A propensity score matched study: Predictive signs of anastomotic leakage after gastric cancer surgery and the role of CT

Background: Anastomotic leakage is a life-threatening postoperative complication after gastric cancer surgery. Previous studies have not produced convincing results for the early diagnosis of anastomotic leakage. This study thoroughly investigated the clinical factors and postoperative computed tomography (CT) findings that could facilitate the early diagnosis of anastomotic leak. Methods: Gastric cancer patients who underwent curative gastrectomy and had a CT examination after surgery were included in this study. The authors identified a total of thirty-five cases of anastomotic leakages among 210 gastric cancer patients. Propensity score (PS) matching generated 70 eligible cases (35 cases of anastomotic leakage and 35 cases of no anastomotic leak) on the basis of age, body mass index (BMI), mode of surgery (open or laparoscopic), extent of resection (subtotal or total), and combined resection of adjacent organs. Univariate and multivariate analyses were used to identify the predictive postoperative variables. A nomogram was developed for prospective prediction. Results: Patients with anastomotic leaks were severely complicated by infections, abdominal haemorrhage, and impaired vital organ function (p<0.05). The reoperation rate was significantly higher in patients with anastomotic leakage (p<0.05). The median length of postoperative hospital stay was significantly longer and the overall expenditure was significantly higher in patients with an anastomotic leakage (p<0.001). Four patients died of postoperative complications after developing an anastomotic leakage. There were no deaths among the patients without anastomotic leakages. The logistic regression analysis revealed that the extraluminal gas at the anastomosis site (on CT examination), neutrophilia (NE [≥]85.8%) on postoperative day 1-3 (POD 3), fever (T[≥]38.5 C) on POD 4-7, and neutrophilia (NE [≥]78%) on POD 4-7 were the independent predictive factors for an anastomotic leakage (p<0.05). The nomogram showed that extraluminal gas at the anastomosis site was the most important sign of the anastomotic leak (100 points). Conclusions: On the postoperative CT examination, the patients with extraluminal gas at the anastomosis site are more likely to have an anastomotic leakage, especially in the patients with fever and neutrophilia during POD 4-7. The authors advocate for the routine use of postoperative CT after gastric cancer surgery.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Prevalence of postoperative complications and risk factors
Postoperative complications were quite prevalent after gastric cancer surgery, and the complication rate was almost equal between open and laparoscopic surgery [1].
However, severe complications occurred at a lower rate at high-volume centres than in low-volume centres [2]. Generally, the rate of anastomotic leakage has been reported to be below five percent, or even below two percent in the experienced centres of many Asian countries [1,3].
Nevertheless, anastomotic leaks are still considered severe postoperative complications that aggravate the condition of compromised patients, and the mortality rate of patients with anastomotic leakages is significantly higher than that of patients without anastomotic leakages [4][5][6]. Several scientific reports have explored predicting the postoperative complications of gastric cancer surgery; however, most of these reports were observational studies, and many risk factors were unavoidable in general practice [7][8][9][10][11][12]. For instance, surgeons cannot deny surgical treatment to elderly, diabetic or obese patients with gastric cancer. The overall results had a good consensus at high-volume centres in Asian countries [2]; however, these results are also not practical as we cannot restrict gastric cancer treatment to only Asian patients.

Radiological methods
Computed tomography (CT) scans have been used to detect anastomotic leaks in patients after oesophagectomy, but very few studies have reported on gastric cancer surgery, and the routine use of CT has been controversial [13][14]. No definitive All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04. 25.20080093 doi: medRxiv preprint 7 suggestions exist on whether a postoperative CT or abdominal X-ray with an oral contrast agent should be routinely performed for the early detection of anastomotic leakage [15][16]. Furthermore, the interpretation of CT findings is highly subjective; for example, clinicians and radiologists are still unsure whether the presence of free gas in the abdominal cavity is normal after gastrectomy, whether this free gas is common after laparoscopic surgeries, and what the presence of free gas in the abdominal wall after a laparoscopic surgery could indicate.
The authors considered all these issues and carefully analysed the data. Statistical analyse were performed for the potential postoperative signs of anastomotic leak, including patient body temperature, blood test results (neutrophils) and CT findings.
The statistical calculations were also used to create a nomogram that can be applied to identify compromised patients with anastomotic failure.

Methods
This is a retrospective study, and the primary inclusion criterion was gastric cancer patients who underwent curative gastrectomy and a CT examination after surgery. These were not the consecutive data of all gastric cancer patients who underwent surgery. The study endpoint was the presence of postoperative complications and any complications within one month of discharge from the hospital.
The authors collected all data by comprehensively reviewing the original records of all patients. Altogether, 221 patients were identified, and 11 patients with benign diseases were excluded. Finally, the authors included a total of 210 patients with gastric cancer diseases. All patients underwent curative gastrectomies with All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (Table 1). Hence, there were no significant differences in the clinical parameters between the two groups.
Univariate and multivariate analyses were used to identify the predictive postoperative factors. R language software created a nomogram for future applications.
In addition to the presence of extraluminal gastrointestinal fluid leakage, one of the All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. Nonparametric methods were used to test data with an abnormal distribution.
Statisticians performed PS matching with five variables (age, BMI, mode of surgery, extent of resection, and combined resection) and a match tolerance of 0.02 percent.
The receiver operator characteristic curve was used to find an optimal neutrophil count (NE) cut off value to correctly diagnosis anastomotic leakages. A chi-square test was used to compare the differences between the two groups; the statistician used Fisher's exact test where appropriate. Logistic regression was applied to identify the independent predictive factors for an anastomotic leakage. A p-value less than 0.05 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Types of complication
The authors compared the occurrence of other complications between patients with and without anastomotic leakage ( Table 2). Some low-incidence postoperative morbidity conditions (e.g., different types of infectious complications) were integrated to facilitate a useful statistical calculation. The reoperation rate was significantly higher in patients with anastomotic leakage than in those without an anastomotic leakage. The clinicians observed more haemorrhagic complications, infectious complications, and impaired vital organ function in patients who had an anastomotic failure than for those without an anastomotic leakage ( Table 2). Four patients died of postoperative complications after developing an anastomotic leakage, three of whom had undergone total gastrectomy with a Roux-en-Y anastomosis, and one patient underwent a Billroth I distal gastrectomy. In the distal gastrectomy case, the gastroduodenal anastomosis was the site of leakage; the patient had a severe intra-abdominal infection followed by an abdominal haemorrhage and shock. The patient underwent reoperation but later died 25 days after the first surgery. Among those who underwent total gastrectomy, one patient developed a leakage at the oesophagojejunal anastomosis; the patient underwent reoperation but later died 61 days after the first surgery. The leading cause of death was severe abdominal infection followed by shock, cardiac and respiratory failure and gastrointestinal bleeding. The All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (Table 3).
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04. 25.20080093 doi: medRxiv preprint 12 The surgeons did not find the exact location of the leak in seven patients, even after reoperation for four patients. This finding suggests that these patients potentially had a tiny anastomosis leakage, and severe inflammation at the surgical site created challenging conditions to determine the exact location of the anastomotic leakage.
Interestingly, 28 of the 35 cases patients with anastomotic leakage eventually had digestive content in the drainage tube, which suggests that an adequate draining system at the anastomosis site was crucial not only for diagnosis but also for treatment.

The doctors mainly treated 22 patients by continuous irrigation and avoided
reoperation. The other patients who did not have drainage in situ warranted reoperation or interventional therapy. Drainage was highly important for cases of duodenal stump fistula, as doctors managed all seven of these patients by adequate drainage, none of the patients needed reoperation (Table 3).

Economic burden
The postoperative length of hospital stay was significantly longer for patients with an anastomotic leakage than for those without an anastomotic leakage (p<0.001); the median number of postoperative days (PODs) was 18 days for patients with no anastomotic leak, but the length of stay was almost three times longer (50 days) in patients with an anastomotic leakage (Fig 1). The overall expenditure was significantly different between the two groups (p<0.001). The median total expenditure for patients with no leakage was only 64193.46 RMB (Chinese currency), but the expenditure was almost double (121167.12 RMB) for patients who had an anastomotic leakage (Fig 2). All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Furthermore, the multivariate analysis of overall factors revealed that blood neutrophilia (NE≥85.8%) on POD 3, fever (body temperature (T) ≥38.5°C) on POD 4-7 and suspicious findings on CT were independent predictive factors for an anastomotic leakage (p<0.005, Table 5). The Hosmer-Lemeshow goodness-of-fit test demonstrated a good fit (p>0.05), and the observed number of anastomotic leakage cases was closely aligned with the expected number predicted by the logistic regression model ( Table 6).

Investigation of the postoperative CT parameters
For a more scientific presentation, the authors further analysed different CT parameters. In the univariate analysis, the authors observed a significant association between anastomotic leakage and five CT variables, including pneumoperitoneum, pneumoseroperitoneum (intra-abdominal accumulation of mixed gas and fluid), All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  Table 7). The multivariate analysis of the CT parameters revealed that the accumulation of extraluminal gas at the anastomosis site and seroperitoneum were independent diagnostic parameters of a postoperative anastomotic leakage (p<0.05, Table 8).
Typical images of accumulation of extraluminal gas at different types of anastomosis sites were visible in CT scan (Fig.3, 4, 5). After logistic regression of all significant factors, the R language software created a nomogram with four independent factors, e.g., Extraluminal gas at the anastomosis site, POD 3 Neutrophilia (NE ≥85.8%), POD 4-7 Fever (T≥38.5° C), and POD 4-7 Neutrophilia (NE ≥78%). The presence of extraluminal gas at the anastomosis site was clearly the most critical CT finding ( Fig   6). This nomogram can be used to identify suspicious patients who need further investigations.

Complications
The authors did not calculate the overall complication rate, as this study did not use the consecutive data of all gastric cancer patients. Hence, the results would not represent the natural complication rates of gastric cancer surgery. The occurrence of thirty-five cases of anastomotic leakage among 210 cases is too high for gastric cancer surgery. Nevertheless, many of the postoperative complications were significantly different between the two groups. Hence, these data suggest that once a patient develops anastomotic dehiscence, he or she is likely to experience more severe All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

PS matching
For the analysis of any clinical condition with a low prevalence rate, a major hurdle is the statistical calculation. The analysis needs a substantial cohort to obtain statistically significant results. An anastomotic leakage rate of two percent corresponds to 98 out of 100 patients having a satisfactory recovery, regardless of the scenario. Therefore, PS matching is beneficial not only to standardize retrospective data as prospective data but also to facilitate a better comparison of clinical conditions with low occurrences. Many previous conventional studies have suggested that postoperative complications might be related to age, obesity, mode of surgery, and the extent of resection [7,9,10]. Therefore, the authors incorporated all these factors for PS matching. The authors did not find any other similar studies in previously published literature.

Predictive factors
The early diagnosis of anastomotic failure is crucial for preventing life-threatening postoperative complications of gastrectomy, but clinicians still do not have efficient standard techniques. In this study, univariate and multivariate logistic regression analyses identified simple but useful elements (body temperature, blood neutrophil count) that can detect anastomotic leakages and can be applied at any hospital with minimum resources. A nomogram could help clinicians identify an anastomotic leakage and subsequently decrease the overall burden of patients and hospital.

Problems with CT
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04. 25.20080093 doi: medRxiv preprint Many authors have advocated for barium swallow tests to diagnose suspicious cases of anastomotic leakages after gastrointestinal (GI) surgeries. Nevertheless, the generalized use of this examination is debatable [17][18][19]. Few studies have suggested postoperative CT after gastric cancer surgery, and there are different opinions on the use of oral contrast agents [20][21]. The authors of this study also found there was no apparent benefit of using oral contrast agents to detect anastomotic leakages, as none of the 27 patients who ingested an oral contrast agent 6 hours before the CT scan had an extraluminal contrast leak. This simple comparison showed a better diagnosis rate in patients who received oral contrast agents than in patients who received oral contrast agents (90% vs. 78% of true positive incidence).
However, there are still many unanswered questions, for instance, how much oral contrast agent is needed, what is the optimal concentration, what is the optimal timing to orally ingest the contrast agent? These negative results warrant further investigations about whether the timing of the contrast agent administration and CT examination was inappropriate, or if the concentration was simply too high or too low to be visible on the CT scan. These questions warrant a well-controlled future study regarding whether oral contrast agents are beneficial for diagnosing anastomotic leakages.
Similarly, assessing the discontinuity of an anastomosis was difficult on a CT scan with a slice of 5 mm. A complete circumferential scan of a circular stapled anastomosis was almost impossible to view in a single CT slice; additionally, many types of anastomosis use a linear stapler. Overall, finding a hand-sewn anastomosis All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04. 25.20080093 doi: medRxiv preprint the detection of compromised patients. Nevertheless, the results of this study identified some significant findings from the postoperative CT examinations, which were either unknown or not well-described in previous publications.

Blood routine and fever
Elevated white blood cell (WBC) and neutrophil (NE) counts are well-known laboratory findings for inflammation after surgery. However, to the best of our knowledge, no previous studies have focused on these simple parameters, probably due to these factors having a high sensitivity but low specificity for a particular type of complication, especially for low-incidence postoperative complications such as anastomotic leakages. Standardizing the timing of these findings was difficult in this retrospective analysis. The authors tried to overcome this challenge by creating a time range and noting which findings occurred the most on POD 1 to 3 and POD 4 to 7.
This time range was subjectively determined, and further research with a more reliable cut off should be conducted, which demands a large cohort.
Similarly, regarding body temperature, the authors did not find any previous studies that described whether the presence of a fever was associated with a specific postoperative complication of gastric cancer surgery. Doctors need concrete answers in terms of whether a specific body temperature at a specific time is a determinant for a particular postoperative complication, such as an anastomotic leakage. To perform an objective calculation, the authors divided the patients' body temperatures into two categories (e.g., T ≥38 and ≥38.5°C) and recorded the presence of these two levels of body temperature in two different intervals of time, e.g., during POD 1 to 3 and POD All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.

Other findings
There is still debate about whether abdominal drainage is necessary for patients undergoing gastric cancer surgery. The routine use of an abdominal drainage tube has not been suggested in general practice [22], and some researchers advocate for modified drainage procedures [23]. The authors observed that most patients were diagnosed with an anastomotic leakage through the presence of digestive content in the drainage tube and were mainly treated by continuous irrigation; other patients All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. Similarly, the opinions on using abdominal drain amylase to detect early anastomotic leakage in upper GI surgeries are mixed [24][25][26]. The authors noted elevated drain amylase in patients with anastomotic leak, especially for those with duodenal stump fistulas and leakages at the gastroduodenal anastomosis. Due to the absence of this test in many patients, the authors do not have a definitive answer about advocating for routine tests to evaluate drain amylase, A large-scale study is necessary to determine a better cut off drain amylase value for the diagnosis of anastomotic leakages, and perhaps different cut off values are needed for different types of anastomotic leakages.
For example, a much higher drainage amylase level is anticipated for leakages at the duodenal stump than at the oesophagojejunal anastomosis. The author also observed that a low level of postoperative serum albumin was more prevalent in patients with an anastomotic leakage than in patients without an anastomotic leakage, which was consistent with previous reports for lower GI surgeries [27]. Perhaps the combination of drain amylase and serum albumin levels along with the independent factors of this All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04. 25.20080093 doi: medRxiv preprint 21 study may predict the anastomotic leakage more accurately than each of the factors alone. A well-controlled prospective study is also necessary to investigate this hypothesis.

Conclusion
The (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04. 25.20080093 doi: medRxiv preprint  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  Fig. 2 Difference of expenditure between two groups All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  Fig. 3 Gatroduodenal anastomosis All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 29, 2020.  Fig. 6 Nomogram for the prediction of anastomotic leakage All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.