Abstract
Background/Aim: The relationships between local recurrence (LR), the development of distant metastases (DM) and prognosis in patients with rectal cancer remain unclear. Patients and Methods: In 606 patients who underwent curative resection, the role of LR was assessed retrospectively by time-dependent multivariate Cox models with inverse probability of treatment weighting taking into account competing risks. Results: Patients with LR had more DM than patients without LR (49/79, 62% vs. 86/524, 16.4%; p<0.001); 37% of LR-associated DM developed before or at LR, 63% after diagnosis of LR. Fifty-five percent of patients without DM at diagnosis of LR later developed DM. In these patients, the incidence of DM significantly exceeded the incidence in patients without LR. DM risk was most strongly associated with preceding LR and stage UICC III and II. Conclusion: There is a causal link between LR and DM in patients with rectal cancer.
Total mesorectal excision (TME) significantly reduces the incidence of local recurrence (LR) in patients with rectal cancer, and pre-operative or postoperative radiotherapy (RT) or chemoradiotherapy (CRT) can further reduce the risk of LR (1-3). However, evidence of the influence of LR reduction on distant metastases (DM) and overall survival (OS) remains inconclusive. Though some non-randomized studies have demonstrated a correlation between LR reduction and improved survival after TME (4-6), reduced LR did not change the incidence of DM (7-9) or improve survival in other studies (1-3).
The underlying question is whether rectal cancer is potentially already a systemic disease at diagnosis, with both LR and DM being independent consequences of advanced and aggressive cancer. Alternatively, LR itself may be the origin of DM. This view is consistent with the higher incidence of DM in patients with LR compared to patients without LR (6, 10, 11), the reduction of DM combined with LR in randomized studies in which RT reduced LR (12), and observations of an association between DM and LR in cancers in other organs (13-16).
The conflicting results may be due to methodological issues. LR and DM must be treated as time-dependent variables in survival analyses to avoid immortal time bias (17). In addition, LR and DM probably interact in their effect on cancer-specific survival (CSS) and OS. Studies describing only the incidence of DM with or without LR are unable to assess the causal role of LR in DM, as they fail to distinguish between LR followed by DM and DM occurring simultaneously with or prior to LR. If the time course of LR and DM is not taken into account, an elevated incidence of DM in patients with LR may partially be a result of immortal time bias. Patients with LR must have survived until LR diagnosis and, therefore, potentially had more time to develop DM than patients without LR.
Therefore, the long-term results of patients with rectal cancer with and without TME were analyzed in the present study to determine the impact of reduced LR after TME on the incidence of DM and OS. This study also shows how the analysis of patients with LR followed by DM compared to patients with DM before or at LR sheds light on whether LR is the source of systemic disease besides the primary cancer.
Patients and Methods
Patient data. A total of 1032 consecutive patients with histologically proven rectal adenocarcinoma underwent surgery at the Department of General and Abdominal Surgery at the University Medical Centre Mainz (Germany) from 1985 to 2007. All cancers with a distal margin ≤16 cm from the anal verge were classified as rectal according to UICC classification (18). Patient data were collected prospectively using a standardized form. Anterior resection (AR; >6 cm above the anal verge), low anterior resection (LAR; anastomosis ≤6 cm from the anal verge), abdominoperineal extirpation (APE), and Hartmann procedures were performed. After introduction of TME with preservation of the autonomic nerves (19) in 1996, neoadjuvant CRT was preferred for patients with an involved mesorectal fascia on pre-therapeutic pelvic magnetic resonance imaging (MRI).
Patients had regular follow-up visits according to a standardized program until the fifth postoperative year following hospital discharge. The number of visits (after 6, 12, 18, and 24 months and yearly thereafter) and imaging procedures did not change during the observation period, except computed tomography (CT) of the chest replaced thoracic X-ray in 1990. The patients' status was updated in 2011 and 2012, including vital status, presence/absence of disease, results of follow-up visits, date and treatment of tumor recurrence, and the date and cause of death if applicable, by contacting the patients and their families, treating physicians, and hospitals. Follow-up ended at date of death or on December 31, 2012, whichever occurred first.
Definitions. Local control was determined by using the time from surgery to confirmed LR, which was defined as clinical, radiological, or histological evidence of a recurrent tumor in the anastomosis, pelvis, or perineum, regardless of DM. DM were defined by radiological evidence of tumor spread, with or without LR. In patients with LR and DM, DM were classified as being present before or simultaneously with LR diagnosis, as opposed to following the diagnosis of LR. OS was defined as the time to death by any cause. CSS was defined as the time to death due to rectal cancer, i.e., tumor recurrence at time of death or death after surgery. Patients who died from other causes were censored at time of death.
Exclusion criteria. Patients with concomitant DM, non-curative resection (R1 and R2), or emergency surgery (n=265) were excluded from the study. As preoperative RT may underestimate tumor stage (20) and chemotherapy may influence the incidence of DM, patients that had undergone adjuvant chemotherapy (n=41), adjuvant CRT (n=79), and preoperative CRT (n=41) were investigated separately. To ensure a homogenous collective, 606 patients with R0 resection were enrolled in the analysis.
Statistical analysis. Univariate analysis of OS and CSS was carried out using Kaplan–Meier curves and log-rank tests for group differences. Multivariate Cox regression was used to assess OS and CSS with age at surgery, TME, UICC stage, and grading (G1/G2 vs. G3/G4) as baseline covariates. LR and DM were time-varying covariates. The interaction between LR and DM allowed for a differential effect of LR depending on DM status.
The theoretical association of LR with DM was assessed using multivariate Cox regression for the cause-specific hazard of DM, with patients censored at the time of death using age at surgery, TME, UICC stage, and grading (G1/G2 vs. G3/G4) as baseline covariates and LR as a time-varying covariate. Inverse probability of treatment weighting (IPTW) (21) in Cox regression accounted for possible confounding factors associated with TME. The propensity for TME was estimated using logistic regression with sex, ASA status, UICC stage, pT1/pT2 vs. pT3/pT4 stage, nodal status, and grading as covariates (doubly robust approach) (22). Cox regression used a robust sandwich estimator for the variance (23). To assess the proportional hazards assumption in Cox regression models, the correlation between scaled Schoenfeld residuals and log time was examined. The cumulative incidence of LR alone, DM, and death was estimated from a multi-state model taking into account competing risks (24). The model was implemented in the R environment for statistical computing version 3.5.2 (25) with package etm (empirical transition matrix) (26) which provides the Aalen-Johansen estimator.
p-Values were unadjusted for multiple comparisons and considered significant if <0.05. A baseline data analysis was carried out in SPSS Statistics (version 23.0, IBM, Ehningen, Germany). The survival analysis was carried out using the R with packages survival (27).
Ethics. All patients provided informed consent for data collection and evaluation. All procedures involving human participants were performed in accordance with the ethical standards of the institutional and/or national ethical committees, as well as the 1984 Helsinki declaration and its later amendments or comparable ethical standards.
Results
Patient characteristics are summarized in Table I. TME was performed in 263 (43.4%) patients, and 343 patients (56.6%) underwent curative resection without TME.
Local recurrence and distant metastases. LR with or without DM was observed in 79 patients (13.0%; Table II). Forty-two patients had symptoms of LR and 37 were asymptomatic. LR was diagnosed by rectoscopy (n=27), CT (n=46), or MRI (n=6). Patients who underwent TME had a significantly lower rate of LR (7.6%) than patients who did not undergo TME (17.2%, p<0.001, Table II). The median time to LR was 14 months (range=3-91 months) in patients without and 17 months (range=6-85 months) in patients with TME. The median survival after LR was 12 months and 18 months for patients who underwent and patients who did not undergo TME, respectively (p=0.529). In univariate analysis, tumor stage, tumor grading, TME, and APE were significantly associated with LR (Table II). In IPTW-weighted Cox regression, the most important independent prognostic factors for risk of LR were preceding DM, followed by tumor stage III and II, APE, and tumor grading, whereas TME was not a significant protective factor (Table III).
Patient characteristics with and without TME.
DM were observed in 135/603 patients (22.3%; Table IV). DM were diagnosed by CT in 85 patients, by abdominal sonography in 38 patients, by x-ray of the chest in 5 patients, and by other methods in 7 patients. Patients who underwent TME had fewer DM associated with LR (10/263, 3.8% vs. 39/340, 11.4%; p=0.0013). The incidence of DM without LR was nearly the same in patients with and without TME (13.6% vs. 14.7%; Table V). Median time to DM in patients who underwent TME and patients who did not undergo TME was 21 (range=2.50-85.9 months) and 21 (range=1.48-85.2 months) months, respectively. The median survival after DM was 16 months without TME and 27 months with TME (p=0.039). In univariate analysis, TME, tumor stage, and LR were significantly associated with DM. Multivariate analyses revealed only three independent risk factors for DM: preceding LR and tumor stage UICC III and II. In contrast, TME was not significantly associated with risk of DM (Table III).
Characteristics of patients of this study divided into subgroups with and without local recurrence. Univariate analyses.
Independent risk factors for local recurrence and distant metastases.
Incidence of DM in patients with and without LR. DM occurred more often in patients with LR (49/79, 62.0%) than in patients without LR (86/527, 16.3%; p<0.001; Figure 1). This was independent of gender, TME, tumor localization, type of surgery, grading, and tumor stage (Table VI).
One-third of DM associated with LR occurred before LR (n=7, 14%, median time to DM 14 months, range 6-31 months) or at the time of LR (n=11, 22%). In two-thirds (n=31, 63%), LR preceded DM (median interval 11.5 months, 2 patients <0.5 months, 29 patients, range=1.6-45.8 months). This distribution of LR-associated DM was similar in patients who underwent or did not undergo TME (Table V). In 31 patients without DM at diagnosis of LR, the cumulative DM risk for LR increased to 55% during the first 5 years of follow-up (Figure 2).
The patients with DM before or with LR and patients whose DM occurred after LR were compared to patients with DM but without LR (Table VII). DM occurred more frequently in patients with DM before or with LR than in patients without LR, but the differences were not significant. In contrast, the incidence of DM was significantly higher after diagnosis of LR than in patients without LR in nearly all tumor categories. The differences were most explicit in patients with early cancer. In patients without LR, the frequency of DM increased, with infiltration of the tumor and lymph node involvement. The incidence of DM was high in patients whose DM appeared after LR in all subgroups and did not exhibit increasing incidence with pT and pN category (Table VII).
The interval from tumor resection to DM differed within the groups: 19 months in patients without LR, 14.7 months in patients with DM before or at LR, and 35.3 months in patients with DM after LR (p=0.046).
Overall survival and cancer-specific survival. The 5- and 10-year OS was 67.3% (95%CI=63.6-71.3) and 47.8% (95%CI=43.5-52.4) for all patients, 61% and 43% for patients who did not undergo TME, and 75% and 56% for patients who underwent TME, respectively (p=0.001; Figure 3). In IPTW-weighted Cox regression, the strongest risk factor for OS was DM (HR=9.717, p<0.001), followed by LR (HR=7.292, p<0.001), UICC stage III (HR1.425, p=0.033), age at surgery (HR=1.067, p<0.001), and APE (HR=1.357, p=0.023). TME/PME was not significantly associated with OS. Figure 3 shows the strong correlation between LR, DM, and death when taking into account competing risks.
The 5- and 10-year CSS was 78.5% (95%CI=75.1-82.2) and 69.7% (95%CI=65.5-74.1) for all patients, 73% and 64% for patients who did not undergo TME, and 87% and 79% for patients who underwent TME, respectively (p<0.001). The independent risk factors for CSS were LR (HR=15.062, p<0.001), DM (HR=14.485, p<0.001), and UICC stage III (HR=1.766, p=0.019). TME/PME was a significant protective factor (HR=0.475, p=0.001).
Adjuvant therapy. The results for 161 patients who underwent adjuvant therapy are given in Table VIII and IX. The LR rates were not significantly different from patients who did not undergo adjuvant therapy, but the rate of DM was higher, probably due to the more advanced tumor stages in these patients. Patients with LR had significantly more DM than patients without LR (Table VIII and IX). In multivariate analyses including these patients, adjuvant treatment was not a significant independent risk factor for LR or DM, OS, or CSS.
Discussion
Of all DM observed after operative therapy for rectal cancer, approximately one-third were associated with LR. In one-third of cases, DM occurred before or simultaneously with LR. In patients without DM at diagnosis of LR, the incidence of later DM increased to 55% and significantly exceeded the incidence of DM in patients without LR. Other than tumor stage III and II, earlier LR was the most important independent risk factor for DM.
Distant metastases in patients with (n=79) and without local recurrence (n=527).
Time to distant metastases (months) following local recurrence in patients without distant metastases before and at the time of local recurrence (n=31).
Cumulative incidence of death, local recurrence (LR) without distant metastasis (DM), and DM estimated from a multi-state model taking into account competing risks.
Characteristics of patients of this study divided into subgroups with and without distant metastasis. Univariate analyses.
This study aimed to investigate the link between LR and DM in patients with rectal cancer in a large sample, with a long-term follow-up. In the group of patients without DM at time of LR, the influence of LR on the later development of DM was demonstrated by an increase in the risk of DM during the first 5 years after diagnosis of LR. The incidence of DM after LR exceeded the incidence of DM in patients without LR in nearly all tumor categories. The influence of increasing tumor infiltration on DM was not more detectable in patients with LR. In particular, patients with LR and low tumor stages developed the highest excess rate of DM, which corresponds to similar observations in patients with LR of other cancers (13). Consistent with this observation, multivariate Cox regression analysis with IPTW revealed only two independent risk factors for DM: earlier LR and tumor stage II and III. These data suggest that, in addition to the primary cancer, LR increases the risk of DM and can be the source of DM.
Incidence of distant metastases in patients with and without total mesorectal excision (TME) depending on time of local recurrence.
The interval from tumor resection to DM was longest in patients with DM after LR. In other studies, this was interpreted as evidence of a causal relationship between LR and DM (16), but this interpretation has to be qualified. By group definition alone, patients with DM after LR are guaranteed not to have DM until after LR develops, whereas no such lower bound on elapsed time exists for patients without initial LR.
The percentage of DM caused by LR was probably relatively small. Out of all DM, the percentage of DM combined with LR was 36%; in 63% of these cases, DM were diagnosed after LR and could have been caused by the LR. DM after LR accounted for 23% (31/135) of all DM. This proportion depends on the LR rate. The high reduction in absolute LR rates by TME in the present study (9.6%) explains the significant reduction in DM in these patients. On the other hand, the lower the reduction in LR, the lower the expected effect on DM and OS. This may explain the results of other studies that found no reduction in DM (7-9, 28) after reduction of LR. It may also be the reason for LR reduction having no impact on OS in randomized studies (3, 12, 29-32).
Distant metastasis in patients with and without local recurrence.
One-third of LR-associated DM occurred before or simultaneously with LR. The incidence of DM was not significantly higher than in patients without LR. This suggests that in these patients, the primary cancer as a source of DM outweighs the influence of LR. The strong association between DM and LR in these patients was shown in multivariate analyses where earlier DM was shown to be an independent risk factor for LR. Therefore, in patients with DM after resection of rectal cancer, diagnostic procedures should include a search for LR. Conversely, in the case of LR, the diagnostic evaluation should clarify systemic metastasis.
Even though the introduction of TME in this study was followed by a significant reduction in LR and DM and by an improvement in OS in univariate analyses, these results were not confirmed in multivariate analyses. DM and LR were more important for OS than the performance of TME/PME. This suggests that improvement of OS in patients with rectal cancer cannot be expected with a reduction in LR alone, and the reduction of DM of primary cancer is more important. A promising therapeutic option is systemic preoperative chemotherapy, which is under investigation in patients with rectal and colon cancer (33, 34).
The present study has several limitations. First, the study design was a retrospective single-center trial with a long recruitment time. During the observation period, the follow-up timing and basic diagnostic methods were not changed, with the exception of replacing chest x-ray with CT. However, the quality of the diagnostic methods has steadily improved. Thus, it is possible that more systemic tumor recurrences were detected in later years. The same percentage of DM without LR in patients who underwent and did not undergo TME speaks against such an effect. The results suggest a causal relationship between LR and DM, but cannot provide compelling evidence of this relationship. This would require a method for detecting systemic micrometastases at the time of diagnosis of primary cancer. In addition, in the beginning of TME surgery, quality was not routinely determined, so consistent data are not available on the influence of TME quality on LR and OS. The aim of this study was not to evaluate TME or follow-up after surgery, but to better characterize the relationship between LR, DM, and OS.
Distant metastasis in patients with and without local recurrence.
Local recurrence and distant metastases in patients after surgery and adjuvant therapy.
Distant metastasis in patients without and with LR after surgery and adjuvant therapy.
In conclusion, our results suggest two mechanisms responsible for the development of DM in patients with rectal cancer. In addition to systemic micrometastases originating from the primary cancer and existing at the time of diagnosis, residual tumor or implanted tumor cells at the site of resection are a possible source of DM. The percentage of DM caused by LR seems to be relatively low, explaining the missing effect of reduced LR on prognosis in some studies. On the other hand, these mechanisms may explain the high risk of DM and unfavorable prognosis of patients with LR of rectal cancer (35, 36).
Footnotes
Authors' Contributions
T.J. and A. H. participated in the design of the study, U.G., T.T. and A. H. provided clinical data, A.L., M.B. and D.W. performed statistical analyses and interpretation of data, A.L. and D.W. generated figures, T.J. and D.W. drafted the manuscript, which was critically revised by all Authors.
Conflicts of Interest
The Authors have no conflicts of interest to disclose. The Authors confirm that they were not supported by any grants, equipment, or organizations.
- Received May 10, 2019.
- Revision received May 22, 2019.
- Accepted May 23, 2019.
- Copyright© 2019, International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved
