Adjuvant chemotherapy in rectal cancer patients who achieved a pathological complete response after preoperative chemoradiotherapy: a systematic review and meta-analysis

This study evaluated the prognostic impact of ACT in patients who achieved a pathological complete response (pCR). Articles published from January 1990 to September 2018 were searched in EMBASE, PubMed, Ovid, Web of Science, and Cochrane Library. Hazard ratios (HRs) and 95% confidence intervals (CIs) of overall survival (OS) were extracted. Thirteen observational studies were included. There were four National Cancer Database studies with overlapping study periods, thus individual pooled analyses of four different datasets were conducted (n = 3,182, 3,330, 3,575, and 4,739 for pooled analysis sets including Dossa et al., Polanco et al., Xu et al., and Shahab et al., respectively). Although a trend toward improved OS with ACT was observed, statistical significance was not proven (P = 0.09, P = 0.03, P = 0.12, and P = 0.10, respectively). When we performed a stratified analysis comparing the results from single institution and multicenter studies, there was no significant prognostic benefit of ACT. Publication bias was not observed. Routine use of ACT in patients with a pCR could not be warranted from the present meta-analysis. Further study of individual patient data from randomized trials is needed to clarify the role of ACT.

prospective study designed to assess the survival benefits of ACT in these individuals has been conducted. Here, we performed a systematic review and meta-analysis to evaluate the impact of ACT on the survival of rectal cancer patients who achieved ypT0N0 status. The OS outcomes of two groups, with and without the postoperative use of chemotherapy, were compared. Given the paucity of randomized trials appertaining to pCR status, the present pooled analysis provides clinical insights into the role of ACT in patients who achieved remarkable tumor eradication following CRT.

Methods and Materials
Literature search strategy. A systematic search of electronic databases was conducted to identify studies that analyzed OS in locally advanced rectal cancer patients treated with ACT. The search process followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. The outcomes of the present meta-analysis were reported according to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) criteria (Supplementary Materials Appendix 1) 16 . Published articles that compared ACT and non-ACT groups of rectal cancer patients treated with neoadjuvant CRT plus surgery from January 1990 to September 2018 were identified by searching the EMBASE, PubMed, Ovid, Web of Science, and Cochrane Library databases. Hand searches were also performed to identify other potentially eligible studies, but no additional studies were added. The following keywords, synonyms, and combinations thereof were used as search terms: [ study selection. Two independent authors (YJL and MK) performed the searches and assessed study eligibility. Studies comparing OS between postoperative ACT (intervention group) and observation alone (comparator group) in pCR patients were selected. Exclusion criteria applied during the selection process were as follows: (1) conference abstracts; (2) unstructured papers, such as editorials, comments, and letters; (3) case reports and review articles; (4) studies not reporting the survival outcome of pCR patients; (5) lack of information regarding a comparator group; and (6) insufficient OS data to extract hazard ratios (HRs) and 95% confidence intervals (CIs). The name of the institution or database included in the final set of eligible studies was reviewed. When multiple studies were based on the same data, the one with a longest-duration study period and the largest number of patients was selected. The study selection process was verified independently by a third investigator (YK). www.nature.com/scientificreports www.nature.com/scientificreports/ Data extraction. The data were extracted independently by two authors (YJL and MK). When discrepancies occurred, the authors discussed to reach a consensus. Some authors of potentially eligible studies were contacted via e-mail to request required data, and one study replied 17 . Risk of bias assessment. A risk of bias assessment was conducted independently by two authors (YJL and MK). Since all of the included studies were non-randomized observational studies, the Risk of Bias Assessment tool for Nonrandomized Studies (RoBANS) was used to assess the following six domains: the selection of participants; confounding variables; intervention measurement; blinding of the outcome assessment; incomplete outcome data; and selective outcome reporting 18 . Regarding potential discrepancies between the two authors, a consensus was obtained after further review and discussion. In addition, a third investigator (YK) verified the results. statistical analysis. The primary outcome of interest was OS. The HRs and 95% CIs from each study were either extracted directly from original papers or calculated using Kaplan-Meier OS curves based on the method of Tierney et al. 19 . HRs were calculated using a random-effects model with the inverse variance method. Cochrane Q tests and the I 2 index were used to evaluate heterogeneity. Funnel plots with Egger's regression tests were used to examine publication bias. An additional stratified analysis comparing results from single institution and multicenter studies was performed. RevMan software (ver. 5.3; Cochrane Collaboration, Copenhagen, Denmark) was used for all pooled analyses.  Table 1. Characteristics of 13 studies for the present pooled analysis. a Each of these studies of NCDB was included in the different sets of pooled analyses respectively. b Twenty-two (6 and 16 with and without ACT, respectively) death events were reported. c Six (3 and 3 with and without ACT, respectively) death events were reported. www.nature.com/scientificreports www.nature.com/scientificreports/ Results study characteristics. A flowchart of the study selection process is shown in Fig. 1. Among the 3,021 studies identified initially, the titles and abstracts of 1,838 studies were screened. Irrelevant articles, in terms of their structure or content, were excluded, and 95 manuscripts were reviewed. Further screening then identified 16 studies with sufficient OS data on a potentially eligible pCR population. Of these, three studies that had duplicate data (i.e., from the same institution) but shorter follow-ups, were excluded. Since four different National Cancer Database (NCDB)-based studies overlapped in terms of study period, we included all studies and separately conducted pooled analyses on each of them. A total of 13 studies were eligible 17,20-31 . Table 1 lists the characteristics of included studies. We investigated five single-institution 17,20-23 and four multicenter studies [24][25][26][27] [28][29][30][31] . The time interval between CRT and surgery was typically 1-3 months, and the total radiation dose was 45-50 Gy. Although the details of the chemotherapy regimens used were not included in the NCDB-based studies, most other studies reported fluoropyrimidine-based regimens. Most patients underwent total mesorectal excision. The summarized results regarding risk of bias revealed that the overall quality of the included studies was relatively high ( Table 2).

Stratified analysis.
To compare the results from single institution and multicenter studies, a stratified analysis was performed. Using the pooled analysis set III, which included the study of Xu et al., ACT did not result in significant survival benefits in either multicenter (HR 0.68, 95% CI 0.42-1.11, P = 0.12) or single-institution (HR 0.90, 95% CI 0.38-2.14, P = 0.81) studies (Fig. 3). Other subgroup results based on pooled analysis sets I, II, and IV also showed comparable results (data not shown). publication bias. There was no publication bias in the overall pooled results (P = 0.167, P = 0.210, P = 0.225, and P = 0.365 for pooled analysis sets I, II, III, and IV, respectively) ( Supplementary Fig. 1). No significant asymmetry was observed between the subgroup results of the multicenter and single-institution studies (data not shown).

Discussion
The present meta-analysis reviewed multicenter analyses and single institutional series to evaluate whether rectal cancer patients who achieved a pCR after preoperative CRT could benefit from ACT. Despite a trend toward better survival with ACT, statistical significance was not consistent in the different sets of overall pooled analyses. The stratified analysis comparing the results from single institution and multicenter studies failed to show a significant prognostic benefit. This study provides an updated perspective on the optimal postoperative strategy in cases with a marked treatment response after neoadjuvant CRT.
Based on ambiguous guidelines, patients with more advanced disease and a poorer response to CRT are potential candidates for ACT. A higher ypT and/or N stage are indicative of unfavorable tumor biology, which   13 . In contrast, ACT has often been recommended in down-staged patients after CRT. Based on the responsiveness www.nature.com/scientificreports www.nature.com/scientificreports/ to preoperative cytotoxic treatment, eradicating micrometastatic disease and further beneficial effects arising from ACT can be expected. A meta-analysis of relevant trials showed that the 5-year OS was significantly higher following ACT in patients with ypT0-2N0M0 (i.e., down-staged) (odds ratio 0.57, 95% CI 0.38-0.85) 10 32 .
A pCR, characterized as maximal down-staging that can be achieved by CRT, yields a high 5-year survival rate of ~85-90% [33][34][35] . The lack of prospective studies to evaluate the effect of ACT is because a pCR is only achieved in ~10-20% of all rectal cancer patients 14,15 . Additionally, treatment adherence to ACT in this subpopulation is mostly poor given the possibility of treatment-related toxicity, financial burden, and a patient or clinician's preference for less aggressive treatment 36 . In this clinical setting, a meta-analytic approach with a larger number of patients can help detect a small or absent treatment effect.
The results from this meta-analysis suggest there is no clear evidence to support survival benefits of ACT for pCR patients. The trend toward a favorable prognosis with ACT was mainly based on NCDB studies, and other exploratory analyses excluding the population-based data failed to obtain any reproducible results. Although a recent meta-analysis from a Chinese group suggested the potential survival benefit of ACT in pCR patients 37 , only one NCDB study of Polanco et al. 29 . was included without considering the other three NCDB studies 28,30,31 . Specifically, the propensity score matching method used by Polanco et al. may have produced additional selection bias due to limited patient information, such as underlying morbidity or immediate mortality, within a few months after completion of treatment. Chang also noted that the survival benefit of a certain treatment can be overestimated in the statistical setting of NCDB data 38 . Therefore, we believe that the trend toward better survival in the ACT group, with the underlying predominance of one NCDB-based study, should be interpreted with caution.
The study of Maas et al. analyzing individual-level data of pCR patients from 13 institutions could not prove any benefits of ACT in terms of survival or tumor recurrence (HR 0.94, 95% CI 0.53-1.69 for DFS) 27 . Multicenter studies from Taiwan, Italy/Spain, and Australia 24-26 failed to identify a survival difference, as did a recent multi-institutional study from Korea (n = 118 for pCR, from personal communication) 39 . The I-CNR-RT trial could not analyze outcomes of pCR patients due to the low incidence of mortality events 2 . To date, there is little randomized evidence to demonstrate the survival benefit of ACT in this patient population.
To assess the clinical value of ACT, potential adverse effects also need to be considered. Among the 13 studies, only two reported ACT-related complication events 20,26 . In the study of Tay et al., 13% of study population stopped ACT due to significant toxicities, such as diarrhea, nausea, and vomiting 26 . Time to closure of ileostomy was longer in patients with ACT, suggesting inferior quality of life and related medical morbidities 20 . Although life-threatening complication is not common (~1-2%), grade 3-4 acute toxicities were reported in more than one third of patients after the use of multi-agent ACT regimens 40,41 . Regarding the inconsistent survival outcomes in the present pooled analyses including four different NCDB-based studies, it is questionable whether the therapeutic benefit of ACT is sufficiently expected to risk treatment-related toxicities and quality-of-life problems in patients with a marked response to CRT.
This study had several limitations. First, all of the included studies were observational and inevitably suffer from confounders, such as selection bias and heterogeneity in sample characteristics and treatments 42 . Of the included studies, the method of Tierney et al. 19 to extract HR and 95% CI data was used in 5 studies 20,22,23,25,26 . The differences between the data indirectly extracted from survival curves and the original data can induce additional bias. Due to insufficient data, the effect of ACT on DFS or treatment-related toxicities was not evaluated in the present meta-analysis. Most of the studies included herein did not account for the presence of underlying comorbid illnesses or postoperative complications. Use of 5-fluorouracil, 5-fluorouracil/leucovorin (FL), and capecitabine chemotherapy was reported in most studies, but it remains unclear whether the current results are also applicable to other regimens, such as multi-agent folinic acid/fluorouracil/oxaliplatin (FOLFOX) and folinic acid/ www.nature.com/scientificreports www.nature.com/scientificreports/ fluorouracil/irinotecan (FOLFIRI). Nevertheless, this large-scale meta-analysis provides clinically useful insights into the prognostic role of ACT in this small-sized population who achieved a pCR.
This meta-analysis could not warrant the survival benefits of ACT in patients who achieved a pCR, suggesting that routine use of ACT should not be recommended in this subset of patients with rectal cancer. Further pooled analysis of individual patient data from existing randomized trials is needed to establish guidelines for ACT in conjunction with contemporary neoadjuvant treatments.