AJCC indicates American Joint Committee on Cancer; UICC, Union for International Cancer Control.
Size of the data marker is proportional to the precision of the hazard ratio estimate.
eTable. Comparison of baseline variables between postoperative observation (OB) and adjuvant chemotherapy (AC) groups in the original and matched datasets
Customize your JAMA Network experience by selecting one or more topics from the list below.
Polanco PM, Mokdad AA, Zhu H, Choti MA, Huerta S. Association of Adjuvant Chemotherapy With Overall Survival in Patients With Rectal Cancer and Pathologic Complete Response Following Neoadjuvant Chemotherapy and Resection. JAMA Oncol. 2018;4(7):938–943. doi:10.1001/jamaoncol.2018.0231
Does adjuvant chemotherapy provide a survival benefit in patients with rectal cancer with pathologic complete response following neoadjuvant chemoradiotherapy and resection?
In this propensity score matching analysis of a cohort from a large national data set, adjuvant chemotherapy was associated with significantly improved 5-year survival in patients with rectal cancer who achieved a complete pathologic response following neoadjuvant chemoradiotherapy and resection compared with those who did not receive adjuvant chemotherapy. Patients with clinical stage T3/T4 and node-positive disease benefited most from adjuvant chemotherapy.
Adjuvant chemotherapy was associated with improved overall survival in patients with a complete pathologic response after neoadjuvant chemoradiotherapy for locally advanced rectal cancer following surgery.
Adjuvant chemotherapy (AC) in patients with rectal cancer with pathologic complete response following neoadjuvant chemoradiotherapy (nCRT) and resection is recommended by treatment guidelines. However, its role in this setting is equivocal because data supporting benefits are lacking.
To compare the overall survival (OS) between AC and postoperative observation (OB) in patients with rectal cancer with pathologic complete response following nCRT and resection.
Design, Setting, and Participants
We identified a cohort of patients with rectal cancer and a complete pathological response (ypT0N0) after nCRT in the National Cancer Database between 2006 and 2012. Patients who received AC were compared with OB patients by propensity score matching. Overall survival was compared using the stratified log-rank test and stratified Cox regression model. The outcomes after AC vs OB were also evaluated in patient subgroups. The data analysis was completed in June 2017.
Adjuvant chemotherapy and OB.
Main Outcomes and Measures
We identified 2764 patients (mean [SD] age, 60.0 [12.3] years; 40% female) with clinical stage II or III resected adenocarcinoma of the rectum who had received nCRT and were complete responders (ypT0N0M0). Of this cohort, 741 patients in the AC group were matched by propensity score to 741 patients who underwent OB. The AC cohort had better OS compared with the OB cohort (hazard ratio, 0.50; 95% CI, 0.32-0.79). The 1-, 3-, and 5-year OS rates were 99.7%, 97.1%, and 94.7% for the AC group and 99.2%, 93.6%, and 88.4% for the OB group (P = .005). In subgroup analysis, patients with clinical stage T3/T4 and node-positive disease benefited most from AC (hazard ratio, 0.47; 95% CI, 0.25-0.91).
Conclusions and Relevance
Adjuvant chemotherapy was associated with improved OS in patients with pathologic complete response after nCRT for resected locally advanced rectal cancer. This study supports the use of AC in this setting where there is currently paucity of data.
Current recommendations for treatment of patients with locally advanced rectal cancer (stages II and III) include preoperative neoadjuvant chemoradiotherapy (nCRT) followed by surgical resection and adjuvant chemotherapy (AC), independent of pathological stage.1,2 However, up to 40% of patients with rectal cancer achieve a clinical complete response after nCRT, and approximately 15% to 27% of patients will have pathological complete response (pCR).3-6
Patients with pCR after resection have improved overall survival (OS) as well as local and distant recurrence-free survival in comparison with those with residual viable disease.7-9 Despite this favorable prognosis and with limited data supporting the use of AC in patients with pCR,10,11 most guidelines, including those of the National Comprehensive Cancer Network, recommend at least 4 months of postoperative fluorouracil-based systemic chemotherapy.1,12 Moreover, recent reports and clinical trials suggest no additional survival benefit with AC in patients with locally advanced rectal cancer receiving nCRT.13-17 All of these studies, however, had limitations: Some closed due to poor accrual, and others did not specifically focus on patients with pCR.
Given the potential toxicity, adverse effects, and associated cost of systemic AC, it is important to assess the actual benefit of AC in patients with pCR. Hence, this study aimed to examine the association of AC following pCR with OS in patients with rectal cancer in a large national cohort of patients.
We queried the National Cancer Database (NCDB) between 2006 and 2012. The NCDB is a national cancer registry that captures nearly 70% of incident cancer cases in the United States. Information on patients, tumors, treatments, and outcomes are collected from more than 1500 Commission-on-Cancer–accredited cancer programs.18 This study was approved by the institutional review board at the University of Texas Southwestern. Dataset analysis uses deidentified data, and therefore consents are not needed.
We identified patients who received a diagnosis of nonmetastatic clinical TNM stages T1 to 2N+ or T3 to 4N0/+ rectal adenocarcinoma (International Classification of Diseases for Oncology, Third Edition, topographical code C20 and morphological codes 814, 821, 822, 825, 826, 848, 849, and 857). Patients who achieved pCR (ypT0N0) following nCRT were included in this study. For this study, we were cautious not to use stage category TxNx as inclusion criteria, as previously reported by other authors, because this staging does not accurately represent patients with pCR.6 We excluded patients who died within 90 days following resection (n = 33 [1.2%]) as a means to account for postoperative complications. We also excluded patients without follow-up data (n = 15 [0.05%]).
We abstracted patient data including age, sex, race/ethnicity, median household income, insurance type, and Charlson comorbidity/Deyo score. Tumor information such as year of diagnosis, histologic subtype, and clinical and pathological International Cancer Control/American Joint Committee on Cancer TNM stages was abstracted and analyzed. We also abstracted data on the receipt of neoadjuvant (preoperative) or adjuvant (postoperative) chemotherapy and chemoradiotherapy including time of initiation of treatment since diagnosis. Operative details including type of resection, as well as postoperative information such as length of stay, 30-day unplanned rehospitalization, and 90-day postoperative mortality, were recorded. We also abstracted patient follow-up and vital status data.
After applying patient inclusion and exclusion criteria, we used multiple imputation chained equations19 to impute missing facility teaching status (n = 13 [0.05%]), race/ethnicity (n = 37 [1.4%]), income (n = 27 [1.0%]), insurance type (n = 22 [0.8%]), time from cancer diagnosis to resection (n = 33 [1.2%]), clinical T and N stages (n = 25 [0.9%]), length of stay (n = 284 [11%]), and 30-day rehospitalizations (n = 86 [3%]). We generated 20 imputed data sets.
We used matching to reduce imbalance in patient and treatment characteristics between patients receiving AC and patients undergoing postoperative observation. We matched patients exactly on clinical T and N stage groups and discrete periods from the time of diagnosis to the time of resection. We matched on the former because preoperative stage is an important predictor of long-term outcome, and on the latter so as not to bias survival time between the 2 groups due to variable time spent in preoperative therapy. Next, we matched AC and observation by propensity score within strata of clinical stage and time to resection. Propensity score, or the probability of receiving AC given pretreatment information (here pretreatment implies variables available prior to initiating AC or observation), was estimated within each imputed data set, using a multiple logistic regression model. Propensity scores were then combined across all 20 imputed data sets using the rules by Rubin.20(pp75-79) Specifically, each patient’s propensity scores were averaged across all imputed data sets, and the resultant mean propensity score was used for matching. Preoperative information included in estimating the propensity score were year of cancer diagnosis, facility teaching status, age, sex, race/ethnicity, insurance type, income, comorbidity score, clinical stage, time to resection, type of resection, length of stay, and unplanned 30-day rehospitalizations. We used the nearest-neighborhood method within a caliper (0.25 of the standard deviation of the propensity score) and without replacement to perform a 1:1 match between AC and observation.21 We evaluated balance in the covariates after matching using standardized differences.22
We estimated OS using the Kaplan-Meier method. We compared OS between matched AC and observation groups using a stratified log-rank test. The association between OS and postoperative treatment approach, AC vs observation, was evaluated using a Cox proportional hazards model stratified on matched patient pairs. We explored for heterogeneity in treatment effects within clinical T and N tumor stage group using tests for interaction.23 A Cox regression analysis was also performed to estimate the treatment effect of AC vs observation in each patient subgroup.
In our primary analysis, we used the mean of propensity scores estimated in each imputed data set to match between AC and observation. In a sensitivity analysis, we performed matching by propensity score and estimated the difference in OS between AC and observation, in a similar manner to that described herein, within each imputed data set. Then, using the rules by Rubin,20 we calculated the combined point estimate of the treatment effect and its variance. This approach takes into account the variability between imputed data sets.24
We also conducted a sensitivity analysis using complete case analysis (as compared with multiple imputation in our primary analysis). In this analysis, we excluded missing data for variables available prior to initiating AC or observation (17% of data excluded). Matching patients in AC and observation groups was performed in a similar manner to our primary analysis: exact matching on stage and time to resection followed by matching by propensity score within stage and time to resection strata. Balance in baseline covariates was evaluated using standardized differences. Complete case analysis did not appreciably change the magnitude, direction, or significance of the treatment effect (data not shown); therefore, we report results of the primary analysis only.
In this study, we used a significance level of .05 and 2-sided hypothesis tests. Apart from propensity score matching, which was implemented in R, version 3.3.2 (R Foundation), using the package Matching,25 all analyses were conducted using STATA, version 14.
Between 2006 and 2012, we identified 2764 patients with clinical stage II and III resected adenocarcinoma of the rectum who received nCRT and were ypT0N0M0 in the surgical specimen (Figure 1). Fifteen and 33 patients were excluded due to missing data and absent 90-day mortality, respectively. After exclusions, 1944 patients (72%) underwent observation postoperatively and 772 patients (28%) received AC. From the AC group, 741 patients (96%) were matched with 741 patients who underwent observation. The baseline characteristics of the observation and AC groups before and after propensity score matching are described in the eTable in the Supplement. Generally, an absolute standardized difference of 0.1 implies a negligible difference between treatment groups; the largest standardized difference was 0.06 in the matched data set.22
In the unmatched (original) patient population, AC patients were more likely to receive a diagnosis in the 2010 to 2012 period (522 [68%] vs 1109 [57%] for observation; P < .001), receive care at an academic institution (360 [47%] vs 820 [42%]; P = .04), be younger (P < .001), have private insurance (477 [62%] vs 988 [51%]; P < .001), a more advanced clinical stage (cT3-T4/N+, 345 [45%] vs 742 [39%]; P = .002), a shorter time from diagnosis to resection (P < . 001), and a shorter length of stay (<1 week, 420 [62%] vs 982 [56%]; P = .01).
With a median follow-up of 39 months (interquartile range, 26-55 months), the AC group was associated with better OS compared with the observation group (hazard ratio [HR], 0.50; 95% CI, 0.32-0.79; P = .005) (Figure 2). The 1-, 3-, and 5-year OS rates were 99.7%, 97.1%, and 94.7% for the AC group and 99.2%, 93.6%, and 88.4% for the observation group (P = .005). Our sample size (1482 matched patients) had an 88.5% power to detect a 5% difference in OS at 5 years using a significance level of .05.24 We derived similar results, in direction and magnitude, from our sensitivity analysis in which the treatment effect was combined across imputed data sets (HR, 0.51; 95% CI, 0.42-0.62).
Subgroup analysis demonstrated that patients with clinical stage T3/T4 benefited from AC (HR, 0.48; 95% CI, 0.30-0.77) (Figure 3). Those with clinical T3/T4 and node-positive disease had the largest benefit with AC (HR, 0.47; 95% CI, 0.25-0.91); however, the benefit was not statistically significant among T/N stage subgroups (P = .77 for interaction).
Over the past few years, there has been a great deal of impetus investigating the benefit of adjuvant therapy in resected rectal cancer after nCRT. For instance, Sainato et al13 examined the role of AC (fluorouracil and leucovorin ×6 cycles) vs observation in a randomized clinical trial of 655 patients with locally advanced rectal cancer. This group reported a lack of benefit from AC in terms of OS, disease-free survival, and incidence of distant metastasis. However, the study was limited by low adherence in the AC group, as well as a lack of power to identify subgroups that would benefit from AC. Similarly, the European Organisation for Research and Treatment of Cancer (EORTC) 22921 trial was a large 4-arm randomized clinical trial that investigated the role of nCRT vs preoperative radiation alone and the value of AC vs observation in patients with resectable locally advanced rectal cancer.10,14 Although the initial 5-year results suggested a survival benefit in the AC group, the 10-year OS and disease-free survival were not significantly different when compared with the observation group. These studies questioned the role of AC in patients who received nCRT irrespective of their final pathological staging. Two subsequent randomized clinical trials were designed to assess the role of AC vs observation in patients who received nCRT. The Dutch PROCTOR-SCRIPT trial included 437 patients with locally advanced rectal cancer who received preoperative short-course chemoradiotherapy (25 Gy) followed by R0 resection.16 This study reported no difference in OS and a statistically nonsignificant benefit in disease-free survival and recurrence rate in the AC group compared with observation alone. This study closed prematurely due to poor accrual. The CHRONICLE study, another phase 3 trial comparing observation vs adjuvant capecitabine with oxaliplatin in patients who received preoperative fluorouracil with radiation therapy (45 Gy), also closed due to poor accrual.15 The authors did find improvement in disease-free survival with adjuvant capecitabine with oxaliplatin, but the result was not statistically significant. This was an expected finding given the small number of patients and consequent low power.
Maas et al17 conducted a pooled analysis of 13 different studies that evaluated whether the effect of AC was changed based on the ypT stage in patients with rectal cancer. Maas et al17 showed that patients with residual tumors benefited from AC and that there was no benefit for patients who achieved a pCR. The largest benefit was for patients who had ypT1-2 tumors.
Beyond these clinical trials that include all stages of rectal cancer after nCRT, only 2 single-institution studies attempted to assess the benefit of AC specifically in patients with pCR. García-Albéniz et al26 reported a small series of 26 patients with ypT0N0M0 disease who underwent observation after resection with a 5-year disease-free survival and OS of 96% and 100%, respectively. In a retrospective cohort of pCR patients, Gamaleldin et al27 compared pCR patients: 83 patients who received AC with 46 who underwent observation. The authors did not find a difference in oncologic outcomes among groups.
It is clear that patients who achieve a pCR have better oncologic outcomes overall. The question still remains as to whether the addition of postoperative AC provides a benefit in this cohort of patients. Recognizing that there are limited data supporting the use of AC in patients with pCR after nCRT, and faced with the unrealistic possibility of completing a randomized clinical trial given the failure of previous randomized studies, we sought to answer this question using a propensity-matched analysis of a large national cancer data set. Contrary to previously expressed expectations,28,29 the present analysis showed that the AC group was associated with better OS compared with the observation group (HR, 0.50; 95% CI, 0.32-0.79) ( Figure 2), a 6.3% improvement in 5-year survival. Our analysis included more than 700 patients in each arm, which to our knowledge is the largest matched cohort of patients with pCR reported in the literature. As expected, patients in both arms had a favorable prognosis; nonetheless, the AC group had better 5-year OS compared with the observation group (94.7% vs 88.4%). This is the first study that has reported a significant benefit for AC in patients with pCR. We believe the reason for these results compared with previous reports lies in the larger sample size of our study, which has adequate power to detect a statistically significant difference at 5 years. The study by Maas et al17 compared 290 patients who achieved a pCR after nCRT and received AC with 608 patients who did not. They found no statistical significance but did note that their sample was not sufficiently powered to detect such a difference.17 This is similar to all previously reported trials (closed due to poor accrual) that did not include sufficient patients to detect possible differences.13,15,16
Our study also shows in subgroup analysis that patients with clinical stage T3/T4 with node-positive disease had a larger benefit from AC (HR, 0.47; 95% CI, 0.25-0.91). This seems intuitive given that nodal disease is a major marker for systemic recurrence in colorectal cancer.30-32 These data are different from data presented by Maas et al,17 which showed that patients with ypT1-2 tumors had better outcomes after AC compared with ypT3-4 and node-positive tumors. Our specific group analysis was based on clinical stage vs ypT-stage as analyzed by Maas et al.17
In agreement with other studies, we found a low rate of adherence to National Comprehensive Cancer Network guidelines for the management of rectal cancer pertaining to AC. The magnitude of low adherence, however, was much greater than previously reported. During the study period, only 28% of patients received the AC as recommended by such guidelines. One reason may be that our data do not reflect patients who were offered systemic chemotherapy and declined, or completed less than 2 cycles and stopped due to toxicity. Poor adherence to guideline recommendations has been described in an analysis of Surveillance, Epidemiology, and End Results Program data from Medicare by Haynes et al,33 who reported that only 61.5% of all patients undergoing nCRT received adjuvant fluorouracil after resection. The adherence was even lower in patients with pathological stage I, reaching only 48.3%. No specific rate for AC in patients with ypT0N0 disease was described for this subset, but it would be expected to be even lower. Our data likely reflect current practice in the United States, which includes a trend of omitting AC in patients receiving nCRT, especially in those patients with pCR.
Although our study findings reflect a statistically significant reduction of death at 5 years for patients who received AC, we recognize that this survival benefit, 6.3% at 5 years, may be considered marginal by some clinicians given the good overall long-term prognosis for patients with pCR. Yet, differences even smaller than the ones observed here are often considered advantageous in various other malignant neoplasms when considering adjuvant therapy, including breast and colon cancer. The ultimate decision to offer and deliver AC to patients with pCR should be based on a thorough discussion with individual patients considering the preoperative clinical staging, overall health status, potential adverse effects/toxicities, expectations, and actual survival benefit. However, our data should serve to better inform patients. Furthermore, as new chemotherapeutic agents continue to emerge for the management of rectal cancer in the neoadjuvant or adjuvant setting, this difference in outcomes between AC and observation might also improve.
The present study has a number of limitations that emanate from the retrospective nature of data collection. Although the patients were matched based on perioperative factors in an attempt to minimize bias, unknown confounders not captured in the data set might produce residual bias in the results. Patients who develop postoperative complications or who do not survive the postoperative period are less likely to receive adjuvant therapy, leading to what is known as immortal time bias. The ideal scenario would be an intent-to-treat comparison between observation and adjuvant therapy. Unfortunately, it was not possible to reliably determine patients who were selected for adjuvant treatment from the NCDB or who developed postoperative complications preventing the receipt of adjuvant therapy. Therefore, we included variables such as length of stay and rehospitalization in the propensity score as surrogates for postoperative complications. In addition, we excluded all patients who died within 90 days from the operation to minimize the immortal time bias. Our study reports survival benefit of AC at 5 years, but the longer-term survival benefit is still uncertain and may vary as previously reported in the cohort of the EORTC 22921 trial. Finally, the NCDB does not show the type of chemotherapy used or numbers of treatments provided such that no specific recommendations could be made in this particular area.
This study demonstrates that systemic AC is associated with improved OS in patients with resected locally advanced rectal cancer with pCR after nCRT.
Accepted for Publication: January 22, 2018.
Corresponding Author: Patricio M. Polanco, MD, Division of Surgical Oncology, Department of Surgery, University of Texas Southwestern Medical Center, 5323 Harry Hines Blvd, Dallas, TX 75390 (email@example.com).
Published Online: April 19, 2018. doi:10.1001/jamaoncol.2018.0231
Author Contributions: Drs Polanco and Mokdad had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Polanco and Mokdad contributed equally as first authors.
Study concept and design: Polanco, Mokdad, Huerta.
Acquisition, analysis, or interpretation of data: Polanco, Mokdad, Zhu, Choti.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important intellectual content: Polanco, Mokdad, Zhu.
Statistical analysis: Polanco, Mokdad, Zhu.
Obtained funding: Polanco.
Administrative, technical, or material support: Polanco, Mokdad.
Study supervision: Polanco, Choti, Huerta.
Conflict of Interest Disclosures: None reported.
Funding/Support: Dr Polanco’s effort for this research project was supported in part by the VA North Texas New Investigator Program, Department of Veterans Affairs.
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Meeting Presentation: This study was presented at the 2017 Gastrointestinal Cancer Symposium of the American Society of Clinical Oncology; January 19-21, 2017; San Francisco, California.
Additional Contributions: David Primm, MA, University of Texas Southwestern Medical Center, helped edit the manuscript. No compensation was received for this contribution.