Less is more—the best test for anastomotic leaks in rectal cancer patients prior to ileostomy reversal

Preliminary research was started on October 1, 2018. Formal literature search was done until December 31, 2018. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2009) [6], an a priori protocol for this review was constructed and registered on January 10, 2019 in the PROSPERO database, an international prospective register of systematic reviews. It is accessible online via https://​www.​crd.​york.​ac.​uk/​prospero/​display_​record.​php?​ID=​CRD42019107771 under register number CRD42019107771.

Prospective and retrospective clinical trials of cross-sectional, cohort, or case-control design were considered for this diagnostic test accuracy review. No language restrictions were applied. The reviewed setting is the diagnostic process prior to ileostomy reversal, for detection of asymptomatic anastomotic leaks after mid to low rectal resection and formation of a temporary ileostomy for rectal cancer. Preliminary searches had shown a scarcity of studies solely comprised of rectal cancer patients. Thus, studies with mixed underlying pathologies were included if the majority of the cohort (> 50%) consisted of rectal cancer patients. Types of anastomoses are end-to-end, side-to-end, and colonic J-pouch. Eligible index tests for assessment of anastomotic integrity were antegrade or retrograde contrast enema, digital rectal examination, and endoscopic imaging such as flexible sigmoidoscopy or rigid proctoscopy. The target condition was asymptomatic anastomotic leak. Outcome after ileostomy reversal serves as reference standard, preferably supported by other factors such as laboratory findings or further testing. Changes in clinical management alone, such as a delay in the reversal of the ileostomy in a non-blinded clinical study, were considered unfavorable reference standards by the authors of this review, as those characteristics are heavily influenced by the results of the index tests and thus are prone to bias. Studies reporting only whether patients had ileostomy reversed, without postoperative follow-up on the outcome, were not deemed eligible for this review.

A search of the electronic databases MEDLINE, Cochrane Library, Google Scholar, clinicaltrials.gov, CINAHL, and Web of Science, as well as cross search of references of relevant articles, was performed by two authors (SL, SE) in cooperation with an institutional database researcher. Data from an additional primary study, recently conducted by authors of this review, was included prior to its publication [7].

As an example, the search strategy used for PubMed (MEDLINE) is presented in Table 1. Advanced search options including synonyms, truncations, and combinations were used. Search results were checked for inclusion of all literature that was found in preliminary searches.

Two review authors (SL and SE) independently assessed the search results for eligibility. After removal of duplicates, the results were screened by title and abstract. Eligible studies were then checked by full text for inclusion in the review and meta-analysis. Definitive inclusion was then decided by discussion (SL, SE, and FH).

Two review authors (SL and SE) independently extracted data using a data extraction form, tailored to this review question. A third review author (FH) evaluated any discrepant judgments. Data extraction of true positive, false positive, false negative, and true negative values to generate a 2 × 2 discrepancy table was performed for each study.

Comparison of the index tests by analyzing differences in their respective HSROC curve was the primary aim of this meta-analysis. Investigating the influence of methodological quality was the secondary aim. For the calculation of the HSROC model parameters, the bivariate model parameters and the confidence and prediction regions of the SAS macro MetaDAS (version 1.3), provided by the Cochrane Collaboration [8], were used. This analysis was performed with SAS software, release 9.4 (SAS Institute Inc., Cary, NC, USA). Where HSROC parameters could not be calculated, informal analysis of paired data was performed. For this, index test results of comparative studies are plotted in the same diagram and connected with a dotted line to aid visual interpretation. Further comments on the statistical analysis are provided as supplementary material [Addendum 1].

The risk of bias in individual studies was assessed using QUADAS-2 [9] by two review authors (SL and SE). The tool content was tailored to the question of this systematic review [Addendum 2]. Differences were resolved by discussion and consensus between the review authors. All signaling questions that were changed, added, or omitted are listed in Table 2. For subgroup analysis, number of domains marked with high risk were calculated for each study, and studies were accordingly split into two as equal as possible sized groups.

Results of the QUADAS-2 assessment for each included study are shown in Fig. 1. The overall methodological quality of studies using CE as index test was poorer than that of studies also assessing DRE or EP.

As predicted by preliminary searches, there was a scarcity of studies that comprised solely of rectal cancer patients. Seven of the included 13 studies had patients with other underlying pathologies, such as other malignant dieseases, inflammatory bowel disease, or diverticulitis. However, of the 4 largest included studies with more than 200 patients each, 2 exclusively consisted of rectal cancer patients, and of the total 1903 patients in all included studies, 1619 (85%) were rectal cancer patients.

For CE, an HSROC curve, representing sensitivity in relation to specificity, was calculated (Fig. 2). Too few studies were eligible to calculate HSROC parameters for EP and DRE. 

There were 8 studies with one or less domains marked as high, thus comprising the lower risk of bias group, and 5 studies with two or more domains marked as high, thus comprising the higher risk of bias group. Subgroup analysis showed the overestimation of CE accuracy measures in the group with a higher risk of bias. There was higher sensitivity for equal specificity and vice versa (Fig. 3).

Since HSROC parameters for EP could not be calculated, informal analysis by paired data was performed (Fig. 4). This showed equal or superior accuracy measures of EP in all studies. There were 4 comparative studies [4, 7, 12, 16], but for one (Shalabi 2016) sensitivity could not be calculated due to a 0% incidence. Specificity of EP was superior to CE. In all assessed cases, no leak found by CE was overseen by EP.

HSROC parameters for DRE could also not be calculated; thus, again paired data were used to perform informal analysis (Fig. 5). In all studies, there were superior or equal accuracy measures of DRE compared to CE. For 2 (MacLeod 2004; Shalabi 2016) out of the 5 available studies [4, 12, 16, 19, 20] sensitivity could not be calculated due to a 0% incidence. In all assessed cases, no leak found by CE was overseen by DRE.

There were 3 studies available [4, 12, 16]. Sensitivity could not be calculated in one study due to a 0% incidence (Shalabi 2016). Paired data analysis of DRE and EP showed concordance of all findings and thus equal sensitivity and specificity (Fig. 6).

All diagnostic tools have unsatisfactory sensitivity in routine testing, and prevalence of asymptomatic anastomotic leakage is relatively low. To assess whether routine imaging of patients who had an uneventful postoperative course is useful, or if diagnostic tools should be reserved for those patients who developed anastomotic leakage after the initial operation, studies providing individual data on those two groups were assessed. There were only 3 available studies, with a total of 518 eligible patients without anastomotic leakage following rectal resection and 103 patients with previous anastomotic leak. Only one of those studies had intended a direct comparative assessment of these groups. Mean prevalence of asymptomatic anastomotic leak prior to ileostomy reversal in the group without previous leak was 0.5% (± 0.8%); thus, approximately 200 patients have to be tested to find one leak. In contrast, in the group that developed a leak after the initial operation, 16.4% (± 8.0%) showed an asymptomatic leak prior to ileostomy reversal. Here, only approximately six patients have to be tested in order to find an asymptomatic anastomotic leak.

This review and meta-analysis aimed at comparing the index tests after resection of rectal cancer. However, compromises had to be made to aquire sufficient data for statistical analysis. Overall, rectal cancer patients comprised 85% of included patients. Furthermore, there were too few studies assessing EP and DRE to fulfill the primary objective of the meta-analysis, a formal comparison of the index tests. Overall, methodological quality of the eligible studies was moderate and most studies were retrospective. Blinded prospective clinical trials are completely missing. The overall evidence on which this systematic review and meta-analyis is based is fragile and conclusions drawn from it must be well-considered. The primary aim, to deduct an evidence-based algorithm for the diagnostic process, could not be fulfilled.

The collected evidence clearly points towards the omission of CE wherever an EP is available, since no additional information is gained. Of all 3 investigated tests it is also the one most associated with patient discomfort. Additionally, radiation exposure must be considered. There are limitiations to the evidence, as mentioned above. To gain stronger evidence, prospective and larger studies are needed. In our opinion, although desirable, it appears unlikely that those studies will be made. Thus, a pragmatic synopsis of existing evidence, patient comfort, and economic and risk-benefit considerations must be made. Another use for CE, the prediction of fecal incontinence, has been shown to be of little value [17, 34]. Although knowledge concerning diagnostic procedures for the detection of anastomotic strictures is very limited, in our experience this condition can be sufficiently assessed by EP, very low anastomoses even by DRE.

Since the use of CE is not supported by evidence, it may be omitted as a routine screening test for asymptomatic anastomotic leak prior to ileostomy reversal in rectal cancer patients when EP and DRE are available.