Despite developments in surgical techniques and perioperative care, anastomotic leakage (AL) occurs up to 20% after restorative rectal cancer (RC) resection,1 and remains a severe complication.2–5 AL is associated with increased mortality,6–8 a negative impact on survival, and leads to more reinterventions with subsequently higher health care costs.9,10 In addition, half of the patients with symptomatic AL will end up with a permanent stoma.11 This might be either an initial or secondary defunctioning stoma or end-colostomy after salvage surgery. A permanent stoma is an unintended outcome for a patient who expected restoration of bowel continuity, which likely contributes to inferior quality of life.12,13

Considerable heterogeneity exists in the clinical presentation of AL, which ranges from occult leakages to severe sepsis, and it is debated to which extent this correlates with a permanent stoma.14,15 Furthermore, several patient and leakage-related factors, as well as surgical characteristics for treatment of the primary RC, might influence the chance of healing of an AL and the risk of permanent stoma. Although AL has been studied extensively, long-term outcomes in terms of restoration of bowel continuity is an understudied topic as previous studies mainly focussed on the identification of risk factors, prevention, and early diagnosis of AL.7,16,17 This emphasizes the need to explore predictive factors related to the restoration of bowel continuity.

This study aimed to develop and validate a prediction score for 1-year stoma-free survival (STOMA score), using data from a large international retrospective cohort study that included patients with AL after RC surgery. The STOMA score can be used in clinical practice for the purpose of patient counseling or in the research setting for future intervention studies.

METHODS

The “TreatmENT of AnastomotiC Leakage after rEctal” cancer resection (TENTACLE-Rectum, Supplemental Digital Content 1, https://links.lww.com/SLA/E780) study is an international multicentre retrospective cohort study encompassing patients who developed AL after RC resection, who were operated between the January 1, 2014 and December 31, 2018. The study was reported according to the “Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis” guidelines (Supplemental Digital Content 1, https://links.lww.com/SLA/E780).18 All centres performing RC surgery were eligible to participate without limitations based on case volume or geographic location. In total, the collaborative group consists of 216 centres from 45 countries. The study was reviewed and approved on October 17, 2019 by the Research Ethics Committee of the Radboud University Medical Centre Nijmegen. According to Dutch law, informed consent was not required for observational studies. All participating centres adhered to their own legislation regarding approval and informed consent procedures. The full study protocol has been published,14 and the study is registered in the Clinical Trials registry: NCT04127734.

Patient Selection

Patients were included if they were aged 18 years or older and diagnosed with AL within 1 year after RC resection with the formation of a primary anastomosis with or without defunctioning stoma for either primary RC, regrowth (ie, after watch-and-wait strategy), or as completion surgery after local excision between 2014 and 2018. Exclusion criteria were emergency RC resection, resection for benign disease, or recurrent RC.

Definitions

The international consensus about the definition of the rectum was used to include homogeneous patients with RC. This definition encompasses tumors with their lower border at or below the level of the sigmoid take-off.19 AL was defined according to the definition of the International Study Group of Rectal Cancer: “a defect of the integrity of the intestinal wall at the anastomotic site (including leakage originating from the suture and staple lines of neorectal reservoirs).”20 This definition includes a pelvic abscess near the anastomosis, without a clear bowel wall defect.

Data Collection, Verification, and Validation

Local investigators collected data pseudonymized in an online database (www.castoredc.com) and individual data were only traceable and accessible for the participating centres. Data verification and quality validation were performed to substantiate that all consecutive cases were included and to minimize inconsistencies and missing data (Supplemental Digital Content Material 1, https://links.lww.com/SLA/E780). To reduce bias due to missing data, multiple imputation with chained equations was performed.21 Information about handling of missing data (Supplemental Digital Content Table 3, https://links.lww.com/SLA/E780) can be found in Supplemental Material (Supplemental Digital Content Material 2, https://links.lww.com/SLA/E780).

Outcome

The outcome of this study was 1-year stoma-free survival, which was defined as being alive without a defunctioning stoma or end-colostomy 1-year after RC surgery.

Predictors for Stoma-free Survival

The selection of potential clinically relevant predictors for stoma-free survival was done based on a literature review and expert opinion among the lead investigators. Predictors selected through the literature review consisted of patient demographics (eg, age and comorbidity), disease-related and perioperative factors (eg, metastasis and abdominal approach), and leakage-related factors at diagnosis (eg, ischemia). Literature review and subsequent confirmation by the lead investigators yielded the inclusion of the following predictors: age, American Society of Anesthesiologists (ASA) classification, clinical M-disease, neoadjuvant therapy, abdominal approach, defunctioning stoma created at index surgery, multivisceral resection, postoperative day of AL diagnosis, fistulas, retraction afferent colon, abdominal contamination, ischemia bowel wall, anastomotic defect circumference, and reactivation leakage.5,22–28 In addition, 4 predictors with substantial clinical relevance were identified merely on expert opinion, comprising: sex, body mass index, transanal total mesorectal excision, and clinical setting of AL diagnosis. Based on this selection process, 18 predictors were included in the analysis. The predictors are depicted in Table 1, and additional information concerning sample size calculations and predictor selection can be found in Supplemental Materials (Supplemental Digital Content Materials 3 and 4, https://links.lww.com/SLA/E780).

TABLE 1 - Clinically Relevant Predictors for Stoma-free Survival in Patients With AL After RC Surgery* Demographic factors Surgical and diagnostic factors Leakage-related factors Sex Abdominal approach Fistula(s) Age Defunctioning stoma created at index surgery Retraction afferent colon BMI TaTME Abdominal contamination ASA classification Multivisceral resection Ischemia bowel wall Clinical M-disease Clinical setting diagnosis AL Anastomotic defect circumference Neoadjuvant therapy Postoperative day of AL diagnosis Reactivation leakage *A more detailed description regarding the selection of predictors can be found in the Supplemental Materials, Supplemental Digital Content, https://links.lww.com/SLA/E780.

BMI indicates body mass index; TaTME, transanal total mesorectal excision.

Definitions Predictors

The clinical setting of AL diagnosis was included to make a proxy of the patient's clinical condition at the time of diagnosis and was categorized into: intensive care unit or high-dependency care unit, surgical ward, emergency department, and out-patient clinic. Defect circumference was classified based on the degree of anastomotic dehiscence measured endoscopically: 0% to 25% (mild), 25% to 50% (moderate), and 50% to 100% (severe). Abdominal contamination was defined as a spill or leakage of bowel content into the abdominal cavity confirmed at reoperation. Anastomotic fistulas could either be present as a postoperative iatrogenic complication or as a secondary infection due to chronic pelvic sepsis, with tracks to organs or structures (eg, vagina, small bowel, and skin). Reactivation leakage was defined as AL that was diagnosed after the closure of a defunctioning stoma, even though diagnostic workup before stoma closure showed intact anastomosis.

Statistical Analyses

The study deviated from the original analysis plan as described in the study protocol,14 for the development of a prediction model according to the “Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis” guidelines (Supplemental Digital Content 1, https://links.lww.com/SLA/E780). The total cohort was dived into a development cohort (2014–2017) and a temporal validation cohort (2018). The model was developed based on a multivariable logistic regression model that predicts 1-year stoma-free survival following AL after RC resection. All 18 a priori predictors were included in the final multivariable model. Restricted cubic spline functions were used to test for the nonlinearity of the continuous variable (ie, age).

Internal validation with bootstrap resampling (500 replicates) was applied to reduce the optimism of the prognostic model. The obtained shrinkage factor was used to correct the regression coefficients, which contributes to generalizability and reduction of overfitting of the model. Based on the final bootstrapped multivariable regression analysis, a nomogram was created. In the development cohort, the model's performance was assessed with discrimination [concordance (c)-index] and calibration. The flexible calibration curve allows the examination of calibration across a range of predicted values. A curve close to the diagonal line (ie, perfect calibration) indicates that the predicted (x-axis) and observed probabilities (y-axis) correspond well.

To assess the model's predictive performance in another cohort with similar patients, external validation was performed using a temporal approach.29–31 Temporal validation was done with a cohort of patients who underwent RC resection in 2018. The pooled performance strategy (Rubin’s rule) was used to pool performance measures.32 The internally validated model was implemented in a web application that provides patients’ 1-year stoma-free survival predictions. All analyses were carried out in R version 4.1.3 (R Foundation for Statistical Computing).

RESULTS Patients

In total, 2710 patients were included in the database. A total of 211 patients were excluded based on: incorrect year of RC resection (n = 189), AL diagnosis beyond 1 year from index surgery (n = 21), and absence of AL (n = 1). This resulted in 2499 patients with AL, of whom 1954 were included in the development cohort and 545 in the validation cohort. Figure 1 presents the flowchart of patient inclusion.

FIGURE 1:

Flowchart of patient inclusion.

Data Quality Validation

After correlating the expected with the uploaded cases, all 216 centres included their consecutive cases within the range of the expected number of patients with AL between 2014 and 2018. Of the 2499 patients, 164 cases (7%) from 33 different centres (15%) were validated and the overall accuracy was 96.6%. Hospital characteristics (eg, annual case volume) can be found in Supplemental Tables (Supplemental Digital Content Tables 1 and 2, https://links.lww.com/SLA/E780).

Baseline Characteristics

Table 2 presents the baseline characteristics in the development and validation cohorts, which were predominantly comparable. Small proportional differences were found in the abdominal approach and configuration of the anastomosis. In the validation cohort, less defunctioning stomas were created during primary RC resection (66.4% vs 61.1%), and abdominal contamination was reported more frequently at AL diagnosis (31.9% vs 36.7%). Median postoperative day of AL diagnosis did not differ between cohorts, which was after 8 days [interquartile range (IQR): 4–18] in the development cohort, and after 7 days (IQR: 4–15) in the validation cohort.

TABLE 2 - Baseline Characteristics Development and Validation Cohort Development cohort (2014–2017); N = 1954; n (%) Validation cohort (2018); N= 545; n (%) Patient demographics  Age (yr); median (IQR) 65 (57–72) 64 (57–72)  Sex   Female 540 (27.6) 154 (28.3)   Male 1414 (72.4) 391 (71.7)  BMI (kg/m2)   Underweight (<18.5) 91 (4.7) 30 (5.5)   Normal (18.5–24.9) 579 (29.6) 169 (31)   Overweight (25.0–29.9) 738 (37.8) 193 (35.4)   Obese (>30) 380 (19.4) 119 (21.8)   Missing 166 (8.5) 34 (6.2)  ASA classification   ASA-I 302 (15.5) 80 (14.7)   ASA-II 1098 (56.2) 290 (53.2)   ASA-III/IV 508 (25.9) 162 (29.7)   Missing 46 (2.4) 13 (2.4) Tumor characteristics  Clinical T-classification   T0 6 (0.3) 4 (0.6)   T1 73 (3.7) 10 (1.8)   T2 390 (20) 117 (21.6)   T3 1206 (61.7) 340 (62.4)   T4 190 (9.7) 57 (10.5)   Missing 89 (4.6) 17 (3.1)  Clinical N-classification   N0 716 (36.6) 218 (40)   N1 590 (30.2) 182 (33.4)   N2 393 (20.1) 110 (20.2)   N+ 125 (6.4) 23 (5.1)   Missing 130 (6.7) 12 (2.2)  Clinical M-disease   M0 1536 (78.6) 428 (78.5)   M1 150 (7.7) 43 (7.9)   Missing 268 (13.7) 74 (13.6)  Neoadjuvant therapy   None 839 (42.9) 241 (44.2)   Radiotherapy only 238 (12.2) 57 (10.5)   Chemotherapy 41 (2.1) 7 (1.3)   Chemoradiation 836 (42.8) 240 (44)  Tumor distance from the anorectal junction (mm); median (IQR) 60 (32–90) 60 (30–82) Surgical characteristics  Abdominal approach   Laparoscopic 1181 (60.4) 357 (65.5)   Robot-assisted 179 (9.2) 58 (10.6)   Laparotomy 593 (30.3) 130 (23.9)   Missing 1 (0.05) —  TaTME   No 1599 (81.8) 433 (79.4)   Yes 355 (18.2) 111 (20.4)   Missing — 1 (0.2)  Specification approach   Open (TATA) 82 (23.1) 13 (11.7)   Transanal platform 243 (68.5) 90 (81.1)   Missing 30 (8.4) 8 (7.2)  Configuration anastomosis   End-to-end 1184 (60.6) 382 (70.1)   Side-to-end 604 (30.9) 138 (25.3)   Other* 81 (4.1) 10 (1.8)   Missing 85 (4.4) 15 (2.8)  Multivisceral resection   No 1781 (91.1) 494 (90.6)   Yes 127 (6.5) 41 (7.5)   Missing 46 (2.4) 10 (1.9)  Splenic flexure mobilization   No 630 (32.2) 183 (33.6)   Yes 1014 (51.9) 294 (53.9)   Missing 310 (15.9) 68 (12.5)  Defunctioning stoma created at index surgery   No 656 (33.6) 212 (38.9)   Yes 1298 (66.4) 333 (61.1) Diagnostic characteristics  Clinical setting diagnosis AL   Surgical ward 1324 (67.8) 387 (71.0)   ICU/HC 84 (4.3) 24 (4.4)   ED 198 (10.1) 51 (9.4)   Out-patient clinic 346 (17.7) 81 (14.9)   Missing 2 (0.1) 1 (0.2)  Postoperative day of AL diagnosis; median (IQR) 8 (5–18) 7 (4–15) Leakage characteristics  Leakage location   Circular 1090 (55.8) 337 (61.8)   Side-to-end 183 (9.3) 47 (8.6)   Missing 681 (34.9) 161 (29.6)  Anastomotic defect circumference   0%–25% 433 (39.7) 139 (41.3)   25%–50% 230 (21.1) 79 (23.4)   50%–100% 142 (13.0) 55 (16.3)   Missing 285 (26.2) 64 (19)  Ischemia bowel wall   No 1406 (72.0) 376 (69.0)   Yes 197 (10.1) 64 (11.7)   Missing 351 (17.9) 105 (19.3)  Retraction afferent colon   No 1426 (73.0) 402 (73.8)   Yes 76 (3.9) 23 (4.2)   Missing 452 (23.1) 123 (22.6)  Fistula(s)   No 1721 (88.1) 473 (86.8)   Yes 130 (6.7) 47 (8.6)   Missing 103 (5.2) 25 (4.6)  Abdominal contamination   No 1160 (59.4) 294 (53.9)   Yes 623 (31.9) 200 (36.7)   Missing 171 (8.7) 51 (9.4)  Reactivation leakage   No 1253 (64.1) 354 (64.9)   Yes 130 (6.7) 31 (5.7)   Missing 571 (29.2) 160 (29.4) Mortality  Mortality within 1 yr after index surgery   No 1738 (88.9) 485 (89.0)   Yes 103 (5.3) 27 (4.9)   Missing 113 (5.8) 33 (6.1) Outcome  Stoma-free survival   No 891 (45.6) 252 (46.2)   Yes 880 (45.0) 238 (43.7)   Missing 183 (9.4) 55 (10.1)

*Other = colon pouch, coloplasty, ileal pouch-anal anastomosis.

BMI indicates body mass index; ED, emergency department; HC, high-dependency care; ICU, intensive care unit; TaTME, transanal total mesorectal excision; TATA, Transanal Abdominal Transanal Resection.

Predictors for One-year Stoma-free Survival

In the development and validation cohorts, 1-year stoma-free survival was 45.0% and 43.7%, respectively. Table 3 shows the univariable and multivariable odds ratios (ORs) of the 18 tested predictors for stoma-free survival in the development cohort. Presented multivariable ORs are after internal validation. The most important predictors for a stoma at 1 year in the univariable analysis were: age (IQR: OR 1.21, 95% CI: 1.07–1.36), ASA-classification III/IV (OR: 1.48, 95% CI: 1.11–1.98), clinical M1-disease (OR: 2.08, 95% CI: 1.44–3.01), setting of diagnosis AL at the intensive care unit/high-dependency care (OR: 1.64, 95% CI: 1.02–2.63), open resection (OR: 1.58, 95% CI: 1.29–1.94), degree of anastomotic dehiscence (moderate: OR: 2.15, 95% CI: 1.55–2.97 and severe: OR: 4.05, 95% CI: 2.65–6.20), ischemia (OR: 2.53 95% CI 1.83–3.50), retraction of the afferent colon (OR: 2.85, 95% CI: 1.71–4.72), abdominal contamination (OR: 2.33, 95% CI: 1.90–2.85), and reactivation leakage (OR: 1.71, 95% CI: 1.20–2.43). Predictors for not having a stoma at 1 year were: setting of diagnosis AL at the out-patient clinic (OR: 0.66, 95% CI: 0.52–0.85) and transanal total mesorectal excision (OR: 0.71, 95% CI: 0.56–0.90). The following predictors did not reach statistical significance but contributed to the prediction of 1-year stoma-free survival: body mass index, multivisceral resection, neoadjuvant therapy, and postoperative day of AL diagnosis. In the multivariable analysis, predictors that remained significant for a stoma at 1 year were: age (OR: 1.22, 95% CI: 1.06–1.41), open resection (OR: 1.31, 95% CI: 1.04–1.65), degree of anastomotic dehiscence (moderate: OR: 1.72 95% CI: 1.21–2.45, severe: OR: 2.53, 95% CI: 1.53–4.19), ischemia (OR: 1.51 95% CI: 1.03–2.21), abdominal contamination (OR: 1.81, 95% CI: 1.41–2.32), reactivation leakage (OR: 1.50 95% CI: 1.02–2.20), and creation of a defunctioning stoma at index surgery became significant (OR: 1.31, 95% CI: 1.04–1.66).

TABLE 3 - STOMA-scores Predictive Accuracy in the Development Cohort Predictor Univariable model; OR (95% CI) Multivariable model; OR (95% CI)* Sex  Male 1.00 (reference) 1.00 (reference)  Female 1.19 (0.97–1.46) 1.14 (0.90–1.43) Age (yr); median (57–72 IQR)† 1.21 (1.07–1.36) 1.22 (1.06–1.41) ASA classification  ASA-I 1.00 (reference) 1.00 (reference)  ASA-II 1.15 (0.90–1.50) 1.08 (0.81–1.44)  ASA-III/IV 1.48 (1.11–1.98) 1.12 (0.80–1.59) BMI  Normal 1.00 (reference) 1.00 (reference)  Underweight 1.41 (0.90–2.22) 1.30 (0.79–2.14)  Overweight 1.08 (0.86–1.34) 1.13 (0.89–1.43)  Obese 0.95 (0.73–1.24) 0.90 (0.68–1.21) Clinical M-disease  M0 1.00 (reference) 1.00 (reference)  M1 2.08 (1.44–3.01) 1.80 (1.19–2.72) Neoadjuvant therapy  None 1.00 (reference) 1.00 (reference)  Radiotherapy 1.05 (0.79–1.41) 1.17 (0.84–1.62)  Chemotherapy 1.61 (0.83–3.13) 1.10 (0.52–2.36)  Chemoradiation 1.03 (0.85–1.25) 1.13 (0.89–1.42) Abdominal approach  Laparoscopic 1.00 (reference) 1.00 (reference)  Robot-assisted 0.83 (0.60–1.14) 0.86 (0.60–1.23)  Laparotomy 1.58 (1.29–1.94) 1.31 (1.04–1.65) Defunctioning stoma created at index surgery 1.04 (0.86–1.26) 1.31 (1.04–1.66) TaTME 0.71 (0.56–0.90) 0.79 (0.61–1.04) Multivisceral resection 1.36 (0.94–1.98) 1.18 (0.78–1.78) Clinical setting diagnosis AL  Surgical ward 1.00 (reference) 1.00 (reference)  Intensive care/high care unit 1.64 (1.02–2.63) 1.22 (0.72–2.06)  ED 0.89 (0.66–1.20) 1.01 (0.73–1.42)  Outpatient clinic 0.66 (0.52–0.85) 0.75 (0.56–1.01) Postoperative day of AL diagnosis, median (5–18 IQR)† 1.00 (0.97–1.03) 1.02 (0.99–1.06) Anastomotic defect circumference  0%–25% 1.00 (reference) 1.00 (reference)  25%–50% 2.15 (1.55–2.97) 1.72 (1.21–2.45)  50%–100% 4.05 (2.65–6.20) 2.53 (1.53–4.19) Ischemia bowel wall 2.53 (1.83–3.50) 1.51 (1.03–2.21) Retraction afferent colon 2.85 (1.71–4.72) 1.30 (0.70–2.42) Fistula(s) 1.33 (0.92–1.92) 1.10 (0.73–1.68) Abdominal contamination 2.33 (1.90–2.85) 1.81 (1.41–2.32) Reactivation leakage 1.71 (1.20–2.43) 1.50 (1.02–2.20)

*Presented odds ratios after internal validation.

†For continuous variables, odds ratios represent interquartile range odds ratios.

The odds ratio presented gives insight into the importance of predictors, which are expressed on a relative scale. These can be considered as a representation of the contribution to the predicted risk. A causal relation between predictor and outcome or the magnitude of the effect is not necessarily presented by the odds ratios.

BMI indicates body mass index; ED, emergency department; TaTME, transanal total mesorectal excision.

STOMA Score After Internal and Temporal Validation

The STOMA score was developed using a multivariable logistic regression modeling consisting of 18 clinically relevant predictors for 1-year stoma-free survival. After internal validation, the c-index was 0.70 (95% CI: 0.67–0.73). The nomogram is presented in Supplemental Figure (Supplemental Digital Content Fig. 1, https://links.lww.com/SLA/E780). After temporal validation, the c-index was 0.71 (95% CI: 0.66–0.76). The scores’ flexible calibration (Fig. 2) curve shows that predicted probabilities correlated with the observed probabilities across the entire risk range, indicating near-perf