Sorafenib Population Pharmacokinetics and Skin Toxicities in Children and Adolescents with Refractory/Relapsed Leukemia or Solid Tumor Malignancies

Acute myeloid leukemia (AML) accounts for approximately 20% of the acute leukemias in children and adolescents (1, 2). Although the overall survival rate in patients with AML is improving, 30% to 40% of patients experience recurrence, which is associated with poor prognosis and survival rates of less than 30%. Similarly, whereas the overall survival of children with solid tumor malignancies is approximately 75%, that for children with recurrent or metastatic solid tumors is also below 30% (3). For better salvage of such patients and to improve first-line therapy, newer agents with different mechanisms of action are needed.

Sorafenib is an orally available multikinase inhibitor that blocks pathways with possible roles in the development and progression of AML and solid tumor malignancies, involving C-RAF, B-RAF, c-KIT, FMS-like tyrosine kinase 3 (FLT3), platelet-derived growth factor receptors α and β, vascular endothelial growth factor receptors 1–3, or multiple intracellular kinases (4, 5). Sorafenib is metabolized, in part, to the active metabolite sorafenib N-oxide by cytochrome P450 3A4 (CYP3A4) and to sorafenib glucuronide by uridine diphosphate-glucuronosyltransferase family 1 member A9 (UGT1A9; ref. 6). Clinical studies have demonstrated the efficacy of sorafenib as a single agent or in combination with cytotoxic chemotherapy in treating pediatric and adult AML (7–11). Sorafenib is approved for treating adults with hepatocellular carcinoma, renal cell carcinoma, and differentiated thyroid cancer (12–15). We have previously reported that sorafenib administered concurrently with clofarabine and cytarabine showed activity in children and adolescents with relapsed or refractory AML, especially in those with FLT3-internal tandem duplication (FLT3-ITD; ref. 9). In addition, antitumor activity of sorafenib alone or in combination with bevacizumab and cyclophosphamide was seen in recurrent/refractory pediatric solid tumors (16, 17). However, hand–foot skin reaction (HFSR) and/or skin rash were frequent side effects of sorafenib treatment and were exacerbated by the administration of concomitant agents (9, 17–22). These dermatologic adverse effects, especially HFSR, are associated with poor quality of life, and their pathogenesis and strategies for prevention and optimum supportive care are unclear (23).

To develop better management for children treated with sorafenib, we evaluated the pharmacokinetics and skin toxicity profile of the drug in pediatric patients with relapsed/refractory leukemia or solid tumor malignancies.

Patients aged ≤31 years with relapsed or refractory leukemia and those with relapsed or refractory solid tumors aged ≤21 years at the time of original diagnosis, irrespective of the number of prior salvage regimens, were eligible for the protocols RELHEM (NCT00908167) and ANGIO1 (NCT00665990), respectively, as described previously (9, 17). The protocols were approved by the institutional review board of St. Jude Children's Research Hospital (Memphis, TN) and the studies were conducted in accordance with Declaration of Helsinki. Written informed consent was obtained from all patients or from their legal guardians with assent from the patients as appropriate.

The first 12 patients enrolled on RELHEM received deescalating doses of sorafenib (on days 1–28) and clofarabine and cytarabine (on days 8–12) together (the concurrent regimen) at the dose levels shown in Supplementary Table S1 and described previously (9). In subsequent cohorts of patients, sorafenib was administered sequentially with clofarabine and cytarabine (the sequential regimen) with interpatient dose escalation of clofarabine or deescalation of clofarabine or sorafenib (Supplementary Table S1). In the sequential regimen, sorafenib was administered on days 1 to 7 and days 15 to 28 and clofarabine and cytarabine were administered once daily on days 8 to 12. Once the MTD was established, enrollment was expanded to 12 additional patients to enable further assessment of the toxicity, pharmacokinetics, and efficacy.

The first cohort of patients for ANGIO1 received escalating doses of sorafenib (on days 1–21) with fixed doses of bevacizumab (on day 1) and cyclophosphamide (on days 1–21; Supplementary Table S1; ref. 17). Once an MTD of sorafenib was established, the bevacizumab dose was escalated. After the MTD had been established, enrollment was expanded by up to 25 additional patients with solid tumors or hematologic malignancies.

The Common Terminology Criteria for Adverse Events (CTCAE) Version 3.0 was used for toxicity evaluations. Dose-limiting toxicities (DLT) were evaluated in the first course and included any grade 3 or higher nonhematologic toxicities related to therapy except for the following: grade 3 elevations of amylase, lipase, or total bilirubin or grade 3 or 4 elevations of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) that resolved within 7 days of holding sorafenib; grade 3 hypokalemia, hypocalcemia, hypophosphatemia, or hypomagnesemia that was correctable with oral supplements; grade 3 hypertension that was well controlled with oral medication; or grade 3 or 4 infection or fever, as described previously (9, 17). For HFSR, grades and symptoms were based on the 2008 consensus panel recommendations (24).

For the concurrent regimen of RELHEM, blood samples were collected before and at 2, 4.5, and 7.5 hours after sorafenib administration on days 7 and 12 of course 1. For the sequential regimen, blood samples were obtained before and at 0.5, 1, 2, 4, 6, 8, and 24 hours after sorafenib administration on day 1; the evening dose of sorafenib was held on day 1 to permit pharmacokinetic assessments. Serial blood samples were also obtained on day 7, and a pretreatment trough concentration was obtained on day 15 ± 2 days for the sequential regimen.

In the ANGIO1 protocol, samples were collected before the first dose of sorafenib on day 1 and at 0.5, 2, 4.5, 6, 7.5, 24, and 48 hours after sorafenib administration. After the first dose, sorafenib was withheld until the morning of day 3 (after the blood sample at 48 hours was obtained). Blood samples were also obtained during course 1 before sorafenib treatment on days 7, 13, and 21.

The concentrations of sorafenib, sorafenib N-oxide, and sorafenib glucuronide were measured in plasma (heparinized) by using a validated HPLC-based method with tandem mass spectrometric detection (6).

The PPK and individual post hoc estimates of sorafenib, sorafenib N-oxide, and sorafenib glucuronide were determined by nonlinear mixed-effects modeling with NONMEM 7.3 (ICON Development Solutions), using the iterative two-stage, stochastic approximation expectation-maximization, and importance sampling methods, sequentially. A linear three-compartment model (one compartment for each component) with zero-order absorption, an absorption lag time, and first-order elimination was used to model the data. A diagram of the model along with the parameters estimated is shown in Supplementary Fig. S1. In addition, the individual post hoc parameter values were used to estimate the AUC, the maximum concentration (C_max), and the steady-state (SS) trough concentration on day 7 of therapy (i.e., the concentration immediately before the first dose on day 7) for sorafenib, sorafenib N-oxide, and sorafenib glucuronide. Each individual's pharmacokinetic study was subdivided into two occasions. Occasion 1 was the first serially sampled pharmacokinetic study in each individual. This was day 1 (after the first dose of sorafenib) in all individuals except those on RELHEM with concurrent treatment, for whom this occurred on day 7 with 4 serial samples. Occasion 2 included samples after the serially sampled study to the end of the course. This typically occurred on day 7 (for those who had a pharmacokinetic study on day 1), along with trough samples on days 14 and 21. The interindividual variability (IIV) and interoccasion variability (IOV) of the parameters was assumed to be log-normally distributed. A proportional residual error model was used with assumed normal distribution of the residuals.

The covariates age, body surface area (BSA), serum chemistries (ALT, AST, albumin, alkaline phosphatase, bilirubin, and creatinine), estimated glomerular filtration rate via the St. Jude equation (25), concomitant medications that were inhibitors of organic anion-transporting polypeptide (OATP) 1B1 (micafungin, vancomycin, and posaconazole; refs. 26–28), study day (day 1 vs. day 7), study protocol (RELHEM vs. ANGIO1), and concurrent use of clofarabine and cytarabine in RELHEM versus other regimens (sequential use of clofarabine and cytarabine in RELHEM and ANGIO1) were evaluated to determine their significance in explaining pharmacokinetic variability. These covariates were considered significant in a univariate analysis if their addition to the model reduced the objective function value (OFV) by at least 3.84 units (P < 0.05, based on the χ² test for the difference in the −2 log-likelihood between two hierarchical models that differ by 1 degree of freedom) and if the covariate term was significantly different from zero (P < 0.05 by a t test).

Cox proportional hazard models were used to evaluate the association of the time from starting sorafenib treatment until the first occurrence of grade 2 or higher HFSR with the following clinical and pharmacologic variables: treatment regimen (concurrent regimen in RELHEM or sequential regimen in RELHEM vs. ANGIO1), age, BSA, incidence of grade 2 or higher rash (as a time-dependent predictor indicating whether rash has previously occurred), and the SS trough levels of sorafenib, sorafenib N-oxide, and sorafenib glucuronide. Each of these variables was first evaluated as the sole predictor in a Cox model. Then, all possible models with two of these variables were considered and evaluated with the Bayes information criterion (BIC). BIC values were transformed into evidence weights that represented the probability of being the best model among the considered set of models (29). Because the sorafenib glucuronide SS trough concentration was available only for a subset of individuals (55 of 74), it was not considered in the BIC model selection analysis. Individuals were considered censored at the end of the course or when sorafenib therapy was stopped. In addition, Fine–Gray models were used to evaluate factors that were associated with the development of grade 2 or higher rash, treating development of grade 2 or higher HFSR before rash as a competing event. Rash was modeled as a time-dependent covariate predicting HFSR, and HFSR was modeled as a competing event for rash because sorafenib was primarily discontinued for HFSR.

The clinical features of the 37 patients enrolled on RELHEM (12 patients on the concurrent regimen and 25 on the sequential regimen) and the 45 patients enrolled on ANGIO1 (19 patients in the dose-escalation phase and 26 in the expansion phase) are summarized in Supplementary Table S2.

As reported previously, among the 12 patients treated with the concurrent regimen of RELHEM (9), 2 of 4 patients in stratum 1 and 1 of 2 patients in stratum 2 had grade 3 HFSR and/or rash as DLTs with 200 mg/m² sorafenib (Table 1). All DLTs were due to skin toxicities. No DLTs were observed in stratum 1 with 150 mg/m² sorafenib (6 patients), but grade 2 HFSR (in 3 patients) and skin rash (in 3 patients) were still frequent at this dose level. Overall, HFSR and skin rash of grade 2 or higher were each observed in 8 (67%) of the 12 patients, respectively. In the 21 evaluable patients treated with the sequential regimen of RELHEM, HFSR and skin rash of grade 2 were seen in 4 patients (19%) and 6 patients (29%), respectively (Table 1). No DLTs, including grade 3 skin toxicities, were observed at any dose level, with dose level 3 (sorafenib, 200 mg/m² twice daily; clofarabine, 40 mg/m²; and cytarabine, 1,000 mg/m²), which was determined to be the MTD.

During the phase I portion of ANGIO1, 2 of 4 DLTs were HFSR (Table 1). Among 4 patients treated at the dose level 2, two patients had DLTs; one patient had grade 3 HFSR and the other experienced grade 3 elevated lipase (17). Therefore, the MTD for sorafenib was determined to be 90 mg/m²/dose twice daily. At dose level 3 (in the bevacizumab escalation), one of 6 patients developed grade 3 thrombus, and at dose level 4, one of six patients developed grade 3 HFSR and anorexia. Although the MTD was not reached, the recommended phase II dose was defined as sorafenib 90 mg/m²/dose twice daily, bevacizumab 15 mg/kg, and cyclophosphamide 50 mg/m²/dose. Among the 19 patients in the phase I portion, 4 (21%) developed grade 2 (n = 2) or grade 3 (n = 2) HFSR, and three (16%) developed grade 2 rash. Among 26 patients treated in the expansion arm, 6 (23%) developed HFSR of grade 2 (n = 4) or grade 3 (n = 2). No grade 2 or higher skin rash was observed in the expansion arm.

A total of 74 patients enrolled on either RELHEM (35 patients, 11 treated with the concurrent regimen and 24 with the sequential regimen) or ANGIO1 (39 patients) had evaluable pharmacokinetic studies in course 1. The PPK parameters for plasma concentrations of sorafenib (total number of plasma concentrations: n = 721), sorafenib N-oxide (n = 662), and sorafenib glucuronide (n = 474) for the base model (with only BSA-normalized parameters) are listed in Table 2, and their concentration-versus-time plots, along with the population-estimated curve for each individual study (the concurrent and sequential arms of RELHEM and ANGIO1) are shown in Fig. 1. In RELHEM, sorafenib day 12 exposures were maintained relative to the day 7 exposure with the concurrent regimen. With the sequential regimen, the sorafenib model-predicted day 12 exposures (after 5 days of stopping sorafenib for clofarabine and cytarabine administration), as well as the observed day 15 exposures, were up to three orders of magnitude lower than the corresponding exposures with the concurrent regimen. The sorafenib exposure in ANGIO1 was lower than that in RELHEM.

We evaluated each of the covariates and compared them to the base model. Among the covariates, age, study protocol (RELHEM vs. ANGIO1), concurrent administration with clofarabine and cytarabine (vs. others), bilirubin, creatinine, and OATP1B1 inhibitors, each significantly improved the model fit relative to the base model (Supplementary Table S3). OATP1B1 inhibitors were only given to RELHEM patients as infection prophylaxis (Supplementary Table S4). Sorafenib population apparent clearance decreased with increasing age from 3.8 L/hour/m² in a patient aged 1 year to 1.8 L/hour/m² in a patient aged 15 years (P = 1e−5), and the sorafenib population apparent clearance was 37% lower in ANGIO1 than in RELHEM (P = 9e−5; Fig. 2A). In addition, a sorafenib population apparent clearance of 3.1 L/hour/m² with a creatinine level of 0.2 mg/dL decreased to 1.5 L/hour/m² with a creatinine level of 0.8 mg/dL (P = 7e−6; Fig. 2B). Sorafenib N-oxide population apparent clearance was 65% lower in patients who received concurrent administration in RELHEM than in other patients (P = 7e−4; Fig. 2C). Sorafenib N-oxide and glucuronide population apparent clearance were also lower in patients with higher bilirubin, declining from 25.7 and 0.098 L/hour/m² when the bilirubin level was 0.2 mg/dL to 10.4 and 0.064 L/hour/m² when the bilirubin level was 1.0 mg/dL, respectively (P = 1e−4; Fig. 2D and E). Sorafenib population apparent clearance was 50% higher and the sorafenib glucuronide population apparent clearance was 22% lower in individuals who received OATP1B1 inhibitors (P = 3e−3; Fig. 2F and G). The covariates, age, study protocol, creatinine, and OATP1B1 inhibitors accounted for 22%, 43%, 59%, and 13%, respectively, of the IIV in sorafenib apparent clearance; concurrent administration with clofarabine and cytarabine and bilirubin accounted for 49% and 12%, respectively, of the IIV in sorafenib N-oxide apparent clearance; and bilirubin and OATP1B1 inhibitors accounted for 24% and 13%, respectively, of the IIV in sorafenib glucuronide apparent clearance.

We next considered combinations of covariates. Relative to the base model, the model that included age, study protocol (RELHEM vs. ANGIO1), and concurrent administration with clofarabine and cytarabine (vs. others) accounted for 67% of the IIV in sorafenib apparent clearance and 49% of the IIV in sorafenib N-oxide apparent clearance (Table 2; Supplementary Table S3). The only laboratory result that remained significant relative to the above multivariate model was the bilirubin level. This accounted for an additional 8% and 26% of the IIV in sorafenib N-oxide and sorafenib glucuronide apparent clearance, respectively, relative to the previous model.

A summary of the base and final covariate model results and goodness-of-fit plots for sorafenib, sorafenib N-oxide, and sorafenib glucuronide are shown in Table 2 and Supplementary Fig. S2, respectively. The apparent clearance of sorafenib N-oxide decreased over the period from day 1 to day 7 with daily administration (from a median 27.3 L/hour/m² to a median 12.3 L/hour/m², P = 2e−5; Supplementary Fig. S3A). The apparent clearance of sorafenib N-oxide correlated significantly with that of sorafenib (r = 0.7, P = 7e−13), whereas the apparent clearance of sorafenib glucuronide did not correlate with that of the parent compound (r = 0.06; P = 0.6; Supplementary Fig. S3B and S3C).

A summary of the post hoc individual estimated exposure parameters for sorafenib, sorafenib N-oxide, and sorafenib glucuronide on days 1 and 7 is presented in Supplementary Table S5.

The determinants of sorafenib-induced HFSR were evaluated using Cox proportional hazards regression analysis. The rate of the first grade 2–3 HFSR was 4.72 times higher (95% CI = 1.81–12.33; P = 0.0015) with concurrent administration of clofarabine and cytarabine compared with ANGIO1. However sequential administration of clofarabine and cytarabine versus ANGIO1 was not significantly associated with HFSR (HR = 0.72; 95% CI = 0.22–2.35; P = 0.591; Fig. 3A). Next, exposure to sorafenib or its metabolites were considered. Descriptively, patients experiencing grade 2–3 HFSR had higher sorafenib SS trough concentrations when compared with those experiencing grade 0–1 HFSR (median: 4.1 vs. 3.3 mg/L, respectively). A 1,000 ng/mL increase in the sorafenib SS trough concentration was associated with a 1.45-fold increase in the HFSR rate (95% CI = 1.18, 1.78; P = 0.0004; Fig. 3B), so the upper quartile concentration associates with an HFSR rate that is 2.16 times that of the lower quartile (95% CI = 1.41–3.32). Patients experiencing grade 2–3 HFSR also had higher sorafenib N-oxide SS trough concentrations when compared with those experiencing grade 0–1 HFSR (median: 0.74 vs. 0.47 mg/L, respectively). In this case, a 100 ng/mL increase in the sorafenib N-oxide SS trough concentration was associated with a 1.04-fold increase in the HFSR rate (95% CI = 1.00, 1.09; P = 0.0275; Fig. 3C), so that the upper quartile concentration associates with an HFSR rate that is 1.39 times that of the lower quartile (95% CI = 1.03, 1.86). In addition, each square-meter increase in BSA was associated with an increase in the rate of HFSR by a factor of 2.63 (95% CI = 1.18–5.88; P = 0.0186), and after experiencing rash, patients experienced a 2.97-fold increase in the rate of HFSR (95% CI = 1.13–7.79; P = 0.0268). Of the 9 patients who had both rash and HFSR, rash was seen before HFSR in 8 patients by a median of 1 day (range, 0–13 days).

We next considered models with two predictors. By BIC, the four models with the greatest probability of being the best among those considered are the model with treatment regimen (concurrent or sequential administration with clofarabine and cytarabine vs. ANGIO1) and BSA (47.9% probability); the model with BSA and sorafenib SS trough levels (10.8% probability); the model with treatment regimen and sorafenib SS trough levels (9.1% probability); and the model with prior occurrence of rash and sorafenib SS trough levels (8.8% probability) as predictors of HFSR (Supplementary Table S6). This analysis indicates that HFSR is associated with the concurrent administration with clofarabine and cytarabine, higher sorafenib SS trough levels, larger BSA, and previous occurrence of rash.

The above analysis indicated that rash may be an indicator of an increased risk for subsequent development of HFSR. Therefore, we performed Fine–Gray modeling analyses to evaluate factors that were associated with the development of grade 2–3 rash while treating development of HFSR before rash as a competing event. In this analysis, treatment regimen was significantly associated with the development of rash. Patients treated with concurrent administration of clofarabine and cytarabine developed rash at 21.4 times the rate of ANGIO1 patients (95% CI = 4.71–97.24; P = 0.0001). Patients treated with sequential administration with clofarabine and cytarabine regimen developed rash at 4.48 times the rate of ANGIO1 patients (95% CI = 0.88–22.91; P = 0.071). A 1,000 ng/mL increase in the sorafenib SS trough concentration was associated with a 1.21-fold increase in the rash rate (95% CI = 0.98–1.48; P = 0.074), and a 100 ng/mL increase in the sorafenib N-oxide SS trough concentration was associated with a 1.05-fold increase in the rash rate (95% CI = 1.01–1.10; P = 0.018). In a BIC model selection analysis, treatment regimen appeared in all of the top four models (Supplementary Table S6).

We performed pharmacokinetic simulations to evaluate alternative sorafenib administration schedules (once a day and every other day) that would minimize the number of patients whose sorafenib SS trough concentrations reached levels associated with HFSR (Fig. 4). Our simulations indicated that both the once-daily and every-other-day schedules of sorafenib 200 mg/m² would minimize exposure to sorafenib SS trough concentrations associated with HFSR in most patients. Specifically, the median sorafenib SS trough concentration decreased from 5,743 to 2,564 or 934 ng/mL when the administration schedule changed from twice-daily to once-daily or every-other-day. This decrease in SS trough concentration corresponds to a 3.35-fold reduction in the HFSR rate when changing from twice-daily to once-daily administration (95% CI: 1.71–6.55) and a 6.71-fold reduction when changing from twice-daily to every-other-day (95% CI: 2.26–16.88). Of relevance to patients with AML and FLT3-ITD mutations, which are a target of sorafenib (5), we found that with once-daily or once-every-other-day administration, the estimated SS trough concentrations at the 10th percentile were well above the sorafenib concentration of 143 ng/mL (308 nmol/L) that inhibited phospho-FLT3 of leukemia samples in a plasma inhibitory assay (30). We used this strategy to treat a patient with FLT3-ITD–positive AML with single-agent sorafenib, and remission was maintained for approximately 1 year (Supplementary Table S7).

We have reported the PPK of sorafenib and its metabolites sorafenib N-oxide and sorafenib glucuronide, as well as the skin toxicity, in pediatric patients who were treated with a combination of sorafenib and clofarabine and cytarabine (for leukemia) or a combination of sorafenib and bevacizumab and cyclophosphamide (for solid tumor malignancies). Concurrent administration of sorafenib with clofarabine and cytarabine was associated with skin DLTs in 3 of 6 patients when sorafenib was administered at 200 mg/m²/day. Sequential administration of clofarabine and cytarabine administration and administration of lower sorafenib dose with bevacizumab and cyclophosphamide decreased the onset of HFSR. In the PPK analysis, older age, the study protocol (ANGIO1 vs. RELHEM), and a higher creatinine level were associated with decreased sorafenib apparent clearance and administration of OATP1B1 inhibitors was associated with higher sorafenib apparent clearance. A higher bilirubin level, concurrent administration with clofarabine and cytarabine, and repeated drug administration for 7 days were associated with decreased sorafenib N-oxide apparent clearance. In addition, OATP1B1 inhibitors and increased bilirubin were associated with lower sorafenib glucuronide apparent clearance. In the exposure–toxicity analysis, grade 2–3 HFSR was significantly associated with concurrent administration of clofarabine and cytarabine, higher sorafenib SS trough levels, greater BSA, and previous occurrence of rash and grade 2–3 skin rash was associated with concurrent administration of clofarabine and cytarabine.

This article reports the first PPK model for sorafenib in children. Notably, age was a significant covariate of sorafenib apparent clearance, with younger children exhibiting higher drug clearance than older children. Sorafenib is partly metabolized by CYP3A4 to the active metabolite sorafenib N-oxide and by UGT1A9 to sorafenib glucuronide (6). In young children, CYP450-catalyzed metabolism is increased (31), which could explain the increased sorafenib apparent clearance in younger patients in our PPK model. Greater BSA was associated with higher incidences of HFSR, which may reflect the association of reduced sorafenib clearance (higher exposure) with older age and increasing BSA. The differences in sorafenib apparent clearance between studies could be due to the different cancer types (solid tumor malignancies, which often involve organs, vs. leukemia), the concomitant administration of OATP1B1 inhibitors in RELHEM patients, or differences in concomitant chemotherapy. Because of the small number of each solid tumor subtypes, we were unable to analyze based on the disease subtypes. However, this is the largest available population of pediatric patients with collected pharmacokinetic data. Although urinary excretion plays a minor role in sorafenib and metabolite elimination (accounting for <20% of the total), increased creatinine was associated with decreased apparent clearance of sorafenib. Recently, it has been shown that the clearance of drugs that are substrates for OATP uptake transporters, such as OATP1B1 and OATP1B3, which are located on the sinusoidal membrane of hepatocytes, decreases as kidney function declines (32). We previously demonstrated that sorafenib is a substrate for OATP1B1 and OATP1B3, which may also explain why creatinine is a covariate in the PPK model (33).

We have shown that sorafenib N-oxide also plays an important role in sorafenib-associated skin toxicities, as well as in antileukemia activities as an active metabolite (9, 34). The apparent clearance of sorafenib N-oxide was decreased with higher bilirubin levels and repeated drug administration for 7 days, which could be associated with sorafenib glucuronide, with its high SS concentrations in the liver competing with sorafenib N-oxide for biliary excretion via the efflux transporters ABCB1 and/or ABCC2 (35). It is unclear why there was a difference in sorafenib N-oxide apparent clearance in patients receiving concurrent administration with clofarabine and cytarabine when compared with those in the other group.

In PPK analysis, several variables were associated with sorafenib glucuronide apparent clearance. Bilirubin accounted for a portion (24%) of the variability in the apparent clearance of sorafenib glucuronide. We also showed that concurrent administration of OATP1B1 inhibitors in RELHEM patients was associated with reduced sorafenib glucuronide apparent clearance (higher exposure) and a concomitant increase in sorafenib apparent clearance (lower exposure). This can be explained by inhibition of enterohepatic recirculation of sorafenib glucuronide (by inhibiting liver uptake), which accounts for approximately 50% of the circulating sorafenib parent compound (36). This is the first study to demonstrate a drug–drug interaction between OATP1B1 inhibitors and sorafenib.

The population estimates of sorafenib apparent clearance reported in this study are similar to those reported in several other population pharmacokinetic studies in adults. Specifically, our population estimated apparent clearance of sorafenib in a 10-year-old (the median age of our population) was 2.46 and 1.77 L/h/m² in RELHEM and ANGIO1, respectively. The population-estimated apparent clearances of sorafenib reported by Widemann and colleagues (ref. 16; median patient age, 14 years) and Hornecker and colleagues (ref. 37; normalized to a BSA of 2.1) were 3.36 and 1.0 L/hour/m², respectively. In addition, the population estimated apparent clearance of sorafenib reported by Jain and colleagues (ref. 36; normalized to a BSA of 2.1 and accounting for the fact that their estimate addressed enterohepatic recirculation) was 1.94 L/hour/m². Note that none of these studies measured metabolites.

HFSR is a common adverse event with sorafenib, occurring in approximately 60% of adults treated with this drug, and studies have aimed to identify risk factors for this toxicity, especially for the more severe grades that can decrease the quality of life of patients. Grade 2 or higher HFSR is reported to occur in 36% of patients and typically manifests within the first 2 weeks of treatment (38). The corresponding information on HFSR in pediatric populations is limited, and we observed an increased incidence of grade 2 or higher HFSR (67%) when sorafenib was administered concurrently with clofarabine and cytarabine, with onset occurring during the second week of treatment. The clofarabine and cytarabine regimen is frequently associated with skin toxicity, including palmar–plantar erythrodysesthesia (20, 21), and probably contributed to the increased incidence of HFSR observed during concurrent treatment. In the sequential regimen, the drug holiday between days 8 and 14 led to exposures that were up to three orders of magnitude lower than the corresponding exposures with the concurrent regimen, which can be associated with lower incidences of HFSR. In the ANGIO1 trial, administration of lower doses of sorafenib and bevacizumab and cyclophosphamide, a chemotherapy regimen which has not been associated with skin toxicity, probably contributed to the decreased incidences of HFSR. Although it can be difficult to tease out the role of sorafenib exposure in skin toxicities in the presence of other chemotherapy, sorafenib trough level appeared in three of the top four models by the BIC model selection criteria. One of these top models used treatment regimen and sorafenib trough levels as predictors of HFSR. Therefore, sorafenib exposure, specifically day 7 SS trough concentration, is very likely to be a significant factor associated with HFSR.

In our study, patients with higher SS trough concentrations of sorafenib and sorafenib N-oxide experienced more grade 2 or higher HFSR. Even with the sequential administration of sorafenib in RELHEM or the lower doses of sorafenib in ANGIO1, approximately 20% of patients still experienced HFSR, supporting the importance of trough concentrations. These results are in line with those reported for two previous studies in adults with solid tumors who received single-agent sorafenib therapy, which found an association between a minimal-threshold sorafenib concentration of 5 mg/L or 5.8 mg/L and the incidence of grade 2 or higher HFSR (38, 39). In another study, a nonlinear mixed-effect model was constructed to link sorafenib administration to the risk of HFSR in adults with solid tumors (40). The risk of HFSR increased with the number of doses administered per day (e.g., thrice daily > twice daily > once daily), and this factor was more influential than the total daily dose (range, 200 to 1,600 mg/day). This finding suggests that administering higher daily doses of sorafenib on fewer occasions could minimize the risk of severe HFSR. However, this strategy would need to take into consideration the saturable absorption of sorafenib above single doses of 400 mg (37). Sorafenib is a potent inhibitor of FLT3-ITD–positive AML, and a sorafenib plasma concentration of 143 ng/mL was required to inhibit phospho-FLT3 in a plasma inhibitory assay (30). In this study, the described population model was used to simulate sorafenib exposure with different administration schedules. These simulations suggest that sorafenib can be administered at a dose of 200 mg/m² with a reduced frequency of once a day or every other day while maintaining trough concentrations above the threshold needed for the inhibition of FLT3-ITD and avoiding higher concentrations that are associated with HFSR. A similar modeling approach could be used to define alternative sorafenib dosing regimens for children with solid tumors. Further model refinements and the incorporation of physiologically based mechanisms can be used to define exposure–toxicity or exposure–response relationships and to develop alternative dosing strategies for sorafenib (41).

In conclusion, we evaluated the pharmacokinetics and skin toxicity profile of sorafenib administered in combination with clofarabine and cytarabine (either concurrently or sequentially) or bevacizumab and cyclophosphamide. We developed a PPK model for sorafenib and its metabolites in children that identified clinical covariates affecting their apparent clearances. We used the pharmacokinetic model to characterize associations between drug exposure and the dose-limiting toxicity HFSR. In addition, our pharmacokinetic model simulations suggested that an alternative sorafenib treatment regimen has the potential to decrease HFSR while maintaining efficacy, especially in patients with FLT3-ITD–positive AML. This regimen warrants further investigation.

H. Inaba reports receiving commercial research grants from Bayer/Onyx. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Inaba, J. Rubnitz, S.D. Baker

Development of methodology: H. Inaba, J.C. Panetta, S.D. Baker

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Inaba, F. Navid, S.M. Federico. E.D. Eisenmann, A. Vasilyeva, S. Shurtleff, C.-H. Pui, T.A. Gruber, J. Rubnitz

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Inaba, J.C. Panetta, S. Pounds, L. Wang, L. Li, Y.-D. Wang, R.C. Ribeiro, J. Rubnitz, S.D. Baker

Writing, review, and/or revision of the manuscript: H. Inaba, J.C. Panetta, S. Pounds, F. Navid, S.M. Federico, S. Shurtleff, C.-H. Pui, T.A. Gruber, R.C. Ribeiro, J. Rubnitz, S.D. Baker

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Inaba, J.C. Panetta, E.D. Eisenmann, A. Vasilyeva, C.-H. Pui

Study supervision: H. Inaba, S.D. Baker

We thank the clinical, research, and pharmacokinetic nurses at St. Jude Children's Research Hospital who participated in this study. The research reported in this article was supported by the National Cancer Institute of the NIH [R01CA138744 (to S.D. Baker) and P01CA21765], by ALSAC, and by Bayer/Onyx. We thank Keith A. Laycock, PhD, ELS (St. Jude Children's Research Hospital, Memphis, TN) for editorial assistance.

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.