The I-SPY 1 population had clinical and biologic high-risk features: median age 49 years, 41% HR negative, 90% intermediate or high grade, and 91% high risk by the 70-gene profile.10a
For patients with biologically high-risk invasive breast cancers, 3-year RFS is better for those who experience a pCR after neoadjuvant chemotherapy than for those who do not.
The observation of Simpson's paradox based on the estimated hazard ratios in is associated with well-known characteristics of breast cancer. Of the three tumor subtypes in , the tumors that are least sensitive to either adjuvant or neoadjuvant chemotherapy are those that are HR positive/HER2 negative.5,19,20
However, these tumors have the best prognosis in the absence of chemotherapy. Despite the fact that these tumors have the lowest rate of pCR (9% v
36% and 41%), patients with these tumors tend to have a better RFS than patients with triple-negative or HER2-positive tumors, both overall and within pCR and no pCR categories. So although our relatively small study is not sufficient by itself to conclude that the population parameters have the same relationship as their estimates, this observation is consistent with what we understand about the biology of breast cancer and is to be expected. Other researchers should categorize tumors by molecular and receptor characteristics in relating pCR and longer-term end points such as RFS and overall survival.
These results provide additional insights into previous neoadjuvant trials that have examined the relationship between pCR and RFS.21
For example, although most neoadjuvant chemotherapy trials have shown that pCR is associated with favorable outcome, this was not seen in the National Surgical Adjuvant Breast and Bowel Project (NSABP) B-27 neoadjuvant trial. In that trial, although pCR improved significantly when paclitaxel was added to a doxorubicin-based chemotherapy regimen, the improvement was not of the same magnitude as that for RFS or overall survival.7
The possibility of a Simpson's paradox relationship suggests that analyzing only overall pCR and survival rates may have underestimated the true predictive effect of pCR in NSABP B-27.
The rate of pCR in the HR-positive/HER2-positive subset was lower than in the HR-negative/HER2-positive subset in the I-SPY TRIAL (in the absence of neoadjuvant trastuzumab). The sample sizes in our study were small, but this effect has also been observed consistently in three large randomized phase III neoadjuvant trials: NeoALTTO (Neoadjuvant Lapatinib or Trastuzumab Optimization Study),22
and NEOSPHERE (Neoadjuvant Study of Pertuzumab and Herceptin in an Early Regimen Evaluation)24
; the effect was less than that mentioned in a personal communication with G. von Minckwitz on August 19, 2011, and more than that mentioned in the CHER-LOB (Preoperative Chemotherapy Plus Trastuzumab, Lapatinib or Both in HER2-Positive Operable Breast Cancer) phase II trial.25
In the nonrandomized MD Anderson Cancer Center neoadjuvant series,26
pCR rates were also lower in the HR-positive/HER2-positive subset than in the HR-negative/HER2-positive subset, and the subsequent RFS rates were also lower in this group. Our trial results are limited by the small number of patients who received neoadjuvant trastuzumab. Ongoing and future trials will help us better understand whether different treatment approaches will be needed to improve outcome for HER2-positive disease on the basis of HR status.
High rates of proliferation, as measured by Ki67, increased the likelihood of pCR. However, Ki67 is highly correlated with receptor subsets and appears to improve the ability to predict response within all subsets except HER2 positive.
Clinical trials for breast cancer have historically contained a mix of receptor subsets. Given our findings and those of others that tumor biology is different for these subsets, comparisons should be anchored within molecular subsets rather than across whole trial populations. Sufficiently powered analyses of subsets on the basis of receptor status or molecular profiles (which are highly correlated with receptor status) should be a planned feature of future trials if we hope to extract the maximal value from them.10a
Molecular profiles at baseline may provide the opportunity to identify, beyond HR positivity alone, patients likely to have a good survival outcome as shown by I-SPY 1 molecular analyses.10a
Getting a drug approved for marketing is estimated to take more than 10 years and $1 billion.27,28
To shorten the period of time for drug development (what has been referred to as knowledge turns29
), a new approach is needed. The neoadjuvant approach gives us the opportunity to use response to therapy as an early evaluation end point. Fortunately, whether chemotherapy is given before surgery or after does not influence RFS and overall survival,30
but giving chemotherapy before surgery offers an opportunity both to evaluate response to therapy and to alter subsequent treatment if the tumor is downstaged.9
More importantly, it informs us about the success of standard and experimental therapies. For example, in the MD Anderson Cancer Center neoadjuvant trial, in which HER2-positive patients were randomly assigned to receive paclitaxel followed by fluorouracil, epirubicin, and cyclophosphamide plus or minus trastuzumab, the significant increase in pCR in the trastuzumab arm31
26% for patients who did not receive trastuzumab) in just 34 patients foreshadowed the results of the adjuvant trastuzumab trials, which accrued more than 9,000 patients and were completed 1 year later.32,33
Several trials are now evaluating novel agents in the neoadjuvant setting, including poly (ADP-ribose) polymerase inhibitors, CALGB 40601, NEOSPHERE, and GeparQuinto. Our results support this approach and further emphasize that evaluation should be performed by receptor or molecular subset.
Despite optimal therapy, many patients still remain at substantial risk for disease progression. To make more rapid progress in finding agents that may reduce risk for these women, we have incorporated the observations from I-SPY 1 TRIAL into the design of the I-SPY 2 TRIAL. That trial uses pCR as an end point and focuses on women with biologically higher risk disease—in the context of either standard marker subsets or molecular subsets of breast cancer—and uses an adaptive design to screen phase II agents in combination with the taxane portion of chemotherapy.34
The goal is to test these novel agents for their ability to improve pCR rates for 10 subsets of breast cancer defined by both standard and molecular markers.35
By focusing on high-risk patients and analyzing data by subsets, the neoadjuvant setting can provide a unique opportunity to accelerate learning.
In an effort to develop a noninvasive measure of pCR, I-SPY 1 also included multiple imaging parameters12
and demonstrated that change in MR volume is the imaging marker that best predicts pCR, has the potential to serve as a noninvasive measure of response to therapy, and may further accelerate evaluation of novel agents in combination with standard treatments. In I-SPY 2, MR volume is being used to provide information about response to chemotherapy between regimens, information that cannot otherwise be obtained without surgical resection.
The conclusions of our study are limited by the short follow-up time. It is well known that the HR-positive patients remain at risk for recurrence for many years, and early recurrence data may not reflect the overall outcome. However, the Oxford Overview Analysis36
of early breast cancer trials strongly suggests that the benefit of chemotherapy is reflected by the distant disease-free survival at 5 years; that is, the survival curves diverge during the first 5 years, but after year 5, they are parallel, suggesting that the survival benefit from chemotherapy is likely to be manifest in the first 5 years. With the median follow-up for the I-SPY 1 TRIAL currently at 3.9 years, current rates of RFS are likely to reflect primarily the benefit from chemotherapy.
The outcome for patients in I-SPY 1 is also affected by hormonal therapy given adjuvantly, although molecular predictors of sensitivity to endocrine therapy have been described and will likely further refine our ability to focus on patients most at risk for disease progression at the time of diagnosis.37,38
In conclusion, the I-SPY 1 TRIAL shows that the ability of pCR to predict outcome is substantially improved when analyzed within tumor subsets. This predictive power is sufficiently strong that pCR can be used as an early indicator of RFS. Indeed, pCR is the primary end point of the next generation study, the I-SPY 2 TRIAL, which is designed to identify agents early in the drug development cycle that improve the rates of pCR and for which molecular subsets this is so.