A novel nomogram to predict the risk of anastomotic leakage in patients after oesophagectomy

Anastomotic leakage (AL) is one of the most common serious complications after oesophageal surgery and causes prolonged hospital stay, intensive care unit (ICU) admission, and even mortality [1‐5]. The incidence of AL is approximately 6–24% in patients undergoing oesophagogastrostomy [2, 6], with an increased postoperative mortality rate of 7.2–18.2% compared with that of 3.1–6.2% in patients without leakage. The cost burden was two-times higher for high-grade leakages in an analysis of cervical oesophagogastric anastomotic leaks after transhiatal oesophagectomy [7]. The early detection of AL is crucial since delayed treatment is associated with significant morbidity, a prolonged hospital stay, and mortality. The common risk factors for postoperative AL are generally believed to be Sex, age, and the general condition of the patient (long-term malnutrition in patients with oesophageal cancer and a state of negative nitrogen balance, which is often accompanied by chronic wasting diseases such as anaemia and hypoproteinaemia that arise from the body’s poor tolerance to surgery). Moreover, the anastomotic type, anastomotic approach, anastomotic site, anastomotic tension, blood supply to the anastomotic region, intrathoracic infection, and postoperative digestive tract emptying are also related to AL [8]. Since there are many relevant risk factors, the use of accurate predictive tools and early intervention may be the most effective preventive actions for postoperative AL. Although relevant guidelines for this research area exist both internationally and domestically, there is a lack of validated tools based on national characteristics. This study aimed to develop a simple and effective prediction tool to perioperatively estimate the risk of AL for oesophageal cancer patients using readily available characteristics at the start of therapy.

The current study developed a novel tool to predict the risk of AL after oesophageal surgery based on 2 years of data from indigenous Chinese patients. On the one hand, the prediction of the model focused on selecting predictors. When determining variables, the complexity of the model should be adjusted to achieve a balance between overfitting and insufficient fitting. LASSO regression analysis was used instead of univariate analysis to find independent risk factors. We believe that principal component analysis is only applicable to samples with few dimensions. LASSO is another data dimensionality reduction method, and it is suitable not only for linear data but also nonlinear data. LASSO can directly compress insignificant variable coefficients to 0 [15]. On the other hand, the accuracy and consistency of the model were evaluated in the trial, and the internal verification results showed that the nomogram can be widely used for AL prediction.

Two previous reports have made similar attempts to identify the risk of AL. One study was conducted by Haga Y et al. [16] in 2011, who created a prediction scoring system, “Estimation of Physiologic Ability and Surgical Stress” (E-PASS), to predict the postoperative morbidity and mortality of elective gastrointestinal surgery. The study included nine variables to calculate a numerical score. However, their research included various kinds of gastrointestinal surgery. There were only 292 oesophageal surgeries among the 6005 cases (4.8%), which may not fully reflect AL after oesophageal surgery. The other study was conducted in 2012 by Noble F et al. [17], who combined blood-borne inflammatory markers, including postoperative CRP levels, white cell counts and albumin levels, as predictors of AL and major complications. They reported a sensitivity of 95%, a specificity of 49% and a diagnostic accuracy of 0.801. The same nomogram was recently validated by Bundred J. et al., who reported an accuracy of 0.77 [18]. However, the three indicators used by Nobel F and Bundred J were laboratory indicators on the 4th day after surgery, which may be closely related to the anastomotic fistula itself. Their results can be used well for the treatment 4 days after surgery.

Previous studies have mainly focused on postoperative laboratory biomarker data, which is quite different from our study. The main advantage of the current study is that our nomogram is easier and more intuitive than existing methods, and our nomogram includes different parameters to consider both pre- and intraoperative factors. The current scoring tool is able to clarify the combination of surgical factors and patient factors by focusing on active smoking, CRP levels and intraoperative parameters, such as anastomotic type, reconstruction route and the presence of intraoperative cardiac arrhythmia. In addition, of the 45 empirical clinical parameters, 7 parameters with significant effects in the multivariate analysis were selected, and the weighting of each parameter was significant, which could reflect the significant influence of these factors on the predicted value.

In this study, approximately 24% of patients experienced AL after surgery. In a previous study, David T et al. [19] showed that a large number of preoperative comorbidities, advanced pathologic stage, postoperative arrhythmia, history of oesophagogastric surgeries, and active smoking history were risk factors for developing cervical oesophagogastric ALs, while a side-to-side stapled cervical oesophagogastric anastomosis was a protective factor [19]. A meta-analysis revealed that diabetes, preoperative serum albumin < 35 g/L, respiratory diseases, hypertension, preoperative neoadjuvant radiotherapy, stage III and IV oesophageal cancer, manual anastomosis and posterior sternal neck anastomosis were risk factors for AL after oesophageal cancer surgery [5].

We agree that patients with lower oesophageal carcinoma who received nCRT cannot be considered in the same way as patients treated for cervical cancer when analysing AL. However, we did not find a statistically significant difference based on the current data when including the factors of nCRT treatment and the location of the tumour (see Table 1). There is still some controversy about nCRT and AL. Although Briel et al. [20] reported that oesophagocolic anastomoses after nCRT were a risk factor for AL results from other studies did not support this finding [21‐23].

Despite the fact that 37.2% of cases were T1-3N1-2M0 and T4aN0 − 1 M0, which were indications for nCRT according to the 2018 Chinese guidelines for diagnosis and treatment of oesophageal carcinoma [24], only 12.4% received nCRT because the use of nCRT before surgery in patients with locally advanced oesophageal cancer has only been slowly accepted in the last two years in China. Previously, it was believed that nCRT had a significant survival benefit for patients with locally advanced oesophageal cancer. However, nCRT was not recommended for patients with early oesophageal cancer because of its high surgical resection rate, and the addition of nCRT does not improve the R0 resection rate and increases the risk of postoperative death. In addition, even though nCRT appears to be associated with a higher R0 resection rate, there are still people who are reluctant to use preoperative nCRT because they are afraid of missing the appropriate surgical timing due to concerns about disease progression.

Urschel JD [25] and Pierie JP [26] suggested that an insufficient blood supply to the anastomotic stoma was the main risk factor for AL. Cooke found that postoperative arrhythmias that caused a low-flow state predicted leakage based on multivariate analysis [19]. Among our patients, those who experienced intraoperative arrhythmia were predisposed to AL, possibly for the same reason. The surgical approach is divided into a retrosternal reconstruction route and mediastinal reconstruction route. From an anatomical point of view, the length of the incision is longer with the posterior sternal approach than with the mediastinal approach, which inevitably leads to oesophageal free ends. An anastomotic route that is too long could cause excessive anastomotic tension, lack of blood supply to the tissue, and an elevated incidence of AL. Other potential risk factors, including smoking history and diabetic microangiopathy, in theory, may reduce microperfusion of tissues [19, 27]. The probability of postoperative AL is higher in men than in women; however, this difference is not clinically significant, which is consistent with the results from Vaporciyan AA [11]. The anastomotic techniques include hand-sewn anastomosis and device anastomosis. Hand-sewn anastomosis may cause variations in needle spacing and unequal ligation strength, which facilitates anastomotic stoma occurrence. CRP is an inflammatory indicator that can reflect the preoperative inflammatory condition of the body. High preoperative CRP levels may also increase the probability of developing AL by affecting tissue perfusion. Additionally, some studies have reported the utility of serum CRP levels for predicting postoperative inflammatory complications of oesophageal cancer before any clinical signs or symptoms appear [28‐30].

Therefore, an accurate predictive model will help physicians assess the risk of AL and provide timely interventions. In actuality, it is difficult to predict AL in individual patients due to uncontrollable factors, such as the type of surgical procedure. However, early assessment, suitable examinations and multi-faceted interventions may still be the most effective method for preventing AL.

There are several limitations in our current study. First, the database was from a single, high-volume institute. Second, the study also has some limitations associated with its retrospective nature. Third, some prognostic parameters (such as pathological stage) and other important factors (such as drug history) were not included in our analysis. Finally, because external validation could not be conducted, we performed an internal validation using bootstrap testing.

A novel risk score for the prediction of AL using Sex, diabetes history, anastomotic type, reconstruction route, smoking history, CRP level and the presence of cardiac arrhythmia was developed and internally validated. This nomogram offers a useful tool for clinicians to assess the risk of AL in individuals after surgery. The model focuses on preoperative and intraoperative data, which can be acquired earlier and more easily. Additionally, a prospective and international study is required to further validate the new nomogram.