Exposure–Response Relationships of the Efficacy and Safety of Ipilimumab in Patients with Advanced Melanoma

Cytotoxic T-lymphocyte antigen-4 (CTLA-4) is a key immune checkpoint molecule that downregulates T-cell activation by binding to its ligands, B7-1 (CD80) and B7-2 (CD86), on antigen-presenting cells (1). Ipilimumab is a fully human IgG₁ monoclonal antibody that blocks CTLA-4 from binding to its ligands, thereby augmenting antitumor immune responses (2). Consistent with its proposed mechanism of action, ipilimumab has been shown to increase the percentage of activated CD4⁺ and CD8⁺ T cells in peripheral circulation, with a concomitant decrease in naïve CD4⁺ and CD8⁺ T cells, in patients with advanced (unresectable stage III or IV) melanoma (3). Ipilimumab has shown a statistically significant improvement in overall survival (OS) in 2 randomized controlled, phase III trials of patients with advanced melanoma; one with ipilimumab monotherapy at 3 mg/kg in previously treated patients (4) and the other with ipilimumab at 10 mg/kg in combination with dacarbazine in previously untreated patients (5).

Ipilimumab monotherapy has been extensively studied in phase II clinical trials of patients with advanced melanoma, most of whom had received prior therapy for metastatic disease (6–8). The results of one of these trials (CA184-022) suggested that ipilimumab elicits a dose-dependent effect on both efficacy and safety endpoints, where the best overall response (BOR) rates were 11.1%, 4.2%, and 0% for ipilimumab at doses of 10, 3, and 0.3 mg/kg, respectively (6). Across ipilimumab clinical trials, the most common treatment-related adverse events were immune-related (irAEs; refs. 4–8), i.e., adverse events that are inflammatory in nature and consistent with an immune-based mechanism of action (2, 9). The incidence of irAEs of any grade, and irAEs of grade 3 and 4, showed a dose-dependent increase with ipilimumab in study CA184-022 (6). These efficacy and safety outcomes correlated with the results of population pharmacokinetic analyses, which showed greater achievement of the target trough concentration during the induction dosing period (every 3 weeks for 4 doses) for ipilimumab at 10 versus 3 mg/kg, and with the results of pharmacodynamic analyses, which showed a greater mean rate of increase in absolute lymphocyte count (ALC) at 10 mg/kg compared with lower doses (6).

Ipilimumab continues to be evaluated in melanoma and other tumor types, including castration-resistant prostate cancer and non–small cell lung cancer (2, 10, 11). To better understand the risk-benefit profile of ipilimumab monotherapy across the dose range evaluated in clinical studies, we conducted a retrospective analysis of pooled data from phase II trials including all patients for whom pharmacokinetic samples were available. More than 450 patients with advanced melanoma were included in the analyses, which were done to characterize ipilimumab exposure–response relationships for important measures of efficacy and safety. Specifically, the exposure–response relationships for tumor response and overall survival were analyzed to characterize efficacy and those for irAEs were analyzed to characterize safety.

The patients included in the current analyses had participated in one of the following 4 completed phase II clinical trials which evaluated ipilimumab monotherapy in advanced melanoma: CA184-004, bmyCA184-007, CA184-008, and CA184-022 (6–8, 12). Study CA184-004 investigated potential biomarkers of clinical activity in patients who received ipilimumab at 3 or 10 mg/kg (12), and study CA184-007 evaluated the impact of prophylactic oral budesonide on the rate of grade 2 or more diarrhea in patients who received ipilimumab at 10 mg/kg (7). Both of these studies included previously treated and previously untreated patients. In study CA184-022 (6), patients intolerant to or who progressed on prior therapy were randomized to ipilimumab at 0.3, 3, or 10 mg/kg, whereas in study CA184-008, only one dose of ipilimumab (10 mg/kg) was evaluated in previously treated patients (8). In all 4 of these studies, ipilimumab was given every 3 weeks for up to 4 doses (induction phase), followed by a maintenance phase (every 12 weeks beginning at week 24) in eligible patients. The first tumor assessment was carried out at week 12 (end of induction dosing). All patients, or their legal representatives, gave written informed consent to participate in their respective trials. A summary of patient demographics and laboratory values for the exposure–response analyses is given in Supplementary Table S1.

The exposure–response analysis of efficacy was characterized by 2 measures of tumor response and by OS, and the exposure–response analysis of safety included toxicities characterized as irAEs. The two measures of tumor response were: (i) BOR of partial response (PR) or complete response (CR) by mWHO criteria, and (ii) clinical activity of PR and CR by immune-related response criteria (irRC), an exploratory endpoint defined to capture the delayed tumor responses that are sometimes observed with immunotherapy (13). The irRC are derived from mWHO criteria and include new lesions in the assessment of total tumor burden; in contrast to standard criteria, the irRC do not necessarily characterize patients with new lesions as having progressive disease (13).

The exposure–response analysis of BOR by mWHO criteria was conducted with data from patients in studies CA184-007, CA184-008, and CA184-022 for whom summary measures of ipilimumab exposure and BOR assessment were available (N = 354; ∼73% of the total treated). Data from study CA184-004 were not included in the exposure–response analyses of BOR as tumor assessments in this trial were not adjudicated by an independent review committee (IRC), and patients who underwent biopsy of the index lesions were censored for efficacy. Similarly, the exposure–response analysis of clinical activity by irRC was conducted with data from patients in studies CA184-007, CA184-008, and CA184-022 for whom summary measures of ipilimumab exposure and the assessment of clinical activity were available [N = 419; 86% of the total number of patients treated/randomized (487)].

The exposure–response relationships for efficacy (OS) and safety (irAE) were analyzed with data from patients in studies CA184-004, CA184-007, CA184-008, and CA184-022 for whom summary measures of ipilimumab exposure were available [N = 498; 88% of the total number of patients treated/randomized (569)]. The safety endpoint reflected the worst reported grade of an irAE in any one of the categories: gastrointestinal, hepatic, skin, endocrine, and other.

The measure of ipilimumab exposure employed in the exposure–response analyses was ipilimumab steady-state trough concentration (Cminss) during the induction phase, as determined by a previously developed population pharmacokinetic model (6), using data from the 4 phase II studies. Pharmacokinetic concentrations in samples from patients enrolled in the 4 phase II studies were analyzed by a validated ELISA assay that had a lower limit of quantification of 0.4 μg/mL (14). Model validation in a previously published linear 2-compartment PPK model with zero-order intravenous infusion and first-order elimination have shown that it provided a good description of the entire ipilimumab serum concentration-time profile for the 3 and 10 mg/kg doses (15). The observed and model-predicted PK profile for patients who received ipilimumab at 3 or 10 mg/kg is presented in Supplementary Fig. S1. Steady-state peak and time-averaged concentrations were not selected for these analyses, as they are highly correlated with Cminss (correlation coefficient > 0.9). Moreover, the pharmacologic rationale for Cminss was considered to be stronger than the other 2 summary measures of exposure, based on the assumption that T-cell activation in the induction phase of treatment is maximized by blocking the interaction of CTLA-4 with its ligands over the entire dosing interval.

The exposure–response analysis of tumor response was characterized by proportional-odds logistic regression models relating Cminss to the probability of achieving BOR of (CR or PR), as defined by mWHO criteria as well as clinical activity by irRC, and based on IRC adjudicated tumor size measurements. The logit (log-odds) are given by:

where β₀ and β are scalar and vector parameters that represent baseline odds and the effect of Xi on achieving BOR or clinical activity, respectively. The model parameters were estimated by maximum likelihood. This model formulation assumes that the predictor variables Xi have proportional effects on the odds of Pr(BOR) or Pr(clinical activity).

Model development was conducted in 3 stages. First, a base model was developed to establish the existence and functional form of a relationship between ipilimumab Cminss and probability of a tumor response. Second, the covariate effects that may potentially modulate the exposure–response relationship were examined in a full model. Third, the final model was developed by backward elimination to retain only the covariates that were significant at a 1% level. The final model was evaluated by assessing the agreement between the observed and predicted proportion of tumor response and the associated 90% model prediction intervals (PI) in each dose group (0.3, 3, and 10 mg/kg). The model-predicted proportions (and associated 90% PI) in each dose group were obtained by simulation using the model- predicted probability of response of the patients in each dose group.

The full model examined the potential effects of the following covariates: body weight, age, gender, Eastern Cooperative Oncology Group performance status (ECOG PS), baseline serum lactate dehydrogenase (LDH) levels, concomitant budesonide, prior systemic anticancer therapy, metastatic stage (M-stage), prior immunotherapy, HLA-A*0201 status, and prior interleukin (IL)-2 therapy.

The exposure–response analysis of OS was characterized by a Cox proportional–hazards (CPH) model relating ipilimumab Cminss to the hazard of death. The hazard function is expressed as |$\lambda(t)=\lambda_0(t)\exp(\bbeta^T{\rm\bf x}_i)$|⁠, where λ₀(t) is the baseline hazard function and x_i is a vector of predictor variables. The parameter vector β is estimated by maximum partial likelihood.

We also assessed the potential effects of the following covariates on the exposure–response relationship: age, weight, gender, baseline ALC, HLA-A*0201 status, baseline LDH levels, prior systemic anticancer therapy, prior immunotherapy, prior IL-2 therapy, ECOG PS, and M-stage at study entry.

The CPH model was developed in 3 stages: (i) a base model was developed to establish the existence and functional form of the exposure–response relationship between OS and ipilimumab exposure (Cminss); (ii) a full model was developed to assess the effect of all the covariates simultaneously; (iii) the final model was developed by retaining the covariates that were significant at a 1% level, with appropriate functional forms of their relationships with OS. The CPH model was evaluated by comparing model-predicted cumulative probability of OS versus time with that obtained by Kaplan–Meier (KM) analyses.

The exposure–response of irAEs was characterized by a proportional-odds ordinal logistic regression model relating Cminss to the probability of experiencing grade 2 or more or grade 3 or more irAEs. The logit (log-odds) of grade 2 or more and grade 3 or more irAEs are given by:

where βGrade2, βGrade3 and β are scalar and vector parameters that represent baseline odds of grade 2 or more and grade 3 or more irAEs and the effect of Xi [i.e., Cminss, log(Cminss)] on achieving irAEs, respectively. The model parameters were estimated by maximum likelihood. This model formulation assumes that the predictor variables Xi have proportional effects on the odds of probability of irAEs.

Model development was conducted in 3 stages, similar to that for the exposure–response analysis of tumor response. Covariate-parameter relationships were examined for the baseline covariates of body weight, age, gender, ECOG PS, baseline LDH levels, concomitant budesonide, prior systemic anticancer therapy, metastatic stage, prior immunotherapy, HLA status, and prior IL-2 therapy. The final model was evaluated by assessing the agreement between the observed proportion of irAEs by dose and the associated 90% model PI.

Cminss was found to be a statistically significant predictor of the probability of a BOR (CR or PR) by mWHO criteria (P < 0.001), and the probability was found to increase with an increase in log–transformed Cminss (Table 1 and Fig. 1A). However, none of the covariates examined were retained in the final model as none had statistically significant effect on the exposure–response relationship for BOR by mWHO criteria.

The odds of a patient achieving a BOR by mWHO criteria increased by 2.41-fold for a 2.7-fold increase in Cminss [log(Cminss) increases by one unit for approximately 2.7-fold increase in Cminss]. The observed responders at 0.3, 3, and 10 mg/kg were 0%, 6%, and 11.7%, respectively, consistent with model-estimated probabilities of a BOR at median Cminss for 0.3, 3, and 10 mg/kg groups of 0.6%, 4.9%, and 11.6%, respectively. Figure 1A shows the results of model evaluation that compares observed and predicted proportion of mWHO responders at each dose level. The results from a predictive check showed that there is good agreement between the model-predicted probability of BOR and the observed proportion of BOR responders. Estimates of covariate effects from the full exposure–response model for BOR are presented in Supplementary Fig. S2, and the results suggest that Cminss had a greater effect on achieving a tumor response than other covariates.

Cminss was found to be a statistically significant predictor of the probability of clinical activity by irRC (P < 0.001), and the probability increased with higher log–transformed Cminss (Table 1 and Fig. 1B). However, as with BOR by mWHO criteria, none of the covariates examined were retained in the final model as none had statistically significant effect on the exposure–response relationship for clinical activity by irRC.

The odds of a patient achieving clinical activity by the irRC increased by 1.76-fold for a 2.7-fold increase in Cminss. The observed clinical activities by irRC at 0.3, 3, and 10 mg/kg were 6.3%, 15.0%, and 25.1%, respectively, consistent with the model-estimated probabilities of irRC responses at median Cminss for the 3 dose groups of 4.5%, 15.1%, and 25.6%, respectively. These results suggest that the probability of achieving clinical activity by the irRC increased with dose of ipilimumab. Figure 1B shows the observed proportion and model-predicted median proportion/probability of an irRC response versus Cminss. The results from a predictive check showed that there is good agreement between the model-predicted probability of irRC responses and the observed proportion of irRC responses. Estimates of covariate effects from the full exposure–response model for tumor response by irRC are presented in Supplementary Fig. S3, and similar to BOR by mWHO criteria, Cminss seemed to have a greater effect on achieving a tumor response than other covariates.

From the Kaplan–Meier analysis, OS improved with increasing ipilimumab exposure (Cminss) and dose as illustrated in Fig. 2, although OS seems to be more closely associated with exposure than dose. Almost all the patients in the 2nd and 3rd tertiles were from the 10 mg/kg dose group (approximately 90% and 99%, respectively), whereas most patients in the 1st tertile were from the 0.3 and 3 mg/kg dose groups (approximately 29% and 49%, respectively). Patients in the highest tertile of Cminss seemed to have markedly better OS than patients in the lower tertiles of Cminss, suggesting the existence of an exposure–response relationship even within the 10 mg/kg dose group. However, it is important to note that the Kaplan–Meier analysis for OS by Cminss does not account for the potential effect of confounding variables (such as LDH levels), and therefore model-based analyses were conducted.

Estimates from the full exposure–response model for OS are presented in Supplementary Fig. S4, and the results suggest that baseline LDH levels and Cminss had greater effects on the risk of death relative to other covariates. The results of the final model show that Cminss, baseline LDH, and ECOG PS were significant predictors of OS (P < 0.001). Importantly, HLA-A*0201 genotype did not have a significant impact on exposure–response OS in this analysis. OS improved with increasing Cminss, decreased with increasing LDH, and was worse for patients with ECOG PS more than 0. Patients in the 5th percentile of Cminss (1.75 μg/mL) had an OS hazard ratio of 1.52 relative to patients with median Cminss of 43.9 μg/mL. Patients in the 95th percentile of LDH levels (846 IU/L) had an OS hazard ratio of 3.27 relative to those with median LDH levels (206 IU/L), which was less than the top limit of normal (225 IU/L). Furthermore, the estimated OS hazard ratio for patients with an ECOG PS more than 0 was 1.72 relative to patients with an ECOG PS of 0. The hazard ratios at the 5th and 95th percentiles of the continuous predictors Cminss and LDH, relative to their median values, indicate that the magnitude of the effect was greater for LDH than for Cminss (Table 2).

At the median Cminss determined for the 0.3 mg/kg dose of ipilimumab, OS was estimated to be 0.85-fold and 0.58-fold lower than at the median Cminss for the 3 mg/kg and 10 mg/kg doses, respectively. The results of the final CPH model are presented in Fig. 3. In the final model, the effects of the continuous predictors Cminss and LDH and the categorical 1 predictor ECOG PS were evaluated by a 4 × 2 stratification of the data according to whether Cminss and LDH were above or below the median values in patients with ECOG PS of 0 or more than 0 for the dataset. For all of the stratified groups, there was generally good agreement between the model-predicted cumulative probability of OS and the corresponding Kaplan–Meier curves (the latter lying within the 90% model PI; Supplementary Fig. S5). The predicted distribution of OS in patients by LDH level, ECOG PS, and ipilimumab exposure (median Cminss at 0.3, 3 or 10 mg/kg) indicates that the effect of LDH on the risk of death is greater than that for ECOG PS and Cminss (Fig. 3). Patients with greater ipilimumab exposure seem to have lower risk of death at any given value of LDH and ECOG PS relative to patients with lower exposure at the same value for LDH and ECOG PS.

As depicted in Fig. 4, the probability of experiencing grade 2 or more and grade 3 or more irAEs seemed to increase with increasing ipilimumab exposure; however, patients who did not experience an irAE during the induction period were not likely to experience an irAE during maintenance therapy with ipilimumab. The Kaplan–Meier analysis of irAEs show that the majority of events occurred during the induction dosing period (Supplementary Figs. S6 and S7).

Estimates from the full exposure–response irAE model are graphically presented in Supplementary Fig. S8. The results suggest that Cminss had the largest effect on the occurrence of an irAE among the covariates tested. The results of the final model (developed by backward elimination) suggest that, among the covariates tested, only ipilimumab Cminss was identified as a significant predictor for the occurrence of an irAE (P < 0.001). HLA-A*0201 genotype was not found to have a significant impact on exposure–response irAEs in this analysis.

The exposure–response analysis indicated that the probability of experiencing a grade 2 or more and grade 3 or more irAE increased with increasing Cminss (Table 1). The point estimates of the odds ratio for 5th:median Cminss (1.75 μg/mL) and 95th (103 μg/mL):median Cminss (43.9 μg/mL) were 0.124 and 1.74, respectively; therefore, the odds of a patient experiencing a grade 3 or more irAE (vs. no irAE, grade 1 or grade 2) are expected to increase by 1.74-fold for a Cminss increase from 43.9 (median) to 103 μg/mL (95th percentile). Model-predicted probabilities of experiencing a grade 2 or more irAE at the median Cminss (5th, 95th percentiles) for the 0.3, 3, and 10 mg/kg treatment groups are approximately 9.8% (5.6, 14), 33% (19, 43), and 51% (34, 62), respectively, and the corresponding probabilities of experiencing a grade 3 or more irAE are approximately 3.3% (1.8, 4.9), 13% (6.8, 19), and 24% (14, 33; Fig. 4). The results from the predictive check indicate that there is good agreement between the model-predicted probability of grade 2 or more and grade 3 or more irAEs and the observed proportion of grade 2 or more and grade 3 or more irAEs.

The current pooled analyses of data from phase II trials in advanced melanoma represent the first quantification of exposure–response relationships for ipilimumab efficacy and safety endpoints in any tumor type. Specifically, the purpose of this study was to determine if there is an association between pharmacokinetics (exposure) and key clinical outcomes with ipilimumab, and to determine baseline patient characteristics that may impact these relationships. While there are 2 completed phase III clinical trials of ipilimumab in patients with advanced melanoma (4, 5), data from the 4 completed phase II studies included herein provide the largest dataset across a consistent patient population and treatment regimen, and provide a range of evaluated ipilimumab doses. Exposure–response analyses for ipilimumab at 10 mg/kg in combination with dacarbazine in patients with previously untreated advanced melanoma (5) are the subject of separate publications.

One of the objectives of the exposure–response analyses for efficacy was to characterize the relationship between ipilimumab exposure and objective response. Using either mWHO criteria or the exploratory irRC, the exposure–response analyses indicated that Cminss during the induction phase was a statistically significant predictor of response. By mWHO criteria, the probability of BOR increased with a higher log-transformed Cminss, even though the median Cminss at 10 mg/kg was well above the target concentration (20 μg/mL) required for maximal inhibition of CTLA-4 binding to its ligands (6). While a similar association was observed using irRC, model-based estimates indicated that the probabilities of a tumor response at the median Cminss for 0.3, 3, and 10 mg/kg doses were greater when evaluated by the irRC (4.3%, 15.0%, and 25.6% for irRC vs. 0.6%, 4.9%, and 11.6% for BOR by mWHO criteria, respectively). This observation reinforces the potential value of the irRC as a tool for characterizing the efficacy of immunotherapies, where durable objective responses are sometimes seen in patients who are characterized as having disease progression by mWHO criteria. Collectively, regardless of which efficacy criteria are used, these results suggest that the probability of achieving an objective response is greater with higher ipilimumab exposure. None of the covariates tested, including baseline LDH levels, had a statistically significant effect on the exposure–response analyses for tumor response.

An exposure–response analysis was conducted to characterize the relationship between ipilimumab exposure and OS using a CPH model. The analysis showed that there was a statistically significant relationship between ipilimumab Cminss and the hazard ratio for OS. Patients at the 5th percentile of Cminss (1.75 μg/mL) had an OS hazard ratio of 1.52 relative to patients with median Cminss (43.9 μg/mL), and the OS of patients at the 95th percentile of Cminss (103 μg/mL) had an OS hazard ratio of 0.552 relative to patients at the median Cminss. These results suggest that OS improves with increasing Cminss and ipilimumab dose.

The OS of patients with advanced melanoma is known to be associated with several prognostic factors, including ECOG PS and baseline serum LDH (16, 17). LDH has been shown to be an independent prognostic factor for survival even after accounting for site and number of metastases, and has been incorporated into the current American Joint Committee on Cancer (AJCC) staging classification (16). The Kaplan–Meier analysis with respect to dose should be interpreted as being suggestive of a qualitative dose response rather than a definitive quantification of the dose response, as the pooled phase II data included in the analyses are not balanced with respect to prognostic factors across doses. Similarly, the differences in OS in Kaplan–Meier analyses by Cminss tertiles should be interpreted as being indicative of a qualitative exposure–response, because this univariate analysis does not account for potential imbalances in risk factors between the Cminss groups. However, the multivariate Cox-proportional hazards model does include the key prognostic factors, and this model provides an estimate of the effect of Cminss on reducing the risk of death after adjusting for the presence of these risk factors.

Among the covariates tested, baseline serum LDH levels and ECOG PS were found to be significant predictors of OS. Our results further show that HLA-A*0201 status does not significantly impact exposure–response relationships for OS or safety, consistent with the results of a retrospective analysis of ipilimumab clinical trial data showing that efficacy and safety outcomes were independent of HLA-A*0201 status (18). Thus, while the phase III trial of ipilimumab monotherapy at 3 mg/kg enrolled only patients that were HLA-A*0201-positive (4), the results collectively suggest that ipilimumab efficacy and safety are independent of HLA-A*0201 status.

The risk of death increased with increasing LDH levels, and was higher for patients with ECOG PS more than 0 compared with those with an ECOG PS of 0. LDH seemed to have a greater effect on the risk of death than Cminss or ECOG PS in our exposure–response analyses. Although baseline LDH levels do not have a clinically meaningful impact on the systemic clearance of ipilimumab, patients with lower baseline LDH levels tended to have higher ipilimumab concentrations, which may have impacted OS. However, LDH levels and Cminss were accounted for in the exposure–response model and both were found to be independent covariates for OS. Interestingly, Cminss seemed to be more closely associated with OS than treatment group (dose). Based on the exposure–response OS analysis, a higher dose of ipilimumab may also provide a survival benefit for patients with poor prognostic features, for example, elevated baseline LDH levels. This result is consistent with the improved OS in phase III trials of ipilimumab in advanced melanoma, where patients with normal and elevated LDH levels were enrolled (4, 5).

The exposure–response analysis for safety showed that the probability of experiencing an irAE of grade 2 or more or grade 3 or more, and the probability of first occurrence of an irAE at a given time point, increased with increasing Cminss over the evaluated dose range. The model predicted that the probabilities of grade 3 or more irAEs were approximately 3%, 13%, and 24% at the median Cminss of the 0.3, 3, and 10 mg/kg doses. These data are consistent with the frequencies of grade 3 and 4 irAEs reported in the phase III trial of ipilimumab monotherapy at 3 mg/kg, which ranged from 10.2% to 14.5% (4). IrAEs may reflect ipilimumab's immune-based mechanism of action and the majority occur during the induction dosing period. They can be severe and life-threatening, but most were reversible with treatment guidelines developed in ipilimumab clinical trials (4–8, 19). There were no study drug-related deaths in CA184-007 (7), one at 3 mg/kg in CA184-022 (6), and one at 10 mg/kg in CA184-008 (8). In the first phase III trial, 12 deaths in 511 treated patients (2.3%) were related to ipilimumab at 3 mg/kg (4). However, there were no study drug-related deaths in the subsequent phase III trial in 247 patients who received ipilimumab at 10 mg/kg plus dacarbazine (5). Thus, while greater ipilimumab exposure increases the frequency of grade 3 or more irAEs, the availability of treatment guidelines for the management of irAEs may prevent this increased incidence from causing a higher rate of treatment-related deaths.

The results of the current analyses show a positive association between ipilimumab exposure, and by extension dose, and key clinical outcomes. An ipilimumab dose of 10 mg/kg produced a greater Cminss with numerically higher tumor responses and an increased probability of improved OS, although these outcomes were associated with a greater likelihood of developing an irAE. Given the association of Cminss with both efficacy and safety measures, individualized dosing might be an effective approach for the ipilimumab treatment of patients with advanced melanoma in the future, provided it is possible to determine a range of ipilimumab exposure that is optimal with respect to both efficacy and safety with data being collected in an ongoing phase III dose optimization trial (11). Both the 3 and 10 mg/kg doses of ipilimumab have been shown to improve OS in phase III clinical trials of patients with advanced melanoma (4, 5). While the current results provide evidence for the potential of better efficacy outcomes with greater ipilimumab exposure, further clinical studies of the risk-benefit profile are needed to determine the most effective dose of ipilimumab.

A. Roy is employed (other than primary affiliation; e.g., consulting) as a group director, clinical pharmacology and pharmacometrics at Bristol-Myers Squibb and has ownership interest (including patents) in Bristol-Myers Squibb. T.-T. Chen is employed (other than primary affiliation; e.g., consulting) as a adjunct assistant professor at Columbia University and has ownership interest (including patents). J. Weber is a consultant/advisory board member of Bristol-Myers Squibb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Feng, A. Roy, E. Masson, T.T. Chen, J. Weber

Development of methodology: Y. Feng, A. Roy, E. Masson, R. Humphrey, J. Weber

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Humphrey, J. Weber

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Feng, A. Roy, E. Masson, T.T. Chen, R. Humphrey, J. Weber

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Humphrey

Writing, review, and/or revision of the manuscript: Y. Feng, A. Roy, E. Masson, T.T. Chen, R. Humphrey, J. Weber

Study supervision: R. Humphrey

The authors thank StemScientific for providing editorial and writing assistance.

This study was supported by Bristol-Myers Squibb Co.

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.