Trials and Tribulations in Rare Cancer Clinical Research

Precision medicine, the application of novel systemic therapies on the basis of biomarkers, has been remarkably successful in advancing cancer treatment. With such progress, a new problem arises, namely, that of the ability to identify sufficient patients with the biomarker to complete a clinical trial. The time component to enroll a randomized trial becomes more acute for rare cancers, such as salivary gland carcinomas, peripheral T-cell lymphomas, or sarcomas of soft tissue and bone. All these groups of cancers represent molecularly diverse groups of cancers, but even as a group, each is so rare collectively that randomized trials are very challenging to complete.

THE TAKEAWAY

• In the article that accompanies this editorial, Gronchi et al¹ extended their findings from a larger randomized trial comparing less toxic neoadjuvant chemotherapy (trabectedin) with standard anthracycline-ifosfamide treatment for patients with primary extremity high-grade myxoid liposarcoma and found the less-toxic therapy noninferior, using a relatively broad hazard ratio for noninferiority. Although clinically intriguing, there are important caveats interpreting these trial data, given the difficulties in assigning type 1 error and power of the overall data set since the extension had a different end point from the initial trial.

The natural tension between study accrual rate versus the statistical power needed to claim a meaningful difference between treatment arms is worth further discussion, and it applies to phase II and phase III studies alike. In the article that accompanies this editorial, Gronchi et al,¹ who examined standard versus experimental neoadjuvant chemotherapy for primary soft tissue sarcoma, provided an example of the challenges of a clinical trial examining rare cancer subtypes.

Sarcomas represent approximately 0.9% of adult cancers and about 70 different cancer histologies with many more molecular subtypes, underscoring the challenges in studying these cancers.² The neoadjuvant trial under discussion examined patients with five different types of primary soft tissue sarcomas. One standard of care, as defined through previous randomized trials led by the Italian Sarcoma Group, is the use of three cycles of anthracycline-ifosfamide-mesna (AIM) before radiation and surgery for high-risk primary sarcomas.³ The novelty of the present trial was to investigate whether combinations of drugs thought better for a specific sarcoma subtype (histology-tailored therapy [HTT]) were superior to traditional, generally more toxic AIM. However, the original study was written with one-sided P values, in which the hypothesis was that HTT was better than AIM.⁴ However, the converse was seen in the final analysis, that is, survival with AIM was numerically better than that with HTT.⁵ However, since the study design did not include this comparison prospectively, the study was technically a negative one; HTT was not superior to AIM.

Clinicians debated whether these data could be used to argue for AIM chemotherapy as a standard of care for patients with these histologies. Some clinicians agreed that the study is a negative one. A different group indicated that the data were prospectively collected and that the outcome would not have changed with a proper two-sided comparison between arms, so that AIM was indeed an appropriate standard of care. A third group declared that the study should not be considered definitive since a no treatment control arm was omitted.

Examining the data at the end of the primary trial, a subset of patients appeared to do no worse with HTT than with AIM chemotherapy, specifically those with myxoid-round cell liposarcoma treated with the drug trabectedin. After analyzing the data from the 64 patients from the original trial, the study was reopened and random assignment continued. The study design continued with 1:1 random assignment, but the study design was changed to an open-ended Bayesian design. The end point for the study extension was determination of whether there was a <5% probability that HTT was noninferior to AIM with an odds ratio of 1.25, analyzing overall outcomes after every 10 events (disease recurrence or metastasis). HTT (trabectedin) was declared noninferior to AIM for myxoid-round cell liposarcoma after 37 more patients were randomly assigned and treated, for a total of 101 patients, with 15 events in 56 patients (27%) treated on the AIM arm and seven events in 45 patients (16%) treated on the trabectedin arm.

The implications of such a finding are important, specifically that patients with one of the more common sarcoma subtypes can benefit from the less toxic trabectedin than AIM. The concerns regarding the trial design are equally clear; it was not possible to determine type 1 or type 2 error from a study in which patients were enrolled with two different end points and two different study designs. There were relatively few events (22) on which to make the determination of noninferiority. Nonetheless, these data were collected from prospectively randomly assigned patients.

Are this study sample size and small number of events enough to provide convincing evidence of the noninferiority of trabectedin compared with AIM for high-grade myxoid-round cell liposarcoma? While the survival curves and the result of a prospective, randomized trial indicate noninferiority of the less toxic treatment, study design issues complicate the interpretation. Statisticians serve as guides for clinicians to help make decisions for their patients. In this case, the nature of the question changed between the initial study and its extension. This means that the clinician is left without the ability of the statistician to provide that guidance although the Bayesian noninferiority criterion was achieved. Nonetheless, clinicians treating these patients will need to either interpret and apply these data or discard them, as it is unlikely that a sizable study of neoadjuvant therapy will be examined again in this diagnosis in the foreseeable future, unless new drugs that are very active in metastatic disease are found. Similarly, for the larger study as a whole, should the standard of care for neoadjuvant therapy for soft tissue sarcomas be AIM, given the one-way nature of the analyses from the original study design? That decision also needs to be based on the other existing trials and the primary data from this study.

Noninferiority designs often times involve a 1,000 or more patients to show that a new treatment, often times a biosimilar, is not worse than the existing approved agent. In this study, the hazard ratio (HR) for noninferiority was 1.25, a coefficient that is more commonly used to identify superior therapy rather than noninferior ones.⁶ However, to enroll such a study in a rare cancer with a sample size of 1,000 could take a generation. The practical nature of the larger target HR is imposed by the practicalities of accrual. Similarly, the difference between outcomes between a control group and an experimental arm of a clinical trial has to be large to enroll the trial in a reasonable amount of time. The Oncology Center of Excellence of the US Food and Drug Administration and other regulatory authorities have provided guidance regarding the nature of studies examining rare cancers that support a disease-specific approach, which has yielded a large number of approvals for therapies using a wide variety of designs.⁷ The strength of the data supporting these approvals is a topic of discussion.^8,9

Phase II trial designs can also suffer from design issues, even when the goal is to simply look for activity of a treatment. Investigators writing their own trials all too often take the total available budget divided by an estimated cost per patient, yielding a study size, and back-calculate type 1 and 2 errors and an activity target, the converse of the method that a study hypothesis should be composed. Since the power of a study increases less than linearly with the size of the trial, it is of particular importance to identify agents that provide a chance for substantial benefit, to avoid having a study open for so long that the question is no longer interesting. A successful example of such a study is the phase II trial of atezolizumab that led to regulatory approval in the United States for the agent for the ultra-rare sarcoma subtype alveolar soft part sarcoma.¹⁰

For some cancers, for example, standard-risk germ cell tumors, the cure rate is fortunately high enough to make randomized studies difficult to conduct to change the standard of care. In the case of rare cancers, outcomes are much poorer and randomized trials are still feasible, but designs have to be customized to account for the challenges of accrual and expectations of outcomes. Setting a high bar for outcomes and including enough sites to complete a study are two important ideas to make rare cancer therapeutic trials as efficient as possible. All we need now are agents that meet those lofty aspirations.

AUTHOR'S DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Trials and Tribulations in Rare Cancer Clinical Research

The following represents disclosure information provided by the author 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/jco/authors/author-center.

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