Introduction

Anastomotic leakages (AL) complicate between 3 and 17% of colorectal surgeries involving an anastomosis and are associated with morbidity, utilization of resources and even mortality [1,2,3,4]. AL can worsen long-term oncological outcomes, particularly local recurrence, following colorectal cancer resections [5].

Enhanced recovery after surgery (ERAS) protocols has improved perioperative outcomes of colorectal surgery and contributed to prompt recognition of postoperative complications including AL [6, 7]. Early diagnosis of AL can lead to timely treatment and potentially better outcomes [8, 9]. Conversely, reassurance of absence of AL may facilitate early discharge from hospital or reversal of defunctioning ileostomy [10,11,12].

C-reactive protein (CRP) is a serum protein which is elevated during an inflammatory or infective process and is elevated in AL [13,14,15,16,17]. Since the last meta-analysis of serum CRP in AL in 2013 [17], the number of studies has tripled. We amalgamate current data to understand CRP in the early diagnosis of AL and calculate a cut-off CRP level.

We aimed to conduct a meta-analysis to evaluate the association between serum CRP level and AL after colorectal surgery, and to determine a cut-off CRP value for AL.

Materials and methods

Design and study selection

Eligibility criteria, methodology, and investigated outcome parameters were defined in a review protocol. The methods of this study followed standards of Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) [18].

All comparative studies investigating serum CRP levels in patients with and without an AL following colorectal resections involving an anastomosis were included. An AL was defined as radiological or operative evidence of defect in the enteric wall at the site of the anastomosis. We considered AL treated conservatively or surgically. Adult male and female patients (18 years or greater) who had open, laparoscopic, or robotic colorectal resection for benign or malignant colorectal pathologies including inflammatory bowel disease, symptomatic diverticular disease, colorectal cancer, or other indication were considered. Elective and emergency colorectal procedures were considered.

Studies reporting postoperative serum CRP values for patients with and without ALs or septic complications were included. Studies not reporting AL outcomes separately from other septic complications were excluded.

Outcome measures

The primary outcome parameter was mean CRP level in mg/L on post-operative days (POD) one to seven.

Search strategy

Thesaurus headings, search operators, and limits were used to develop a search strategy and the search was carried out by two independent authors (DEY, EP) via MEDLINE, EMBASE, CINAHL, and CENTRAL databases (latest search 15 June 2020). The World Health Organization International Clinical Trials Registry http://apps.who.int/trialsearch/, ClinicalTrials.govhttp://clinicaltrials.gov/ and ISRCTN Register http://www.isrctn.com/ were queried for unfinished or unpublished studies. The search terms and strategy are in Appendix 1.

Selection of studies

Two reviewers (DEY, EP) independently executed a preliminary review of titles and abstracts identified through the literature search. Full-text analysis of remaining studies was undertaken and data extraction of studies meeting our inclusion criteria was carried out. Discrepancies were discussed with a third author (SH).

Data extraction and management

An electronic data extraction spreadsheet was prepared in accordance with Cochrane’s recommendations for intervention reviews. The reviewers independently extracted the following data from the studies:

  • Study-related data (first author, publication year, country of research, journal of publication, study design, surgical procedure, surgical approach, and sample sizes)

  • Demographic and clinical information (age, gender, body mass index, use of neoadjuvant radiotherapy, smoking status, cancer staging, level of anastomosis, site of anastomosis)

  • Outcome data

Discrepancies discussed with another author (SH).

Assessment of risk of bias

Two authors independently assessed the methodological quality and risk of bias (DY, EP) using the Newcastle-Ottawa scale (NOS) [19]. The NOS allows authors to evaluate observational studies, specifically considering the method of study group selection, comparability of the groups, and determination of the outcome. The highest score (nine points) denotes lowest risk; moderate risk scores seven or eight, while a high risk of bias would fetch six points. Disagreements were adjudicated by a third author (SH).

Summary measures and synthesis

The primary outcome was mean serum CRP measurements. Thus, mean difference (MD) was calculated between AL and non-AL. Where mean values were not available, the method described by Hozo et al. was used to estimate mean and standard deviation (SD) based on median and interquartile range (IQR) values [20].

The unit of analysis was the individual patient. Where available, attrition and other missing data was recorded. Authors were contacted where information for our outcome was not reported. Our calculations followed the intention-to-treat principle.

One author (DY) used Review Manager 5.3 software to perform the meta-analysis [21]. The calculations were independently analysed by another author (SH). Random-effects modelling was used for analysis. Forest plots with 95% confidence intervals (CI) were used to display the results of each of the calculations.

Cochran Q test (Χ2) was used to assess heterogeneity between studies. To quantify heterogeneity, I2 values were calculated. An I2 value of less than 50% suggests heterogeneity may not be important in this analysis; between 50 and 75% suggests moderate heterogeneity and between 75 and 100%, there may be substantial heterogeneity. Funnel plots were constructed to screen for publication bias where more than ten studies were available for any single outcome.

Leave-one-out sensitivity analysis was performed to gauge the influence of each study on overall effect size and heterogeneity.

For the secondary objective of this study, we performed a ROC curve analysis using MedCalc 13.0 software. We used the method described by DeLong et al. [22] to calculate standard error of the area under the curve (AUC) and an exact Binomial Confidence Interval for the AUC. We calculated associated sensitivity and specificity for all possible threshold values of CRP level and determined the optimal criterion value as cut-off value of CRP for an AL.

The method described by DeLong et al. [22] was used to analyse the ROC curves. MedCalc 13.0 software was used to determine the standard error of the Area Under the Curve (AUC) and to calculate an exact Binomial Confidence Interval for the this. For each threshold value of CRP level, sensitivity and specificity were calculated to understand the best cut-off value for CRP in AL.

Results

The literature search strategy resulted in 1102 articles (Fig. 1). A total of 1008 articles were excluded as they were irrelevant to our research question. Ninety-four potentially eligible studies were further evaluated of which 71 studies were excluded: 38 did not provide serum CRP for AL patients, 10 were review articles, 7 were letters to editor, 6 did not define CRP values (instead utilizing ratios or other inflammatory markers), 3 stated pre-operative CRP values, 3 defined a CRP level as dichotomous with variable cut-off points, 3 reported on the same data set, 2 did not provide numerical data for analysis, and the remaining 2 did not have full text available. Therefore, 23 comparative studies were deemed appropriate for inclusion (Fig. 1). They were all observational studies, with twenty prospective cohort, two retrospective cohort, and one retrospective case-matched cohort comparison study reporting a combined total of 6647 patients who had colorectal resections with primary anastomosis, amongst whom 482 had AL (Table 1) [3, 4, 14, 15, 23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41]. Table 1 summarizes data for the included studies (country of origin, journal of publication, study design). Table 2 shows the characteristics of the study populations. All patients underwent either emergency or elective laparoscopic, robotic, or open colorectal surgery for cancer, diverticular disease or inflammatory bowel disease or other indication (Table 2). The pooled mean time to diagnosis of AL was 7.70 ± 1.91 days.

Fig. 1
figure 1

PRISMA flow chart

Full size image
Table 1 Baseline characteristics of the included studies
Full size table
Table 2 Baseline characteristics of included population. NR not reported. Mean age in years. SD standard deviation. BMI body mass index (kg/m2). XRT: preoperative radiotherapy. ASA III/IV: American Society of Anaesthesiologists grade III or IV. Level of anastomosis is denoted in centimetres. *: values provided are median (range). IQR: Interquartile range
Full size table

Methodological appraisal

Table 3 summarizes the NOS methodological assessment of the studies. Twelve studies had low risk of bias and 11 studies had moderate risk of bias.

Table 3 Risk of bias assessment based on Newcastle-Ottawa Scale for non-randomized studies [Wells 2020]
Full size table

Outcome synthesis

Figures 2 and 3 summarize the results of the outcome calculations.

Fig. 2
figure 2figure 2

Forest plots of comparison of CRP on post-operative days. a Day 1, b Day 2, c Day 3, d Day 4, e Day 5, f Day 6, and g Day 7. The solid squares denote the mean difference (MD). The horizontal lines represent the 95% confidence intervals (CIs), and the diamond denotes the pooled effect size. M-H, Mantel Haenszel test

Full size image
Fig. 3
figure 3

Funnel plots of comparison of serum CRP on post-operative days. a Day 1, b Day 2, c Day 3, and d Day 4

Full size image

CRP on POD 1

Fourteen studies (2830 patients) were included. Mean serum CRP levels in the AL and no AL groups were 114.45 ± 32.51 and 95.82 ± 29.48, respectively. AL was associated with higher mean CRP level when compared with no AL (MD 15.19, 95% CI 5.88–24.50, p = 0.001). Heterogeneity between studies was moderate (I2 = 67%, p = 0.0002) (Fig. 2a).

CRP on POD 2

Fourteen studies (4559 patients) were included. Mean serum CRP level in AL group was 201.55 ± 29.90 and 145.36 ± 30.67 in the no AL group. AL was associated with higher mean CRP compared to no AL (MD 51.98, 95% CI 37.36–66.60, p < 0.00001). Heterogeneity between studies was significant (I2 = 77%, p < 0.00001) (Fig. 2b).

CRP on POD 3

Twenty studies (5598 patients) were included. Mean serum CRP level in AL and no AL groups were 224.09 ± 51.38 and 122.78 ± 32.05, respectively. AL was associated with higher mean CRP level on POD 3 when compared with no AL (MD 96.92, 95% CI 67.96–125.89, p < 0.00001). There was significant heterogeneity between studies (I2 = 91%, p < 0.00001) (Fig. 2c).

CRP on POD 4

Eleven studies (2955 patients) were included. Mean CRP level in the AL group was 203.84 ± 38.40 whereas it was 104.58 ± 17.06 in the group without AL. AL was associated with higher mean CRP than non-AL (MD 93.15, 95% CI 69.47–116.84, p < 0.00001). There was significant heterogeneity between studies (I2 = 86%, p < 0.00001) (Fig. 2d).

CRP on POD 5

Seven studies (1838 patients) were included. Mean serum CRP level in the AL group was 187.49 ± 35.20 while it was 65.31 ± 23.76 in the no AL group. AL was associated with higher mean CRP level on POD 5 when compared with no AL group (MD 112.10, 95% CI 89.74–134.45, p < 0.00001). There was significant heterogeneity between studies (I2 = 58%, p < 0.00001) (Fig. 2e).

CRP on POD 6

Nine studies (3473 patients) were included. Mean serum CRP level in the AL group was 176.9 ± 32.62 while it was 70.59 ± 20.04 in the no AL group. AL was associated with higher mean CRP level than non-AL (MD 98.38, 95% CI 80.29–116.46, p < 0.00001). There was moderate heterogeneity between studies (I2 = 53%, p < 0.00001) (Fig. 2f).

CRP on POD 7

Eight studies (2143 patients) were included. Mean serum CRP level in AL and no AL groups were 189.29 ± 25.31 and 77.73 ± 23.79, respectively. AL was associated with higher mean CRP level on POD 7 when compared with no AL group (MD 106.41, 95% CI 75.48–137.35, p < 0.00001). There was significant heterogeneity between studies (I2 = 80%, p < 0.00001) (Fig. 2g).

Sensitivity analysis

Leave-one-out sensitivity analysis did not demonstrate any difference in the direction of pooled effect size and no particular study caused skewing. Funnel plots for POD 1 through 4 did not suggest publication bias (Fig. 3).

ROC curve analysis

Outcomes are presented in Fig. 4 and Table 4.

Fig. 4
figure 4figure 4figure 4figure 4

Receiver operator curves (ROC) for serum C-reactive protein (mg/l) on different post-operative days (POD). a POD1 at cut-off value 110; b POD2 at cut-off value 184; c POD3 at cut-off value 148; d POD4 at cut-off value 123; e POD5 at cut-off value 115; f POD6 at cut-off value 105

Full size image
Table 4 Summary of pooled AUC analysis results for serum CRP levels by post-operative day. AUC area under the curve; CI confidence interval
Full size table

CRP on POD 1

A cut-off CRP level of 110 was shown through ROC analysis to have a sensitivity of 60% (95% CI 32–84%) and specificity of 73% (95% CI 45–92%). The AUC was 0.66 (95% CI 0.47–0.82, P = 0.1110).

CRP on POD 2

A cut-off CRP level of 184 was shown through ROC analysis to have a sensitivity of 71% (95% CI 42–92%) and specificity of 100% (95% CI 77–100%). AUC was 0.91 (95% CI 0.74–0.98, P < 0.0001).

CRP on POD 3

A cut-off CRP level of 148 was shown through ROC analysis to have a sensitivity of 95% (95% CI 75–99%) and specificity of 95% (95% CI 75–95%). AUC was 0.95 (95% CI 0.83–0.99, P < 0.0001).

CRP on POD 4

A cut-off CRP level of 123 was shown through ROC analysis to have a sensitivity of 100% (95% CI 72.0–100%) and specificity of 100% (95% CI 72.0%–100%). AUC was 1.00 (95% CI 0.85–1.00, P < 0.0001).

CRP on POD 5

A cut-off CRP level of 115 was shown through ROC analysis to have a sensitivity of 100% (95% CI 63–100%) and specificity of 100% (95% CI 63–100%). AUC was 1.00 (95% CI 0.79–1.00, P < 0.0001).

CRP on POD 6

A cut-off CRP level of 105 was shown through ROC analysis to have a sensitivity of 100% (95% CI 66–100%) and specificity of 100% (95% CI 66–100%). AUC was 1.00 (95% CI 0.82–1.00, P < 0.0001).

CRP on POD 7

A cut-off CRP level of 96 was shown through ROC analysis to have a sensitivity of 100% (95% CI 59–100%) and specificity of 100% (95% CI 59–100%). AUC was 1.00 (95% CI 0.77–1.00, P < 0.0001).

Discussion

After colorectal surgery, AL can worsen patient morbidity and mortality outcomes [1,2,3,4, 42]. Post-operative serum CRP level can be utilized to predict occurrence of an AL following colorectal resection with primary anastomosis. [8, 43]. We performed a meta-analysis of 23 comparative studies reporting a total of 6647 patients undergoing colorectal resections and primary anastomoses, of whom 482 had ALs. Meta-analysis showed AL was associated with significantly higher serum CRP level on POD 1 through 7 compared to patients who did not have AL. The heterogeneity between studies was moderate in the analysis of CRP level on POD 1, 5, and 6 indicating variable reporting by included studies on these POD. Heterogeneity was high regarding analysis of CRP level on POD 2, 3, 4, and 7 indicating our findings on these days may be less robust.

Our ROC curve analysis determined a threshold CRP level of 148 mg/l on POD 3 with sensitivity and specificity of 95%, and cut-off CRP levels of 123 mg/l on day 4, 115 mg/l on day 5, 105 mg/l on day 6, and 96 mg/l on day 7 for AL with sensitivity and specificity of 100%. We believe our meta-analysis is currently the most comprehensive meta-analysis of literature with inclusion of nearly 7000 patients pooled from 23 studies and i++ndependent MD analyses of CRP levels on 7 consecutive PODs and determined cut-off points on each day. We have demonstrated sensitivity and specificity of 100% associated with our cut-off values on POD 4 to 7 which are higher than those reported by previous meta-analyses Previous meta-analyses have investigated the utility of serum CRP in diagnosing either a post-operative infectious complication or AL. Singh et al. conducted a meta-analysis of 2483 patients who had colorectal resections across seven studies, and the authors found the most sensitive and specific CRP level cut-off values were 172, 124, and 144 mg/L on POD 3, 4 and 5, respectively, with pooled sensitivities of 76%, 79%, and 72% and pooled specificity of 76%, 70%, and 79%, respectively [17]. Our provided cut-off values are nearly comparable with findings of Singh et al. [17] and our higher sensitivity and specificities on the aforementioned PODs further confirms the robustness of these cut-off CRP values. Adamina et al. calculated pooled ROCs and found the best sensitivity and specificity profile of CRP on POD 4 cut-off of 96 mg/L (sensitivity 76%, specificity 61%), but the study was hampered by heterogeneity of the study populations, with different cut-offs for different types of operations (POD 4 cut-off 123 mg/L for open colonic cancer resection (sensitivity 68%, specificity 75%)) [44]. In 2015, Warschow et al. presented a meta-analysis of 1832 patients across six studies and determined the best specificity and sensitivity were on POD 4 with a cut-off level of 135 mg/L, which demonstrated an odds ratio of 11.7 against those who did not have infectious complications. However, a sub-group analysis for AL was not done [45]. Although their cut-off value was lower than our cut-off value, demonstration of the highest sensitivity and specificity on POD4 is consistent with our findings. Gans et al. analyzed the post-operative CRP in 2215 patients who had abdominal surgery; their meta-analysis calculated the threshold CRP on POD 3 of 159 mg/L provided the best sensitivity and specificity (77% sensitivity and 77% specificity) for post-operative infectious complications [46]. Cousins et al. performed a meta-analysis encompassing 2692 patients across 11 studies and demonstrated a cut-off CRP of 130 mg/L or less on POD 3 to have a pooled negative predictive value of 96.7% [47]. Considering that our sample size is much larger in comparison to previous meta-analysis, our findings are much less susceptible to type 2 error. Therefore, we encourage use of our reported cut-off values in prediction of AL.

A number of studies utilized ROC to determine a cut-off CRP level on POD 3 for AL and reported cut-off CRP levels ranging from 149 to 245 mg/L [15, 36,37,38]. Our determined value of POD 3 falls within those reported ranges. Our cut-off value on POD 4 is similar to Ortega-Deballon et al. who found a CRP level of 125 mg/L on POD 4 was the best cut-off point for AL [40]. However, our threshold value on POD 5 is lower than what reported by Reynolds et al. who determined a cut-off point of 132 mg/L on POD 5 [36].

Time to diagnosis of AL in the literature varies, but is typically reported as between seven and 10 days after operation [28]. In our analysis, pooled mean time to diagnosis of AL was 7.70 days ±1.91, with some citing diagnosis as early as 1 day and some as late as 30 days [28, 29]. Thus, in terms of a clinical application of this study, our cut-off levels would still potentially give a diagnostic advantage if CRP level was used as a cue towards further investigations to diagnose AL or reassurance to facilitate earlier discharge.

Randomized controlled trials in the context of AL, a postoperative outcome as compared to an intervention, is not possible. Therefore, the current study represents the best possible available evidence (level 2). Nevertheless, future studies are required to address shortcomings of available evidence. The included studies did not report use of preoperative radiotherapy, level of anastomosis, height of anastomosis, or whether the anastomosis was hand-sewn or stapled. We were therefore unable to consider our outcomes in relation to these potential confounders. Moreover, we were not able to analyse our findings with respect to other important confounder such as emergency or elective nature of surgery, benign or malignant pathology, or the presence of sepsis in the initial operation which can potentially have independent impact on the outcomes. Considering the findings of our study, we encourage use of our cut-off CRP values on POD 4 through 7 as a decision-making tool to predict AL in patients with primary anastomoses after colorectal surgery. The cut-off CRP values, albeit 100% sensitive and specific, warn of the presence of AL, but do not diagnostic in themselves.

Any interpretation of these results should be tempered by the limitations of the study. The studies included were all observational, which are liable to selection bias. Many baseline characteristics of study populations were not reported by the included studies. Twelve studies had moderate risk of bias. Some studies reported their data using median and interquartile range (IQR) or total range and an estimation of mean and standard deviation were calculated using an equation described by Hozo et al. [20], which is a potential source of bias.

Conclusions

This meta-analysis demonstrated AL is associated with significantly higher serum CRP levels on POD 1 through 7 compared with those with no AL after colorectal surgery. Considering the sensitivity and specificity of our determined cut-off CRP levels (100%), we do not hesitate to recommend use of our cut-off CRP levels on POD 4 through 7 to predict AL in order to allow prompt investigation and treatment or reassurance. Future studies should report the outcomes with respect to use of preoperative radiotherapy, level of anastomosis, height of anastomosis, or comparing hand-sewn and stapled anastomoses.