The use of adjuvant chemotherapy is not associated with recurrence or cancer-specific death following curative resection for stage III rectal cancer: a competing risks analysis

The use of adjuvant chemotherapy (AC) following resection of stage III rectal cancer (RC) has been recommended based largely on randomised trial evidence of favourable outcomes from its use in colon cancer [1]. Nonetheless, the use of AC in stage III RC remains controversial [2, 3]. While a Cochrane meta-analysis demonstrated a significant benefit for AC both in terms of disease-free survival (DFS) and overall survival (OS) [4], a subsequent systematic review comparing AC versus observation following surgery alone in locally advanced RC [5] reported improved OS in the chemotherapy group in only two of six studies [6‐12]. Notably, both of these reviews were limited by their inability to analyse outcomes by tumour stage (i.e. stage II versus stage III tumours). Moreover, many of the included trials were conducted prior to total mesorectal excision (TME) being implemented in routine care, whilst only in a few trials was multimodal treatment, including neoadjuvant (chemo)radiotherapy, actually delivered. Based on this, it has been argued that the results from these systematic reviews based mainly from retrospective evidence [13] cannot apply to contemporary RC management as they include patients with a substantially higher risk of both local pelvic and systemic recurrence, thus potentially enhancing the likelihood of AC impacting favourably on survival [14].

Recently, the efficacy of AC in protecting against recurrence in stage III colon cancer has been questioned [15]. Suggested reasons why such treatment may no longer be beneficial include increasing adoption of a standardized technique for resecting colon cancer, improvements in the assessment of the extent of tumour burden present prior to surgery using modern imaging methods, and the introduction of high-quality multidisciplinary patient care. In routine clinical practice at Concord Hospital, we have previously shown no significant difference in the competing risks incidence of either recurrence or colon-cancer-specific death between stage III patients who received AC and those who did not [16]. Similarly, it is conceivable that the efficacy of AC in stage III RC may be also influenced by the same factors previously noted, particularly by the increasing adoption of TME [17], as none of the previously mentioned randomised trials in RC indicate the surgical technique employed [6‐12].

Previous studies on the topic have also been limited by their assessment of disease recurrence using so-called naïve Kaplan–Meier censoring. If recurrence and cancer-specific death are assessed by Kaplan–Meier analyses [18] in which patients who die of causes other than RC have their survival times censored, this violates the assumption that censored patients would continue to have the same risk of recurrence and cancer-specific death as if they had not died of the other cause [18‐20]. Obviously, patients who die of another cause cannot continue to be at risk of recurrence or death due to RC. The consequence of this violation is incorrect estimates of the incidence of recurrence and RC-specific death rates. Competing risks techniques will produce more accurate and precise estimates of recurrence and cancer-specific death rates and should be the modelling approach for such survival analyses.

The aim of this study was to compare the competing risks incidence of recurrence and RC-specific death between patients who did and did not receive AC and therefore clarify any additional (and potentially causal) beneficial association with the use of AC on long-term outcomes following resection. Our hypothesis was that oncological outcomes would be significantly better in those who received chemotherapy than in those who did not.

This study was based on analysis of a prospective database of consecutive patients with RC who were operated on by specialist colorectal surgeons, all members of the Concord Hospital Colorectal Surgical Unit [21, 22]. The data set included patient characteristics, comorbidity, mode of presentation, imaging, surgical management, postoperative complications, pathology and follow-up and has the approval of the Sydney Local Health District Ethics Committee (CH62/62011-136-P Chapuis HREC/11/CRGH206).

All resections were performed by mobilisation of the rectum along strict anatomical planes employing a standardized technique adopted in 1981 [23], equivalent to the technique of TME of the rectum as promoted by Heald et al. [17]. Resections performed between 1995 and 2019 inclusive were selected for analysis. Comorbidity and complications were as defined previously [24]. Examination of the resected specimen was performed or reviewed by one pathologist (CC) with a special interest in colorectal cancer using a standard protocol [25]. Only adenocarcinomas (including mucinous and signet ring carcinomas) of the rectum (defined as those tumours whose distal edge was within 15 cm from the anal verge) were included in the data set. All pathological characteristics analysed were looked for in every specimen and their presence or absence was recorded explicitly. Analysis was confined to rectal tumours with nodal metastasis, no identifiable systemic disease (pTNM stage III) and no histological confirmation of tumour transection present in the proximal, distal or deep lines of resection in the operative specimen. Such patients were defined as having had a potentially curative resection. There were no missing data on any variable except cause of death (five patients), cardiac morbidity (21 patients because scoring on the New York Heart Association classification requires that the patient be mobile and not demented) and an American Society of Anesthesiologists (ASA) score (one patient not recorded). The modal category was substituted in the missing ASA score variable, whilst missing cardiac morbidity data were excluded in relevant analyses; these had no material effect on the results. Patients were excluded if they had had a prior colorectal cancer resected, or if they had inflammatory bowel disease, familial adenomatous polyposis coli or if their cause of death was unknown.

Patients without distant metastasis but with clinically suspected locally advanced RC (typically with a threatened or involved circumferential resection margin, anterior tumour position or mesorectal or side wall nodal involvement) were identified by either computed tomography, magnetic resonance imaging or endorectal ultrasonography and were considered for preoperative radiotherapy with or without supplementary chemotherapy.

AC was considered routinely for all patients at a multidisciplinary team meeting, but prescribed selectively taking into account the patient’s age, comorbidity, the presence of adverse pathological features [26], postoperative complications, social circumstances and the patient’s wishes. The chemotherapy regimens utilised varied but were in accordance with best practice at the time and for the most part were bolus injections of 5-fluorouracil and folinic acid administered daily in 5-day blocks and repeated every month for six months (Mayo Clinic regimen [27]); 5-FU and leucovorin repeated weekly for six doses with a 2-week rest between (Roswell Park regimen [28]); semi-monthly 22-h 5-FU infusion with leucovorin (de Gramont [29]); or modified oxaliplatin, folinic acid and 5-FU every 2 weeks (FOLFOX [30]). As the database was originally oriented principally towards surgical treatment, no record was kept of reason/s why individual patients did or did not receive chemotherapy or of the chemotherapy agents or regimens used in individual patients or of dose alteration, complications, interruption or cessation of treatment.

From 1203 RC resections performed in the period 1995 to 2019 inclusive, successive exclusions were 835 tumours considered not stage III, six patients with previous colorectal cancer, seven who had developed a new primary, six with a synchronous cancer, six with inflammatory bowel disease or familial adenomatous polyposis coli and five patients whose cause of death was unknown. There were 338 patients remaining with a single RC suitable for analysis. The clinical characteristics of patients who had AC are compared with those who did not in Table 1. By the close of study in December 2020, 178 patients (52.7%) had died; the median time to death was 3.3 years (range 0.01–20.9). Of the 160 patients remaining alive or lost to follow-up, the median survival time was 11.1 years (range 0.4–22.5).

This observational study of prospectively recorded data from 25 years of routine hospital practice, analysed by competing risks regression, found that AC did not result in significantly diminished recurrence nor reduced RC-specific death among the 208 consecutive stage III patients who received chemotherapy. Fifty variables which potentially could have influenced the selection of patients for chemotherapy or influenced oncological outcomes were examined, yet none was found to have a confounding effect to explain the absence of any impact from chemotherapy on recurrence or cancer-specific death. These findings are not directly comparable with the existing literature because no published studies have used recurrence or rectal-cancer-specific death as primary endpoints and none have used competing risks methods of analysis [6‐12]. Rather, such studies used simply OS or DFS as endpoints. In our patients, OS was significantly longer in those who received chemotherapy, even after adjustment for relevant confounders. DFS was initially longer in those who received chemotherapy, but this difference attenuated to borderline statistical significance after adjustment for confounders.

In early cancer trials studying the impact of AC [32], recurrence was assessed by naïve Kaplan–Meier censoring [18] which treats competing events (such as non-cancer death) as censored observations. This results in inaccurate recurrence estimates, which is avoided by the use of the competing risks method [18‐20]. In the present study, the use of competing risks methods showed an absence of any association between AC and recurrence as the principal endpoint. After multiple regression adjustment for the possibility of a suppressor effect from any factor which was related to patient selection to receive chemotherapy, there was still no indication of a significant effect.

In this study, only OS showed a significant independent protective effect of chemotherapy, a finding which is consistent with a previous population-based longitudinal study [33]. This seems difficult to reconcile with our finding of no association between chemotherapy and recurrence. The explanation lies in appreciating that recurrence is a necessary condition for cancer-specific death in a patient following resection of a stage III cancer; if the patient has not developed a recurrence, they cannot die of RC. In OS, the ‘failure event’ is death from any cause, so if chemotherapy is not associated with recurrence, and as recurrence is a necessary condition for cancer death, then diminished OS cannot be a consequence of not having chemotherapy but must arise from some other cause. Similarly, for DFS (or recurrence-free survival) where the principal failure event is either cancer recurrence or death from any cause [34], if chemotherapy is not associated with recurrence, then diminished DFS cannot be a consequence of not having chemotherapy but rather must arise from some other cause. In contrast, the outcome in competing risks analysis of recurrence is a single event (recurrence) rather than a combination of events (recurrence or death due to any cause). Unsurprisingly, the use of OS or DFS as endpoints would yield misleading results when there is no association between chemotherapy and recurrence in stage III RC.

Clinical practice guidelines published by the American Society of Colon and Rectal Surgeons [35] and the Association of Coloproctology of Great Britain and Ireland [36] both advocate that AC should be considered for patients with stage III RC if systemic chemotherapy has not been given preoperatively. However, both guidelines recognise that their recommendations are based on extrapolation of data from adjuvant colon cancer therapy given the paucity of data specifically for RC, especially those managed contemporarily with TME surgery and neoadjuvant therapy. Such uncertainty has led to the current recommendation by the (Australian) National Health and Medical Research Council that the uncertain benefits of AC should be acknowledged, even in patients regarded as ‘high-risk’ [37]. Undoubtedly, it would be desirable that modern trials be repeated based on contemporary surgical and medical management, the results of which could ‘replace’ those of dated trials which remain controversial. However, because of the widespread acceptance of AC and for ethical reasons, it is unlikely that new randomised trials will be conducted. Instead, future investigations on this topic would have to utilise prospective observational designs as in this present study, coupled with analysis of oncological outcomes by competing risks techniques. Also, such studies have the advantage of being conducted in routine practice rather than in the rarefied environment of a randomised trial.

A limitation of this retrospective study was that no detailed record was kept for reason/s why individual patients did or did not receive chemotherapy, nor of the dose alterations, complications or interruption or cessation of treatment. It is also acknowledged that chemotherapy regimens have changed over time, and more recent addition of oxaliplatin as doublet therapy may have influenced results. However, as chemotherapy treatment used established agents and was administered according to conventional guidelines at the time, it is difficult to explain how its application in routine clinical practice should not show at least some tendency toward a protective effect if such an effect is true.

This study of prospectively recorded data on consecutive patients having a resection for stage III RC in a routine hospital setting found no significant difference in the prevalence of recurrence or cancer-specific death between patients who had received AC and those who had not. The failure to find a protective effect from AC may be attributable to standardized, anatomically based surgery with contemporary preoperative staging and multidisciplinary management. It is recommended that the endpoints of both cohort studies and randomised trials addressing this topic should consider both recurrence and cancer-specific death and that contemporary competing risks methods should be used in future analyses.