This was a prospective observational, single-center study that included adults over 18-years-old who underwent urgent or elective colorectal resection, regardless of the surgical approach (open or laparoscopic), indication (benign or malignant), and option for a protective stoma. The study was conducted in the colorectal division of a non-academic hospital accredited by Joint Commission International^® and included about 500000 inhabitants. The data were collected between March 1, 2017 and August 31, 2019. The local ethics committee approved the study, and potential participants provided written informed consent before inclusion.

CAL was defined in accordance with the following criteria[29]: (1) Clinical: Enteric discharge from abdominal drain or wound, rectovaginal fistula, or anastomotic defect found by digital examination; (2) Radiological (CT): Extravasation of endoluminally administered contrast, intra-abdominal collection around the anastomosis, presacral abscess near the anastomosis or perianastomotic air, and free intra-abdominal air; and (3) Surgical findings (reoperation): Necrosis of the anastomosis or signs of peritonitis and anastomotic defect.

Faced with clinical deterioration and/or serum biomarker increase, patients underwent further imaging with abdominopelvic CT scan (and water-soluble contrast enema if colorectal anastomosis was present). Once diagnosed, anastomotic leakage was classified into two categories: (1) Minor: Patients with CAL and Clavien-Dindo grade I or II, requiring no active intervention (radiological or surgical intervention) (Grade A of the International Study Group of Rectal Cancer definition); and (2) Major: All other patients with CAL[30,31]. Definitions of other postoperative complications, such as pneumonia, urinary tract infection, paralytic ileus, and surgical wound infection, are available in Supplementary material 1A (Definitions).

Patients were excluded from the study if they were younger than 18-year-old, pregnant, unable to give or not providing written informed consent, R0 resection with anastomosis not having been performed, or presence of inflammatory bowel disease.

Prospective data were collected and recorded in an electronic database according to the study protocol (Supplementary material 1B – Study protocol). Five biomarkers were measured in the first 5 posto-perative days (POD), including white blood cell (WBC) count, eosinophil cell count (ECC), CRP, CLP, and procalcitonin (PCT). Clinical criteria, such as abdominal pain and clinical condition, were also assessed. Blood samples were analyzed at the Leiria Hospital Centre laboratory, according to the techniques described in Supplementary material 1C (Laboratory). The 90-d follow-up included data of all postoperative complications, the length of hospital stay, and the readmission rate. Discharge criteria are available in the Supplementary material 1B (Study protocol). All patients received prophylactic antibiotic accordingly to hospital infection control committee protocol.

Data were analyzed by using standard descriptive statistics and graphical analysis. One-way analysis of variance was performed to compare the differences in mean biomarkers’ values across the three relevant groups of patients (G1 – no complications; G2 – complications not related to CAL; and G3 – CAL). Chi-squared tests were conducted to assess the association between other categorical variables and the patients. Receiver operating characteristic (ROC) curve analysis was employed to evaluate each biomarker as an appropriate classifier to detect CAL early. The area under the receiver operating characteristic curve (AUROC) was used to establish the diagnostic performance of the studied biomarker. Liu’s method was used to establish the threshold value of each biomarker, and its sensitivity (SS) and specificity (SP) were defined[32]. The negative likelihood ratio (NLR) and positive likelihood ratio (PLR), and the negative predictive value (NPV) and positive predictive value (PPV) were computed by combining the observed incidence of CAL with the estimated SS and SP at the optimum cut-off value.

The added value of combining two different biomarkers, observed on POD3 or POD5, as a classifier to predict early CAL was explored. Regression models (probit, logit, and complementary log-log) were used to analyze binary dependent variables, and the observed CAL status (0/1) in a pairwise manner of all biomarkers included in our study: WBC, ECC, CRP, PCT, and CLP. Several potential classifiers of CAL were built, applying a non-linear combination of two different biomarkers. To minimize overfitting, the “leave-one-out” methodology was adopted[33]. The AUROC graph was used to select the classifier (defined by the model and the combination of two biomarkers) with the best predictive diagnostic performance. Liu’s method was adopted to select the cut-off value for CAL.

The expected reduction in time to CAL diagnosis obtained by using one biomarker or a pairwise combination of biomarkers was estimated. This was the difference between the observed and the expected mean time to CAL diagnosis, if a specific classifier is used. The expected time to CAL diagnosis was computed by using the following expression: S × d1 + [(1 – S) × d2], where S is the SS of the classifier, d1 is the POD of the classifier yielding a positive cut-off value for CAL, and d2 is the day of diagnosis if the classifier provides a false-negative result (time to CAL diagnosis estimated in the dataset). The statistical methods of this study were reviewed by Óscar Lourenço from the Faculty of Economics, CeBER, University of Coimbra, Portugal. All data management and statistical analyses were conducted with Stata Statistical software (Release 16; StataCorp, College Station, TX, United States).

During the study period, we included 458 consecutive patients who underwent colorectal resection, and 62 (13.5%) were ruled out [exclusion criteria (n = 31), no consent (n = 15), no anastomosis (n = 16)] as shown in Figure 1. Patient characteristics, divided into three groups (G1, G2, and G3, as previously defined), are shown in Table 1. Age, the Charlson Comorbidity Index, and American Society of Anesthesiologists grade seem to affect CAL onset.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Flow diagram of patients according to the study protocol. POD1: Postoperative day 1.

Table 2 summarizes the main operative characteristics. Eighty-two percent of patients had a laparoscopic approach, and the most common procedures performed were right colectomy (n = 196; 49.5%) and sigmoid colectomy/rectosigmoid resection (n = 74; 18.7%). The surgical approach (P < 0.001), the volume of blood loss (P < 0.001), the occurrence of intraoperative complications (P < 0.001), and the duration of the procedure (P = 0.011) were significantly related to the development of CAL.

In this study, CAL developed in 25 of 396 patients (6.3%) and was more frequent in men than women (68% vs 32%). Twenty-three patients with CAL (92.0%) were diagnosed during the first hospital admission. The mean ± SD and median time for CAL detection were 9.0 ± 6.8 d and 8 d (interquartile range = 7), respectively. Anastomotic leak was significantly associated with a longer hospital stay (median of 21 d vs 7 d and 13 d, in G1 and G2 patients, respectively; P < 0.001), the readmission rate (20% vs 6.4% and 5.4%), and the reoperation rate (12% vs 3.2% and 1.8%). Table 3 provides a summary of 90-d morbidity and mortality rates. Based on the Clavien-Dindo classification, grades III and IV complication were significantly higher in the G3 cohort (84.0% vs 17.0%; P < 0.001) (Table 4).

Table 5 outlines the intraoperative and postoperative details of patients with CAL (G3) based on the CAL classification (minor vs major). Seven patients (28.0%) were managed nonoperatively and two (8.0%) underwent radiologic drainage of intraabdominal collections. The remaining 16 patients (64.0%) required surgical intervention. Of the 16 reoperated patients, 10 (56%) had an anastomosis takedown with an end stoma and 6 (44%) received a defunctioning stoma. The 90-d mortality rate was 0.8%, representing 3 patients with CAL.

Abdominal pain: Abdominal pain was markedly higher and persistent from POD3 onwards in G3 patients (Figure 2A). The AUROC for abdominal pain on POD3, POD4, and POD5 was 0.77, 0.84, and 0.83, respectively, as shown in Supplementary Table E (Supplementary material 2A) and Figure 3A. The predictive effect was higher on POD4 with an estimated AUROC of 0.84.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Distribution of rates of abdominal pain (A) and clinical condition (B). G1: No complications; G2: Complications not related to colorectal anastomotic leakage; G3: CAL. POD: Postoperative day.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Area under the receiver operating characteristic curve of colorectal anastomotic leakage for clinical criteria. A: Abdominal pain from postoperative day 1 to postoperative day 5; B: Clinical condition from postoperative day 1 to postoperative day 5. POD: Postoperative day.

Clinical condition: The clinical condition was worse in G3 compared with G2 patients, and it was significantly different after POD3 (P = 0.001). The overall postoperative trend was a declining clinical condition, as shown in Figure 2B. The AUROC for the clinical condition on POD3, POD4, and POD5 was 0.62, 0.81, and 0.90, respectively, as shown in Supplementary Table E (Supplementary material 2A) and Figure 3B. The prediction effect was higher on POD5 with an estimated AUROC of 0.90.

WBC count and ECC: During the first five POD, WBC in G3 patients was higher than that in patients without CAL and was significantly different on POD2, POD4, and POD5 (P = 0.01 for each day). On the other hand, ECC was lower in G3 patients and significantly different on POD1 and POD5 (P = 0.04 and P = 0.01, respectively), as presented in Supplementary Figures 1 and 2 (Supplementary material 2B). Overall, the postoperative course showed a sustained trend for both blood cell counts, except for ECC on POD5. The AUROC for WBC and ECC from POD1 to POD5 is presented in Supplementary Figures 3 and 4, respectively ( Supplementary material 2B). The predictive effects of blood cell count were better on POD5. On POD5, when ECC was greater than 250 cells/μL, the AUROC, SS, and SP were 0.70, 89.0%, and 43.0%, respectively, as shown in Table 6.

CRP, PCT, and CLP: The mean values of CRP, PCT, and CLP increased promptly after surgery in all groups. CRP decreased in G1 patients and remained elevated in patients with a complicated postoperative course, but was significantly higher than in G3 patients. On POD5, the mean CRP level in G3 patients was significantly higher than that in G1 patients (195.5 ± 139.9 mg/L vs 59.5 ± 43.4 mg/L; P < 0.00001) (Figure 4A). Patients with major CAL had a higher mean CRP level than those with minor CAL (251.45 mg/dL vs 107.64 mg/dL; P = 0.01) (Table 5). On POD3, POD4, and POD5, the overall diagnostic accuracy of CRP to detect CAL was expressed by an AUROC of 0.76, 0.76, and 0.81, respectively (Figure 5A). On POD5, the optimum cut-off value of 96.8 mg/L was estimated, resulting in an SS and SP of 78%, an NPV of 98%, and a PPV of 19% (Table 6).

Open in New Tab   Full Size Figure   Download Figure

Figure 4 C-reactive protein (A) and calprotectin (B) levels. Values are the mean ± SE. G1: No complications; G2: Complications not related to colorectal anastomotic leakage; G3: CAL; ▲: P statistically significant (P < 0.05).

Open in New Tab   Full Size Figure   Download Figure

Figure 5 Area under the receiver operating characteristic curve of colorectal anastomotic leakage. A: For C-reactive protein from postoperative day 1 to postoperative day 5; B: For calprotectin from postoperative day 1 to postoperative day 5; C: For calprotectin from postoperative day 3 to postoperative day 5; D: For C-reactive protein from postoperative day 3 to postoperative day 5. Left: All leaks; Right: Major leaks; POD: Postoperative day.

The PCT level tended to be stable from POD3 onwards. The mean values were higher in G3 patients than in patients without CAL, but without statistical significance [on POD5, 0.23 ± 0.08 ng/mL vs 0.22 ± 0.07 ng/mL; Supplementary Figure 5 (Supplementary material 2C)]. The AUROC on POD3, POD4, and POD5 was 0.57, 0.50, and 0.61, respectively, as shown in Supplementary Figure 6 (Supplementary material 2C). The best predictive effect was on POD5. When PCT was greater than 0.39 ng/mL, the SS and SP were 44.0% and 79.0%, respectively (Table 6).

In the first 5 POD, the mean CLP value tended to follow the pattern of CRP, although it was not as pronounced (Figure 4B). The mean CLP value was significantly higher in G3 patients from POD2 onwards. On POD3, the mean values of G1 vs G3 patients were 5.26 ± 3.58 μg/mL vs 11.52 ± 6.81 μg/mL (P < 0.00005). On POD3, POD4, and POD5, the CLP AUROC was 0.78, 0.67, and 0.65, respectively, as presented in Table 6 and Figure 5B. On POD3, a cut-off value of 6.57 μg/mL yielded a sensitivity of 71.0% and a specificity of 72.0% (Table 6).

Finally, when we analyzed the best predictors (CRP and CLP) for major CAL, the AUROC of CRP was 0.74 and 0.88 for POD3 and POD5, respectively. CLP was a better predictor of CAL than CRP at POD3, with an AUROC of 0.92 (Figure 5C and D).

Combination of biomarkers: Tables 7 and 8 present the AUROC of several possible classifiers of CAL, built with the Probit model, on POD3 and POD5, respectively. The combination of CRP and CLP on POD3 showed the best performance, with an AUROC of 0.82 (Table 7). Of note, on POD5, the combination of CRP and ECC also generated good predictive performance (AUROC = 0.81). However, with the aim of early CAL diagnosis, we chose the combination of CRP and CLP on POD3. Thereafter, we determined the probability of CAL, based on the computed equation P (CAL) = F [-3.0842 + (0.094 × CLP_D3) + (0.0059 × CRP_D3)], where F is the cumulative standard normal distribution. Applying Liu’s method, this classifier had an optimum cut-off point of 0.055, evidencing the existence of CAL above 0.055 on POD3, with an SS and SP of 86% and 75%, respectively. For hypothetical patient X on POD3 with CRP and CLP plasma levels of 137.4 mg/L and 8.75 μg/mL, respectively, the computed probability of CAL is high (score = 0.074). By adopting this classifier, the time to CAL diagnosis is estimated as 3.8 d [(0.86 × 3) + (0.14 × 9.0)], which represents a 5.2-d reduction compared with the baseline results.

One strength of this study was its prospective design and independent data collection model, which minimized observer bias. Second, it was one of the largest monocentric sample size published to date. Based on the recent meta-analyses of Waterland et al[46] van Helsdingen et al[47], only two monocentric prospective studies have enrolled more than 400 patients. Furthermore, we analyzed five biomarkers, including plasma CLP, which was first studied by Reisinger et al[18]. Third, we chose a comprehensive definition of CAL, recently defined by van Helsdingen et al[29] to include all patients with CAL, minimizing selection bias. We did not exclude minor CAL from the cohort, which also affected the predictive effect of the analyzed biomarkers. In addition, to keep the biomarkers optimum cut-off values in AUROC analysis both standardized and reproducible, we adopted Liu’s method. This method defines the optimum cut-off point as the point maximizing the product of SS and SP[48]. These reasons may explain some differences in biomarkers’ diagnostic accuracy in this study. Fourth, we tried to adapt the study protocol to daily practice, making its enforcement easier in the future. Hence, we included all patients undergoing colorectal resection, even those with a diverting ostomy. In addition, clinicians were not blinded to the daily biomarkers’ results and might use those data according to the study protocol. Finally, we proposed a predictive tool based on the combination of two biomarkers that improved CAL diagnostic accuracy. Adoption of this tool in daily practice might shorten the time to CAL diagnosis and management. Moreover, the data from this study provide information for the development of more complex mathematical predictive models, including machine learning methods.

This study had several limitations. First, the monocentric design may limit the external validity of the results. Second, our sample had some grade of heterogeneity, because the study population included benign and malignant disorders, elective and urgent procedures, and anastomosis within different levels of the colon and rectum. Third, we designed a phase I diagnostic study and estimated cut-off values for early CAL detection. However, we should change the direction of interpretation, running from the diagnostic test result toward the CAL diagnosis. To address this issue, we are performing a new multicentric prospective phase II diagnostic study, using the predictive tool and defined biomarkers cut-off values[49]. Fourth, plasma CLP measurement is expensive and these kits are not easily accessible in daily clinical practice. Finally, our study did not address the cost-effectiveness of biomarkers’ measurement. It is crucial to estimate the economic burden of CAL, including the cost related to a delayed diagnosis, the high rate of false positives, and unjustified reoperations or frequent readmissions.

Colorectal anastomotic leakage (CAL) is a major complication in abdominal surgery. Prompt diagnosis can reduce morbidity and mortality associated with this condition. Serum biomarkers have been proposed as predictors of CAL.

Biomarkers such as C-reactive protein (CRP) and white blood cell (WBC) count are frequently requested in the postoperative period of colorectal surgery. However, the usefulness of these and other biomarkers remains unclear.

To assess the role of CRP, WBC, eosinophil cell count, calprotectin (CLP), and procalcitonin in the first 5 postoperative days (PODs) after colorectal surgery, and thus, discuss in what order these biomarkers can be employed in clinical practice.

From March 2017 to August 2019, we measured and analyzed five serum biomarkers daily in 396 patients who underwent colorectal surgery. The area under the receiver operating characteristic curve, specificity, sensitivity, positive predictive value (PPV), and negative predictive value (NPV) were used to estimate the best predictive diagnostic performance.

CRP had an NPV of 98% on POD5. The combination of CLP and CRP measurement presented a high diagnostic accuracy (AUCROC = 0.82) on POD3. We identified a reduction of 5.2 d to the diagnosis of CAL.

The combination of CRP and CLP demonstrated good diagnostic accuracy. These tests can likely be used to reduce time to CAL detection.

Further studies should test a warning index score built from selected predictive variables as biomarkers CRP and CLP.