Accuracy, Feasibility, and Clinical Impact of Prospective Bayesian Pharmacokinetically Guided Dosing of Cyclophosphamide, Thiotepa, and Carboplatin in High-Dose Chemotherapy Free

Cyclophosphamide, thiotepa, and carboplatin (CTC) are alkylating agents widely used in high-dose combination chemotherapy regimens with bone marrow or peripheral blood cell transplantation (1). These three compounds are administered simultaneously in the high-dose CTC regimen, often applied in the treatment of metastatic breast cancer and relapsing germ cell cancer (2–6).

Cyclophosphamide is a prodrug that requires enzymatic bioactivation to manifest its cytostatic activity. After administration, cyclophosphamide undergoes a sequence of activating and inactivating pathways, with ∼75% to 80% being activated to 4-hydroxycyclophosphamide (4OHCP). Various cytochrome P450 isoenzymes are involved in the bioactivation of cyclophosphamide, of which CYP2B6 has the highest activity. 4OHCP is very unstable and decomposes into phosphoramide mustard, the ultimate alkylating metabolite. 4OHCP plasma levels are expected to reflect the intracellular levels of phosphoramide mustard. Circulating phosphoramide mustard in plasma does not contribute to cytotoxicity because it is largely ionized at physiologic pH and does not enter cells. The metabolism of cyclophosphamide shows autoinduction, which leads to increased rate of bioactivation of cyclophosphamide after repeated administrations. Simultaneous administration of thiotepa has been shown to cause inhibition of cyclophosphamide bioactivation (7).

Thiotepa is rapidly oxidatively desulfurated to yield its active metabolite N,N′,N-triethylenephosphoramide (tepa), a reaction catalyzed by the cytochrome P450 isoenzyme subfamilies 3A, 2B, and 2C (8–10). Tepa has pharmacologic properties similar to parent thiotepa and augments its effect (10–12). Because tepa has a longer elimination half-life than thiotepa, it contributes significantly to therapeutic response and toxicity (8). Cyclophosphamide has been shown to induce the conversion of thiotepa to tepa (13).

The pharmacokinetics of carboplatin are relatively simple, with glomerular filtration accounting for almost all drug elimination (14). In patients with normal renal function, 60% to 70% of the dose are excreted into urine within the first 24 hours (14). The remainder of the drug binds irreversibly to body proteins and tissue. The free, ultrafilterable platinum fraction is considered pharmacologically active.

Substantial differences in pharmacokinetics of CTC between individuals have been established, resulting in markedly different exposures, as expressed by the area under the plasma concentration-time curve (AUC), to the individual drugs and their metabolites in patients treated at the same dose levels (15). In high-dose chemotherapy protocols, such as the CTC regimen, doses of the individual compounds are chosen maximal to maximize the benefit of therapy. Doses are limited by the occurrence of serious nonhematologic organ toxicities, such as severe mucositis, ototoxicity, neuropathy, cardiotoxicity, and hepatotoxicity, as seen in the high-dose CTC regimen (15, 16). Relationships between exposure to CTC (and their metabolites) and toxicity have been established. An inverse correlation between the cyclophosphamide AUC and both treatment-related cardiotoxicity and (event-free) survival in women with breast cancer have been reported (17, 18). In addition, an indication for a relationship between the AUC of 4OHCP and phosphoramide mustard and veno-occlusive disease (VOD) of the liver was found (15). The AUC of both thiotepa and tepa after high-dose therapy have been associated with response and nonhematologic toxicity (19, 20), elevation of transaminases (15), and occurrence of mucositis (15, 21). Toxicities such as nephrotoxicity, ototoxicity, central nervous system toxicity, and peripheral nervous system toxicity have been associated with a higher carboplatin exposure in high-dose regimens (15, 22–26). These relationships form the rationale to design a strategy to decrease variability in exposure to decrease morbidity without comprising efficacy.

In this study, we exploited the benefits of therapeutic drug monitoring (TDM) of CTC in chemotherapy. Population pharmacokinetic models describing the pharmacokinetics of cyclophosphamide and 4OHCP (13), thiotepa and tepa (13, 27), and carboplatin (28), as developed in our institute, formed the basis for the TDM strategy. The primary aim of the study was to investigate the ability to obtain target exposures of 4OHCP, thiotepa and tepa, and ultrafilterable carboplatin based on real-time pharmacokinetic follow up. In addition, the technical feasibility of this approach was evaluated. The secondary aim was to evaluate whether patients receiving pharmacokinetically guided individualized doses in the CTC regimen had less toxic events compared with a historical reference population receiving conventional doses without pharmacokinetically guided dose (PGD) adaptations.

Patients and Treatment. Patients were included in several clinical studies that employed the CTC high-dose chemotherapy regimen with peripheral blood progenitor cell transplantation (2–6). Patients either had high-risk primary breast cancer and received high-dose chemotherapy as part of their adjuvant treatment or had advanced breast, germ cell, or ovarian cancer. Two different high-dose CTC schedules were administered. The full-dose CTC regimen (2, 3, 5) consisted of 4 days of chemotherapy with cyclophosphamide (1,500 mg/m²/d) as a 1-hour infusion immediately followed by carboplatin [target AUC 5 mg × min/mL/d as calculated with the Calvert et al. (29) formula using the Cockcroft-Gault (30) formula for estimating creatinine clearance] as a daily 1-hour infusion and thiotepa (120 mg/m²/d) divided over two 30-minute infusions (the second daily dose of thiotepa was administered 12 hours after the first dose). The “tiny” CTC regimen (tCTC) was identical to the CTC regimen, except that it incorporated two thirds of the dose of each agent (4, 6). Patients received either one or two courses of CTC or two or three courses of tCTC, when possible, every 4 weeks.

Mesna (500 mg) was administered six times daily for a total of 36 doses, beginning 1 hour before the first cyclophosphamide infusion. All patients received antiemetics both prophylactically and as indicated, which usually included dexamethasone and granisetron. Patients received prophylactic antibiotics, including ciprofloxacin and amphotericin B p.o., starting 4 days before chemotherapy. Approximately 60 hours after the last thiotepa infusion, the peripheral blood progenitor cells were reinfused. The details of the CTC and tCTC regimens have been published previously (2–6).

All protocols were approved by the Committee of Medical Ethics of the Netherlands Cancer Institute and written informed consent was obtained from all patients.

Sampling and Analyses. During the 4-day CTC course, blood samples were collected from a double lumen i.v. catheter inserted in a subclavian vein. Samples were collected before the start of the infusions on all 4 days of chemotherapy. Complete pharmacokinetic profiles were assessed on two separate days, always including day 1 and day 3 or 4. On these two days, samples were taken at 30 minutes after the start of cyclophosphamide infusion and at 60 (end of cyclophosphamide infusion and start of carboplatin infusion), 90, 120 (end of carboplatin infusion and start of thiotepa infusion), 150 (end of thiotepa infusion), 165, 180, 210, 285, 390, and 660 minutes. On day 5, an additional sample was collected ∼22 hours after the last cyclophosphamide infusion.

After blood sampling, samples were immediately placed on ice. Plasma was separated by centrifuging the sample at 3,500 × g for 3 minutes at 4°C. A 500 μL volume of plasma was immediately added to 50 μL of a 2 mol/L semicarbazide solution and incubated for 2 hours at 35°C for the stabilization of 4OHCP. 4OHCP is an unstable compound and therefore requires immediate derivatization to a more stable derivative (31). Plasma ultrafiltrate was prepared immediately by transferring 500 μL plasma in an Amicon micropartition system with a 30-kDa YMT-14 membrane (Millipore Corp., Bedford, MA) and centrifuging the system at 2,500 × g for 20 minutes. All samples were stored at −70°C until analysis.

Analysis of platinum in ultrafiltrate was done using flameless atomic absorption spectrometry (32). Accuracy and within-day and between-day precision were <10%. In the first 20 patients, thiotepa and tepa concentrations were quantified with a validated gas chromatographic assay (33). Accuracy and within-day and between-day precision were <6% for both compounds. Later, we developed a rapid analysis method based on high-performance liquid chromatography-mass spectrometry/mass spectrometry to simultaneously quantify cyclophosphamide, 4OHCP, thiotepa, and tepa in one sample (31). From the time that this method was operable, cyclophosphamide dose adaptations could be done. Accuracy and within-day and between-day precision of this method were <15% for all compounds.

Definition of the Target Exposure. For cyclophosphamide dose adaptations, a 4OHCP target AUC (AUC_4OHCP) was defined because 4OHCP is the activated component of the prodrug cyclophosphamide. Because both thiotepa and tepa have similar alkylating activities, the sum of both thiotepa and tepa AUC (AUC_{TT + T}) was targeted. For carboplatin, the ultrafilterable AUC_CA was targeted. Because no quantitative relationships have been established between toxicity/efficacy and exposure to cyclophosphamide, thiotepa, carboplatin, and their relevant metabolites, the optimal dose of the three compounds in the CTC regimen is not known. To define a safe and effective target exposure for 4OHCP, thiotepa and tepa, and carboplatin, their median values as obtained in complete conventional tCTC and CTC courses in a reference population were calculated. This reference population was similar to the population used for development of the population pharmacokinetic models used in this study and consisted of 51, 42, and 42 patients for 4OHCP, thiotepa and tepa, and carboplatin, respectively. Because the “median” patient in this regimen did not experience extreme toxicities, the median AUC is expected to be a safe target exposure. Target AUCs for complete tCTC and CTC courses were 105 and 140 μmol/L × h for AUC_4OHCP, 276 and 374 μmol/L × h for AUC_{TT + T}, and 13.3 and 20 mg × min/mL for AUC_CA.

Population Pharmacokinetic Analyses. Bayesian dose adaptations were done using population pharmacokinetic models for cyclophosphamide and 4OHCP (13), thiotepa and tepa (13, 27), and carboplatin (28) as developed with the nonlinear mixed effect modeling program NONMEM (double precision, version 1.1; ref. 34).

Carboplatin. Ultrafilterable platinum data were described with a two-compartment model with first-order elimination clearance (Cl_CA), volume of distribution (V_CA), and distribution rate constants k₁₂ and k₂₁ (28). A total of 42 patients receiving 67 courses of tCTC or CTC were used as a reference population for estimating the population pharmacokinetic variables for this model. In Fig. 1, the model used for carboplatin is incorporated. In Table 1, the population pharmacokinetic variables of carboplatin as used in the model are summarized.

Thiotepa. The population pharmacokinetic model used in the pharmacokinetic analyses of thiotepa was published recently (13). In this model, the distribution of both thiotepa and tepa was described to take place over two compartments with first-order elimination from the central compartment. The conversion of thiotepa to its metabolite tepa was induced in the presence of cyclophosphamide (total thiotepa clearance increased with 10-25% during a course). The pharmacokinetic variables estimated for thiotepa were apparent inducible clearance leading to formation of tepa (Cl_TTind), apparent noninducible clearance (Cl_TTnonind), volume of distribution (V_TT), and distribution microconstants k₁₂ and k₂₁. For tepa, these variables were elimination rate constant (k_T), volume of distribution (V_T), and distribution microconstants k₃₄ and k₄₃. The induction of thiotepa elimination was modeled using a hypothetical enzyme compartment (ENZ_TT) in which the amount of enzyme increased linearly in the presence of cyclophosphamide with a zero-order rate constant of k_ENZTT. Cl_TTind was directly proportional to the amount in this hypothetical enzyme compartment. This model was part of an integrated model (13), describing a mutual drug-drug interaction of cyclophosphamide and thiotepa, as shown in Fig. 1 and Table 2, which is described further in the section on cyclophosphamide.

Because the influence of cyclophosphamide on the pharmacokinetics of thiotepa was only recognized during this study, dose adaptations for thiotepa in the first 31 patients of this study were done using a similar model as described above, only without the inductive effect of cyclophosphamide (27). This model has also been published previously. Dose adaptations in the final patients of our study were done with the improved model (13).

Cyclophosphamide. The pharmacokinetic model developed for cyclophosphamide (13) also contains thiotepa because it has been shown that thiotepa inhibits the conversion of cyclophosphamide to 4OHCP. This model was developed using plasma concentration-time data of cyclophosphamide, 4OHCP, thiotepa, and tepa from 49 patients receiving 86 courses of CTCor tCTC. Pharmacokinetics of cyclophosphamide were described with a two-compartment model and 4OHCP with a one-compartment model. Cyclophosphamide was eliminated by a noninducible route (Cl_CPnonind) and an inducible route (Cl_CPind), the latter leading to formation of 4OHCP. The apparent clearance of the inducible route leading to 4OHCP was directly proportional to a hypothetical amount of enzyme (ENZ_CPact). Autoinduction led to a zero-order increase (k_enzCP) in amount of this enzyme during treatment in the presence of cyclophosphamide. The influence of thiotepa on the bioactivation of cyclophosphamide was modeled as a thiotepa concentration-dependent reversible deactivation of the enzyme resulting in the formation of an inactive enzyme (ENZ_CPinact). Mass transport between both activated and inactivated enzyme pools was modeled using an association and dissociation rate constant (k_ass and k_diss), corresponding to deactivation by association of the active enzyme with thiotepa and reactivation by the dissociation of inactive enzyme and thiotepa.

For more details on the model, we refer to ref. 13. The model is schematically represented in Fig. 1 with the involved pharmacokinetic variables summarized in Table 2.

PGD Strategy. Prospective dose interventions took place both during and between courses. During days 1 and 2 of the first course, patients received the standard CTC dose. Blood samples were collected on day 1. Plasma concentrations of cyclophosphamide, 4OHCP, thiotepa, tepa, and carboplatin were determined immediately after collection in the sample collected until 285 minutes after the start of the course. The concentrations in these samples were available the next day. Based on these data, Bayesian estimates were generated for the individual pharmacokinetic variables of cyclophosphamide, 4OHCP, thiotepa, tepa, and carboplatin. These estimates were obtained with the POSTHOC option of NONMEM taking both population pharmacokinetic variables and individual data into account (34). Based on the estimated individual pharmacokinetic variables, an appropriate dose for days 3 and 4, to approach the defined target values, was calculated.

For dose adjustment during a course, the defined target values were used without correction for the exposure already obtained during days 1 and 2 of the course to prevent the administration of excessive high or low doses after adjustment to compensate for very low or very high exposures on the first 2 days of the course.

Before second and third courses, Bayesian estimates were generated based on all results of the previous course(s). The doses for days 1 and 2 of these courses were again based on approaching the defined target exposures. During second and third courses, doses were again adjusted on days 3 and 4 based on the concentrations in samples collected at day 1 of that course and data of previous course(s), which is similar to the procedure employed during course 1. In case it was proven impossible to adjust the dose during the course, the dose of days 1 and 2 was maintained for the other days of the course.

Pharmacokinetic Validation. To evaluate the obtained AUCs after dose adaptations, individual pharmacokinetic variables of cyclophosphamide, 4OHCP, thiotepa, tepa, and carboplatin in a specific course were calculated by individual fits of the data using the developed models. With this method, we obtained independent values of the pharmacokinetic variables, unbiased toward the population model. Based on these individual pharmacokinetic variables, the AUCs of 4OHCP, thiotepa and tepa, and carboplatin were calculated for that specific course. Subsequently, the total course AUC the patient obtained by dose adjustment (AUC_adapted) was compared with the AUC the patient would have obtained in case the standard dose was administered during the whole course (AUC_standard). Both AUC_adapted and AUC_standard were compared with the defined target AUC.

Toxicity. When the dose of at least one compound was adapted in a course, the toxicity data of this patient were included in the analysis. Toxicity was scored during CTC chemotherapy and after each course and was graded according to the National Cancer Institute Common Toxicity Criteria version 2.0 (35). Because some toxicity were infrequent, they were registered in a dichotomous (e.g., no toxicity or toxicity of any grade) or descriptive way. Toxicity data were compared with toxicity data reported in a reference population of 43 patients who received 75courses of high-dose CTC with standard doses as published previously (15). Patient characteristics of this population were similar as the population in our study.

Significance of the difference in toxic outcome between the two groups was compared using the χ² test.

A total of 46 patients have been included in the study, who received a total of 108 CTC or tCTC courses. Because the infrastructure for cyclophosphamide dose adaptations only became available during the studies, dose adaptations for cyclophosphamide were only done in the last 26 patients (56 courses). In Table 3, baseline patient characteristics are summarized. These characteristics were comparable with those in the populations used for development of the pharmacokinetic models of CTC.

The proposed dosing strategy for CTC seemed to be technically feasible in clinical practice. Of all patients, full pharmacokinetic profiles could be obtained of day 1 and day 3 or 4 of the course, which could be used for the adjustment of the dose and the evaluation of the dose adjustment. In six patients, the planned overnight analyses of thiotepa and tepa in the samples obtained at day 1 was not possible due to technical problems with the method of analysis (33), preventing a dose adjustment during the course. This problem did not occur with the rapid high-performance liquid chromatography-mass spectrometry/mass spectrometry (31).

Carboplatin. Dose adaptations for carboplatin were done in 46 patients. A total of 65 adaptations were done during the CTC courses and 43 adaptations before the start of a new course. In Fig. 2A, the results of the dose adaptations of carboplatin are graphically shown. The individualized doses approach the target with a greater precision compared with the situation in which a conventional dose was administered. In Table 4, the accuracy and precision of the PGD adaptations are presented. Overall, after Bayesian dose adaptations during a course, target AUC_CA levels were approached with a mean precision of 13% versus 30% for standard creatinine clearance–based dosing. Adaptations between courses resulted in target AUC_CA levels with a mean precision of 19% versus 31%, respectively. Highly deviating exposures from the target AUC were effectively prevented by dose adaptation, which can be concluded from Table 5. A total of 95% of the doses adapted during a course resulted in exposures within ±25% of the target compared with 69% without adaptations. Doses of six patients, who would have received exposures with >50% deviation from the target (range, 51-127%), were successfully adjusted during the course to reach exposures <30% of the target (range, 6-29%). With dose adaptations before the second and third courses, doses of six patients who would have obtained deviations >50% (range, 51-82%) were adapted to obtain exposures in the range 9% to 51%. Median percentage carboplatin dose change during courses was 6.8% (range, −54% to 28%). Between courses, the median change in dose was −11% (−37% to 18%).

Thiotepa. The results of the thiotepa dose adaptations can be divided into two parts. In the first 31 patients (78 courses), the dose adaptations were based on a pharmacokinetic model without recognition of induction of thiotepa metabolism by cyclophosphamide (27). In the following 13 patients (22 courses), the full model as shown in Fig. 1 and Table 2 was used (13). Because of the different models applied, the obtained results are presented separately.

Using the first model, a total of 34 thiotepa dose adaptations were done before start of a second or third course and 41 adaptations were done during a course. From Fig. 2A, it is clear that the dose adaptations of thiotepa resulted in a better approximation of the target compared with conventional dosing. In Table 6A, the accuracy and precision of the PGD adaptations are presented. Overall, after Bayesian dose adaptations during a course, target AUC_{TT + T} levels were approached with a mean precision of 16% versus 40% for standard dosing. Adaptations between courses resulted in target AUC_{TT + T} levels with a mean precision of 35% versus 61%. Median percentage dose change during courses was −9% (range, −51% to 42%). Between courses, the median change in dose was −4.4% (−47% to 46%). Results obtained with this model were slightly positively biased.

Using the full model, a total of 17 dose adaptations were done during courses and 6 between courses. Results are presented in Table 6B. Overall, after Bayesian dose adaptations during a course, target AUC_{TT + T} levels were approached with a mean precision of 16% versus 26% for standard dosing. Adaptations between courses resulted in target AUC_{TT + T} levels with mean precision of 14% versus 24%. Median percentage dose change during the course was 6.2% (range, −39% to 34%). Between courses, the median change in dose was 0% (−24% to 17%). Results obtained with this model were slightly negatively biased (obtained exposures were lower than the target exposures), although the number of patients was low.

With both models, highly deviating exposures from the target AUC were effectively prevented by dose adaptation, as can be seen in Table 5. Only 10% of the doses adapted during a course resulted in exposures not within ±25% of the target compared with 40% in case of no dose adaptation. Exposures >50% above the target were successfully prevented with during-course dose adjustments because all 9 patients who would have had these exposure (range, 50–149%) were within 35% of the target after the adjusted doses (range, −5% to 35%). With adaptations before courses 2 and 3, exposures >50% were prevented in 11 patients (range, 52–203%) to result in exposures within the −11% to 103% range.

Cyclophosphamide. A total of 39 cyclophosphamide dose adaptations were done during courses and 17 adaptations before start of a new course. In Fig. 2C, the results of the dose adaptation of cyclophosphamide are graphically shown, and in Table 7, the results are summarized. Overall, after Bayesian dose adaptations during a course, target AUC_4OHCP levels were approached with a mean precision of 19% versus 29% for standard dosing. Adaptations between courses resulted in target AUC_4OHCP levels with mean precision of 19% versus 23% for standard dosing. Median percentage dose change during the course was 0% (range, −47% to 55%). Between courses, the median change in dose was −3% (−23% to 24%). Also for cyclophosphamide, extremely high deviations from the target exposure were prevented (Table 5) because 85% of the adapted doses resulted in exposures within ±25% of the target AUC compared with 67% in case of no adaptations. Doses in four patients who were to receive exposures with deviations from the target >50% (range, −55% to 89%) were successfully adjusted during the course to receive exposures within ±31% of the target value (range, −30% to 24%).

Toxicity. In general, there were few toxic events in the individualized patient population and no toxic deaths were encountered. In comparing the toxicities of our population with those of the reference population (15), we focused on the occurrence of serious toxicities with a possible relationship with exposures to cyclophosphamide, thiotepa, carboplatin, or their metabolites. In Table 8, the toxicities in the population with adapted doses are summarized and compared with those obtained in the reference population receiving fixed doses.

VOD occurred in three patients after their second CTC course, with ascites and grades 3 to 4 toxicity of alanine aminotransferase, aspartate aminotransferase, and bilirubin. Because the occurrence of VOD has been correlated with 4OHCP exposure (15), it is important to notice that cyclophosphamide doses were not adapted in these three patients. In the 26 patients of our study population in which cyclophosphamide doses were adapted, not a single case of VOD occurred. In the reference population, VOD occurred in two patients after a second and third course of tCTC. It therefore seemed that the occurrence of VOD was decreased in patients receiving adapted doses of cyclophosphamide.

All other serious events were of similar frequency and severity in the adapted dose group compared with the conventional dosed group. Three patients (also receiving adapted cyclophosphamide doses) developed hemorrhagic cystitis after a first CTC course, a second CTC course, and a third tCTC course. In the reference population, two patients developed this toxicity. During our study, four patients developed sinus-tachycardia grade 1 after a first (n = 3) or second (n = 1) tCTC course. In two of these patients, cyclophosphamide dose adaptations were done. In the reference population, grade 1 cardiotoxicity was seen in three patients. One patient, not receiving cyclophosphamide dose adaptations, developed pneumonia, a side effect ascribed to cyclophosphamide treatment. Two patients receiving CTC without cyclophosphamide dose adaptations had dyspnea grade 2 after a first and second course, whereas three patients receiving tCTC with adapted cyclophosphamide doses had dyspnea grade 2 after their first tCTC course. In the reference population, pulmonary toxicity grades 1 and 2 was reported in six courses. One patient developed mucositis grade 4 after a first tCTC course; therefore, treatment was discontinued. Mucositis grade 3 occurred after a first CTC course (n = 2), a first tCTC course (n = 1), and after a second tCTC course (n = 1). In the reference population, grade 3 mucositis was observed after single courses of CTC in six patients. Grade 3 neuropathy was observed in four patients (seven courses) after first, second, and third courses. In the reference population, grade 3 neurotoxicity was never observed. Exposures to carboplatin in the patients with grade 3 neuropathy were not extremely deviating from the target exposures. A relatively high incidence of ototoxicity was observed. Grade 2 ototoxicity after first, second, and third courses was reported in 11 patients (12 courses). Similar results were obtained in the reference population.

Toxicity in high-dose CTC chemotherapy may be severe and sometimes life threatening. Individualized dosing may be applied to minimize interpatient variability in drug exposure to maximize the benefit of therapy while keeping the risk of serious adverse effects at an acceptable level. Especially for anticancer agents like cyclophosphamide and thiotepa that display a high interpatient pharmacokinetic and pharmacodynamic variability, and for which clear exposure-response relationships have been identified, it may be beneficial to monitor plasma drug concentrations to improve treatment outcome. However, patients may only benefit from an adapted dose when the adaptation is done in an early stage of treatment. Real-time TDM therefore requires rapid sample analysis with results available the next morning and accurate and reliable assays for the determination of all analytes. These techniques are available in our laboratory for CTC, allowing fast dose adaptations already from the third day of the course on.

The primary objective of this study was to prospectively validate the performance of the applied Bayesian dose adaptation strategy for CTC in approaching desired plasma levels. This study showed that adaptation of the CTC doses resulted in less variability in exposure between patients compared with conventional dosing. Moreover, patients with a highly deviating pharmacokinetic profile were effectively recognized and their dose was adjusted to approach the target exposure. Dose adaptations during a course generally led to a better approximation of the target exposure compared with adaptations between courses, which may be explained by substantial intrapatient course-to-course variability.

Initial carboplatin doses are usually already individualized based on renal function of a patient. Appropriate a priori doses can be based on the Calvert formula using an individuals glomerular filtration rate, as determined using the ⁵¹Cr EDTA clearance, as input (29). Due to imprecise estimations of glomerular filtration rate, using creatinine clearance estimated by the Cockcroft-Gault (30) formula as in our study, this dosing method may still result in large individual differences in carboplatin exposure (28). Retrospective studies showed that TDM of carboplatin is a possibility for a better approximation of a desired carboplatin exposure (28, 36). In our study, we confirmed this prospectively. Highly precise Bayesian estimations of carboplatin pharmacokinetic variables can be obtained using only a few blood samples (28, 36).

The dose adaptations for thiotepa using the primary developed model resulted in more precise but slightly positively biased results. The recognized interaction between cyclophosphamide and thiotepa, with cyclophosphamide inducing thiotepa metabolism (13), seemed to be responsible for a significantly increased clearance of thiotepa during a CTC course (10–15%), resulting in more tepa formation in time. Because tepa has a longer half-life than thiotepa, AUC_{TT + T} was therefore increased. Not taking this interaction into account resulted in an underestimation of AUC_{TT + T} at the end of a CTC course with concomitant dose adaptations that were too low. Adaptation of thiotepa doses with the newly developed model, taking into account the induction of metabolism by cyclophosphamide, however, resulted in slightly negatively biased results due to overestimation of the tepa concentrations. More patients should be included to prospectively validate the accuracy of this model.

TDM of cyclophosphamide faces the challenge of a prodrug undergoing a complex metabolism, producing both active and inactive products (7). Because the kinetics of cyclophosphamide itself may not be predictive for the activation of this drug (37–40), we based dose adjustments of cyclophosphamide on the pharmacokinetic profile of 4OHCP. A complicating factor in the individualization of cyclophosphamide dose is the nonpredictable variation in enzyme activity during a course. This enzyme activity is influenced by both autoinduction and thiotepa and becomes apparent only until after the second day of treatment and cannot be fully predicted from data of day 1. Estimating the extent with which cyclophosphamide induces the conversion of thiotepa to tepa during a course is similarly complicated when only day 1 data are available.

It should be clear that approximation of desired target exposures is a surrogate end point because we are mainly interested in the clinical results of the dose individualization. Therefore, the secondary objective of the study was to evaluate the incidence and severity of toxicity obtained in the individualized dosing group compared with patients receiving conventional doses. The previously established pharmacokinetic-pharmacodynamic relationships of CTC only accounted for exposures to individual compounds and not their combined effects. Nevertheless, the compounds have overlapping toxicities; therefore, adapting the doses of all three agents simultaneously was done. In our study, no clear difference in toxic outcome was detected in patients receiving adapted doses compared with patients receiving conventional doses. Only the incidence of VOD seemed to be reduced in patients who received an adapted cyclophosphamide dose. A factor highly contributing to these findings is the heterogeneity of response in patients. Observed relationships between CTC pharmacokinetics and toxicity, as outlined in INTRODUCTION, were not very strong but significant. Therefore, large number of patients will be necessary to establish significant differences in pharmacodynamic outcome between individualized and reference populations. Moreover, the relatively low number of severe toxic events in both our population and the reference population necessitates large patient populations to show a beneficial effect of TDM.

In this prospective study, we also showed the feasibility of the proposed dosing strategy in clinical practice. In this study, all planned dose adaptations were done, with the exception of a few adaptations for thiotepa due to bioanalytic problems. In general, for fast and routine dose adaptations, trained personnel and short lines among doctors, technicians, and clinical pharmacists are aprerequisite. A trained technician for the bioanalysis as well as a clinical pharmacologist able to perform and interpret the mathematically complex pharmacokinetic calculations have proven to be indispensable for rapid dose individualizations.

In conclusion, Bayesian PGD of CTC has proven to be technically and logistically feasible in the short interval of a 4-day course. The dosing strategy led to a marked reduction in the variability of exposures to 4OHCP, thiotepa and tepa, and carboplatin, especially when dose adaptations were done during a course. Our study did not show a clear benefit for individualized dosing in the CTC high-dose chemotherapy regimen compared with standard dosing in terms of reduced toxicity, although an indication for reduction of the occurrence of VOD was found.