Alternative statistical designs cannot fully mitigate the limitations of traditional clinical trials in rare cancers. Creative study designs that integrate early clinical data and correlative outcomes from concomitant translational and laboratory models to evaluate the efficacy of druggable targets can potentially expedite access to novel therapies for these patients.
See related article by Hunter et al., p. 6095
Rare cancers such as chronic myelomonocytic leukemia (CMML) represent a significant therapeutic unmet need. At this time, allogeneic stem cell transplant is the only available treatment for CMML with a curative potential (1). Developing rational therapies becomes challenging when the rarity of the disease prohibits enrollment of the number of patients needed for adequately powered clinical trials (2). Alternative trial designs have been proposed as a mechanism to mitigate the limitations of small patient samples (3). However, innovative research designs that integrate human, murine, in vitro and next-generation sequencing (NGS) data to synthesize a comprehensive analysis could constitute the next frontier to solve this statistical conundrum and rapidly bring life-saving therapies to patients with rare cancers.
In this issue of Clinical Cancer Research, Hunter and colleagues present the results of a phase I/II multi-institutional trial of ruxolitinib in CMML (4). The study's premise is that CMML lacks FDA-approved therapies besides hypomethylating agents (HMA) due to being a rare disease; however, an improved understanding of its pathobiology can identify therapeutic targets. The authors' previous work demonstrated hypersensitivity to GM-CSF as a molecular pathway in CMML and established the preclinical rationale for JAK1/2 kinase inhibition as an attractive therapeutic target (5, 6). The phase I portion of the study employed a rolling ‘6' design and enrolled 24 patients while the phase II portion employed a Simon 2 stage design. Ruxolitinib was administered in two divided doses of 40 mg/day in the phase II component. A panel of 49 recurrently mutated somatic mutations in CMML and cytokine secretion patterns were also correlated with response to therapy. Simultaneously, 49 patient-derived xenograft murine models (PDX) were created from 13 patient samples and randomized 1:1 to receive pharmacokinetically equivalent doses of ruxolitinib (60 mg/kg) twice daily or vehicle. The primary endpoint in the PDX models was overall survival, with secondary endpoints including leukemic engraftment, splenic weight, and hematologic parameters.
The median duration of treatment (DOT) was 152 days and 38% (19/50) patients achieved a clinical response by MDS/MPN IWG criteria. Among 23 patients with baseline splenomegaly, 43.5% (10) achieved a splenic response. Responses were also noted among patients that had received prior HMA therapy. A nonstatistically significant longer median OS (mOS) was observed in responding patients compared with those with stable or progressive disease. There was a significant association between DOT and clinical benefit from ruxolitinib. Compared with other patients, the 31 (62%) patients that achieved a clinical response or stable disease and had DOT ≥ 3 months had a greater improvement in the MPN – Symptom Assessment Form (SAF; 43.2% vs. 23.1%, P = 0.05), a symptom score reflecting quality of life, and an improved mOS (32.8 months vs. 7 months, HR 0.36, P < 0.001). The toxicity profile was tolerable and expected for ruxolitinib, with the only grade 3/4 adverse events being anemia in 5 (10%) patients and thrombocytopenia in 3 (6%) patients (4).
The somatic mutations reported in the study were consistent with the most frequently reported cluster of mutations in CMML, mainly TET2 (62%), SRSF2 (47%), ASXL1 (20%), and NRAS (20%); however, no individual mutation or functional class of mutations correlated with the observed responses to ruxolitinib. Similarly, the cytokine profiles failed to yield relevant information and did not correlate with response to ruxolitinib. In this scenario, additional datapoints provided by the generated PDX models proved to be useful to confirm the clinical results. The authors previously established that murine PDX CMML models result in matching immunophenotype and genetic mutations in engrafted myeloid cells valuable for in vivo drug evaluation (7). Remarkably, in this study, the PDX derived from patients that responded to ruxolitinib demonstrated greater improvement in median OS and greater reduction in leukemic engraftment when treated with ruxolitinib compared with vehicle. The mice generated from responding and nonresponding patients uniformly showed a decrease in splenic volume as may be expected from ruxolitinib therapy compared withvehicle. Supporting a key observation in the human trial, PDX mice derived from patients that experienced a clinical response or stable disease and had DOT ≥ 3 months had a significant improvement in mOS, reduction in leukemic engraftment, and reduction in splenic volume. Data obtained from the PDX model proved to be crucial for the conclusions of the study (4).
The limitations to conduct traditional prospective clinical research in rare diseases are well known and will persist in the era of precision medicine as the treatment paradigm shifts from cancer type to molecular targets in smaller groups (8, 9). Some statistical designs have been reported in the literature to overcome the problems related to power and sample size in rare diseases, such as factorial studies, two group crossover studies, n-of-1-trials, and alternating designs, with limited success (9–11). A variety of adaptive statistical designs using Bayesian methods or historical controls are also increasingly used. While treatment differences can be identified, the wide confidence intervals make it difficult to estimate the true magnitude of the treatment effect. Thus, in instances where traditionally powered studies are unfeasible or impossible to enroll and complete within a stipulated timeframe, the adoption of more creative study designs could provide a quicker and stronger answer. In particular, the adoption of correlative outcomes beyond the traditional biomarker driven endpoints could be a game changer.
We envision a future for clinical trials in rare cancers where information obtained from multiple sources is leveraged and integrated bioinformatically into tangible results that can provide the required information to multiple agencies (e.g., FDA, clinical trials networks) to make decisions regarding applicability and approval of specific agents in rare cancers. In Fig. 1, we describe a model for such integrative studies that can be immediately adapted. It encompasses the information from phase I and phase II study design as well correlates obtained from next-generation sequencing and biomarker-driven associations, which are already frequently incorporated as a standard in modern clinical trial designs. In addition, we suggest that clinical studies in rare cancers should always seek out translational laboratory correlates that inform disease pathophysiology and express the druggable target, that can then be utilized as a therapeutic model to recapitulate observed clinical responses in the ongoing human trials. While PDX models are ideal, they may not always be feasible. Depending on the cancer type, other innovative models such as fluorescence-activated cell sorting (FACS) performed pretreatment and posttreatment or patient derived cell lines (in hematologic malignancies) and organoids (in solid tumors) generated and subjected to the same treatment as in the clinical trials, may achieve similar data to confirm or support observed clinical responses. Furthermore, proteomic and metabolomic profiles can be analyzed before and after treatment to provide a specific proteomic/metabolic signature associated with response to therapy and can also be adapted into clinical practice.
While we await results from the expanded phase II portion on this study and further data examining the question of whether ruxolitinib can truly modify the disease pathobiology in CMML, one thing is certain, incorporation of novel laboratory correlatives (e.g., PDX models) and newer omics platforms into traditional clinical trial design is an exciting prospect to look forward to. We hope this is a trend that is here to stay and correlative translational experiments inform future innovative and integrative clinical trial designs.
A. Shastri reports grants from Kymera Therapeutics, other support from Janssen Pharmaceuticals, and other support from OncLive outside the submitted work. No disclosures were reported by the other author.