The Editors have satisfied themselves that all authors have contributed significantly to this publication
Multicentre analysis of oncological and survival outcomes following anastomotic leakage after rectal cancer surgery†
Article first published online: 11 AUG 2009
Copyright © 2009 British Journal of Surgery Society Ltd. Published by John Wiley & Sons, Ltd.
British Journal of Surgery
Volume 96, Issue 9, pages 1066–1075, September 2009
How to Cite
den Dulk, M., Marijnen, C. A. M., Collette, L., Putter, H., Påhlman, L., Folkesson, J., Bosset, J.-F., Rödel, C., Bujko, K. and van de Velde, C. J. H. (2009), Multicentre analysis of oncological and survival outcomes following anastomotic leakage after rectal cancer surgery. Br J Surg, 96: 1066–1075. doi: 10.1002/bjs.6694
- Issue published online: 11 AUG 2009
- Article first published online: 11 AUG 2009
- Manuscript Accepted: 21 APR 2009
- Quality Assurance Fellowship of the European Society of Surgical Oncology
The association between diverting stomas and symptomatic anastomotic leakage after rectal cancer surgery was studied, as well as the impact of leakage on local recurrence, distant metastasis, and disease-free, overall and cancer-specific survival.
Data from the Swedish Rectal Cancer Trial, Dutch TME trial, CAO/ARO/AIO-94 trial, EORTC 22921 trial and Polish Rectal Cancer Trial were pooled (n = 5187). All eligible patients without distant metastases at the time of low anterior resection were selected (n = 2726); overall survival was studied in patients aged 75 years or less (n = 2480). Multivariable models were used to study the association between diverting stomas and anastomotic leakage, and between leakage and recurrence or survival.
Some 9·7 per cent of patients were diagnosed with a symptomatic anastomotic leak; diverting stomas were negatively associated with leakage (11·6 per cent without and 7·8 per cent with a stoma; P = 0·002). Anastomotic leakage was negatively associated with overall survival in the multivariable analysis (hazard ratio (HR) 1·29 (95 per cent confidence interval 1·02 to 1·63); P = 0·034), but not with cancer-specific survival (HR 1·12 (0·83 to 1·52); P = 0·466).
Diverting stomas were associated with less symptomatic anastomotic leakage. Oncological outcome was not significantly influenced by leakage, but overall survival was reduced. Copyright © 2009 British Journal of Surgery Society Ltd. Published by John Wiley & Sons, Ltd.
Surgery is the cornerstone in the treatment of rectal cancer. Widespread propagation of standardized total mesorectal excision (TME) has improved overall survival1, 2. However, TME may be associated with an increased risk of developing anastomotic leakage3 with attendant morbidity and mortality in the postoperative period4, 5. Leaks may be associated with decreased local control6–11 and survival7, 12, 13. Therefore, the rate of (symptomatic) anastomotic leakage has been considered as one of the quality indicators of surgical performance14.
Studies to identify risk factors for anastomotic problems and methods to reduce symptomatic leaks are clearly important15, 16. At the end of last century, two small randomized trials tested the hypothesis that a diverting stoma reduces the incidence of anastomotic leakage17, 18. Although both trials showed fewer anastomotic leaks with stoma use, the difference was not statistically significant. A larger randomized trial concluded that a diverting stoma significantly reduced the risk of symptomatic anastomotic leakage19.
In this study, the data from five large European randomized clinical trials were pooled to study the association between the creation of a diverting stoma and the rate of symptomatic leakage after a (low) anterior resection for rectal cancer. In addition, the impact of anastomotic leakage on the rate of local recurrence, distant metastasis, disease-free survival, overall survival and cancer-specific survival was investigated.
Patient and treatment variables of the following five trials were pooled: Swedish Rectal Cancer Trial20, Dutch TME trial21, German CAO/ARO/AIO-94 trial22, European Organization for Research and Treatment of Cancer (EORTC) 22 921 trial23 and the Polish Rectal Cancer Trial24. The period of inclusion, randomization arms and number of included patients are shown in Table1. From this pooled database of treatment variables, all eligible patients treated with a low anterior resection and without distant metastases at time of surgery were selected. In the Swedish Rectal Cancer Trial no data on stomas were available, although stomas in that trial were rarely used as very high anastomoses were usually created. The patients from that trial were thus excluded from all analyses related to stomas. The fifth edition of the tumour node metastasis (TNM) classification of malignant tumours was used to determine the TNM stage25. To control the analyses of overall survival, disease-free survival and cancer-specific survival for different age limits allowed in the various trials, those analyses were restricted to patients aged 75 years or less.
|Swedish Rectal||1987–1990||Preop. 5 × 5 Gy RT||1180|
|Cancer Trial||Surgery alone|
|Dutch TME trial||1996–1999||Preop. 5 × 5 Gy RT with TME||1861|
|CAO/ARO/AIO-94 trial||Postop. CRT|
|EORTC 22921 trial||1993–2003||Preop. 45 Gy RT||1011|
|Preop. 45 Gy RT and postop. CT|
|Preop. CRT and postop. CT|
|Polish Rectal Cancer Trial||1999–2002||Preop. 5 × 5 Gy RT with TME||312|
|Preop. CRT with TME|
In the included trials, only symptomatic anastomotic leakages were documented. Anastomotic leakage was defined as clinically apparent leakage such as faecal discharge from pelvic drain or abdominal wound, or radiologically, endoscopically or surgically proven anastomotic leakage in symptomatic patients such as those with peritonitis.
Endpoints, variables and statistical analysis
The χ2 test was used for comparisons of categorical variables. Univariable and multivariable logistic regression analyses were performed with the following variables to study their association with anastomotic leakage: sex, age, distance of tumour from anal verge, TNM stage and presence of a stoma. The multivariable analysis was adjusted for trial and randomization arms.
To study the effects of anastomotic leakage on local recurrence, distant metastasis, overall survival, disease-free survival and cancer-specific survival, Cox regression analyses were used, stratified for trial and randomization arms. The following confounders were first studied by univariable analysis: sex, age, distance of tumour from anal verge, TNM stage and circumferential resection margin (CRM) involvement. Variables with P ≤ 0·100 were then entered in the multivariable Cox regression models. A positive CRM was defined as microscopic or macroscopic tumour in the resection margin (not available in the Swedish Rectal Cancer Trial). Time to local recurrence, distant metastases and overall survival were calculated as the time from surgery to local recurrence, distant metastases and death respectively. For overall survival, the analyses were performed first for all selected patients and then with a landmark selection excluding all patients who died within 90 days after surgery to correct for short-term mortality associated with anastomotic leakage itself. Disease-free survival, defined as time from surgery to first event of local recurrence, distant metastases or death, and cancer-specific survival, defined as time from surgery to death due to rectal cancer, were studied only using the landmark selection excluding patients with 90-day postoperative mortality. The probability of local recurrence is reported as cumulative incidences, with death as the competing risk; cancer-specific survival is reported as 1 - cumulative incidence, with death from causes other than rectal cancer as the competing risk26.
Data were analysed with the statistical package SPSS® version 14.0 for Windows® (SPSS, Chicago, Illinois, USA). A two-sided P value ≤ 0·050 was considered to be statistically significant.
In total, 5187 patients were included in the Swedish Rectal Cancer Trial, Dutch TME trial, German CAO/ARO/AIO-94 trial, EORTC 22 921 trial and the Polish Rectal Cancer Trial. Reasons for exclusion and number of patients are shown in Fig.1. Of 1962 patients who had a procedure other than low anterior resection, 1749 were treated with an abdominoperineal resection. For the analyses, 2726 patients (52·6 per cent) were included. Patient and disease characteristics of these patients are shown in Table2. The median follow-up was 5·9 (range 0·2–14·9) years. Overall, disease-free and cancer-specific survival was studied in 2480 of these 2726 patients, who were aged 75 years or less.
|No. of patients (n = 2726)|
|≤ 60||1008 (37·0)|
|> 70||711 (26·1)|
|Swedish Rectal Cancer Trial||430 (15·8)|
|Dutch TME trial||1132 (41·5)|
|German CAO/ARO/AIO-94 trial||495 (18·2)|
|EORTC 22921 trial||502 (18·4)|
|Polish Rectal Cancer Trial||167 (6·1)|
|Distance of tumour from anal verge (cm)|
|≥ 5·0||2197 (80·6)|
|< 5·0||500 (18·3)|
In total, 264 patients (9·7 per cent) were diagnosed with anastomotic leakage. No information on stoma construction was available for the Swedish Rectal Cancer Trial (n = 430). Therefore, these patients were excluded in the analyses relating to stomas, for which 2296 patients were studied. In 1226 patients (53·4 per cent) a stoma was constructed; in 1067 patients (46·5 per cent) no stoma was created; status was unknown for three patients (0·1 per cent). Symptomatic anastomotic leakage occurred in 124 patients (11·6 per cent) without a stoma, whereas it was diagnosed in 96 patients (7·8 per cent) with a stoma (P = 0·002).
Table3 shows the results of the univariable and multivariable analyses for risk factors associated with anastomotic leakage. From the univariable analyses, both sex and the presence of a diverting stoma were selected for entry in the multivariable analysis (P ≤ 0·100). Trial and treatment arms were entered in the analysis as adjustment. Female sex and the presence of a diverting stoma were both independently associated with a reduced chance of developing symptomatic anastomotic leakage.
|Univariable analysis||Multivariable analysis|
|Odds ratio||P||Odds ratio||P|
|M||1·56 (1·18, 2·07)||1·64 (1·20, 2·24)|
|61–70||1·00 (0·74, 1·34)||0·975|
|> 70||0·95 (0·69, 1·32)||0·780|
|Distance of tumour from anal verge (cm)||0·949||—|
|< 5·0||0·99 (0·71, 1·38)|
|II||1·14 (0·83, 1·57)||0·418|
|III||1·15 (0·85, 1·57)||0·362|
|Yes||0·65 (0·49, 0·85)||0·62 (0·47, 0·82)|
The anastomotic leakage rates per trial and randomization arm are shown in Table4. In none of the trials was a significant difference found between the randomized treatment arms.
|n||Anastomotic leak rate (%)*||Univariable analysis|
|Swedish Rectal Cancer Trial||0·283|
|Surgery only||209||18 (8·6)||1·00|
|5 × 5 Gy RT + surgery||221||26 (11·8)||1·41 (0·75, 2·67)|
|TME alone||578||65 (11·2)||1·00|
|5 × 5 Gy RT + TME||553||54 (9·8)||0·85 (0·58, 1·25)|
|Preop. CRT||241||39 (16·2)||1·00|
|Postop. CRT||248||36 (14·5)||0·88 (0·54, 1·44)|
|EORTC 22921 trial||—|
|Preop. RT||122||0 (0·0)||NE|
|Preop. CRT||125||0 (0·0)||NE|
|Preop. RT + postop. CT||122||4 (3·3)||NE|
|Preop. CRT + postop. CT||133||4 (3·0)||NE|
|Polish Rectal Cancer Trial¶||0·657|
|Preop. CRT||81||8 (10)||1·00|
|Preop. 5 × 5 Gy RT||83||10 (12)||1·25 (0·47, 3·35)|
Of patients without anastomotic leakage, 1·3 per cent (33 of 2446) died within 30 days of surgery, whereas the 30-day mortality rate after anastomotic leakage was 5·7 per cent (15 of 263) (P < 0·001). For one patient with anastomotic leakage, no details on death status were available.
Anastomotic leakage and local recurrence
Anastomotic leakage was not associated with local recurrence in the univariable analysis and therefore not entered in the multivariable analysis; the 5-year local recurrence rate was 8·8 (95 per cent confidence interval (c.i.) 7·6 to 10·0) per cent for patients without anastomotic leakage and 12·0 (95 per cent c.i. 7·4 to 16·5) per cent for those with a leak (P = 0·103). The cumulative incidence of local recurrence with death as competing risk for patients with and without anastomotic leakage is depicted in Fig.2a.
Anastomotic leakage and distant metastasis
The univariable analysis for the association between anastomotic leakage and distant metastases was not significant; the rate of distant metastasis at 5 years was 25·6 (95 per cent c.i. 23·7 to 27·3) per cent and 27·5 (95 per cent c.i. 21·4 to 33·6) per cent respectively for patients without and with anastomotic leakage (P = 0·480). Therefore, no multivariable analysis with anastomotic leakage was performed for distant metastases.
Anastomotic leakage and overall survival
First, the analyses were performed with all selected patients. Anastomotic leakage was significantly associated with a worse overall survival rate in both the univariable analysis (hazard ratio (HR) 1·49 (95 per cent c.i. 1·20 to 1·84); P < 0·001) and the multivariable analysis (HR 1·48 (95 per cent c.i. 1·19 to 1·83); P < 0·001). The 5-year overall survival rate was 74·4 (95 per cent c.i. 72·4 to 76·4) per cent for patients without anastomotic leakage compared with 66·4 (95 per cent c.i. 60·1 to 72·7) per cent for those with leakage (P < 0·001).
Table5 shows the results of both the univariable and multivariable analyses for risk factors associated with overall survival, excluding patients who died within 90 days of surgery (n = 52). The 5-year overall survival rate was 75·5 (95 per cent c.i. 73·4 to 77·4) per cent for patients without anastomotic leakage versus 71·5 (95 per cent c.i. 62·2 to 77·8) per cent for those with a leak (P = 0·030). Male sex, age above 70 years, advanced TNM stage and postoperative anastomotic leakage were associated with diminished overall survival in both univariable and multivariable analyses. Kaplan–Meier curves for overall survival are shown for all patients in Fig.2b and excluding patients who died in the first 90 postoperative days in Fig.2c.
|n||Univariable analysis||Multivariable analysis|
|Hazard ratio||P§||Hazard ratio||P¶|
|Sex||< 0·001||< 0·001|
|M||1526||1·43 (1·23, 1·67)||1·33 (1·14, 1·56)|
|Age (years)||< 0·001||< 0·001|
|61–70||984||1·16 (0·98, 1·38)||0·084||1·23 (1·04, 1·46)||0·016|
|> 70||447||1·86 (1·54, 2·25)||< 0·001||2·06 (1·70, 2·49)||< 0·001|
|Distance of tumour from anal verge (cm)*||0·466||—|
|< 5·0||464||1·08 (0·88, 1·32)|
|TNM stage†||< 0·001||< 0·001|
|II||712||2·11 (1·70, 2·63)||< 0·001||2·08 (1·67, 2·26)||< 0·001|
|III||858||3·93 (3·21, 4·81)||< 0·001||4·02 (3·28, 4·92)||< 0·001|
|Yes||81||1·63 (1·11, 2·39)||0·013||1·17 (0·79, 1·72)||0·442|
|Unknown||499||1·09 (0·76, 1·56)||0·651||0·94 (0·64, 1·40)||0·774|
|Yes||220||1·29 (1·02, 1·63)||1·29 (1·02, 1·63)|
Anastomotic leakage, stomas and overall survival
When the analyses for overall survival were repeated with the two variables of anastomotic leakage and stomas in the model, both were significantly associated with a worse overall survival (data not shown). However, no statistical significant interaction between anastomotic leakage and stomas could be demonstrated (P = 0·255). Patients with a stoma had an increased risk of death (multivariable model: HR 1·24 (95 per cent c.i. 1·04 to 1·48); P = 0·015). Fig.3a shows Kaplan–Meier curves for overall survival separately for patients with/without anastomotic leakage and with/without stomas. Fig.3b shows the curves for overall survival excluding patients who died within 90 days of surgery. The difference between Fig.3a and Fig.3b is due to early postoperative mortality. Patients without anastomotic leakage and without a stoma fared better than the other three groups in the long term. For patients with no anastomotic leakage and no stoma, with no leakage and with a stoma, with anastomotic leakage and without a stoma, and with anastomotic leakage and with a stoma, the 90-day mortality rate was 1·3, 1·9, 8·9 and 5·8 per cent respectively. The difference in 90-day postoperative mortality was significant only between patients with and those without anastomotic leakage (P < 0·001).
Anastomotic leakage and disease-free and cancer-specific survival
Anastomotic leakage was associated with a worse disease-free survival rate: HR 1·26 (95 per cent c.i. 1·02 to 1·56); P = 0·033) in the univariable analysis and HR 1·24 (95 per cent c.i. 1·01 to 1·56); P = 0·040) when adjusted for sex, age and TNM stage. The disease-free survival curve is shown in Fig.2d. The 5-year disease-free survival rate was 66·9 (95 per cent c.i. 64·9 to 68·9) per cent for patients without anastomotic leakage and 60·6 (95 per cent c.i. 53·7 to 67·5) per cent for those with leakage (P = 0·033). The estimates of the cumulative incidence for cancer-related mortality with death from causes other than rectal cancer as competing risk are shown in Fig.2e. No significant association was found between cancer-specific survival and anastomotic leakage (HR 1·12 (95 per cent c.i. 0·83 to 1·52); P = 0·466); the 5-year cancer-specific survival rate was 80·6 (95 per cent c.i. 78·8 to 82·4) per cent for patients without and 79·5 (95 per cent c.i. 73·6 to 85·4) per cent for patients with anastomotic leakage (P = 0·466).
In this study, patient data from five large randomized European trials for rectal cancer were pooled. Although the decision to create a stoma was left to the discretion of the surgeon, and each individual trial was not designed to study anastomotic leakage, the present results are interesting owing to the large number of patients included from several European countries with a long and well documented follow-up. Anastomotic leakage was significantly reduced when a diverting stoma was created. Interestingly, leaks were associated with decreased disease-free and overall survival rates, but oncological outcome measures (local recurrence, distant metastases and cancer-specific survival) were not affected.
Apart from the early consequences after a leak, such as sepsis-related death, anastomotic failure has been reported to be associated with decreased local control6–11 and survival7, 12, 13. However, the association between anastomotic leakage and local control cannot be confirmed in all studies: in a population-based cohort study in Norway (1958 patients), anastomotic leakage did not result in an increased local recurrence rate27. In the present study, anastomotic leakage was associated with both reduced disease-free survival and overall survival rates. Interestingly, when excluding early postoperative mortality, overall survival in the groups with and without anastomotic leakage was very similar in the first 4 years. After 4 years, however, overall survival decreased in patients who had a symptomatic leak. In the present analysis, no association was found between anastomotic leakage and cancer-specific survival, although in other studies such an association was demonstrated7, 12, 13. Apparently, patients in the pooled database who developed anastomotic leakage had a higher chance of dying than those without anastomotic leakage, but mainly owing to causes other rather than rectal cancer.
The observed consequences of anastomotic leakage—early and late morbidity and mortality—stress the importance of decreasing the incidence of (symptomatic) anastomotic leakage. One of the options is to create a diverting stoma. Recently, Matthiessen and colleagues19 performed a randomized trial in 234 patients who underwent a low anterior resection. Patients were randomized between a diverting loop stoma and no stoma. In this study it was found that a diverting stoma decreased the rate of symptomatic anastomotic leakage. Hüser and co-workers28 did a systematic review and meta-analysis of 27 retrospective and four randomized clinical trials on the role of a diverting stoma in low rectal cancer surgery. They concluded that a diverting stoma reduces the rate of clinically relevant anastomotic leakage and is thus recommended in operations for low rectal cancer. Nevertheless, it should not be forgotten that stoma closure is also associated with morbidity and mortality29, 30. In addition, one in five diverting stomas is never closed31.
In the present analysis, patients without leakage and without a stoma had a better survival than those with no leakage and with a stoma. As the pooled studies did not randomize between stoma and no stoma (the decision to create a stoma was left to the discretion of the surgeon), there is probably a selection bias here. However, this reflects daily clinical practice and it may be possible that patients with a stoma had more co-morbidity than those without a stoma. Even so, patients with a stoma had less symptomatic leakage, and postoperative mortality after anastomotic leakage tends to be lower with a stoma (5·8 versus 8·9 per cent), although this was not statistically significant. Owing to the above-mentioned bias, the question of whether the presence of a stoma (as an isolated variable) might improve overall survival cannot be answered by this study.
Many observational studies have examined the association between preoperative treatment and anastomotic leakage. In national population-based studies in Sweden and Norway, preoperative radiotherapy was found to be associated with anastomotic leakage27, 32. Similarly, in a case–control study using the Swedish Cancer Registry, preoperative radiotherapy was found to be a risk factor for anastomotic leakage15. In randomized trials, however, there is no association between anastomotic leak and short-course radiotherapy16, 33. Owing to different treatment protocols and other variance, anastomotic leakage rates cannot be fairly compared across trials, although comparison within each trial is valid. In none of the five randomized trials discussed here was there a significant difference in anastomotic leak rate due to preoperative treatment, but trials are notorious for not necessarily reflecting real practice. Indeed, based on the real-life observational studies15, 27, 32, there are clearly other (confounding) factors that affect the selection of patients for preoperative radiotherapy contribute to the higher risk of leak.
Anastomotic leakage cannot be avoided but its consequences can be limited by a diverting stoma28, 34. Apart from a diverting stoma, some have found that placement of a pelvic drain can limit the consequences of anastomotic leakage16, although others could not find such an association35. Nevertheless, prompt diagnosis and treatment of anastomotic leakage is necessary to limit morbidity and mortality. Standardized postoperative surveillance results in early identification of and reduced mortality from symptomatic anastomotic leakage4.
M.d.D. is supported by a Quality Assurance Fellowship of the European Society of Surgical Oncology. The authors are grateful to all institutes that participated in the Swedish Rectal Cancer Trial, Dutch TME trial, German CAO/ARO/AIO-94 trial, EORTC 22921 trial and the Polish Rectal Cancer Trial. The authors declare no conflict of interest.
- 25TNM Classification of Malignant Tumors (5th edn). John Wiley: New York, 1997., .
- 33Swedish Rectal Cancer Trial participants. Initial report from a Swedish multicentre study examining the role of preoperative irradiation in the treatment of patients with resectable rectal carcinoma. Br J Surg 1993; 80: 1333–1336.