Targeted-Agent Continual Reassessment Method: A Novel Bayesian Enrichment Design for Phase I Trials of Molecularly Targeted Therapies

Abstract

PURPOSE

Novel therapies targeting specific genomic alterations are a promising treatment approach for relapsed/refractory cancer. Patients with specific alterations may be more likely to respond. Trial designs should maximize opportunities for such patients to enroll on these trials. Existing designs do not enrich for patients with specific alterations. We developed the adaptive Targeted Agent-Continual Reassessment Method (TARGET-CRM) to optimize dose finding and enrich for patients with specific alterations, and applied it in a pediatric phase I trial.

METHODS

Patients were stratified to cohort A (unspecified tumors) or cohort B (rare genomic alterations). The TARGET-CRM design permits cohort B patients to immediately enroll at one dose level below the currently evaluated dose level instead of waiting for an open slot at the current dose level. Using simulations, we compared the operating characteristics (accuracy—the proportion of trials in which the true maximum tolerated dose [MTD] was identified/recommended; safety—the dose-limiting toxicity [DLT] rate; the proportion of cohort B patients enrolled) of the TARGET-CRM, standard CRM, and 3 + 3 designs across various scenarios.

RESULTS

The proportion of enrolled patients who were cohort B was higher for TARGET-CRM (90%-100%) compared with CRM (approximately 85%) and 3 + 3 (approximately 79%). The DLT rate and rate the true MTD was recommended were similar for TARGET-CRM and CRM, differing by only 0%-4% and 0%-4%, respectively. Results were similar regardless of trial sample size and proportion of cohort B patients in the population.

CONCLUSION

In phase I dose-finding trials of targeted agents, the Bayesian adaptive TARGET-CRM design maximizes enrollment of patients hypothesized as most likely to benefit from the targeted agent, while maintaining similar or superior accuracy and safety as the CRM and 3 + 3 designs.

INTRODUCTION

In phase I clinical trials, the objective is to identify the maximum tolerated dose (MTD) or recommended phase II dose of the agent. The rule-based 3 + 3 design¹ is conservative regarding dose escalation, with more patients treated at suboptimal doses. By contrast, novel adaptive model–based designs such as the continual reassessment method (CRM)² have a higher chance to identify the true MTD compared with the 3 + 3 design, per simulation studies.^3-5 Model-assisted designs⁶ such as the modified toxicity probability interval,⁷ Bayesian optimal interval,⁸ and keyboard⁹ designs achieve similar accuracy as model-based designs for identifying the true MTD. Despite superior performance, relatively few oncology trials use adaptive designs. In a recent review of pediatric phase I cancer trials in the United States, only 15% (35/231) used a model-based design.¹⁰ There is an unmet need to improve accuracy and safety of phase I trials through novel adaptive designs rather than rule-based designs.

CONTEXT

• Key Objective

• This study aimed to develop a novel Bayesian adaptive design, the Targeted Agent-Continual Reassessment Method (TARGET-CRM), for phase I dose-finding trials, which can enrich for patients who have a rare targetable mutation. The study also aims to raise awareness of the benefits of Bayesian adaptive designs compared with rule-based designs, especially in pediatric oncology where the 3 + 3 or rolling six designs have been used almost exclusively in phase I trials for decades.

• Knowledge Generated

• Bayesian adaptive designs such as the CRM and TARGET-CRM have demonstrated better accuracy in identifying the maximum tolerated dose (MTD) than rule-based designs. Bayesian adaptive methods are sufficiently flexible to permit enrichment for patients who could benefit from targeted therapy during a phase I trial to identify the MTD.

• Relevance

• Working together, the clinical investigator and biostatistician can simulate multiple trial designs, including TARGET-CRM, to identify the design best suited to the needs of a given phase I trial.

Novel therapies that target specific genomic alterations are a promising treatment approach for relapsed/refractory cancer. The number of US Food and Drug Administration approvals of novel targeted therapy demonstrating clinical benefit have recently increased. Patients harboring specific aberrations targeted by the agent are anticipated to experience higher response rates, but such patients may be rare. For phase I trials of targeted agents, enrollment must be maximized for patients with targeted aberrations, even if the trial's primary objective is dose-finding and not efficacy. Phase I trials with an all-comers design do not enrich for patients with targeted aberrations. The phase I trial of vismodegib (targeting Sonic Hedgehog [SHH]) in children with recurrent/refractory medulloblastoma used an all-comers design,¹¹ and only three patients with SHH-subtype medulloblastoma enrolled. All-comers designs enroll quickly, but patient characteristics are left up to chance. Alternative strategies enrich for patients through investigator selection. In the phase I trial of larotrectinib in pediatric solid tumors harboring NTRK gene fusions, investigators subjectively prioritized enrollment of patients with TRK fusion cancers.¹² Alternatively, enrollment could be stratified by disease subtype. The phase I trial of dasatinib in children and adolescents with relapsed/refractory leukemia had prespecified accrual goals in three strata: chronic myeloid leukemia (CML) in chronic phase; CML in advanced phase or Philadelphia chromosome (Ph)–positive ALL; and Ph-negative ALL or acute myeloid leukemia.¹³ However, such a design might increase accrual duration and be infeasible.

Our design research was motivated by the phase I trial of the stapled-peptide inhibitor of MDMX/MDM2, ALRN-6924, in children with cancer (ClinicalTrials.gov identifier: NCT03654716). The primary objectives were to determine the recommended pediatric phase II dose (RP2D) and safety profile of ALRN-6924. Patients in cohorts A (solid tumors or lymphoma) and B (rare tumors with hypothesized sensitivity to MDM2 or MDMX inhibition) with relapsed or refractory TP53 wild-type solid tumors, central nervous system tumors, or lymphoma received ALRN-6924 (Fig 1A). Patients in cohort C are not considered herein. The key design requirement was enrichment of cohort B patients, while allowing enrollment of patients with more common histologies.

FIG 1.

(A) Study cohorts for the pediatric phase I trial of ALRN-6924. Cohorts A and B use the TARGET-CRM design; cohort C uses the standard CRM design as there are no enrichment features planned for cohort C. (B) Continuous enrollment of cohort B patients using the TARGET-CRM design. TARGET-CRM, Targeted Agent-Continual Reassessment Method.

To meet this design requirement, we developed a novel adaptive Targeted-Agent Continual Reassessment Method (TARGET-CRM) design, which allows a specific subset of patients (eg, cohort B) to enroll at one dose level below the current dose level during the dose-limiting toxicity (DLT) observation period. This flexible framework can be applied to enrich any subset of patients, by diagnosis, tumor subtype, or underserved populations. In this paper, we compare the operating characteristics of the TARGET-CRM, standard CRM, and 3 + 3 designs using trial simulations. TARGET-CRM was implemented in the ALRN-6924 trial and a phase I trial of bromodomain and extraterminal domain inhibitors (ClinicalTrials.gov identifier: NCT03936465); the latter trial is not considered herein.

METHODS

For each design compared below, the primary objective is to identify an agent's MTD. The primary end point is the occurrence of DLT during a fixed DLT observation period. The following operating characteristics of the TARGET-CRM, standard CRM, and 3 + 3 designs were compared: (1) accuracy—proportion of trials where the true MTD was identified; (2) safety—DLT rate; and (3) proportion of enrolled patients from cohort B. Operating characteristics were averaged across 10,000 simulated trials in each scenario for various combinations of the simulation parameters.

3 + 3 Design

Cohorts of three patients are enrolled on the current dose level; if 0, 1, or ≥2 of three patients experienced a DLT, the next three patients are enrolled at a higher dose level, the current dose level, or the trial is stopped, respectively (Data Supplement, Fig S1). Traditionally, while a given cohort of three patients is being evaluated, new patients cannot enroll and are placed on a waitlist.

CRM

Let $d_1<d_2<\dots <d_K$ be the K prespecified dose levels to be evaluated in the trial. For the CRM,² let $F\left(d_k,\beta \right)=$ $\pi \left(d_k\right)$ be the dose toxicity model where the probability of experiencing a DLT strictly increases with dose level $\pi \left(d_1\right)<\pi \left(d_2\right)<\dots <\pi \left(d_K\right)$. A common choice is the empiric model: $F\left(d_k,\beta \right)={{\alpha }_k}^{\beta }$ where ${\alpha }_k$ is the prespecified Bayesian prior DLT probability for dose k. The prior can be specified by the investigator or determined using an algorithm.¹⁴ Let the target DLT threshold be $\pi *$, where $0<\pi *<1$, typically between 20% and 30%.

In a CRM design, a cohort of size X (investigator-specified) is enrolled at the current dose level. When that cohort's DLT observation period is complete, or a DLT is observed, the CRM model computes the Bayesian posterior DLT probabilities using the prior and observed DLT data. The recommended dose for the next cohort is the dose level with posterior DLT probability closest to the target DLT probability: $v_n={\arg \hspace{0.17em}\min }_k\left|F\left(d_k,b_n\right)-\pi *\right|$, where $b_n$ is the posterior mean of $\beta$ after $n$ observed patients. For safety, the trial will not skip dose levels during escalation. Accrual proceeds until the prespecified maximum sample size is attained. The MTD is the dose level with posterior DLT probability closest to the target DLT probability.

TARGET-CRM

The TARGET-CRM is a variation of the CRM design. In a CRM, patients enter a waitlist, and enrollment is suspended during observation for DLTs. If rare cohort B patients arrive during the DLT observation period, they cannot enroll (Fig 1B). TARGET-CRM allows cohort B patients to enroll at any time, at one dose level below the currently evaluated dose level. All patients, including cohort B at a lower dose level, are included in the TARGET-CRM model for dose escalation decisions.

Simulation Methods

Simulations, conducted in R version 3.6.3, compared the operating characteristics of TARGET-CRM, CRM, and 3 + 3. Simulations scenarios varied: true DLT probability, interarrival time, proportion of arriving patients, cohort size, toxicity rate, and total sample size.

Simulated trials had K = 5 dose levels, starting at the lowest dose level $d_1$. Five scenarios were tested, varying the distribution of true DLT probabilities over dose levels. The true MTD was defined as the dose level with true DLT probability closest to the target DLT probability, either $\pi *=$ 20% or ${\pi }^{*}=$ 33% (Table 1). An arriving patient presents to the clinic, is eligible, but not yet enrolled. The interarrival time between patients followed a Poisson distribution, either 15 or 28 days. The DLT observation period was 28 days. The proportion of arriving cohort B patients was 10%, 2%, or 33%.

TABLE 1.

Bayesian Prior DLT Probabilities, and True DLT Probabilities for Five Scenarios, One for Each Simulation Study

Dose Level	Prior DLT Probabilities	True DLT Probabilities
		Scenario S1	Scenario S2	Scenario S3	Scenario S4	Scenario S5
1a	0.049	0.20	0.10	0.04	0.05	0.01
2	0.11	0.25	0.20	0.12	0.10	0.05
3	0.20	0.32	0.25	0.20	0.15	0.10
4	0.31	0.40	0.32	0.28	0.20	0.15
5	0.42	0.45	0.40	0.35	0.30	0.20

NOTE. The target DLT probability is 20%, shown in bold, corresponding to the dose level for the true MTD.

Abbreviations: DLT, dose-limiting toxicity; MTD, maximum tolerated dose.

^a Starting dose level for the trial.

For the CRM and TARGET-CRM, simulations were performed using modified functions from the dfcrm package.¹⁵ Prior DLT probabilities were determined using the getprior() function (empiric model, halfwidth = 0.05, estimated MTD = d₃; Table 1). The cohort size was 1 or 3 patients. Intracohort dose de-escalation was permitted if recommended by the CRM model. For the CRM and 3 + 3 using the all-comers approach, if a dosing cohort was full, cohort A/B patients were placed on a waitlist; we assumed 50% of waitlisted patients would eventually enroll.

In TARGET-CRM, only cohort A patients are waitlisted. For cohort B, two design variations were considered: (1) TARGET-CRM1—cohort B patients arriving during the DLT observation period were allowed to enroll at one dose level below the currently evaluated dose, unless the it was the lowest dose, and then they were waitlisted; and (2) TARGET-CRM2—cohort B enrolled at the current dose.

For the CRM and TARGET-CRM, the trial ended at accrual goal: (1) if N = 18 was planned, ≥six patients treated on at least one dose level; or (2) if N = 30 was planned, ≥nine patients treated on at least one dose level. The minimum sample size was permitted to be exceeded if needed for enough patients to be enrolled/treated on one dose level. For the 3 + 3, the total sample size was governed by its rule-based dose escalation.

RESULTS

Simulation Results

Key operating characteristics are plotted in Figure 2 according to design and scenario. In a visual comparison of Figure 2A versus Figure 2D versus Figure 2G, and Figure 2B versus Figure 2E versus Figure 2H, and Figure 2C versus Figure 2F versus Figure 2I, the operating characteristics appear robust to changes in the proportion of arriving cohort B patients across scenarios.

FIG 2.

Simulated operating characteristics for the TARGET-CRM1, TARGET-CRM2, CRM, and 3 + 3 designs across five scenarios, using an enrollment cohort size of 3 patients. Simulated trials assume: (A-C) 2%, (D-F) 10%, and (G-I) 33% of arriving patients were from cohort B with total N = 18. Metrics are averaged across 10,000 simulated trials per design and per scenario. Metrics presented: (A, D, G) mean rate at which the true MTD was recommended; (B, E, H) mean DLT rate; and (C, F, I) of arriving eligible patients, the mean proportions per trial of cohort A and cohort B patients who enrolled. DLT, dose-limiting toxicity; MTD, maximum tolerated dose; TARGET-CRM, Targeted Agent-Continual Reassessment Method.

N = 18 With 10% of Arriving Patients From Cohort B

The proportion of trials in which the true MTD was identified (accuracy) was similar (within 0%-4%) for TARGET-CRM1 and TARGET-CRM2 compared with CRM (Fig 2D). However, in scenarios S2 and S3, the 3 + 3 was substantially less accurate than TARGET-CRM and CRM (Fig 2D). In scenario S5 where the true MTD was the highest dose level, TARGET-CRM designs achieved accuracy similar to the CRM, although some cohort B enrolled below the currently evaluated dose level. TARGET-CRM1, TARGET-CRM2, and CRM had similar (within 0%-4%) DLT rates (Fig 2E). The 3 + 3 had a DLT rate 3%-6% higher than the Bayesian adaptive designs. In scenario S1, all four designs had DLT rates greater than the 20% toxicity threshold since the starting dose was the true MTD. TARGET-CRM1 and TARGET-CRM2 enrolled a higher mean proportion of arriving cohort B patients (90%-96% and 100%, respectively) compared with the all-comers CRM (approximately 85%) and 3 + 3 (approximately 79%; Fig 2F).

N = 18 With 2% of Arriving Patients From Cohort B

TARGET-CRM1 and TARGET-CRM2 maintained a similar rate at which the true MTD was recommended (accuracy; Fig 2A) and DLT rate (Fig 2B) as CRM, while increasing the proportion of arriving cohort B patients enrolled (Fig 2C).

N = 18 With 33% of Arriving Patients From Cohort B

Results are similar to scenarios with 2% or 10% arriving cohort B (Figs 2G-2I).

In Figure 3 (N = 18, 10% arriving from cohort B), additional operating characteristics are varied: toxicity rate (Figs 3A-3C), cohort size (Figs 3D-3F), interpatient arrival rate (Figs 3G-3I), and total sample size (Figs 3J-3L).

FIG 3.

Simulated operating characteristics for the TARGET-CRM1, TARGET-CRM2, CRM, and 3 + 3 designs across five scenarios. Unless otherwise specified, simulations assume that 10% of arrival patients were from cohort B with a total N = 18, a target toxicity rate = 20%, enrollment cohort size of 3 patients, and an interpatient arrival rate = 21 days. Simulated trials assume (A-C) a target toxicity rate = 33%; (D-F) an enrollment cohort size of 1 patient; (G-I) an interpatient arrival rate of 28 days; and (J-L) a total of 30 patients. Metrics are averaged across 10,000 simulated trials per design and per scenario. Metrics presented: (A, D, G) mean rate at which the true MTD was recommended; (B, E, H) mean DLT rate; and (C, F, I) of arriving eligible patients, the mean proportions per trial of cohort A and cohort B patients who enrolled. DLT, dose-limiting toxicity; MTD, maximum tolerated dose; TARGET-CRM, Targeted Agent-Continual Reassessment Method.

20% Versus 33% Target Toxicity Rate

TARGET-CRM1 and TARGET-CRM2 maintained similar accuracy and DLT rate relative to CRM for trials with a target toxicity rate of 33% versus 20% (Figs 3A and 3B v Figs 2D and 2E). For a 33% target toxicity rate, the accuracy for the Bayesian designs was highest in scenarios S4 (approximately 60%-63%) and S5 (approximately 86%-88%) when the true MTD was the highest dose level.

Effect of Cohort Size

TARGET-CRM1 maintained similar accuracy and DLT rate for trials with an enrollment cohort size of 1 versus 3 patients (Figs 3D and 3E v Figs 2D and 2E). TARGET-CRM2 had lower accuracy with an enrollment cohort size 1 versus 3 for scenarios S3 and S4. TARGET-CRM1 and TARGET-CRM2 had a higher mean proportion of arriving cohort B enrolled for cohort size 1 versus 3 (Fig 3F v Fig 2F).

Effect of Interpatient Arrival Rate

TARGET-CRM1 and TARGET-CRM2 maintained similar rate at which the true MTD was recommended (accuracy) and DLT rate for interpatient arrival rates of 28 (Figs 3G and 3H) versus 15 days (Figs 2D and 2E). TARGET-CRM1 and TARGET-CRM2 had a lower proportion of arriving cohort B enrolled for trials with interpatient arrival rates of 28 (Fig 3I) versus 15 days (Fig 2F).

Effect of Total Sample Size

The rate at which the true MTD was recommended (accuracy), DLT rate, and proportion of arriving cohort B enrolled appear robust to changes in sample size (Figs 3J-3L v Figs 3G and 3I).

Effect on Study Duration

Accrual duration of TARGET-CRM is slightly reduced compared with CRM, since a small number of cohort B patients enroll during the DLT observation period. For scenario 5 (interpatient arrival rate = 15 days, DLT observation period = 28 days; Data Supplement, Table S1), the mean study durations of TARGET-CRM were 362, 358, and 342 days when the cohort B proportion was 2%, 10%, and 33%, respectively. The mean study duration for CRM was 363 days.

ALRN-6924 Trial Design

The trial started on dose level 1 of 5 (Table 2). To evaluate the operating characteristics of TARGET-CRM, statistical simulations were conducted using the following parameter values: target toxicity rate = 20%; cohort size = 3; cohort B comprised 10% of arriving patients; interpatient arrival rate = 15 days; and DLT observation period = 21 days.

TABLE 2.

Dose Levels and Prior Toxicity Probabilities for the Pediatric Phase I Trial of ALRN-6924

Dose Level	Dose of ALRN-6924 (days 1, 4, 8, 11)	Prior Toxicity Probabilities, %	True Toxicity Probabilities, %
			Scenario A1	Scenario A2
Level −1	1.6 mg/kg	2.5	2.5	3
Level 1—starting dose	2.2 mg/kg	5	5	5
Level 2	2.7 mg/kg	10	12	10
Level 3	3.5 mg/kg	15	20	15
Level 4	4.3 mg/kg	25	30	20

NOTE. Also shown are the true toxicity probabilities for two simulation scenarios (A1 and A2), with target DLT probability of 20% shown in bold, corresponding to the dose level for the simulation-true MTD.

Abbreviations: DLT, dose-limiting toxicity; MTD, maximum tolerated dose.

In scenarios A1 and A2, respectively, 33.7% and 58.6% of trials recommended the true MTD, 14.6% and 12.0% experienced a DLT, 34.8% and 34.8% enrolled one or more arriving cohort B patients at one dose level below the current dose, and an average of 0.41 arriving cohort B patients enrolled per trial (Table 3). With 10% of arriving patients from cohort B (eg, an average of two cohort B patients per trial when N = 18), TARGET-CRM enrolled approximately 30% of all arriving cohort B patients. Furthermore, in the simulations, nearly every arriving cohort B patient was able to enroll in the trial. On the basis of these simulation results, TARGET-CRM was selected for the ALRN-6924 trial, which was activated on October 16, 2018, with dose levels and priors from Table 2. The trial is ongoing; clinical results will be reported separately.

TABLE 3.

Operating Characteristics of the TARGET-CRM Design in Two Scenarios on the Basis of 2,000 Simulated Trials

Scenario	Operating Characteristics
	Total Sample Size, Median (range)	Proportion of Trials That Identified the True MTD, %	Proportion of Patients Who Experienced a DLT, %	Trials With ≥ 1 Cohort B Patient Enrolled at One Dose Level Below the Current Dose, %	No. of Cohort B Patients Enrolled per Trial at One Dose Level Below the Current Dose, Mean (SD)	No. of Cohort B Patients Enrolled per Trial Overall, Mean (SD)
A1	18 (18-21)	33.7	14.6	34.8	0.41 (0.61)	2.0 (1.3)
A2	18 (18-19)	58.6	12.0	34.8	0.41 (0.62)	2.0 (1.3)

Abbreviations: DLT, dose-limiting toxicity; MTD, maximum tolerated dose; SD, standard deviation; TARGET-CRM, Targeted Agent-Continual Reassessment Method.

DISCUSSION

Our simulation study showed that our TARGET-CRM design increased the enrollment of eligible enrichment cohort patients while maintaining similar accuracy and safety as the standard CRM. These results held true when the proportion of arriving patients from cohort B was 10% or 33%, and for small (n = 18) and moderate (n = 30) sample sizes. As expected, the adaptive TARGET-CRM and CRM designs have accuracy that is superior to the rule-based 3 + 3 design in many scenarios, as the 3 + 3 design is known to be conservative in dose escalation.⁵

The key feature of the TARGET-CRM design is that it allows continuous enrollment of arriving eligible enrichment cohort patients during the DLT observation period of the current patient cohort. Other designs have a similar feature: the rule-based rolling six allows enrollment of two to six patients concurrently, depending on the number experiencing a DLT or at risk for a DLT¹⁶; and the adaptive time-to-event continual reassessment method is an extension of the CRM, using time without DLT to inform the Bayesian model and allow continuous enrollment.¹⁷ Recently proposed backfill designs^18,19 also allow patients to enroll concurrently at dose levels below the currently evaluated dose during dose escalation to increase the availability of patient slots and to generate additional safety and efficacy data at dose levels below the MTD. Although these designs allow continuous enrollment at or below the currently evaluated dose level, only the TARGET-CRM prioritizes continuous enrollment of a specific subset of patients (eg, enrollment of those with a rare targeted mutation who are most likely to benefit). The TARGET-CRM design can reduce trial duration compared with the standard CRM but this effect is small since the number of cohort B patients enrolling during the DLT observation period is relatively low. As long as cohort B patients form a relatively low proportion of the overall accrual, there is no need to increase the planned sample size for the TARGET-CRM compared with the CRM.

In the ALRN-6924 trial, the investigators wanted to maximize the opportunity for patients with aberrations of MDM2 or MDMX to enroll, although response was not a primary end point of the trial. The ALRN-6924 trial included an expansion cohort to explore efficacy at the RP2D. Alternatively, a composite toxicity/efficacy end point could have been used to simultaneously evaluate DLTs and response to identify a biologically optimal dose.²⁰ However, there are feasibility concerns regarding composite toxicity/efficacy end points when there is patient heterogeneity and limited sample size, as in ALRN-6924.

The TARGET-CRM design is generalizable and can enrich for patient cohorts with any prespecified feature, for example, to prioritize/increase enrollment of underrepresented minorities in clinical trials, thereby improving generalizability of trial results and decreasing disparities.^21-23 Likewise, patients with rare histologies can be enriched. Depending upon the specific context and expected toxicity profile, either the TARGET-CRM1 or TARGET-CRM2 approach can be considered.

Our simulations are fairly comprehensive, but limited to five dose levels, toxicity thresholds of 20% or 33%, and average accrual of one patient every 15 or 28 days. Another potential limitation is that only one set of prior toxicity probabilities was used, although the performance of the designs is unlikely to be strongly influenced by the selection of the prior.²⁴ If the accrual rate is very slow, the operating characteristics of the TARGET-CRM design would converge to those of the standard CRM design, as fewer cohort B patients would arrive and enroll during the DLT observation period. By contrast, if the accrual rate is very high, the TARGET-CRM design might be less accurate than the standard CRM design, as a large number of cohort B patients would be enrolled at lower dose levels, and fewer patients would be enrolled at dose levels closer to the true MTD. When designing a trial with TARGET-CRM, trial-specific circumstances should be explored through simulations.

In summary, the TARGET-CRM design prioritizes enrollment of patients with a rare targetable mutation, increasing enrollment of patients anticipated to benefit. With the TARGET-CRM, trial accuracy and patient safety are maintained compared with the standard CRM. The TARGET-CRM has been successfully implemented in the ongoing pediatric trial of ARLN-6924, illustrating its advantages in phase I trials of molecularly targeted agents and for enrichment of patient subgroups.

AUTHOR CONTRIBUTIONS

Conception and design: Clement Ma, David S. Shulman, Steven G. DuBois, Wendy B. London

Financial support: Wendy B. London

Administrative support: Wendy B. London

Provision of study materials or patients: Wendy B. London

Collection and assembly of data: Clement Ma, Hasan Al-Sayegh, Wendy B. London

Data analysis and interpretation: Clement Ma, David S. Shulman, Hasan Al-Sayegh, Wendy B. London

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).