## Abstract

**Purpose:** We aimed to evaluate the performance of therapeutic drug monitoring (TDM) approach in controlling interpatient variability of carboplatin exposure (AUC) in patients treated with TI-CE high-dose chemotherapy for advanced germ cell tumors and to assess the possibility of using a formula-based dosing method as a possible alternative.

**Experimental Design:** Eighty-nine patients receiving carboplatin for 3 consecutive days during 3 cycles were evaluable for pharmacokinetic study. Blood samples were taken on day 1 to determine the carboplatin clearance using a Bayesian approach (NONMEM 7.2) and to adjust the dose on day 3 to reach the target AUC of 24 mg.min/mL over 3 days. On days 2 and 3, samples were taken for retrospective assessment of the actual AUC. A population pharmacokinetic analysis was also performed on 58 patients using NONMEM to develop a covariate equation for carboplatin clearance prediction adapted for future TI-CE patients, and its performance was prospectively evaluated on the other 29 patients along with different methods of carboplatin clearance prediction.

**Results:** The mean actual AUC was 24.4 mg.min/mL per cycle (22.4 and 26.8 for 10th and 90th percentiles, respectively). The new covariate equation [CL (mL/min) = 130.7 × (Scr/83)^{−0.826} × (BW/76)^{+0.907} × (Age/36)^{−0.223} with Scr in μmol/L, BW in kilograms, age in years] allows unbiased and more accurate prediction of carboplatin clearance compared with other equations.

**Conclusions:** TDM allows controlling and reaching the target AUC. Alternatively, the new equation of carboplatin clearance prediction, better adapted to these young male patients, could be used if TDM cannot be implemented. *Clin Cancer Res; 23(23); 7171–9. ©2017 AACR*.

### Translational Relevance

Therapeutic drug monitoring (TDM) does not represent a routine practice for cytotoxic drugs due to their usual schedule of administration (i.e., day 1–day 21 schedule), which does not allow a dose adjustment within a cycle. Carboplatin is intensified in the TI-CE regimen for germ cell cancer treatment in order to reach a 24 mg.min/mL target AUC over 3 days. The results of this prospective pharmacokinetic clinical study demonstrate the feasibility and the benefit of conducting TDM to control carboplatin plasma exposure to accurately reach the target AUC in patients treated with this high-dose regimen. Moreover, we propose a new formula of carboplatin clearance prediction more adapted to these particular young male patients than the Calvert formula using Cockcroft–Gault for glomerular filtration rate estimation. This new equation (based on age, body weight, and serum creatinine) can be used as a putative alternative in cases where TDM cannot be implemented.

## Introduction

High-dose (HD) chemotherapy and stem-cell rescue have been evaluated in salvage treatment (first or second salvage) or in first-line treatment in the poor-risk group of advanced germ cell tumors according to the International Germ Cell Cancer Collaborative Group. In 2000, Motzer and colleagues demonstrated the benefit of dose-intensified chemotherapy scheme (TI-CE high-dose chemotherapy) with peripheral blood–derived stem-cell (PBSC) support in patients who failed to conventional-dose salvage treatment (1). This TI-CE regimen combines two cycles of paclitaxel plus ifosfamide followed by three cycles of HD carboplatin and etoposide. Substitution of cisplatin by carboplatin allowed dose escalation as toxicity profile of carboplatin is more favorable. Einhorn and colleagues (2) retrospectively analyzed HD as third-line treatment in patients with metastatic germ cell tumors for which both etoposide and carboplatin daily doses were calculated according to body surface area (BSA; i.e., 750 and 700 mg/m², respectively).

After a first clinical trial evaluating HD regimen in first-line therapy for patients with poor-risk germ cell tumors for which carboplatin dose was also calculated according to BSA (total dose of 1,800 mg/m²; ref. 3), Motzer and colleagues adopted AUC dosing for carboplatin. They performed a phase I trial consisting of several levels of carboplatin target total AUC (i.e., corresponding to the sum of 3 daily administrations) from 12 to 32 mg.min/mL using the Calvert formula based on a glomerular filtration rate (GFR) measured with radionuclide-labeled ligand technetium-99m DTPA plasma clearance to calculate the dose (1). Despite this individualization of the carboplatin dose, the mean measured AUC in patients included at the 24 mg.min/mL AUC level was lower than the target and displayed a high variability with observed AUCs ranging from 12 to 48 mg.min/mL. Therefore, Kondagunta and colleagues (4) conducted a second trial in which the GFR was calculated with the Jelliffe's equation based on serum creatinine (Scr), age, and BSA. The value of 24 was chosen as recommended target total AUC, but the mean observed AUC for this level was 21.6 mg.min/mL, with values ranging from 10.9 to 36.7 mg.min/mL. Given the toxicity (i.e., ototoxicity and hematotoxicity) of this HD regimen (5, 6), we first performed a pilot monocenter study in 5 patients (15 cycles) showing the feasibility to control individual carboplatin total AUC using therapeutic drug monitoring (TDM) performed at day 1: the measured total AUC ranged between 24.0 and 26.5 for a target value of 24 mg.min/mL (7). The present TI-CE trial was a national multicenter phase II trial aiming to evaluate the efficacy (primary endpoint) and tolerance of the TI-CE protocol in the treatment of previously treated germ cell tumors in male adult patients. The particularity of this trial is the individualization of carboplatin dose according to TDM in order to reach the target AUC of 24 mg.min/mL. The main objective of the present pharmacokinetic (PK) study was to evaluate the performance of this TDM approach in controlling interpatient variability of carboplatin exposure as well as its feasibility to be conducted at a multicenter level. Secondly, we also aimed to assess the possibility of using a formula-based dosing method as an alternative to the practice of TDM. To this end, we developed and evaluated a covariate model adapted to this specific population of patients based on patient characteristics which are known to influence carboplatin clearance (Scr, body weight, age, and sex) and additionally serum cystatin C (CysC) which has recently been demonstrated to be a marker of carboplatin elimination (8, 9).

## Patients and Methods

### Patients and treatment

The trial (ClinicalTrials.gov reference number NCT00864318) was approved by the ethical committee of Toulouse. The treatment consisted of two cycles (14 days apart) of paclitaxel (200 mg/m², D1 over 24 hours) plus ifosfamide (2g/m²/d D2 to D4) and mesna protection. Leukapheresis for PBSC collection started at day 11 and was done for 3 consecutive days. Patients who mobilized more than a total of 9 × 10^{6} CD34^{+} cells/kg did not undergo a second cycle of paclitaxel/ifosfamide. These two cycles of paclitaxel/ifosfamide were followed by three cycles (14- to 21-day intervals) of HD carboplatin (total AUC over 3 days of 24 mg.min/mL) and etoposide (400 mg/m²) given for 3 consecutive days in each cycle with PBSC support. For the PK study and throughout this article, the HD cycles are named C1, C2, and C3.

In total, 101 male patients with relapsed or refractory germ cell tumors were enrolled in 8 different centers in France, and informed consent was obtained from each patient. Twelve patients were not considered for the PK study due to informed consent withdrawal or premature termination of treatment before carboplatin/etoposide cycles on account of unsupported toxicity, disease progression, or death.

Clinical outcome and toxicity will be reported in detail in a separate manuscript.

### Carboplatin administration, blood sampling and platinum analysis.

In each cycle, carboplatin was administered as a daily 1-hour infusion in 5% dextrose for 3 consecutive days. For each carboplatin administration, three blood samples were collected at 5 minutes before the end of infusion, and 1 and 4 hours after the end of infusion. The three samples were selected according to a limited sampling strategy developed previously (10). After immediate centrifugation of the blood samples at 1,500 *g* for 10 minutes at 4°C, 1 mL of plasma was taken and then ultrafiltered (centrifugation at 4°C for 15 minutes at 1,500 *g*) using the Amicon MPS1 micropartition system with YM-T membrane. Carboplatin levels in the plasma ultrafiltrate (UF) obtained at day 1 of each cycle were measured by means of flameless atomic absorption spectrophotometric analysis according to a previously described method (11) in six different hospital laboratories. A cross-validation procedure was done before the start of the study. The mean coefficients of variation using three spiked plasma UF control samples with nominal values of 0.0525, 1.575, and 31.5 mg/L were 6.1%, 6.5%, and 7.2%, respectively. Each laboratory obtained a value within the interval of ±20%, ±10%, and ±10% for the low-, medium-, and high-level control sample, respectively. These intervals were those used for validation of each run in each laboratory. The determination of carboplatin UF concentrations corresponding to samples obtained at days 2 and 3 of the 3 cycles was centralized in the Institut Claudius-Regaud laboratory.

Of note, these PK samples will also be used to determine etoposide concentrations in order to perform a PK analysis of this drug given in the HD setting. These results will be reported in a future manuscript.

### Carboplatin dose adaptation and PK analyses.

The initial carboplatin dose (day 1 and day 2 of cycle 1 HD: C1D1 and C1D2) was calculated as follows: 8 × CLp where 8 is the daily target AUC (target total AUC divided over 3 days) and CLp is the predicted carboplatin clearance calculated with an equation previously published by our team (8): CLp (mL/min) = 110 × (Scr/75)^{−0.654} × (BW/65)^{+0.625} × (age/56)^{−0.507} with Scr in μmol/L, body weight (BW) in kilograms, and age in years. Although this previous work showed the benefit of including CysC in the equation for CLp prediction, we could not use the equation including CysC because the CysC measurement was not implemented in some centers participating in the trial. We then used the alternative equation without CysC. To limit the risk of overdosing patients, a value of predicted carboplatin clearance of 200 mL/min was set as the superior boundary as this value is close to the largest value observed in our previous multicenter study (i.e., 229 mL/min; ref. 9). Consequently, the maximum daily dose to be administered on C1D1 and C1D2 was 1,600 mg.

Actual carboplatin CL on C1D1 was obtained by Bayesian approach using the NONMEM program (version 7, level 2.0) according to a two-compartment PK model and first-order conditional estimation method with interaction (FOCE-I) as previously described in detail (8). No covariates were considered for the typical value of CL; the data of each patient (i.e., 3 UF carboplatin concentrations and dose) were combined with those of a database composed of 143 patients with rich sampling (10, 12) and 45 other patients with limited sampling (ref. 8; i.e., without implementing the dataset during the clinical trial). Individual values of clearance (CL_{D1}) were obtained for each patient. The carboplatin dose of day 3 of cycle 1 (C1D3) was adjusted to obtain a total target AUC of 24 based on the hypothesis that carboplatin CL was constant over the 3 days of the cycle:

Dose_{D3} (mg) = [24 – (Dose_{D1} + Dose_{D2})/actual CL_{D1}] × actual CL_{D1}.

For the subsequent cycles of treatment, the AUC of 24 mg.min/mL remained the target AUC if no major ototoxicity had been observed; otherwise, the target AUC was reduced to 18 mg.min/mL. The first dose was calculated using the day 1′s actual CL of the preceding cycle, and the same procedure of dose adaptation on D3 based on analysis of D1 concentrations was conducted.

## Overall PK Analysis

After the completion of treatment of all the patients, analysis of whole PK data (3 cycles of 3 days) was performed with three objectives: (i) to measure the actual total AUC and compare it with target AUC; (ii) to quantify the intrapatient PK variability corresponding to carboplatin CL; and (iii) to perform a covariate analysis in order to evaluate the possibility of proposing an equation to predict carboplatin CL in patients with germ cell tumors as a putative substitute to TDM to individualize carboplatin dose.

### Determination of total measured AUC

Bayesian estimation of carboplatin CL was performed retrospectively for days 2 and 3 of the 3 cycles according to the method used for TDM at day 1 of each cycle. Actual total AUC for each cycle was obtained by the sum of daily AUC calculated by dividing each daily carboplatin dose by the corresponding carboplatin CL.

### Study of intrapatient PK variability

The intrapatient variability (intra- and intercycle effects) was first assessed by performing a PK analysis of the whole dataset (89 patients at C1, 80 at C2, and 72 at C3, 3 cycles, 3 days per cycle) including interoccasion variability (IOV) in order to determine cycle-to-cycle and day-to-day variability. The model description is provided in Supplementary Data S1. Secondly, in order to evaluate the clinical relevance of the intrapatient variability, a statistical analysis was performed on the individual carboplatin CL values obtained on each day using a mixed-effects model for repeated measures and testing the contrasts of the marginal linear predictions for the effect of days within cycles and the effect of cycles at days 1, 2 and 3.

### Covariate analysis

Based on our previous work (9), the four covariates used for the prediction of carboplatin CL (i.e., Scr, BW, age, CysC level) were evaluated according to allometric equations: TVCL = θ_{1} × (SCr/mean SCr)^{θ2} × (CysC/mean CysC)^{θ3} × (BW/mean BW)^{θ4} × (Age/mean Age)^{θ5}.

CysC plasma was measured from a frozen serum sample by an automated particle-enhanced nephelometric immunoassay at the Institut Claudius-Regaud. The analyzer (BN ProSpec), as well as the controls, standards, and kits (N Latex CysC) were provided by Siemens.

Because CysC was not available in 2 patients, the analysis was performed on 87 patients. Data-splitting was done randomly to create a model-building dataset (58 patients) and a model-validation dataset (29 patients). Only data of C1D1 of the model-building dataset were considered for the covariate analysis. A stepwise backward elimination of each covariate from the full equation was then performed to test its influence on carboplatin clearance. Full and reduced models (one parameter less) were compared by the Khi-2 test of the difference between their respective objective function values (OFV). OFV is equal to minus twice the log likelihood of the data. This value is an indicator of the goodness-of-fit of the model. An increase of at least 3.84 (*P* < 0.05, one degree of freedom) was required to consider the covariate as having significant impact on carboplatin clearance.

The predictive performance of the covariate equations obtained from the above covariate analysis was prospectively assessed using the model-validation dataset (*n* = 29 patients) for each cycle of treatment. For the jth patient, the relative prediction error [pej(%)] for carboplatin CL was calculated with the following equation: Pej (%) = (CL_{pred} – CL_{actual}) ×100/CL_{actual}, where CL_{actual} is the observed value obtained by Bayesian estimation and CL_{pred} is the value of predicted CL from: (i) covariate equations, (ii) the widely used Calvert formula (13) where Cockcroft–Gault or Jelliffe CL_{CR} was used as a surrogate for the GFR, or (iii) the modified Thomas formula which integrates CysC (9). The mean percentage error [MPE = *N*^{−1}.Σ^{N}_{j}_{=1} (pej), where *N* is the number of patients] and the mean absolute percentage error [MAPE = *N*^{−1}.Σ^{N}_{j}_{=1}**|**pej**|**] was computed as a measure of bias and precision, respectively. For statistical analyses, MPE values were compared with the theoretical value of 0 with the Student *t* test. Linear regression was performed between the actual CL and the CL values predicted according to the formulas stated above.

## Results

### HD chemotherapy and carboplatin dose modifications

The main characteristics of the 89 patients are shown in Table 1. Among these 89 patients, 72 patients (81%) received 3 cycles of HD chemotherapy combining carboplatin and etoposide, 9 patients (10%) received only 2 cycles, and 8 patients (9%) received 1 cycle.

Table 2 shows the mean daily dose and the mean total dose per cycle of carboplatin that were administered to the patients. The initial dose (days 1 and 2 of cycle 1) was capped to 1,600 mg for 9 patients. The mean observed carboplatin CL of these 9 patients was 180 mL/min (ranged between 153 and 223 mL/min). On average, in each cycle of treatment, the dose was decreased on day 3. For 8 patients, carboplatin was not administered on the third day because the target AUC of 24 was obtained after dose administration on days 1 and 2. In terms of the total dose over 3 days, the TDM resulted in an absolute dose change greater than 20% (i.e., lower than –20% or greater than 20%) for 20 of 89 patients at cycle 1 (dose change ranging from –33% to +44%) compared with the total dose calculated using solely the predicted CL (i.e., if TDM had not been performed). In cycles 2 and 3, 23 of 80 and 22 of 72 patients had an absolute change of total dose greater than 20%, respectively (total dose change ranging from –42% to +31%, and ranging from –40% to +24%, respectively).

The target total carboplatin AUC was reduced to 18 mg.min/mL for 9 patients of 80 in cycle 2 and 23 patients of 72 in cycle 3 due to toxicity observed during the intercycle period.

### Overall carboplatin exposure (AUC) per cycle

The structural (i.e., two-compartment model) and statistical (proportional error for both interindividual and residual variabilities) population PK model described data very accurately with residual variability ranging from 18.6% to 19.0% depending on the run. The models were also associated with low η_{CL}-shrinkage (between 5% and 6%) and low ε-shrinkage (between 16% and 17%). Supplementary Fig. S2 shows that the individual weighted residuals versus PRED of the run on C1D1 data (as a representative example) were very close to 0, making the carboplatin CL obtained and corresponding AUC likely to accurately estimate the actual values.

Mean AUCs with ranges were 24.3 (18.8–29.4), 24.6 (21.1–30.4), and 24.2 mg.min/mL (21.5–29.4) for cycles 1, 2, and 3, respectively, at the target total AUC level of 24 mg.min/mL, and 18.0 (16.1–21.4) and 17.8 (16.0–19.8) for cycles 2 and 3 when target AUC was 18 mg.min/mL (Fig. 1).

For the AUC 24 level, the 10th and 90th percentiles of the observed AUC were (22.5; 26.8), (22.3; 26.7), and (21.9; 26.5) at cycles 1, 2, and 3, respectively, indicating that the vast majority of patients had a carboplatin exposure close to target value thank to the TDM-based individual dosing.

To further evaluate the TDM in patients with decreased renal function, Fig. 1 was reproduced by considering only the patients with Scr > 120 μmol/L at C1D1 (Supplementary Fig. S3) and shows the benefit of TDM in these particular patients compared with formulas that tend to underestimate their carboplatin CL.

Among the 17 patients that did not undergo the 3 planned cycles, 6 had unacceptable toxicity and 1 patient died because of treatment. These 7 patients received a mean AUC (min–max) of 23.4 mg.min/mL (22.3–24.8) per cycle which excludes carboplatin overexposure as the reason for their severe toxicity.

### Intrapatient variability

The analysis of the whole dataset with IOV allowed us to estimate intraindividual variability of carboplatin CL: the intercycle variability was 8.6%, and the interday variabilities were 8.5%, 6.3%, and 8.7% at cycles 1, 2, and 3, respectively (Supplementary Table S1). These values confirm that the intrapatient variability of carboplatin CL was sufficiently limited to justify the use of a TDM strategy. Because the IOV does not allow defining a trend of variation, Table 3 presents the mean values (± 95% CI) of individual carboplatin clearances observed at each occasion. Both day- and cycle effects were statistically significant in the statistical analysis. For the cycle effect, there was a trend toward a decrease of carboplatin CL from cycle 1 to cycle 3 whichever the day with mean variations (with range) of –11% (–36% to +19%), –9% (–43% to +33%), and –8% (–47% to +21%) for days 1, 2, and 3, respectively. In contrast, for the intracycle variability (day effect), there was no systematic trend.

### Covariate analysis

A NONMEM analysis of C1D1 concentration data of the model-building dataset (58 patients) was performed. The four covariates (SCr, CysC, BW, and age) were all significant to predict carboplatin CL. The deletion of each covariate from the full model was associated with a significant increase (*P* < 0.05) of the OFV as shown in Supplementary Table S2. The highest increase corresponded to BW followed by Scr, CysC, and age. The full equation (± 95% CI) named the TICE 4-cov equation was:

CL (mL/min) = 128.2 ± 6.0 × (SCr/83)^{minus;0.578 ± 0.419} × (CysC/0.9)^{−0.368 ± 0.357} × (BW/76)^{+0.852 ± 0.218} × (Age/36)^{minus;0.195 ± 0.178}

To further assess the benefit of CysC in the prediction of carboplatin CL in this specific population, the equation without this covariate named the TICE 3-cov was considered for prospective evaluation together with the 4-cov equation.

The TICE 4-cov equation and the TICE 3-cov equation were prospectively evaluated for each cycle of treatment using the D1 data of the model-validation dataset. The precision was calculated and compared with previously published equations as stated in Table 4. The MPE values corresponding to the two new equations as well as the ones for Calvert–Jelliffe were not significantly different from 0 (no significant bias), whereas those of Calvert–CG or modified Thomas formula were (*P* < 0.005). In terms of MAPE, the best results corresponded to the two new equations (TICE 3-cov and TICE 4-cov) or Jelliffe with lower values ranging between 7.3% and 11.1% depending on the cycle and the equation. The similar performance of the two new equations indicates that CysC does not improve the predictive performance of carboplatin CL. The TICE 3-cov equation was therefore considered as the final covariate equation.

By considering the whole dataset at C1 for the target AUC of 24 (Fig. 1), the TICE 3-cov equation appeared more performant than Calvert–Jelliffe equation with 2 versus 8 of 79 patients with AUC below 18 mg.min/mL, respectively, but further prospective evaluation is needed because a part of the data was used to raise the TICE 3-cov equation.

The coefficients of determination (*r*²) of regression were 0.62, 0.63, and 0.52 for the Calvert formula, TICE 3-cov formula, and the equation used for calculation of dose at C1D1 (Thomas formula), respectively.

To better compare the performance of the TICE 3-cov equation and the TDM in a prospective population, Fig. 2 shows the discrepancy (presented as relative percentage error) between total dose obtained with each approach and the ideal dose in the patients of the validation cohort. The ideal dose is defined as the total dose yielding the target AUC given that the daily CLs were known. Therefore, 2 patients at cycle 1 and 4 patients at cycles 2 and 3 could not be included in this analysis because one daily CL could not be calculated (PK samples for D3 missing for example).

When the dose was calculated using CL predicted by the TICE 3-cov equation, the absolute percentage error was less than 10% in 22 patients of 27 in cycle 1, 11 of 25 in cycle 2, and 17 of 25 in cycle 3. With TDM, there were 25 patients of 27, 22 of 25, and 20 of 25 in cycles 1, 2, and 3, respectively.

## Discussion

The concept of dose and plasma exposure of platinum compounds is important for the treatment of germ cell tumors. In case of recurrence or relapse after cisplatin, etoposide, and bleomycin, HD chemotherapy including HD carboplatin (TI-CE) represents one way to achieve durable remissions in approximately one-half of patients because dose escalation of cisplatin is not possible due to irreversible oto- and nephrotoxicity. As for standard carboplatin treatment of other tumor sites, the concept of AUC-based dosing of carboplatin became evident. In the TI-CE HD regimen, the value of 24 mg.min/mL was chosen as target AUC (4) based on toxicity (i.e., 50% of DLT at 28 mg.min/mL) observed in a phase I study (1). However, despite the use of Calvert or Jelliffe formulas to calculate the carboplatin dose, a poor control of AUC was observed in previous studies (1, 4). The PK results of the present phase II study corroborate our previous findings (7), showing that individual dosing of carboplatin based on TDM is feasible and allows for the reaching of the target AUC with good adequacy.

We observed a correlation between predicted carboplatin CL and actual CL with coefficients of determination (*r*²) of 0.62 and 0.52 for Calvert using Cockcroft–Gault and the equation we used for dose calculation at C1D1 respectively, but actual CL was poorly correlated to BSA (*r*² = 0.13, data not shown). However, dosing based on both equations would have been associated with carboplatin overdosing, and above all, with a large interindividual carboplatin plasma exposure (Fig. 1). On the contrary, the TDM really allowed us to control individual AUC. The result is in link with the small intraindividual variability of carboplatin CL within each cycle. Although a statistically significant variation of CL was observed within cycle, the mean change was modest and not clinically relevant with a mean IOV < 9%. Patients with extreme values of AUC are the ones with highest intracycle decrease of carboplatin CL (for example, 30 mg.min/mL in a patient with a 38% decrease of CL at cycle 2). Moreover, as illustrated in Fig. 1, the TDM is the method associated with the lower number of underexposed patients, which is also a very important point for this curative treatment.

In contrast, despite the limited value of intercycle IOV, the decrease of carboplatin CL from cycle 1 to cycle 3 was both statistically and clinically significant. As a consequence of this intrapatient variability from one cycle to the following one, the dose at day 3 was decreased in cycles 2 and 3 for several patients, although the dose given at days 1 and 2 was already calculated according to the individual carboplatin CL observed during the previous cycle. This result emphasizes the need for TDM not only in cycle 1 but also in the two following cycles for controlling individual carboplatin exposure. In addition, the extent of the dose change was not negligible because 20% to 30% of patients needed a total dose modification greater than 20%, reaching more than 40% for some patients.

The study also proved the feasibility of TDM for HD chemotherapy at a multicenter level. Several intersite configurations have been successfully implemented during this clinical trial. For three centers, the laboratory was located within the hospital in which patients were treated. For two other sites, carboplatin assay was performed within laboratories in close proximity to the clinical investigation Unit (same city but different hospitals). For the three remaining centers, samples were sent to the Toulouse Pharmacology laboratory and analyzed during day 2 or early day 3 for the adapted dose to be communicated to clinicians for day 3. Although it is feasible, we recognize that the procedure may be difficult to organize in all institutions. For this reason, we prospected for a way to predict carboplatin CL accurately in order to avoid TDM.

The covariate analysis allowed us to obtain an equation (TICE 4-cov) consistent with our previous results in terms of significant covariates (BW, age, SCr, CysC). However, the coefficient relative to each of the covariate was relatively different to those of the equation we developed and prospectively validated from 357 patients who received carboplatin as part of standard (not HD) chemotherapy (9). Surprisingly, the benefit of combining SCr and CysC was lower in the present study than in the previous work as illustrated by (i) the poor precision of the power coefficient values when the two covariates were associated (Table 4) and (ii) the result of the prospective evaluation showing similar performance of the TICE 4-cov and the TICE 3-cov equation based on BW, age, and SCr. These differences could be explained by the demographical (younger males) and functional (relatively good normal renal function) characteristics of the 89 patients in comparison with the previous population of patients.

Moreover, the age difference between the patients of our study and those of the previous PK studies from which CLp equations were obtained (8, 9, 13, 14) explains the overestimation of CLp when those formulas were used in these young patients of the present study. On the contrary, in the 89 patients of our trial, we observed a trend for CLp underestimation with the Calvert–Jelliffe equation (median value of 22.7 mg.min/mL) at cycle 1 with 10.1% of patients who would have had an AUC below 18 for a target at 24, as it was previously observed in these young patients (4, 15). The TICE 3-cov equation we developed seems to be better in terms of bias but needs further prospective evaluation. Therefore, in situations where it is impossible to perform TDM for this HD chemotherapy, the TICE 3-cov or Calvert–Jelliffe equations should be used to individualize the carboplatin dose for day 1 of cycle 1 but with the knowledge that AUC will not be controlled as accurately as with TDM. Moreover, the use of a maximum predicted carboplatin clearance in these young male patients may lead to underexposure and should be re-evaluated. For instance, in our cohort, 30 of 89 patients had a carboplatin clearance calculated with Calvert–Jelliffe higher than 150 mL/min (maximum allowed value in the ongoing TIGER trial NCT02375204). In these 30 patients, the observed carboplatin CL ranged between 117.7 and 222.6 mL/min emphasizing the benefit of TDM to avoid any under- or overdosing of these patients.

It was neither an objective nor *a priori* a way to evaluate the pertinence of the value 24 mg.min/mL previously proposed by Kondagunta and colleagues (4) as target AUC for HD carboplatin. However, evaluation of the adverse events observed during this clinical phase II study (16) confirmed that the choice of this value is adequate. Ototoxicity has been observed (ongoing analysis) and has, in some cases, led to a decrease of the target AUC to 18 mg.min/mL for cycle 2 or 3. Like ototoxicity, nephrotoxicity is not a usual side effect of carboplatin for standard regimen but has been observed after HD carboplatin treatment. The PK results themselves provided valuable information regarding renal toxicity because carboplatin CL is closely dependent of GFR. A mean decrease of carboplatin CL of 11% from cycle 1 to last cycle was observed for patients who received 3 cycles. The median decrease value (–12%) reveals two main situations: half of the patients without clinically change of carboplatin CL and half of the patients with substantial decrease of GFR. Without evidence of other intercurrent cause, we made the hypothesis that this trend of decrease of carboplatin CL was due to subacute nephrotoxicity. Overall, these results allow to anticipate nonacceptable nephrotoxicity if higher AUC had been chosen.

The benefit of TDM in terms of safety gain is difficult to assess by direct confrontation with previous single-institution TI-CE studies in which different levels of target AUCs were used (1, 4, 6). The percentage of patients who started the intensification part of the protocol and who actually received the three HD cycles (81%) is satisfactory for a multicenter study in a very poor-risk population (16). Retrospective analysis of the measured carboplatin AUC showed that none of the patients with severe toxicity was overexposed to carboplatin.

In conclusion, this study promotes the use of carboplatin TDM for the TI-CE regimen as the only way to really control the individual exposure. This practice could be generalized to other HD carboplatin regimen when carboplatin is given on 3 or 5 consecutive days allowing dose adaptation on the last days of the cycle.

## Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

## Authors' Contributions

**Conception and design:** C. Chevreau, C. Massart, A. Fléchon, E. Chatelut, F. Thomas

**Development of methodology:** C. Chevreau, C. Massart, E. Chatelut, F. Thomas

**Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.):** C. Chevreau, S. Broutin, J. Guitton, B. Lelièvre, J. Ciccolini, C. Massart, A. Fléchon, R. Delva, G. Gravis, J.-P. Lotz, J.-O. Bay, M. Gross-Goupil, A. Paci, S. Marsili, L. Malard, E. Chatelut, F. Thomas

**Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis):** S. Moeung, C. Chevreau, C. Massart, L. Malard, E. Chatelut, F. Thomas

**Writing, review, and/or revision of the manuscript:** S. Moeung, C. Chevreau, S. Broutin, B. Lelièvre, C. Massart, A. Fléchon, R. Delva, G. Gravis, J.-P. Lotz, J.-O. Bay, M. Gross-Goupil, A. Paci, E. Chatelut, F. Thomas

**Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):** C. Massart, A. Fléchon, M. Gross-Goupil

**Study supervision:** C. Chevreau, M. Gross-Goupil, E. Chatelut

**Other (medical oncologist, patients' recruitment, and treatment):** J.-P. Lotz

## Grant Support

The study was funded by a French Programme Hospitalier de Recherche Clinique (PHRC).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked *advertisement* in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

## Acknowledgments

We thank the members of the French GPCO and GETUG Unicancer groups for their active contribution to the conduct of this multicenter TI-CE clinical study.

## Footnotes

**Note:**Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

- Received May 12, 2017.
- Revision received July 13, 2017.
- Accepted September 14, 2017.

- ©2017 American Association for Cancer Research.