Accelerating the Pace of Change: Getting the Right Drugs To the Right Patients

Laura J. Esserman
Helen Diller Family
Comprehensive Cancer Center,
University of California San Francisco

Laura van’t Veer
Helen Diller Family
Comprehensive Cancer Center,
University of California San Francisco


he following article describing the I-SPY2 trial summarizes this multi-year effort and its important research achievements regarding increased efficiency in evaluating and developing promising therapeutic drugs for, as well as improvement in strategies for treatment and care of, individual breast cancer patients. These drug development advances derive from using an efficient master protocol and a platform (umbrella) trial design.

The former involves submission and approval of a protocol which specifies the trial design but allows drug candidates to be subsequently included in a multi-arm trial design when appropriate mechanism of action, dosing, and toxicology data from specific drug candidates become available. It fills the “niche” needed to evaluate breast cancer drug intervention strategies between late recurrence localized disease (not needing chemotherapy) and metastatic disease, focusing on breast cancer populations at risk for early recurrence.

The I-SPY2 phase 2 trial design uses pathologic complete response (pCR) as a primary endpoint and employs an adaptive design to provide higher accrual in the trial arms showing higher pCR responses. The biomarker regulatory science employed by the trial includes those needing IDEs, as well as those for which data exists but from which patient decisions are not yet made, to those that are purely exploratory. Many candidate drugs have shown statistically significant activity versus randomized controls, to allow estimation of smaller/more efficient phase 3 trial designs. The employed precision medicine science methodology and biomarker science is leading to individual patient treatment benefits, such as evidence for those not needing toxic chemotherapy, those not benefiting from initial targeted therapy, and those for which lesser/non-toxic dosing may be adequate treatment. pCR data as an endpoint is adequate thus far to support accelerated approval but not yet a validated surrogate endpoint. Emerging data comparing pCR to multiyear event-free survival could likely result in its validation. Finally, the “customization” of therapy for individual patients provided by this biomarker-based and overall neoadjuvant trial design provides a model approach to other cancers as the biomarker and drug science and data evolve.    

More and more, it is hard to keep up with the accelerating pace of new technology. What seemed impossible just a decade ago – self-driving cars, bionic limbs, or AI on mobile phones (of all places) – is rapidly becoming commonplace.

Then why is drug development so painfully slow?

This is precisely the question the founding investigators of the I-SPY2 trial asked themselves over a dozen years ago. The situation then was much as it generally remains today: new drugs taking longer and costing more to get to market, high failure rates of agents in phase 3 trials, patients frustrated with the lack of new treatment options, etc. At that time, treatment for non-metastatic breast cancer was generally a choice between two different cytotoxic chemotherapy regimens, neither ideal.

But this was also the dawn of targeted therapies ushered in by breakthrough treatments like Gleevec and Herceptin. The approval of Herceptin (trastuzumab) in 2006 fundamentally changed our approach to treatment and outcomes for a specific subtype of aggressive breast cancers that express the HER2 oncogene.

For many of us, this was no cause for celebration. It was simply the end of an excruciating, long wait — the first phase 1 trials of Herceptin took place almost 20 years before. What’s more, it was approved for use in metastatic breast cancer eight years prior, in 1998. This journey started in 1987 with validation of the target. Development of the targeted antibody took only a few years. But many of our patients quite literally died waiting for this life-saving drug.

We resolved to find a way to get new and better treatments to our patients, faster and cheaper. Our goal was not only to design a faster and more efficient clinical trial, but to plant the seeds of a fundamentally new approach to the clinical development process, and to change the culture of expectation. Instead of settling for seismic knowledge turns every fifteen to twenty years, our aspiration was to drive this cycle to under five years.

So far, it seems to be working.

Phase 2 I-SPY2 Trial

The centerpiece of I-SPY is the phase 2 I-SPY2 trial (see Tables 1 and 2), which opened in 2010 and is evaluating neoadjuvant therapy for stage 2 and 3 breast cancers where the risk for recurrence is early (within five years). We picked an early primary endpoint: pathologic complete response (pCR) at the time of surgery, which is the complete disappearance of tumor locally and within nodes. Importantly, incorporating an early read out on treatment efficacy is the first essential step to dramatically reduce the time to assess the impact of new agents. The second step is introducing targeted biologic agents early in the course of disease where cure is still possible, and complete response is likely to translate into better survival. The third step is to focus the trial on the population at risk for early recurrence (by excluding 70-gene low-risk, hormone receptor positive patients, whose risk of recurrence is late and who do not benefit from chemotherapy), and to evaluate new agents prospectively across predefined biomarker signatures (ten total), enabling a precision medicine approach.

To date, I-SPY2 has enrolled more than 2,200 participants and completed the evaluation of eleven different experimental agents or combinations, about two per year. Of these, seven have “graduated” (the I-SPY term for meeting the primary outcome) in at least one biomarker signature, one was halted for toxicity, and three completed enrollment or were stopped for futility. Perhaps most importantly, a woman entering the study today has about double the chances of achieving a good outcome (pCR) compared to when the study began.

Table 1. Features of I-SPY2 Trial

  • Platform trial evaluating multiple agents, from multiple companies, under a master protocol
  • Carried out in neoadjuvant treatment setting of high-risk Stage II and III locally advanced breast cancers, evaluating agents in disease setting where cure possible
  • Involves response adaptive randomization across ten biomarker signatures; up to five agents tested simultaneously against relevant common controls
  • Pathologic complete response (pCR) is primary endpoint; volumetric MRI used to follow patients longitudinally during neoadjuvant treatment.

Table 2. Achievements of I-SPY2 Trial

  • Enrolled >2,200 participants; completed evaluation of 11 agents and agent combinations
  • Successful use of master protocol; agents added under one IND by protocol amendment, not new totally separate protocols for each agent or agent combination
  • Established pCR as early endpoint associated with clinical outcome in high-risk locally advanced breast cancer
  • Precompetitive model applied where no one stakeholder has competitive advantage; simplified contracting with stakeholders
  • Transformed care at participating sites, standardizing approach to neoadjuvant therapy across all involved specialties (surgery, pathology, radiology, medical oncology)
  • Created an example of team science and a learning system
  • With established pCR endpoint and longitudinal evaluation of volumetric MRI, emerging new agents can be rapidly evaluated, strategies for care optimized, and treatments for individual patients will be more effective with lower toxicity.

When I-SPY2 opened for enrollment, it was one of the first, and is now the longest continuously enrolling, “platform trial.” The term “platform trial” refers to a clinical trial with a single master protocol that evaluates multiple experimental treatments (or diseases) in parallel. The master protocol allows individual agents to enter and leave the study seamlessly, with no interruption or new enrollment into the study. New agents are introduced via protocol amendments, eliminating repetitive reviews of the entire protocol by ethics boards and significantly speeding implementation. The infrastructure for the study is built once, and used optimally to facilitate rapid iteration of agents, learning about biomarkers, and collaboration rather than competition. Credit and engagement has nothing to do with writing the protocol, but “chaperoning” an agent through the trial does. The team-like environment this creates enables everyone to work together in more optimal ways toward the goal of rapidly finding better treatment options.

Another key element of I-SPY2 is the use of a precompetitive model, where all stakeholders are treated as equals and no one stakeholder has any competitive advantage over any other. This level playing field is essential for trial conduct and objectivity, and also significantly improves efficiency. To keep all partners equal, I-SPY2 worked with stakeholders to develop a universal standard contract. Potential new partners are informed very early in discussions that the terms of these contracts are non-negotiable. This approach reduces friction in the system and enormous wasted time in starting up evaluation of each agent. The Foundation for the National Institutes of Health, Biomarker Consortiums, initially served as the broker among the academic institutions, companies, and study sponsor. This role was transferred to Quantum Leap Healthcare Collaborative where the trial has continued to flourish, iterate, and innovate.

By eliminating the need for long, drawn-out, back and forth negotiations, I-SPY2 expends very little time and few human resources on contracting. Because of these efficiencies in contracting and protocol amendments, it generally takes less than six months from initial contact with a stakeholder to opening a new arm with high accruing sites ready to enroll.

I-SPY2 can evaluate up to five experimental agents or combinations of agents simultaneously, along with a common control arm. Patient randomization to the arms is based on baseline assessments of their breast cancer molecular subtypes. The adaptive randomization algorithm uses this subtype to preferentially assign patients to an arm (or agent) that has been the most successful against that particular subtype. The trial enriches the specific arm with the patient molecular subtype that responds, resulting in fewer patients needed for evaluation.

Neoadjuvant versus Standard Adjuvant Therapy

A critical feature of I-SPY2 is that it uses neoadjuvant therapy, where systemic chemotherapy is administered prior to surgery; rather than standard adjuvant therapy, where surgery occurs first. The critical advance of the neoadjuvant model is that it allows us to assess tumor response to the systemic treatment at the time of surgery. I-SPY2 goes one further by employing serial MRIs during treatment to longitudinally assess tumor volumes, which then inform the adaptive randomization algorithm. There is really no reason not to gather critical information about response during neoadjuvant care. Failure to do so – adjuvant therapy after the tumor has been removed – leads to very long knowledge turns, the inability to understand how specific treatments benefit various types of tumors, and robs us of opportunities to continually refine our treatments (Figure 1). Adjuvant therapy following surgical resection is the paradigm we’ve been using for decades. If systemic therapy is used, it should be given in a way to maximally inform us. Failure to do this leads to imprecise treatment standards that are difficult to change.

Although pCR has yet to be validated as a surrogate endpoint, it is accepted by the FDA for accelerated approval in the neoadjuvant setting. Based upon a meta-analysis of a dozen neoadjuvant trials, FDA concluded that it was “reasonably likely to predict a clinical outcome”; however, questions remain about the strength of the association between pCR and long-term outcome.

Using pCR as an early endpoint is a central imperative of the I-SPY design, and critical to reducing agent development times. The relative time savings are clearly evidenced in comparing the use of pCR, where the endpoint occurs only around six months following study entry, with traditional outcomes like three- and five-year event-free survival.

We recently submitted a manuscript that analyses the relationship between pCR and three-year event-free and distant relapse-free survival of patients enrolled in I-SPY2 to date. The results are striking: A pCR confers a hazard ratio of 0.2 for both long term outcomes, and is consistent across multiple molecular subtypes despite patients receiving eleven different treatments.

Many refinements in the value of pCR as a surrogate endpoint made by I-SPY2 are likely due to the highly standardized environment created across I-SPY’s network of clinical sites. MRIs, for instance, must be reported in real-time to inform the adaptive randomization engine. A quality assurance program is therefore in place across I-SPY sites, in which sites must demonstrate consistency with centrally analyzed reference images. To date, over 6,000 MRI exams have been performed in I-SPY2, with more than 98% meeting our strict quality standards. A similar approach has been implemented for assessing the primary endpoint, pCR. Like MRI measures, pathology assessments are performed according to highly standardized protocols, and accompanied by extensive site training, qualification testing, and an ongoing quality assurance program.

More recently, we (and others) have seen enormous success with newer immunotherapies, which appear to play an important role in the future of oncology. The success of these agents sets a high bar for the evaluation of new agents if they are to show improvements in overall efficacy. As a result, I-SPY has begun to evolve.

Now that we have established that pCR is associated with excellent outcomes (event-free survival ≥94%), we can use this endpoint for targeted escalation or de-escalation of therapy, which also allows us to shift our focus to getting all patients to pCR, and to do so with less toxicity. New agents are emerging as more effective (or, at least as effective) as chemotherapy combinations with less toxicity. I-SPY will begin to focus on reduction of toxicity as a measurable endpoint. In our first steps, our latest protocol amendment includes the option for patients who achieve a pCR early during their therapy, as measured by MRI, to skip the standard course of anthracycline and cyclophosphamide chemotherapy combinations and proceed directly to surgery, sparing them the often-intense toxicities of these drugs. Longer-term follow-up will allow us to determine if pCR, however you get there, is indeed associated with an excellent outcome. Again, the framework of the trial allows for rapid iteration and innovation.

Through a project with the Imaging Program of the National Cancer Institute, we are currently working with multiple stakeholders to adapt our design to permit patients who fail to show an early response to a given treatment to switch to a different, biologically targeted (and optimized) agent without having to leave the trial. The “treatment switching” feature of this protocol has several aims. First, as standard of care continues to improve, it will provide a reason for patients to consider participating in a trial, as they will have assurances of receiving the best possible chance at a good outcome. Second, it will bring the I-SPY2 workflows into greater harmony with typical clinical care workflows. Finally, it will allow us to continue to evaluate new experimental agents as both first-line and second-line therapies, and provide insight into the impact of serial treatments and the development of resistance within tumors.

When I-SPY was still in the design phase in 2008, there was scant experience with models other than the standard “agent versus control” randomized clinical trial. Although there was vigorous discussion of the potential of using adaptive approaches and Bayesian statistical models in oncology trials, there were limited examples from which to draw. Similarly, the use of a master protocol was largely untested, having been employed only in one or two cases and in limited context.

“A Grand Experiment”

In many ways, I-SPY2 itself has been a grand experiment. The question we set out to answer was: “Are there more efficient and timely ways of getting drugs through phase 2 clinical development?” Today, after eight years of operation, our experiment has yielded an unequivocal YES and we find ourselves asking new questions and designing our next grand experiment. I-SPY has served as a crucible for innovation. Patients will benefit from new agents and combinations that are more specific for their tumor types. Physicians will have the chance to identify the right solutions for their patients before they develop metastatic disease, and have the chance to harness science with new care delivery models. Drug developers have the chance to rapidly test their agents and identify the tumor subtypes for which they work. Regulators will get better data, robust endpoints that can be validated, and efficiency. Investors will receive better return on their investment. Payers will have better outcomes from the resources they invest in healthcare. We will waste less time and energy on our path to better results for everyone. Going forward, we should all commit to working together on better models for driving change and improvement in the system, where everyone wins and resources are harnessed in more optimal ways.

Figure 1: Incorporating an early endpoint into the course of care not only informs the clinical and research community about which treatments are effective for which types of tumors/patients, but also takes years off the drug development cycle. As validation of the surrogate endpoint emerges, the timeline will accelerate even further, enabling use of the surrogate to optimize agents and combinations.