Neoadjuvant Rectal (NAR) Score: a New Surrogate Endpoint in Rectal Cancer Clinical Trials

The idea of using a surrogate endpoint (one that occurs earlier or is more easily attainable) in lieu of a final (or true) endpoint is not a new concept in clinical trial design [10]. Guidelines, considerations, and limitations for establishing surrogate endpoints in clinical trials have been proposed by the National Institutes of Health [11]. A surrogate endpoint requires two-step validation through randomized controlled clinical trials to ensure that (1) it correlates with the true endpoint (aka individual level association) and (2) the effects of treatments that impact the surrogate and true endpoints correlate (aka trial level association) [12].

As previously stated, locoregional relapse was initially a primary form of treatment failure in this disease. However, with the introduction of TME and radiotherapy, local failure is far exceeded by systemic metastatic development. The latter significantly influences disease-free (DFS) and overall survival (OS). As such, these two benchmarks have been the primary endpoints of most major rectal cancer clinical trials in the past three decades [13, 14]. The prolongation of survival as a primary goal of new therapeutic interventions in rectal cancer is laudable, but complicated by several highly effective agents in the metastatic setting, improved supportive care, and opportunities for salvage surgical resection or ablation of oligometastatic disease. Thus, the duration of patient follow-up from completion of rectal cancer treatment intervention until death, as a primary endpoint, can take perhaps 3 to 5 years longer than the original conduct of the entire clinical trial itself. Although slightly shorter in time as an endpoint, the same is true for DFS. In addition to the two-step validation described above, an ideal surrogate endpoint would be achieved at or near the completion of the clinical trial intervention, be highly reproducible across different clinical trial study designs and interventions, and be accomplished without added complexity or cost to the patient or the clinical trial design while correlating with the true endpoints of DFS and OS (Table 1).

With the widespread use of TME, pathologic standardization has become increasingly critical for accurate assessment of nodal involvement, margin status, and pathologic staging. While sentinel nodes and clinical responses have proved largely to be poorly representative of systemic disease, histopathology assessments have become critically important [15, 16]. However, the histopathologic assessment of the tumor specimen (and thus any associated variable desired as a surrogate endpoint) is highly dependent upon the quality of the pathologist review [17]. For local risk of recurrence, it was verified that the circumferential radial margin status serves as a highly valuable surrogate endpoint, even in the era of TME [18, 19]. Attempts to identify a surrogate endpoint for DFS and OS have proved more challenging. However, the introduction of neoadjuvant CRT has offered the opportunity to assess the degree of in vivo treatment effect and downstaging as a potential surrogate for longer term outcomes. One popular endpoint, pathologic complete response (ypCR) represents the ultimate degree of tumor downstaging defined as no histopathologic visible residual tumor remaining after neoadjuvant therapy. This endpoint has been extensively studied in phase III randomized controlled trials [7••, 8, 9••]. However, consistent across these studies was the finding that despite increasing the ypCR, which did correlate with improved local control, OS was not significantly impacted. Pathologic complete response after neoadjuvant CRT is dependent upon the inherent chemo-radiosensitivity of the cancer, bulk of the original tumor, and interval after completion of CRT. While ypCR has been suggested to be a valuable trial surrogate, it has not been endorsed as a validated endpoint in part due to these limitations [20‐22].

Since ypCR represents a binary “all or nothing” histopathologic variable, a continuum of treatment regression has been proposed to represent treatment response as a surrogate for survival endpoints [23]. Tumor regression grade (TRG) requires standardization, as there are at least four different institutional or programmatic versions currently in use (Table 2) [24•, 25•, 26‐29]. All versions remain relatively subjective in pathologic scoring, usually requiring central pathologic review when incorporated into multicenter clinical trials [30‐33]. Early studies involving radiographic imaging modalities (such as MRI, contrast-enhanced cross-sectional imaging, and FDG-PET) demonstrate the potential to serve as surrogates for histopathology, with their own limitations on standardization, reproducibility, and generalizability [34‐39]. While each of these proposed endpoints fulfills some of the criteria for a surrogate endpoint (Table 1), challenges remain in validation and widespread reproducibility.

Valentini and colleagues developed a nomogram for predicting local recurrence, distant metastases, and OS for patients with locally advanced rectal cancer [40••]. The nomogram for OS takes into consideration the clinical tumor (cT) stage, pathologic tumor  (pT) stage, pathologic nodal (pN) stage, patient age, adjuvant chemotherapy administration, surgery type (APR vs. LAR), dose of radiotherapy, and gender [40••]. The OS nomogram had a very respectable c-index (0.70) supporting a strong correlation which was derived from external validation in five European rectal cancer clinical trials [7••, 8, 41‐43].

The NAR score was developed as a short-term clinical trial surrogate endpoint to take variables associated with treatment effect beyond ypCR into consideration yet simple enough to support a diversity of clinical trial designs. The NAR score is calculated based on data supported by the Valentini nomogram for OS, but only using the clinical T stage and pathologic T and N stages (Fig. 1). Of the eight variables used in the Valentini nomogram, only pN and pT are potentially influenced by neoadjuvant therapy. We include cT in the calculation of the NAR score based on our belief that the degree of tumor downstaging is more important than the absolute pT. The remaining factors from the Valentini nomogram (age, gender, type of surgery, radiation dose, and receipt of adjuvant therapy) would not be influenced by neoadjuvant therapy and thus cannot contribute to a useful surrogate endpoint for assessing neoadjuvant treatment. The NAR formula, importantly, serves as a pseudo-continuous variable with 24 possible discrete scores from 0 to 100 with higher scores representing a poorer prognosis. The formula also standardizes downstaging by incorporating the treatment effect on the T stage, which accounts for bulky or large tumors regressing, but not to the point of ypCR. The relative weights of 5 for pN and 3 for downstaging of T were suggested by the Valentini nomogram and reflect the relative importance of these variables. The constant 12 assures that all scores are positive inside the brackets. Squaring the numerator transforms the score to a more uniform measure per unit change. The scaling factor 9.61 in the denominator ensures that the final scores range from 0 to 100. The NAR score is meant to be used in clinical trials as a surrogate endpoint for survival. The score is designed to be particularly sensitive to changes in factors that are affected by neoadjuvant therapy. Changes in mean NAR scores between treatment arms as a result of intervention should translate to changes in OS. Importantly, the score uses both clinical and pathologic factors that are universally available in rectal cancer clinical trials, obviating the need for additional trial infrastructure, cost, time, or effort. The score was not designed or intended for individual patient use or to provide prognosis as part of clinical care. In that regard, the original Valentini nomogram is better suited for that purpose.

After establishing the NAR score calculation, it was validated using the NSABP R-04 clinical trial patient dataset [44••]. The NSABP R-04 study involved 1479 patients with stages II or III rectal cancer and randomized them to one of four neoadjuvant CRT arms testing different radiosensitizers including (1) continuous infusion 5-fluorouracil, (2) continuous infusion 5-fluorouracil with weekly oxaliplatin, (3) oral daily capecitabine, or (4) oral daily capecitabine plus weekly oxaliplatin. Outcomes were analyzed in a 2 × 2 factorial design to assess the relative differences between infusional 5-FU vs. capecitabine and oxaliplatin vs. no oxaliplatin [45]. Continuous NAR score was significantly associated with OS (HR/unit 1.04; 95 % CI 1.03–1.05; p < 0.0001) [44••]. NAR scores in the NSABP R-04 trial dataset were categorized as low (NAR <8), intermediate (NAR = 8–16), and high (NAR >16) based on tertiles of the observed scores. These categories were significantly associated with OS (p < 0.0001) with 5 year OS values of 92, 89, and 68 %, respectively (Fig. 2). OS was also predicted by ypCR in this analysis, but continuous NAR score had a stronger association as measured by Akaike’s information criterion (p < 0.0001). The NAR score was subsequently and independently further validated in an international clinical trial dataset providing further evidence of utility as a short-term surrogate [46].

The NAR score represents the next logical step in defining a short-term surrogate clinical trial endpoint in rectal cancer study design. It is undergoing trial level validation currently and otherwise meets all the requirements for a surrogate endpoint outlined in Table 1. The NAR score is now poised to serve as the primary endpoint of upcoming phase II studies being designed within NRG Oncology and the NCI National Clinical Trials Network. It has already been adopted as a secondary endpoint by several ongoing phase I and II studies testing novel radiosensitizers and other neoadjuvant interventions in rectal cancer, including the incorporation of induction chemotherapy or total neoadjuvant therapy. With the development of this surrogate endpoint and incorporation into rectal cancer clinical trials, we anticipate more rapid scientific progress to the benefit of our patients and their families.