In 1998, the National Surgical Adjuvant Breast and Bowel Project (NSABP) began enrollment of the B-18 trial, which randomly assigned women to four cycles of doxorubicin and cyclophosphamide given either pre- or postoperatively.⁶ OS was not improved, but the effects of chemotherapy on the primary tumor were intriguing. pCR in the breast was seen in 13% of women treated preoperatively, and these women seemed to have improved outcomes (DFS hazard ratio [HR], 0.47; OS HR, 0.32). Subsequent studies have confirmed the correlation between pCR and favorable outcomes in many cancer types including breast,^7,8 rectal,⁹ esophageal,¹⁰ and bladder cancers.¹¹

In light of the strong individual-level association between pCR and survival, many have inferred that improvement in the pCR rate could be considered a surrogate for improvement in survival at the trial level. Unfortunately, a substantial body of literature calls into question the ability of the pCR to fulfill the conditions required to establish a preliminary end point as a surrogate for the final end point.

pCR Rate as a Surrogate End Point

It is a maxim of statistics that correlation does not necessarily imply causation. Similarly, correlation does not necessarily imply surrogacy. In both causation and surrogacy, the relationship must be proven by additional conditions. Prentice¹² stated that the effect of a treatment on a surrogate end point should fully capture any effect on the final end point. A robust method to quantitatively establish the degree to which the effect on the proposed surrogate end point is reflected in changes in the final end point was proposed by Buyse et al.¹³ This method entailed a meta-analysis of multiple trials in which the effects of a treatment on the presumptive surrogate end point were compared with the effect of the treatment on the final end point through regression analysis. The coefficient of determination (R²) provides a quantitative estimate of how well changes in the presumptive surrogate end point predict changes in the final end point. For reference, an R² of 0 indicates no association and an R² of 1 is a perfect association. An R² of 0.75 is often considered to be an acceptable level to determine surrogacy.¹⁴

To evaluate the surrogacy of pCR, Cortazar et al,⁸ with the support of the FDA, performed a meta-analysis of individual patient data for almost 9,440 women with breast cancer from 10 randomized trials who were treated with neoadjuvant therapy followed by surgery. The authors found that the R² values for pCR as a surrogate for EFS and OS were extremely low at 0.03 (95% CI, 0.00 to 0.25) and 0.24 (95% CI, 0.00 to 0.70), respectively. Furthermore, the positive slope of the regression lines implies that treatments with a larger effect on pCR actually correlated with worse survival!

Berruti et al¹⁵ subsequently performed a meta-regression of 29 trials that included 14,641 women treated with neoadjuvant therapy before surgery. Again, minimal association between the effect of the treatment on pCR and the effect on EFS (R², 0.08; 95% CI, 0.00 to 0.47) or OS (R², 0.09; 95% CI, 0.01 to 0.41) was observed. The implications for approving therapies based on an improvement in pCR can be can be seen by looking at the individual studies in which the experimental therapy improved the pCR rate by at least a factor of two (odds ratio ≥ 2, the approximate effect seen in the NeoALTTO and NeoSphere trials) or improved survival by 20% (HR < 0.8, the improvement the ALTTO study was powered to detect) (Table 1). Of the 10 trials with improvement in pCR, four failed to show improvement in OS leading to a false-positive rate of 40%. In addition, three of nine experimental therapies that improved OS > 20% would have been falsely assumed to be ineffective for a false-negative rate of 33%.

Proposed Explanations for the Lack of Surrogacy

Some have argued that the pCR rate might be an effective surrogate in certain breast cancer subtypes, including HER2-positive and triple-negative breast cancers.^8,15,31The I-SPY 1 (Investigation of Serial Studies to Predict Your Therapeutic Response With Imaging and Molecular Analysis) trial found that the association between pCR and relapse-free survival was stronger when analyzing the subsets separately.³¹ This is likely because hormone receptor–positive patients are less likely to achieve pCR but have more favorable outcomes in general. It is true that the correlation between pCR and outcome is stronger within the HER2-positive and triple-negative subgroups. However, the Cortazar et al⁸ study specifically analyzed these subgroups and found that despite the strong correlation, there is no evidence of surrogacy within the HER2-positive or triple-negative subgroups.

Berry and Hudis³² proposed some statistical considerations that merit discussion. First, they state that the Cortazar et al⁸ study lacked the power to rule out the possibility of surrogacy. In fact, the upper 95% CIs of the R² for surrogacy of pCR on EFS in the Cortazar et al⁸ and Berruti et al¹⁵ studies were 0.25 and 0.47, respectively, far short of the 0.75 required to establish surrogacy. They also propose that treatments with a larger impact on pCR might translate to improvements in OS. This is possible. However, it is worth noting that the NeoALTTO study showed one of the largest absolute improvements in pCR, larger even than the NeoSphere trial. Nonetheless, no impact on survival was noted in a trial of more than 8,000 women.⁵

Lack of Surrogacy in Neoadjuvant Rectal Cancer Trials

The lack of surrogacy is seen in most settings in which the surrogacy of pCR is tested.^33–36 Bonnetain et al³³ analyzed pooled individual patient data from two randomized rectal cancer trials evaluating the effect of chemoradiation versus radiation alone before surgery on pCR and OS. pCR was increased from 3.7% to 11.2% (P < .001). However, there was no improvement in OS (5-year OS: 65.9% v66.3%). To test pCR as a surrogate for OS, patients were grouped into trial units and the effect on pCR was compared with the effect on OS in the trial units. For reference, the R² of progression-free survival as a surrogate for OS was 0.88 (95% CI, 0.77 to 1). In contrast, the R² of the pCR as a surrogate for OS was 0.11 (95% CI, 0.0 to 0.44) indicating poor surrogacy.

The Radiation Therapy Oncology Group (RTOG) 0247 (A Randomized, Phase 2 Study of Neoadjuvant Radiation Therapy Plus Concurrent Capecitabine and Irinotecan or Capecitabine and Oxaliplatin for Patients With Locally Advanced Rectal Cancer) study was a randomized phase II trial of preoperative radiation and capecitabine with irinotecan or oxaliplatin for patients with rectal cancer. Owing to a lower pCR rate, the irinotecan arm was rejected.³⁴ Although not powered to detect a statistical difference, further follow-up showed that the irinotecan arm actually had a numerically superior survival (4-year OS: 85% v 75%; P > .05).³⁵ Determining future studies on the basis of pCR would have led to the mistaken decision to reject the capecitabine/irinotecan regimen from further study, because longer follow-up showed this regimen was at least equivalent to capecitabine/oxaliplatin in terms of survival.

Paradox of the pCR

Patients who achieve a pCR have more favorable outcomes than those who do not; it would therefore seem logical that a therapy that increases the number of patients with a pCR should lead to more patients being cured of their disease. However, high-quality evidence suggests that this is not necessarily true. Although it is not entirely clear why improvements in pCR fail to translate into improvements in OS, we propose three clear limitations of the pCR that may lead to the poor surrogacy demonstrated in multiple studies.

First, the pCR measures the effect of a therapy only on the primary tumor. Cancer is not simply the primary tumor, but is also the potential micrometastatic systemic disease. The effect of a therapy on micrometastases outside the local tumor is likely to be the primary driver of improved outcomes. These residual clonogenic cells may be phenotypically or genotypically different from most cells in the primary tumor^37,38and may not respond to therapies that are effective against the primary tumor.³⁹

Second, effective therapies may not necessarily lead to pCR. The clearest example is the low rate of pCR after hormonal therapy despite the dramatic impact on survival. Breakthrough medications of the future may work by affecting circulating tumor cells, premetastatic niches, immune activation, or other mechanisms not measured by the effect on the primary tumor.

Third, the additional pCRs that are achieved from an investigational therapy may simply occur in patients who would have been cured by the standard therapy alone. An investigational therapy that increases the rate of pCR in the group of patients who are likely to have been cured with standard therapy alone will lead to a strong association between pCR and improved survival at the individual level. It may not, however, lead to improved cancer outcomes.

Pertuzumab and other therapies that improve the pCR rate may in fact become effective medications. However, a higher standard should be expected before exposing patients to the risk of severe or fatal toxicities^40–42 and increased costs⁴³from experimental therapies, or abandoning potentially effective therapies.^34,35Unfortunately, the allure of quicker results is driving a major shift in the design of clinical trials. Our analysis of the trials listed on ClinicalTrials.gov⁴⁴ found that pCR is the primary end point of approximately 50% of the currently enrolling phase II rectal cancer trials and 45% of the phase III breast cancer trials of preoperative therapy. Effectively, many ongoing clinical trials could be compromised by the choice to use the pCR as a primary end point despite strong evidence that it does not function as an effective surrogate for survival.

In summary, pCR is a favorable prognostic factor for individual patients undergoing treatment. Within the context of a clinical trial, clinicians may be able to reduce the intensity of subsequent therapy for patients who achieve a pCR. However, a substantial body of data suggests that improvements in the pCR from an experimental therapy cannot be considered a surrogate for improved EFS or OS at the trial level. Promoting therapies on the basis of an increase in the pCR rate may lead to approval of therapies that increase toxicity without improving survival. Conversely, rejecting a therapy on the basis of a failure to improve the pCR may lead to the abandonment of therapies that could have resulted in favorable survival outcomes.