equal for all transitions. The average maturation time or mean transition time (MTT) represented the time a cell spent to pass from the proliferation stage to the circulation pool. The status of the proliferation compartment is dependent on the number of cells in that compartment and on the proliferation rate constant (k_prol). The disappearance of peripheral blood cells from the circulation pool is given by the first-order rate constant k_circ. At steady-state conditions, k_prol and k_circ equal k_tr. A feedback mechanism (FB) imitated the effect of the release of endogenous growth factors as a response to the decrease of cells in the circulation pool. This leads to increased cell proliferation and was modeled by a power function of the ratio between the baseline cell count (ANC_base or PLT_base) and the cell count at time = t (ANC_t or PLT_t) according to (ANC_base/ANC_t)^γ or (PLT_base/PLT_t)^γ, where γ constitutes the feedback constant. The function by which drug concentrations (paclitaxel and carboplatin) affect the proliferation rate of circulating blood cells (E_drug) was modeled using either a linear function, in which E_drug is represented by a slope factor (Slope) and drug concentration (c_drug; Eq. A), or a maximum effect model, in which E_drug is a represented by the maximal drug effect E_max, drug concentration at half maximal drug effect EC₅₀, and c_drug (Eq. B). The linear and E_max models were separately adapted to the concentration-time data of paclitaxel and carboplatin to predict individual neutrophil and thrombocyte counts. Subsequently, a model incorporating the effect of paclitaxel (E_paclitaxel) and carboplatin (E_carboplatin) was constructed according to Eq. C, in which E_drug corresponded to (Slope*c_drug) for the linear model and to (E_max*c_drug) / (EC₅₀ + c_drug) for the E_max model based on what has been published by Sandstrom et al. (31). Proliferation of bone marrow cells (k_prol) was defined to be dependent on the feedback variable (FB = [ANC_base/ANC_t]^γ) and on E_drug:

.

Data simulations using the population pharmacokinetic-pharmacodynamic estimates were used to construct both the average neutrophil and thrombocyte curve following one dose of paclitaxel (175 mg/m²) and carboplatin AUC 5 mg*min/mL and derive the average neutrophil and thrombocyte nadir. For model evaluation, a posterior predictive check with 1,000 simulations of the final pharmacokinetic-pharmacodynamic model was run. Day 14 neutrophil and thrombocyte counts after the administration of paclitaxel and carboplatin were chosen for model evaluation purposes, and the median and 25% and 75% percentiles of the hematologic variables from 1,000 simulations were compared with the real data set.

Threshold models for toxicity and treatment response. Pharmacokinetic-pharmacodynamic relationship was additionally analyzed by the use of threshold models. Threshold models describe pharmacodynamic variables as a function of the time above a certain threshold drug concentration, expressed in hours. Previously published paclitaxel threshold models used cutoff plasma concentrations of 0.05 μmol/L (8, 9), 0.1 μmol/L (16, 19), 0.07 μmol/L (18), and 0.2 μmol/L (12) to describe drug-related hematologic toxicity and/or objective response to the paclitaxel/carboplatin. Threshold concentrations of 0.1 and 0.05 μmol/L paclitaxel were used according to the pharmacokinetic-pharmacodynamic models as described by Huizing (16, 19) and Gianni (8), respectively. Individual values for the time above threshold concentration were derived using Bayesian estimates on the population pharmacokinetic model as outlined above. Unpaired t test for two groups with equal variances was used to assess the relationship between t_{C > 0.1}/t_{C > 0.05} and objective tumor response, time to disease progression (TTP), and hematologic toxicity as assessed by common toxicity grading. Paclitaxel and carboplatin C_max and AUC were similarly analyzed for their relation with objective tumor response and hematologic toxicity.

Patient population and treatment. From the 139 included patients, 1 patient received 1 paclitaxel/carboplatin cycle, 6 patients received 2 cycles, 3 patients received each 3, 4, and 5 cycles, 106 patients received 6 cycles, 6 patients received 7 cycles, 7 patients received 8 cycles, 1 patient received 9 cycles, and 3 patients received 10 cycles. A total of 104 patients had complete pharmacokinetic and toxicity data to be included into pharmacokinetic analysis. The remaining 35 patients had either incomplete pharmacokinetic assessments (including 4 patients with allergic reactions to paclitaxel, in whom paclitaxel infusion was stopped or paused) or incomplete hematologic assessments. From these 104 patients, 34 had measurable disease at the time of study treatment and 70 had nonevaluable disease. From 34 patients with measurable disease, 17 had a complete response, 9 had a partial response, and 2 had stable disease and 6 were patients with progressive disease, for an overall response rate of 76%. All 139 patients had surgical debulking of ovarian cancer, with gross residual disease in 48 patients, microscopic malignant disease in 82 patients, unknown disease status in 2 patients, and no tumor residuals in 7 patients. Patient characteristics were as follows: median age, 54 years (23.7-79.4); median weight, 59 kg (30.5-93); median serum albumin, 41 g/L (24-47); total bilirubin (<16 μmol/L); and serum creatinine (<150 μmol/L) within the reference range and no greater than grade 1 baseline elevations of aspartate aminotransferase or alanine aminotransferase. Dose delays and dose reductions due to hematologic and nonhematologic toxicity are outlined in Table 1.

Basic population pharmacokinetic model. Population variable estimates from the final model are presented in Table 2. Distribution of paclitaxel to the peripheral compartment (KM_TR) was saturated at a plasma concentration of 0.99 μmol/L and nonlinear elimination (KM_EL) at 3.12 μmol/L. The maximal elimination capacity (VM_EL) was estimated at 29.6 μmol/h. Maximum clearance of paclitaxel (at low drug concentrations compared with KM_EL, which can be obtained as V₁*VM_EL/KM_EL) was estimated to be 70 L/h. Carboplatin clearance was estimated at 123 mL/min (corresponds to 7.38 L/h, the former is preferred for comparison with literature data). Goodness-of-fit plots between model-predicted and observed pharmacokinetic variables support the accuracy of the model (plots not shown). No differences between data from the different bioanalytic sites could be shown.

Semiphysiologic modeling of hematologic toxicity. The semiphysiologic model with a linear model to describe E_paclitaxel and E_carboplatin could be fitted to both neutrophil and thrombocyte data (Table 3). According to the precision of variable estimates (SE and interindividual variability), the linear E_drug model was more appropriate than the E_max model for both compounds. For the combined drug models (paclitaxel and carboplatin data modeled simultaneously), NONMEM runs only terminated successfully with the linear model. This might be a consequence of overparameterization, as the E_max model uses one variable more than the linear model. Overall, pharmacokinetic-pharmacodynamic modeling gave more robust results with the linear model compared with the E_max model. The estimates for the combined modeling of E_paclitaxel and E_carboplatin (according to Eq. C) for neutrophil and thrombocyte concentration-time data are outlined in Table 3. The combined model for paclitaxel and carboplatin on neutrophil concentration-time data confirmed that paclitaxel is the main compound responsible for neutrophil toxicity and that carboplatin adds little to neutrophil toxicity. The combined model of paclitaxel and carboplatin on thrombocyte data produced a negligible slope variable for paclitaxel (10⁻¹⁰ μmol/L), suggesting that paclitaxel does not add significantly to thrombocyte toxicity. Relative SEs indicate that all variables were accurately estimated. The MTT for neutrophils was 141 h or 5.9 days and, for thrombocytes, 195 h or 8.1 days. Goodness-of-fit plots between model-predicted and observed cell counts for neutrophils and thrombocytes support the accuracy of the model (plots not shown). Typical curves for neutrophils and thrombocytes are shown in Fig. 1, with average population nadirs of 1,090/μL neutrophils at 11.5 days and 200 × 10³/μL thrombocytes at 15 days. In the real data set, median nadir counts at day 14 were 1,170/μL neutrophils and 214 × 10³/μL thrombocytes (Table 1). The model was successfully validated by the posterior predictive check, as the median day 14 neutrophil and thrombocyte counts fell into the 50% confidence interval of 1,000 simulation runs (1,130-1,330/μL for neutrophils and 194 × 10³ to 214 × 10³/μL for thrombocytes).

Threshold models for toxicity and treatment response. Mean t_{C > 0.1} was 16.4 h. In patients with a complete response, t_{C > 0.1} was 17.8 h and, in patients with progressive disease, 14.0 h (Table 4). t_{C > 0.1} was not statistically higher in patients with at least stable disease compared with patients with progressive disease (18.3 versus 14.0 h; P = 0.08). Mean t_{C > 0.05} was 61.4 h. Patients with a complete remission had a significantly higher t_{C > 0.05} compared with patients with progressive disease (91.9 versus 31.5 h; P = 0.02) as had patients with at least stable disease compared with patients with progressive disease (78.3 versus 31.5 h; P = 0.05). No statistically significant differences were found between patients with at least stable disease compared with patients with progressive disease for paclitaxel and carboplatin exposure as assessed by C_max and AUC (Table 4). As expected, TTP was longer in patients with at least stable disease compared with patients with progressive disease under paclitaxel/carboplatin treatment (71.6 versus 35.0 weeks; P = 0.04). t_{C > 0.1} was not correlated to TTP as assessed by Pearson correlation analysis (r = 0.04; P = 0.83), and patients with t_{C > 0.1} > 16.4 h had a TTP comparable with patients with t_{C > 0.1} < 16.4 h (73.1 versus 67.3 weeks; P = 0.66). t_{C > 0.05} was significantly correlated with TTP (r = 0.36; P = 0.03), and patients with t_{C > 0.05} above the mean of 61.4 h had a significantly higher TTP compared with patients with t_{C > 0.05} < 61.4 h (89.0 versus 61.9 weeks; P = 0.05).

For hematologic toxicity, no significant difference was found for the grade of neutropenia or thrombocytopenia and exposure to paclitaxel as assessed by t_{C > 0.1}, C_max, or AUC. t_{C > 0.05} was significantly higher in patients with severe neutropenia (grade 3/4) compared with patients with absent or mild neutropenia (grade 1/2; 74.1 versus 50.3 h; P = 0.02), and t_{C > 0.05} was significantly higher in patients with any thrombocytopenia compared with patients without thrombocytopenia (82.2 versus 53.8 weeks; P = 0.02). Carboplatin C_max and carboplatin pUF-AUC were highest in patients with severe thrombocytopenia (57.7 mg/L and 2.08 mg/mL*min, respectively) compared with patients with mild thrombocytopenia (47.8 mg/L and 1.76 mg/mL*min, respectively) and without thrombocytopenia (30.7 mg/L and 1.25 mg/mL*min, respectively; P < 10⁻⁴; Table 4). Significant correlations between paclitaxel t_{C > 0.05} and relative neutropenia as well as between AUC carboplatin and relative thrombocytopenia are shown in Fig. 2. The correlations as depicted in Fig. 2 were calculated using linear regression analysis.

Optimization of doses and administration schedules for anticancer agents is desirable not only in the development of new drugs but also for established drugs and drug combinations. One important component in optimizing cancer therapy is to describe the hematologic toxicity of the drugs, which is dose limiting for most anticancer drugs, as a function of its pharmacokinetic behavior. Pharmacokinetic conditions with effect on the toxicity and efficacy of paclitaxel have been the topic of several publications (8, 12, 16, 18), but data are still very limited for the paclitaxel/carboplatin regimen (19, 21) despite its common use. The establishment of a pharmacokinetic-pharmacodynamic relationship may enable the optimization of standard first-line treatment in ovarian cancer patients with the potential for higher response rates and/or less hematologic toxicity.

We found an overall response rate of 76%, in accordance to what has been described in patients with advanced ovarian cancer receiving first-line treatment with paclitaxel/carboplatin (64% in ref. 2, 64% in ref. 32, and 78% in ref. 16). Mean time to progression of 70 weeks was slightly inferior to the 83 weeks as reported by Ozols et al. (3) with the paclitaxel/carboplatin drug regimen. However, the study by Ozols et al. included optimally resected ovarian cancer patients, whereas our series included 48 patients (35%) with gross residual disease after cytoreductive surgery, explaining the inferior TTP.

The pharmacokinetic model provided estimates for carboplatin clearance (123 mL/min) that were in accordance with previously published data (33). For paclitaxel, maximal elimination capacity (VM_EL) was estimated at 29 μmol/h, in accordance to our previous paclitaxel pharmacokinetic and covariate analysis in 168 solid tumor patients (37 μmol/h; ref. 28) but higher than previously described by Karlsson et al. (13 μmol/h for the saturable, noninstantaneous binding model; ref. 14). The higher paclitaxel elimination capacity in our previous pharmacokinetic analysis (28) compared with the present analysis is explained by the mixed gender distribution in the former and the female gender in the latter, backing our previous finding of a higher paclitaxel elimination capacity in males compared with females (28). The lower estimated paclitaxel elimination capacity in the study by Karlsson et al. is probably due to differences in the study population and might be a consequence of a better performance status and general health status with a higher paclitaxel elimination capacity in the present study and our previous report (28). Furthermore, effects of comedication may also be considered, although difficult to assess retrospectively. The pharmacokinetic-pharmacodynamic model adequately described neutrophil and thrombocyte counts after the administration of paclitaxel (175 mg/m²) followed by carboplatin AUC 5 mg/mL*min in 105 ovarian cancer patients with complete pharmacokinetic and toxicity data. The good agreement between final pharmacokinetic variables from the presented 105 ovarian cancer patients with a previously published paclitaxel pharmacokinetic model in a group of 168 solid tumor patients (28) suggests a lack of pharmacokinetic interaction between paclitaxel and carboplatin. This is in accordance to what has been described by Huizing et al. (16) in Caucasian and by Yamamoto et al. (34) in Japanese patients with advanced ovarian cancer. The sequence of drug administration (paclitaxel followed by carboplatin or vice versa) did not exert a significant effect on the level of observed neutropenia in 40 patients with advanced gynecologic malignancies (35). This was also found in 36 patients with metastatic non–small cell lung cancer, where neutropenia following paclitaxel/carboplatin was consistent with that observed in patients receiving paclitaxel alone (21). On the contrary, there is a clinically relevant pharmacodynamic interaction between paclitaxel and carboplatin, which results in a thrombocyte-sparing effect of paclitaxel (20, 36, 37). The paclitaxel-mediated, thrombocyte-sparing effect results in high tolerated carboplatin doses if given together with paclitaxel (21). The thrombocyte-sparing effect of paclitaxel was not found to be caused by pharmacokinetic interactions (20) but rather by an antagonistic interaction on thrombocyte precursor cells (36). Despite the platelet-sparing effect of paclitaxel, exposure to carboplatin remained the strongest predictor for thrombocyte toxicity by conventional descriptive analysis, as both C_max and AUC were significantly lower in patients with mild versus severe or absent versus any (P < 10⁻⁴ for all comparisons; Table 4).

Paclitaxel t_{C > 0.05} was significantly higher in patients with at least stable disease during treatment with paclitaxel/carboplatin compared with patients with progressive disease, and t_{C > 0.05} was also significantly correlated with severe neutropenia and with thrombocytopenia (Table 4). t_{C > 0.05} did not segregate between patients with mild or severe thrombocytopenia, but the latter group included only three patients. The presented threshold analysis as well as graphical plots of relative neutropenia/thrombocytopenia as a function of t_{C > 0.1}/t_{C > 0.05} suggest t_{C > 0.05} to be a better surrogate for hematologic toxicity (Fig. 2) and objective response rate (Table 4) in patients with advanced ovarian cancer receiving paclitaxel/carboplatin. Basically, paclitaxel t_{C > 0.05} was the best predictor for treatment response and neutrophil toxicity, whereas carboplatin exposure was the best predictor for thrombocyte toxicity.

We found a MTT of almost 6 days for blood neutrophils and 8 days for thrombocytes. For neutrophils, a similar MTT of 127 h (5.3 days) has been found for paclitaxel by Friberg et al. (23), also using a linear E_drug model. No data on MTT for thrombocytes have been published for paclitaxel to our knowledge. van Kesteren et al. (38) estimated MTT for thrombocytes at 103 h (4.3 days) after the administration of the anticancer agent indisulam using a linear E_drug model. The higher MTT for thrombocytes as found in the presented patient population may also reflect the thrombocyte-sparing effect of paclitaxel. Typical curves for blood counts as outlined in Fig. 1 found a neutrophil nadir of 1,090/μL at 11.5 days and a thrombocyte nadir of 200 × 10³/μL at 15 days. Neutrophil nadir as found in this study is in accordance to what has been reported by Minami et al. (39) in 17 patients with advanced breast cancer. The authors modeled the time course of leukopenia following a 3-h infusion of 210 mg/m² paclitaxel and found an estimated mean leucocyte nadir of 2,000/μL 11.7 days after drug administration, but they did not report the estimated neutrophil nadir. A leucocyte nadir of 2,000/μL as reported by Minami et al., however, would correspond to a neutrophil nadir of ∼1,200/μL if we assume a neutrophil fraction of 60%. A higher neutrophil nadir of 2,320/μL has been reported by Markman et al. (35) in 40 patients with advanced gynecologic malignancies receiving paclitaxel (175 mg/m²) over 3 h followed by carboplatin AUC (6 mg*min/mL). Comparable baseline neutrophil counts and lacking chemotherapy pretreatment in our and Markman's patient group do not explain the difference in neutrophil nadir found. Nadir evaluation in the study by Markman et al. was done 8 to 10 days following the administration of paclitaxel/carboplatin and may have missed the true nadir in a significant part of patients. Simulation studies as done in the presented study are a potent tool for a more precise nadir assessment, but data validation is important to confirm appropriate simulation procedure.

In conclusion, the presented inhibitory linear population pharmacokinetic-pharmacodynamic model predicted neutropenia and thrombocytopenia after the administration of paclitaxel/carboplatin in patients with ovarian cancer. Paclitaxel t_{C > 0.05} was found to be a good predictive marker for severe neutropenia and clinical outcome, whereas carboplatin exposure was a good predictive marker for thrombocytopenia. Prospective studies should assess the value of therapeutic drug monitoring in this patient group.

C. Dittrich, Ludwig Boltzmann-Institute of Applied Cancer Research, Vienna, Austria. J.B. Vermorken, University Hospital Antwerp, Edegem, Belgium. A. Van Oosterom and E. de Bruijn, UZ Gasthuisberg, Leuven, Belgium. D. de Valeriola and M. Piccart, Institut Jules Bordet, Brussels, Belgium. J. Kralovanszky and E. Toth Nat. Inst. Oncologie, Budapest, Hungary. P. Canal and E. Chatelut, Inst. Claudius Regaud, Toulouse, France. C. Ardiet and B. Tranchand, Centre Leon Bérard, Lyon, France. J. Robert, Institut Bergonie, Bordeaux, France. P. Fumoleau, Centre G-F Leclerc, Dijon, France. R. Féty, Centre R Gauducheau, Nantes Saint Herblain, France.

N. Pavlidis and E. Briasoulis, Ioannina University Hospital, Ioannina, Greece. A. Hanauske, AK St. George, Hamburg, Germany. J. Boos and G. Hempel, University of Munster, Munster, Germany. A. Martoni and E. Piana, Policlinico S. Orsola-Malpighi, Bologna, Italy. R. Sorio and I. Robieux, National Cancer Institute, Aviano, Italy. G. Comella and P. Caponigro, National Tumor Institute, Naples, Italy. C.M. Camaggi and E. Strocchi, University of Bologna, Bologna, Italy. J.H.M. Schellens and M. van de Vijver, The Netherlands Cancer Institute, Amsterdam, the Netherlands. J.H. Beijnen, Slotervaart Hospital, Amsterdam, the Netherlands. E.G.E. De Vries and H. Hollema, University Medical Center, Groningen, the Netherlands. D.J. Richel, Medical Spectrum Twente, Enschede, the Netherlands. H.P. Sleeboom, Leyenburg Hospital, Den Haag, the Netherlands. J. Wanders and A.H.G.J. Schrijvers, NDDO-Oncology, Amsterdam, the Netherlands. W.J.F. van der Vijgh, Free University Medical Center, Amsterdam, the Netherlands, M.A. Izquierdo, Catalan Institute of Oncology, Barcelona, Spain. M. Karlsson, Uppsala University, Uppsala, Sweden. T. Cerny, Kantonsspital St.Gallen, St. Gallen, Switzerland. D.I. Jodrell, University of Edinburgh, Edinburgh, United Kingdom. A.H. Calvert, D.R. Newell, and A. Boddy, University of Newcastle upon Tyne, Newcastle upon Tyne, United Kingdom. Ch. Twelves, Glasgow University, Glasgow, United Kingdom. L.S. Murray, Glasgow University, Glasgow, United Kingdom. H.L. McLeod and J. Cassidy, University of Aberdeen, Aberdeen, United Kingdom.

Centers that did bioanalysis of paclitaxel: E. de Bruijn, UZ Gasthuisberg, Leuven, Belgium. G. Hempel, University of Münster, Münster, Germany. E. Strocchi, Universita di Bologna, Bologna, Italy. J.H. Beijnen, Department of Pharmacy and Pharmacology, Slotervaart Hospital/The Netherlands Cancer Institute, Amsterdam, the Netherlands. M. Izquierdo, Catalan Institute of Oncology, Barcelona, Spain. A. Boddy, University of Aberdeen, Aberdeen, United Kingdom.

Centers that did bioanalysis of carboplatin: R. Féty, Centre R Gauducheau, Saint Herblain, France. E. Chatelut, Centre Claudius Regaud, Toulouse, France. W.J.F. van der Vijgh, Free University Medical Center, Amsterdam, the Netherlands. A. Boddy, University of Aberdeen, Aberdeen, United Kingdom.