Differential effects of dosing regimen on the safety and efficacy of dasatinib: retrospective exposure–response analysis of a Phase III study

Abstract

Purpose

Dasatinib is a prototypic short half-life BCR-ABL1 tyrosine kinase inhibitor. The recommended dose of dasatinib for chronic myeloid leukemia in chronic phase was changed from 70 mg twice daily to 100 mg once daily following a Phase III dose-optimization study. To better understand the superior benefit–risk profile of dasatinib 100 mg once daily, exposure–response was characterized for efficacy (major cytogenetic response) and safety (pleural effusion).

Patients and methods

Dasatinib exposure in patients with chronic myeloid leukemia in chronic phase was determined by population pharmacokinetic analysis of data from seven dasatinib clinical studies (N = 981), including the Phase III dose-optimization study (n = 567). Data from the Phase III study were then used to characterize exposure–response relationships for the four dasatinib treatment regimens investigated (100 mg once daily, 50 mg twice daily, 140 mg once daily, and 70 mg twice daily).

Results

Major cytogenetic response was significantly (P < 0.01) associated with weighted average steady-state dasatinib plasma concentrations, and pleural effusion was significantly associated with trough concentration. Major cytogenetic response was also significantly associated with maintenance of uninterrupted dosing. The 100 mg once daily arm had the lowest steady-state trough concentration of the four dose arms investigated in the Phase III study, and although this arm also had the lowest weighted average steady-state dasatinib plasma concentration, it had the highest dose maintenance.

Conclusion

Dasatinib dose optimization to 100 mg once daily from 70 mg twice daily significantly minimizes adverse events while maintaining efficacy by exploiting differences in the measures of exposure associated with efficacy and safety.

Introduction

In 2010, chronic myeloid leukemia (CML) accounted for 11% of all adult leukemias diagnosed in the United States.1,2 CML is characterized by the fusion of a portion of the ABL1 oncogene on chromosome 9 with the breakpoint cluster region gene (BCR) on chromosome 22 to form the BCR-ABL1 gene.1 This oncogene encodes a constitutively active tyrosine kinase protein (BCR-ABL1) that can activate multiple signal transduction pathways affecting hematopoietic cell growth and survival.3 BCR-ABL1 tyrosine kinase inhibitors (TKIs) (imatinib, dasatinib, and nilotinib) are currently the mainstays of CML treatment.4

Dasatinib is a prototypic short half-life TKI (plasma half-life approximately 4 to 6 hours)5 that inhibits BCR-ABL1 with a potency 325-fold that of imatinib in vitro. It is also active against several imatinib-resistant BCR-ABL1 mutants.6 Dasatinib was initially approved for the treatment of adults with Philadelphia chromosome–positive (Ph+) CML in chronic, accelerated, or myeloid/lymphoid blast phase (CML-CP, -AP, or -BP, respectively) with resistance or intolerance to prior therapy, including imatinib, or Ph+ acute lymphoblastic leukemia (ALL) with resistance or intolerance to prior therapy.7 It has since been approved for the treatment of newly diagnosed adults with Ph+ CML-CP. Dose and schedule were initially explored in a Phase I dose-escalation study in which 84 patients with CML or Ph+ ALL that was resistant or intolerant to imatinib received dasatinib (15–240 mg/day) once daily or twice daily.8 In this study, the 70 mg twice daily dose showed the most notable rates of major cytogenetic response (MCyR) in both CML-CP (four of six patients) and advanced-phase CML (CML-AP or -BP) (ten of 17 patients).8 Based in large part upon the plasma half-life observed in this study, dasatinib 70 mg twice daily was further assessed in five Phase II studies in patients across all phases of CML and Ph+ ALL (the Src/Abl Tyrosine kinase inhibition Activity: Research Trials [START]).9–13 These studies served as the basis for approval of the 70 mg twice daily dose in the United States and Europe for Ph+ CML (CP, AP, and BP) and Ph+ ALL in patients intolerant or resistant to imatinib.14,15

Retrospective analysis data from the Phase I and Phase II studies in patients with CML suggested that pleural effusion, a key adverse event (AE) associated with dasatinib therapy, is less frequent with once daily dosing compared with twice daily dosing and at doses of ≤100 mg/day compared with doses of ≤140 mg/day.16 Additionally, in two of the Phase II studies of dasatinib 70 mg twice daily in patients with CML-CP, the mean total daily dose following dose reductions and interruptions was approximately 100 mg.11,12 Based on these data, the dasatinib dose and schedule were prospectively reassessed in an open-label Phase III study of patients with CML-CP17,18 in which patients were equally randomized to four dasatinib treatment regimens (100 mg once daily, 50 mg twice daily, 140 mg once daily, and 70 mg twice daily). Similar MCyR rates were observed with dasatinib 100 mg once daily (n = 167) and 70 mg twice daily (n = 168) (59% and 55%, respectively), whereas dasatinib 100 mg once daily was associated with significantly lower frequencies of grade 3–4 AEs (30% versus 48%) (P = 0.001), grade 3–4 thrombocytopenia (22% versus 37%) (P = 0.004), and any-grade pleural effusion (7% versus 16%) (P = 0.024) compared with dasatinib 70 mg twice daily.17 Furthermore, fewer patients required dose interruptions and reductions due to toxicity in the 100 mg once daily arm versus the 70 mg twice daily arm.17 These data led to a new approved starting dose of 100 mg once daily for patients with CML-CP.19

The exposure–response (E–R) relationship was characterized with respect to efficacy (MCyR) and safety (pleural effusion) to better understand and quantify the factors underlying the superior benefit–risk balance of dasatinib 100 mg once daily compared with dasatinib 70 mg twice daily in patients with Ph+ CML-CP. A previously developed population pharmacokinetic (PPK) model20 was updated with data from the Phase III dose-optimization study and applied to determine summary measures of individual dasatinib exposure (both steady-state and time-dependent peak, trough, and time-averaged plasma dasatinib concentrations). These exposure measures were used to characterize dasatinib E–R relationships for efficacy (attainment of MCyR) and safety (pleural effusion incidence). MCyR was selected for the exposure–efficacy response analysis because it was the primary endpoint in the Phase III dose-optimization study.17 Pleural effusion was selected for the exposure–safety response analysis because it is the most common fluid retention event reported in patients with CML treated with second-line dasatinib.21

Methods

Data

The pharmacokinetics of dasatinib in subjects with CML-CP was described by a PPK model, developed using data from seven open-label clinical studies in patients with CML (one Phase I dose-escalation study of dasatinib 15–180 mg once daily and 25–120 mg twice daily; five Phase II studies of dasatinib 70 mg twice daily [START-A, -B, -C, -L, and -R]; and the Phase III dose-optimization study).8–13,17 This model was applied to determine the dasatinib exposure of subjects in the Phase III dose-optimization study from the available, sparse dasatinib concentration measurements in these subjects. The dasatinib E–R for MCyR and pleural effusion was characterized using data from subjects in the Phase III dose-optimization study for whom dasatinib exposure could be determined. The studies included in the PPK and E–R analyses are summarized in Table S1. All studies were approved by the relevant Institutional Review Boards and Independent Ethics Committees of each participating institution and were conducted in accordance with the ethical principles of the Declaration of Helsinki.

The PPK analysis dataset included a total of 6457 dasatinib plasma concentration values from 981 patients with Ph+ CML (CP, AP, and BP) or Ph+ ALL that was resistant or intolerant to prior therapy from the seven open-label clinical studies. Baseline demographic and laboratory measurements were recorded and included in the analysis dataset. A summary of patient demographics and laboratory values included in the PPK model are shown in Table S2.

Efficacy and safety E–R analyses were performed on data from 567 (86%) of 662 patients with CML-CP treated with dasatinib in the Phase III dose-optimization study for whom dasatinib exposure could be determined. The median age was 55 years (range 18–84), and 47% (266/567) of patients were male. In total, 34% of patients (191/567) had a history of cardiac disease. Most patients (74%, 417/567) were imatinib-resistant, and the others (26%, 150/567) were imatinib-intolerant.

The primary endpoint for efficacy in the Phase III study was achievement of MCyR after a minimum follow-up of 6 months. Patients were considered to have achieved MCyR if they had ≤35% Ph+ metaphases in bone marrow.17 Safety evaluations included pleural effusion incidence and time to first reported pleural effusion (any grade).17,22 Chest X-rays were performed at baseline, after 6 months of treatment, and as required for detection or monitoring of pleural effusion.

Analyses

PPK model

The PPK model was developed by updating a previously developed model20 with dasatinib plasma concentration data collected in the Phase III dose-optimization study. Covariate effects (age, gender, race, body weight, body mass index, baseline hepatic and renal laboratory parameters, hemoglobin, and white blood cell count) on PK parameters (clearance and volume of distribution) were evaluated by the likelihood ratio test. Only covariate effects that were both statistically significant (P < 0.001) and clinically relevant (defined on the basis of covariate inclusion that resulted in more than a ±20% parameter change) were retained in the final PPK model. Although no formal adjustment was made for multiplicity, the significance level of 0.1% was selected in consideration of the multiple parameter–covariate relationships assessed. The model assumed random interindividual variability (IIV) with a log-normal distribution on all structural model parameters. In addition, an interoccasion variability component (IOV) was used to describe the random variability in relative bioavailability within an individual and between dosing occasions. The difference between observed values and the corresponding model-predicted values was described by a log-normal residual error model. A model evaluation was conducted using visual predictive performance checks on the Phase III study data. The observed dasatinib plasma concentration–time data and corresponding fifth, 50th, and 95th percentiles of the model-based predictions were plotted as a graphical assessment.

The PPK model was applied to determine summary measures of dasatinib steady-state exposure for the nominal dose (steady-state peak, trough, and time-averaged plasma dasatinib concentrations [C_maxss, C_minss, and C_avgss, respectively]) from the maximum a posteriori estimates of individual PK parameters. C_avgss was calculated as the ratio of the steady-state area under the curve to the dosing interval (24 hours for once daily and 12 hours for twice daily). The model also was applied to obtain the time-dependent peak, trough, and time-averaged plasma dasatinib concentrations (C_max, C_min, and C_avg, respectively) given the actual dosing history (including dose interruptions and modifications). The model was developed using NONMEM^® (version VI, level 1.1; Icon plc, Dublin, Republic of Ireland). Diagnostic graphics, exploratory analyses, and postprocessing of NONMEM ^® output were performed using S-PLUS (version 7.0.0 for Linux; TIBCO Software Inc, Palo Alto, CA, USA).

E–R for efficacy: MCyR

The relationship between dasatinib exposure and the probability of achieving MCyR was described by a logistic regression model. The marginal effect of dasatinib exposure on MCyR was first characterized in a base model, followed by examination of effects from patient covariates in a full model. The following patient covariates were examined: age, gender, imatinib failure status (resistant or intolerant), and duration of dose maintenance (uninterrupted duration as percentage of total treatment duration). The final model was developed by backward elimination of covariate effects from the full model and contained effects from both exposure measures and covariates that had statistically significant effects (P < 0.01). Although no formal adjustment was made for multiplicity the significance level of 1% was selected in consideration of the multiple covariates assessed. The exposure measures (C_maxss, C_minss, and C_avgss) for each patient were adjusted to account for dose modification by multiplying these values by the weighted average total daily dose taken by the patient (expressed as a percentage of the nominal dose) to obtain the corresponding weighted average exposures (wC_maxss, wC_minss, and wC_avgss). The weighted average total daily dose of each patient was calculated as the daily dose averaged over the duration of uninterrupted treatment (treatment duration excluding dose interruptions) up to time of MCyR or end of treatment, whichever occurred earlier. The potential effect of dose interruption was assessed with respect to dose maintenance (D_m) (expressed as the percentage of uninterrupted treatment duration). The final model was evaluated by assessing the agreement between the observed proportion of MCyR and the 90% model prediction intervals.

E–R for safety: pleural effusion

The relationship between dasatinib exposure and the time to first occurrence of grade ≥ 1 pleural effusion was described by a Cox proportional hazards model. The marginal effect of dasatinib exposure on the occurrence of pleural effusion was first characterized in a base model, followed by the examination of effects from patient covariates (age, gender, race, and history of cardiac disease) in a full model. The final model was developed by backward elimination of covariate effects from the full model and contained both exposure measures and covariates with statistically significant effects (P < 0.01). Although no formal adjustment was made for multiplicity, the significance level of 1% was selected in consideration of the multiple covariates assessed. The measures of exposure assessed were C_max, C_min, and C_avg, and the model was evaluated by comparing the predicted cumulative probability of pleural effusion with that determined by Kaplan–Meier analysis.

Results

PPK analysis

The dasatinib concentration–time data were well-described by a linear two-compartment PPK model. The model was parameterized in terms of plasma and intercompartmental apparent clearances, apparent volumes of distribution in the central and peripheral compartments, and the absorption rate constant (Table 1). The mean terminal half-life was estimated to be 2.93 hours. The variability in relative bioavailability (IIV of 34.6% and IOV of 37.4%) accounted for a larger portion of overall variability in dasatinib exposure than did the variability in the apparent plasma clearance (28.8%). None of the covariates examined during model development had statistically significant effects (P < 0.001) or clinically relevant effects (>±20% effect) on the PK parameters. Therefore, the final PPK model contained no covariate effects. Evaluation of the diagnostic plots, such as the model predictions versus observations, residuals versus model predictions, and residuals versus time, showed that the model described the observed data well, and assumptions about random variability (ie, IIV, IOV, and residual error) were reasonably satisfied (data not shown).

Table 1.

Final PPK model parameter estimates

Parameter (units)	Estimatea	Standard error (RSE%)	95% CIb
Fixed effects
(CL/F)TV (L/h)	296	6.42 (2)	283–309
(Vc/F)TV (L)	1230	63.7 (5)	1110–1350
(Q/F)TV (L/h)	119	6.05 (5)	107–131
(Vp/F)TV (L)	1030	38.9 (4)	954–1110
KATV (1/h)	2.1	0.15 (7.3)	1.8–2.4
Random effects
ω2CL	0.083 (0.29)	0.016 (19.6)	0.051–0.114
ω2Vc	0.730 (0.85)	0.073 (10.0)	0.587–0.873
ω2KA (fixed)	1.0 (1.0)	–	–
ω2FR	0.120 (0.35)	0.020 (16.2)	0.082–0.158
ω2FR,IOV	0.140 (0.37)	0.008 (5.6)	0.125–0.155
ω2CL,Vc	0.241 (0.98)	0.031 (12.8)	0.181–0.301
Residual error
σL	0.464	0.002 (0.537)	0.459–0.469

Notes:

^aEstimate values in parentheses are standard deviations for estimated variances (first five entries under random effects) and correlation for estimated covariance (last entry under random effects);

^bbootstrap confidence intervals (327 successful out of a total of 500).

Abbreviations: CI, confidence interval; (CL/F)_TV, apparent clearance; KA_TV, absorption rate constant; PPK, population pharmacokinetics; (Q/F)_TV, apparent intercompartmental clearance; RSE%, relative standard error as a percentage of the estimate; (Vc/F)_TV, apparent volume of central compartment; (Vp/F)_TV, apparent volume of peripheral compartment; σ_L, standard deviation of log-additive residual error; ω²_CL, variance of interindividual variability for apparent clearance; ω²_CL,Vc, covariance for apparent clearance and apparent volume of central compartment; ω²F_R, variance of interindividual variability for relative bioavailability; ω²F_R,IOV, variance of interoccasion variability for relative bioavailability; ω²KA, variance of interindividual variability for absorption rate constant; ω²Vc, variance of interindividual variability for apparent volume of central compartment.

Figure 1 shows the observed and model-predicted median (90% prediction intervals) of dasatinib plasma concentration versus time for the four arms of the Phase III dose-optimization study. Overall, the predicted concentrations corresponded well to the observed profiles from the study. The percentage of observations outside the 90% prediction intervals was generally less than 10%, suggesting that the model had no systematic bias with respect to the dose or frequency of dasatinib administration.

Figure 1.

Observed and predicted median concentrations versus time from previous dose for dasatinib in the Phase III study.17,18

Note: Predicted median concentrations have a 90% prediction interval.

The PPK-model–predicted exposures for the 567 patients in the Phase III study show that the C_minss was lowest for the 100 mg once daily regimen and that the C_avgss was similar for the 100 mg once daily and 50 mg twice daily regimens and for the 140 mg once daily and 70 mg twice daily regimens (Table 2). These data showed that for a given daily dose, C_minss tends to be lower for the once daily schedule whereas C_maxss tends to be higher. The weighted average exposure measures showed similar trends. These summary measures and their time-dependent correlates were subsequently applied in the dasatinib exposure–efficacy and exposure–safety analyses after adjustments for dose modifications that were appropriate for each analysis, as described in the Methods section.

Table 2.

Geometric mean (CV%) of weighted average daily dose, dose maintenance, and steady-state dasatinib exposures of patients in the Phase III study

	100 mg once daily n = 146	50 mg twice daily n = 148	140 mg once daily n = 141	70 mg twice daily n = 132
wTDD (% nominal)	94.1 (16)	92.0 (17)	87.9 (16)	85.7 (20)
Dm (%)	87.8 (19)	84.8 (20)	84.4 (20)	85.1 (17)
Cminss (ng/mL)	2.61 (26)	5.00 (24)	3.72 (28)	6.71 (24)
Cmaxss (ng/mL)	54.6 (56)	32.8 (48)	79.7 (55)	47.8 (46)
Cavgss (ng/mL)	13.5 (35)	14.3 (31)	19.7 (32)	20.0 (31)
wCminss (ng/mL)	2.46 (28)	4.59 (29)	3.32 (31)	5.75 (31)
wCmaxss (ng/mL)	51.4 (58)	30.2 (49)	71.2 (58)	40.9 (52)
wCavgss (ng/mL)	12.7 (36)	13.1 (33)	17.5 (34)	17.1 (37)

Abbreviations: C_avgss, steady-state time-averaged plasma dasatinib concentration; C_maxss, steady-state peak plasma dasatinib concentration; C_minss, steady-state trough plasma dasatinib concentration; D_m, percentage dose maintenance duration; wC_avgss, weighted average steady-state plasma dasatinib concentration; wC_maxss, weighted average steady-state peak plasma dasatinib concentration; wC_minss, weighted average steady-state trough plasma dasatinib concentration; wTDD, weighted average total daily dose.

E–R for efficacy: MCyR

During treatment, 63% (358/567) of patients in the Phase III dose-optimization study achieved MCyR. Similar MCyR rates were achieved across the four regimens: 64% (100 mg once daily), 57% (50 mg twice daily), 68% (140 mg once daily), and 64% (70 mg twice daily). Of patients included in the analysis dataset, 61% (217/358) of the responders had dose modifications (including dose escalations or interruptions/reductions due to hematologic or nonhematologic toxicity) before achieving MCyR. The predictor variables investigated in the analysis (D_m, wC_maxss, wC_minss, and wC_avgss) are summarized in Table 2. Dasatinib 100 mg once daily and 70 mg twice daily had mean reductions of 6% and 14% from their nominal daily dose, respectively.

wC_avgss was identified as the most significant exposure predictor for achieving MCyR in the base model. The magnitude of the effect of predictor variables assessed in the full model is shown in Figure S1. The effect of gender was not statistically significant and, therefore, not included in the final model. The final logistic regression model (Table 3) indicated that the probability of achieving MCyR in CML-CP patients was significantly greater (P < 0.01) with increasing wC_avgss (odds increased 2.11-fold for every doubling of the wC_avgss), increasing D_m (odds increased 1.60-fold for every 10% increase of the D_m), and decreasing age (odds decreased 22% for every decade increase in life). Patients with imatinib-resistant disease were less likely to respond compared with patients with imatinib-intolerant disease (odds ratio = 0.52). Given the definition, D_m was considered an independent predictor of exposure (correlation coefficient between wC_avgss and D_m = −0.1).

Table 3.

Parameter estimates for dasatinib exposure–efficacy (MCyR) and exposure–safety (pleural effusion) relations

Predictora	Odds ratio coefficientb	95% CI	P-value	Odds ratio (5th and 95th percentiles: median)
Logistic regression model for dasatinib exposure–efficacy (MCyR) relation
Log2(wCavgss)	2.11	1.52–2.91	<0.001	0.43,1.83
Dm/10	1.60	1.41–1.81	<0.001	0.24, 2.51
Age/10	0.78	0.68–0.90	0.001	2.00, 0.62
Imatinib status (resistant versus intolerant)	0.52	0.40–0.65	<0.001	–
Predictor	Hazard ratio coefficientc	95% CI	P-value	Hazard ratio (5th and 95th percentiles: median)
Cox proportional hazard model for dasatinib exposure–safety (pleural effusion) relation
Cmin	1.22	1.12–1.33	<0.01	0.63, 2.38
Age/10	2.02	1.69–2.43	<0.01	0.13, 3.09

Notes:

^aLog₂(wC_avgss) increases by one unit for every doubling of wC_avgss; D_m/10 increases by one unit for every 10% increase of D_m; Age/10 increases by one unit for every increase of 10 years in age;

^bincrease in odds for every unit increase in the continuous predictor variable, or odds relative to the reference value of the categorical predictor variable;

^cincrease in hazard for every unit increase in the continuous predictor variable.

Abbreviations: CI, confidence interval; C_min, trough plasma dasatinib concentration; MCyR, major cytogenetic response; D_m, percentage dose maintenance duration; wC_avgss, weighted average steady-state plasma dasatinib concentration.

Figure 2 shows the model-predicted probability (95% confidence interval [CI]) of MCyR and an evaluation of the model, comparing the observed proportion of MCyR and model-predicted median proportion of MCyR (90% prediction interval) versus the three continuous predictors (wC_avgss, D_m, and age). Each panel shows that patients with imatinib-intolerant disease had higher response probability. The median predictions followed the trend of the relationships, and most of the observed response proportions were covered by the 90% intervals of the model predictions. The predicted median response rates (with 90% prediction intervals) for patients treated with dasatinib 100 mg once daily, 50 mg twice daily, 140 mg once daily, and 70 mg twice daily were 53% (27%–73%), 20% (7.7%–40%), 60% (47%–80%), and 47% (27%–67%), respectively. This evaluation indicates that, in general, the model described the relationship between MCyR rate and each predictor reasonably well, although response rate in the 50 mg twice daily group was underpredicted by the model.

Figure 2.

Observed proportion and predicted median proportion/probability of MCyR versus continuous predictors: (A) wC_avgss, (B) dose maintenance, and (C) age.

Notes: The symbols represent the proportion of responders, grouped by quartiles of predictors and plotted at the median for the groups; the centered curves and shaded areas represent median values and 95% confidence intervals of the model-predicted response probability, respectively; the vertical bars represent the 90% model prediction intervals of the MCyR rate, grouped by quartiles of predictors and plotted at the median for the groups; the horizontal box shows the distribution of predictors by treatment arm: the interior bar represents the median, the two ends of the box represent the 25th and 75th percentiles, the whiskers represent the fifth and 95th percentiles; the predicted medians have a 90% prediction interval.

Abbreviations: 1×/d, once daily; 2×/d, twice daily; MCyR, major cytogenetic response; wC_avgss, weighted average steady-state plasma dasatinib concentration.

E–R for safety: pleural effusion

Of the 567 patients included in the E–R analysis of the Phase III dose-optimization study, 94 had pleural effusions. Pleural effusion rates were 11.0% (100 mg once daily), 16.2% (50 mg twice daily), 17.7% (140 mg once daily), and 22.0% (70 mg twice daily). The C_min was identified as the most significant predictor of pleural effusion in the base model. The magnitudes of covariate effects on the hazard of pleural effusion assessed in the full model are shown in Figure S2. The effects of cardiac disease history, race (Caucasian/non-Caucasian), and gender were not significant and were removed from the final model. The final model identified age (hazard increased 2.02-fold for every decade increase in life) and C_min (hazard increased 1.22-fold for every 1 ng/mL increase in C_min) as statistically significant risk factors for pleural effusion (P < 0.01) (Table 3). There was no statistically significant (P < 0.01) interaction between the C_min and age, suggesting that age-related variation in C_min effect was not responsible for age-related variation in response.

Figure 3 illustrates the evaluation of the final model with respect to C_min and age (>55 years; ≤55 years) by the four dasatinib regimens in the Phase III study. There was generally good agreement between the model-predicted cumulative probability of pleural effusion and the corresponding pleural effusion probability estimates based on Kaplan–Meier analysis (the latter lying within the 90% model prediction interval), except for a slight overprediction in the 100 mg once daily and >55 years age group and a slight underprediction in the 140 mg once daily and ≤55 years age group. As this discrepancy was not consistent across subgroups, it is unlikely to be due to model misspecification. The underprediction noted in the lower age group is most likely a random effect, whereas the deviation in the higher age group may be due to a weak interaction between age and C_min. However, adding the interaction term into the final model did not produce statistical significance.

Figure 3.

Model-predicted (90% prediction intervals) and Kaplan–Meier estimated time to pleural effusion by age and dasatinib treatment groups.

Discussion

Optimizing therapeutic strategies for BCR-ABL1 inhibitors is an important area of investigation in CML.17,23–25 Given current recommendations for administering BCR-ABL1 inhibitor therapy indefinitely to patients who are responding to and tolerating these agents,4,26 efforts should be directed toward identifying potentially modifiable risk factors associated with the optimization of response and tolerability. The original dasatinib 70 mg twice daily regimen was selected based on the increased probability of achieving more continuous BCR-ABL1 inhibition and observed clinical responses in Phase I studies. Despite the short half-life of dasatinib, responses were nonetheless observed when the drug was administered once daily. We used data from the Phase III dasatinib dose-optimization study to characterize the E–R relationships for efficacy (MCyR) and safety ( pleural effusion) in patients with CML-CP. Our results suggest that it was possible to optimize dasatinib dosage because efficacy and safety were associated with different measures of exposures: achieving MCyR was most closely related to wC_avgss, whereas pleural effusion risk was related to C_min. Changing the dosing interval from twice daily to once daily and reducing the daily dose from 140 mg to 100 mg reduced the C_min and thereby reduced the probability of pleural effusion. Although reduced daily dosing resulted in a nominally lower C_avgss, the effect of the reduced exposure on efficacy was ameliorated by fewer dose modifications and interruptions. These results are consistent with the finding that dasatinib 100 mg once daily was associated with similar efficacy and decreased toxicity when compared with the 70 mg twice daily regimen.18

The dasatinib concentration–time data for patients enrolled in the Phase III dose-optimization study were well-described by a linear two-compartment PPK model with first-order absorption and were consistent with the dasatinib PK from earlier studies. The estimated PK parameters, shown to be time-invariant, agreed reasonably well with previously reported estimates from noncompartmental analyses.20 The variability in dasatinib exposure was found to be mainly due to IIV and IOV in bioavailability. In contrast, none of the examined patient covariates appeared to have a clinically relevant effect on dasatinib pharmacokinetics. These findings support the recommendation that dasatinib can be administered without dose adjustment for body weight, age, gender, or race.21 Dasatinib exposure appears to be dependent on dosing regimen, as demonstrated by applying the PPK model to the four arms of the Phase III study. Such differentiation in dasatinib exposures provided an opportunity to characterize the E–R relationships.

The E–R efficacy analysis results were consistent with previous findings that transient exposure to dasatinib is equivalent to continuous exposure in vitro and that dosing regimens with once daily dasatinib are as effective in achieving rapid and durable clinical responses as twice daily treatment.18,27 The most significant predictor of MCyR was wC_avgss. Because dasatinib pharmacokinetics are linear, patients treated with once daily and twice daily schedules receiving identical total daily doses should have the same probability of achieving MCyR. This expectation was consistent with the observed MCyR rates in the Phase III study (63% [95% CI, 55.7%–70.8%] with 100 mg once daily, 61% [95% CI, 53.5%–68.7%] with 70 mg twice daily, 63% [95% CI, 55.1%–70.2%] with 140 mg once daily, and 61% [95% CI, 53.5%–68.7%] with 50 mg twice daily).18 The analysis also identified D_m as a significant predictor of MCyR, independent of dose modification. Although the predicted transient nature of BCR-ABL1 inhibition with a once daily dosing interval does not negatively affect efficacy, 17 dose interruptions are expected to have an adverse impact, based on our analysis. Patients treated with dasatinib 100 mg once daily had fewer AEs and dose reductions from the nominal daily dose than those treated with 70 mg twice daily (6% versus 14%, respectively), as well as fewer dose interruptions (12% versus 15%). Therefore, maintaining steady dosing may be preferable to higher daily dosing that could lead to dose modifications and interruptions. Analyses for imatinib also revealed that adherence is a critical factor in achieving responses in patients with CML28,29 and that low adherence was likely associated with increased imatinib dose and AEs, although it must be noted that in contrast to dasatinib, imatinib has a prolonged terminal half-life. It is reasonable to believe that any causes of suboptimal adherence, eg, a complex route of administration and drug–food or drug–drug interactions, could negatively impact exposure and hence disease response in real-world settings.

As shown by the data in the Phase III dasatinib study, imatinib-intolerant patients are more likely to achieve MCyR than those with imatinib-resistant disease.17 Our analysis also suggests that younger patients are more likely to achieve a MCyR than older patients (odds decreased 22% for every decade increase in life). Gender had no significant association with response in our analysis. The associations of age and gender with response have been studied in patients with CML receiving second-line imatinib, with inconsistent results.30,31 In one study, age was a significant predictor of hematologic and cytogenetic response; however, in another, neither age nor gender were independent predictive factors for MCyR.30,31

Exposure–safety response analyses performed here identified C_min as having the strongest association with pleural effusion occurrence. In the Phase III study, the lowest C_minss was achieved with the 100 mg once daily schedule, which had the best benefit–risk balance of the four treatment regimens. 17 An ongoing Phase II study (OPTIM) is prospectively evaluating the optimization of residual dasatinib plasma levels in patients newly diagnosed with CML-CP.32 In the OPTIM study, patients began therapy with dasatinib 100 mg once daily, and the C_max and C_min were determined after 7 to 10 days. Patients with C_min <3 nM continued on 100 mg, and those with C_min >3 nM were randomly assigned to continue on 100 mg or to have their dose reduced in 20 mg increments until the C_min (monitored with C_max in 15-day intervals) decreased to <3 nM. After C_min <3 nM was achieved, the C_min and C_max were measured every 3 months. After a median follow-up of 7.2 months, the C_max was found to be associated with response, and the C_min was associated with fluid retention or pleural effusion.32

In addition to the C_min, age was found to be a significant risk factor for pleural effusion. The effect of age is not confounded with exposure, as the PPK analysis showed that exposure to dasatinib did not depend upon age. Furthermore, the Phase III study was well balanced for age across all arms (range of medians, 54–56 years).17 Considering the influence of age on pleural effusion risk and the decreased probability of efficacy, the therapeutic window for older patients (>55 years of age) with CML-CP may be narrower relative to younger patients, and these patients may need more intensive monitoring. Cardiac disease history has previously been identified as a risk factor for pleural effusion.16 However, in this analysis, cardiac disease history was correlated with age (R = 0.4), and the effect of cardiac disease history was not significant after accounting for the effect of age.

Patients receiving proton pump inhibitors (PPIs) or histamine- 2 receptor antagonists were not excluded from this PPK model, although these compounds are likely to reduce dasatinib exposure.21,33 For 399 patients with comedication data available, the PPK model estimated the ratio of median exposure between patients who did and did not receive PPIs or histamine-2 receptor antagonists to be 0.95 for dasatinib C_minss, 0.76 for dasatinib C_maxss, and 0.85 for dasatinib C_avgss. This contrasts with results from a drug–drug interaction study in healthy volunteers, in which prior administration of a PPI reduced dasatinib exposure by approximately 60%.21 The more modest effect of PPIs estimated in the current analysis may be due to confounding factors, including polypharmacy and unknown comedication histories and dosing times.

In general, our data show that opportunities for dose regimen optimization exist when the exposure measure most relevant for efficacy is different from that most relevant for safety. Whether this approach may be applied to other BCR-ABL1 TKIs remains controversial. For example, Larson et al have shown the correlation of imatinib C_min with both response and the occurrence of specific AEs, including fluid retention.34 Two more analyses also found an association of imatinib C_min with clinical response.35,36 Yet, data from two other imatinib studies found no correlation between response and C_min; instead, adherence to the standard imatinib dose was critical to achieve response.37,38 All these analyses examined C_min as the only exposure measure, whereas our analysis examined the model-derived C_maxss, C_minss, and C_avgss and selected the most statistically significant predictor. Our analysis also accounted for dose modification. Because of the inconsistencies among these analyses, further research is needed to better understand the E–R relationship of BCR-ABL1 TKIs as a class. It is possible that the efficacy of those agents with prolonged half-lives may correlate better with steady-state concentrations. It also remains to be determined whether the correlations observed in our analysis are applicable to other TKIs with short half-lives.

In conclusion, our analysis shows that dasatinib efficacy and safety were associated with different measures of exposure, wC_avgss and C_min. PPK analyses of the dasatinib Phase III dose-optimization study confirm that the 100 mg once daily schedule was associated with the lowest C_minss value of the studied schedules, corresponding with improved safety compared with dasatinib 70 mg twice daily.18 Our analyses highlight opportunities for optimizing BCR-ABL1 TKI dosing regimens and provide insight into the factors that should be considered when selecting an appropriate regimen. By quantifying dasatinib E–R relationships, the presented analyses establish a link between dosage regimen and the clinical outcome following drug exposure, providing important insight for optimizing CML therapy.

Supplementary materials

Table S1.

Clinical studies included in the population pharmacokinetic model

Study	Population	Design	Dasatinib dosing regimens	na	PK sampling schedule
CA180-002 (NCT00064233)1	CML(-CP, -AP, -BP) or Ph+ ALL resistant/intolerant to imatinib	Phase I, open-label, multicenter, dose-escalation study	15, 30, 50, 75, 105, 140, and 180 mg once daily (5 days on, 2 days off weekly dosing), 25, 35, 50, 70, 90, and 120 mg twice daily (5 days on, 2 days off weekly or continuous dosing)	83	Once daily (5 days on, 2 days off weekly dosing) Cycle 1, days 1, 5, and 26: predose; 0.5, 1, 1.5, 2, 3, 4, 5, 6, 8, and 24 h (all days); 48 and 72 h (days 5 and 26 only) Twice daily (5 days on, 2 days off weekly dosing) Cycle 1, days 1, 5, and 26: predose; 0.5, 1, 1.5, 2, 3, 4, 6, 8, and 12 h (all days); 5 h (days 1 and 5 only); 24, 48, and 72 h (days 5 and 26 only) Twice daily (continuous daily dosing) Cycle 1, days 1 and 5, and Cycle 2, day 1: predose; 0.5, 1, 1.5, 2, 3, 4, 6, 8, and 12 h (all days); 5 h (Cycle 1, day 1 only); 24, 48, and 72 h (Cycle 1, day 8, and Cycle 2, day 1 only)
CA180-005 (START-A; NCT00101647)2	CML-AP resistant/intolerant to imatinib	Phase II, open-label, multicenter study	70 mg twice daily	41	First 15 treated patients Cycle 1, days 1 and 8: predose; 0.5, 1, 1.5, 2, 3, 4, 5, 6, 8, and 10 h All subsequently treated patients Cycle 1, day 8: predose; between 0.5 h and 3 h
CA180-006 (START-B; NCT00101816)3	CML-BP (myeloid) resistant/intolerant to imatinib	Phase II open-label, multicenter study	70 mg twice daily	29	First 15 treated patients Cycle 1, days 1 and 8: predose; 0.5, 1, 1.5, 2, 3, 4, 5, 6, 8, and 10 h All subsequently treated patients Cycle 1, day 8: predose; between 0.5 h and 3 h
CA180-013 (START-C; NCT00101660)4	CML-CP resistant/intolerant to imatinib	Phase II, open-label, multicenter study	70 mg twice daily	144	Cycle 1, day 8: predose; between 0.5 h and 3 h; between 5 h and 8 h; between 12 h and the next dose
CA180-015 (START-L; NCT00101595)3	CML-BP (lymphoid) resistant/intolerant to imatinib	Phase II, open-label, multicenter study	70 mg twice daily	39	Cycle 1, day 8: predose; between 0.5 h and 3 h
CA180-017 (START-R; NCT00103844)5	CML-CP resistant to imatinib	Phase II, open-label, multicenter study	70 mg twice daily	78	Cycle 1, day 8: predose; between 0.5 h and 3 h; between 5 h and 8 h; between 12 h and the next dose
CA180-034 (NCT00123474)6,7	CML-CP resistant/intolerant to imatinib	Phase III, open-label, multicenter, doseoptimization study	50 or 70 mg twice daily, or 100 or 140 mg once daily	567	Day 15: predose; between 1 h and 3 h; between 5 h and 8 h Day 29: predose

Note:

^aPatients with PK samples included in the analysis.

Abbreviations: AP, accelerated phase; BP, blast phase; CML, chronic myeloid leukemia; CP, chronic phase; Ph+ ALL, Philadelphia chromosome–positive acute lymphoblastic leukemia; PK, pharmacokinetic.

Table S2.

Baseline demographic and laboratory values of patients in the population pharmacokinetic analysis

Covariate	N	Value
Median (range) age, years	981	55 (15–86)
Median (range) weight, kg	961	75 (38–180)
Mean (SD) BMI, kg/m2	905	26.9 (5.4)
Male/female, n (%)	981	484 (49)/497 (51)
Race, n (%)	981
Caucasian		816 (83)
Asian		83 (8)
Black/African American		49 (5)
Other		30 (3)
Unknown		3 (0.3)
Mean (SD) ALT, IU/L	944	26.2 (19.3)
Mean (SD) AST, IU/L	946	27.7 (16.8)
Mean (SD) creatinine clearance, mL/min	926	93.8 (35.4)
Mean (SD) baseline hemoglobin, g/dL	954	11.8 (2.0)
Mean (SD) baseline WBC count, ×103 cells/μL	959	20.6 (30.7)

Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; SD, standard deviation; WBC, white blood cell.

Figure S1

Estimated odds ratios of the full logistic regression model of MCyR.

Notes: The outside edges of the open boxes represent the estimated odds ratio at the fifth percentile of continuous predictor values; the open circles represent odds relative to the reference value of the categorical predictor variable; the outside edges of the shaded boxes represent the estimated odds ratio at the 95th percentile of continuous covariate values; the horizontal lines represent the 95% confidence intervals of estimated effects.

Abbreviations: MCyR, major cytogenetic response; wC_avgss, weighted average steady-state plasma dasatinib concentration.

Figure S2

Estimated hazard ratios of the full Cox proportional hazards model of pleural effusion.

Notes: The outside edges of the open boxes represent the estimated hazard ratio at the fifth percentile of continuous predictor values; the open circles represent hazard relative to the reference value of the categorical predictor variable; the outside edges of the shaded boxes represent the estimated hazard ratio at the 95th percentile of continuous covariate values; the horizontal lines represent the 95% confidence intervals of estimated effects.

Abbreviation: C_min, trough plasma dasatinib concentration.