Esophageal perforation is rare but with high morbidity and often fatal. Even in modern series, the reported 30-day mortality ranges between 18% and 30% [1-4].
Small, contained perforations can be managed expectantly, but those with limited drainage and full wall thickness perforations appear to benefit from endoscopic stent placement , along with pleural drainage, as definitive therapy [ 5,6]. However, failure of nonsurgical treatment, persistent uncontained leak, or worsening sepsis are indications for advanced intervention.
In an effort to classify disease severity at presentation and allow comparison of outcomes, the Pittsburg Severity Score (PSS, range 0-18) was developed. The PSS has since been validated, showing that increasing scores are associated with increased mortality [2-4].
Patients at the high end (>5) have an in-hospital mortality of approximately 40% [4]. Despite such an objective classification system, there is no doubt that there is significant, unquantifiable bias and judgment that complicates treatment decision-making for a given patient with esophageal perforation [7-15].
Therefore, to understand self-bias regarding the use of different advanced interventions in patients with esophageal perforation, the authors have reviewed their clinical series and imputed a PSS to each patient.
| Methods |
> Patients
From 1996 to 2017, 335 patients had esophageal perforations managed at the Cleveland Clinic . Patients with oropharyngeal perforations, leaks after esophagectomy, and those undergoing esophageal surgery for diagnoses other than perforation were excluded, as were patients who were successfully managed expectantly or with endoscopic procedures.
These patients were not analyzed because they were frequently managed by nonsurgical services. These exclusions resulted in a study group of 166 patients managed with advanced surgical interventions.
> Causes of perforation
Spontaneous perforation was defined as Boerhaave syndrome, or foreign body impact. Iatrogenic perforations involved any adverse event related to instrumentation. Malignant perforations were defined as perforations due to malignant disease occurring during treatment. Other causes of perforation included, but were not limited to, aortoesophageal and esophagopleural fistulas, and erosion by cervical surgical material.
> Disease severity at presentation: Pittsburg severity score
The PSS is a clinical score that ranges from 0 to 18, and is calculated at the time of presentation, based on 10 factors [3]. Each factor is weighted as follows: 1 point: age > 75 years, tachycardia (> 100 beats/min), leukocytosis (> 10,000/mL), or pleural or effusion (seen on x-ray, chest computed tomography, or by barium intake); 2 points: fever > 38.5ºC, uncontained filtration (seen with barium ingestion, or by computed tomography), respiratory compromise (rate > 30/min, increasing oxygen requirement, or need for mechanical ventilation), or time until time of diagnosis > 24 hours; and 3 points: presence of cancer or hypotension. Previous studies subcategorized severity into 3 groups: PSS < 2, PSS 3 to 5, and PSS > 5 [4].
> Advanced interventions
Among the 166 patients, there were 74 primary tissue flap repairs (repair), 26 esophagectomies with gastric lift (resection), and 66 esophagectomies with complete bypass of the alimentary tract, with subsequent planned reconstruction (resection-bypass).
• Primary repair with tissue flap (Repair)
Depending on the location of the perforation, a cervical incision, thoracotomy, or supraumbilical median laparotomy was used for access. The edges of the perforated esophagus were debrided to healthy tissue. Full-thickness primary closure was then performed with absorbable monofilament suture, reinforced with nearby muscle or omentum, when available.
• Esophagectomy and gastric lift (Resection)
The approach to performing an esophagectomy with gastric lift , in the setting of a perforation, was similar to that of an elective procedure for cancer. After kocherizing the duodenum and mobilizing the stomach to create a gastric tube, dissection of the esophagus was performed from the thoracic cavity toward the entrance. The affected esophagus was removed through a cervical incision and divided. The esophagogastric anastomosis was then created by stapling the posterior wall and suturing the anterior wall with absorbable monofilament.
• Esophagectomy with diversion and planned reconstruction (Resection-Diversion)
The approach to perform esophagectomy with diversion involved endoscopic or open placement of a gastrostomy tube, followed by a left thoracotomy. The stomach was stapled below the perforation area, and the staple line was reinforced, allowing it to retract into the peritoneal cavity.
The diaphragmatic hiatus was then closed primarily. The mediastinal abscess, if present, was debrided. The esophagus was bluntly dissected to the thoracic inlet and then, through a left cervical incision, brought to the neck.
After removing the affected portion, a terminal esophagostomy was created. Once the patient recovered from sepsis and was rehabilitated, reconstruction was performed using the stomach, or colon in a heterotopic substernal position. The left sternoclavicular joint was typically resected to allow a tension-free anastomosis.
> Data
Clinical characteristics included : demographics, time of presentation, initial therapeutic management, cause of perforation, anatomical location of perforation, preoperative treatment, and perioperative outcomes.
Comorbidities were measured with the Charlson/Deyo score, and the severity of perforation at presentation was quantified with the PSS. Data collection and use for this study were approved for research use by the Cleveland Clinic Institutional Review Board , with patient consent waived.
> Final objectives
The primary endpoint of this study was death from any cause within 90 days of surgery. Vital status was obtained, in part, from the electronic medical record but largely by cross-sectional follow-up using a telephone follow-up script approved by the Institutional Review Board, with patient consent.
For the resection-bypass group, time to reconstruction was the ultimate goal, which was considered in the context of the competing risk of death before reconstruction.
> Statistical analysis
Continuous variables were compared using analysis of variance (ANOVA) for normally distributed variables, and with the Kruskal-Wallis test for non-normally distributed variables. Categorical variables were compared using the chi-square test, or Fisher’s exact test when the frequency was < 5. Comparisons are described in the order of repair versus reconstruction versus resection-bypass.
All-cause mortality was assessed non-parametrically using the Kaplan-Meier method [16]. Differences in survival according to surgical intervention were tested using the log-rank test. Nonparametric estimates of reconstruction and death, before reconstruction, in the resection-bypass group, were obtained with the Andersen method [17].
Random Forest for Classification (RF-C) was used to identify the predictors of the three advanced interventions, using the variables classified in Appendix I SDC (the variables considered for the random forest analysis).
Albumin was excluded because it was absent in 38% of the data, leaving 30 dichotomous, polytomous, ordinal and continuous variables in the analysis. Missing data were pre-imputed, without using on-the-fly outcome information, using the RF imputation methodology, missForest [18]. Calculations were implemented using the random ForestSRC R software , with default settings [18-20]. For that RF-C analysis, 1000 bootstrap trees were grown.
Random Forest for Survival (RF-S ) was used to identify predictors of 90-day mortality using the variables listed in Appendix I SDC. Two RF-C analyzes were performed, the first without PSS, to identify the most important predictors among the individual components of the PSS, and the second including the PSS, to determine its importance as a predictor of 90-day survival, after adjusting by the variables of the first RF-C analysis.
All calculations used the open source R random ForestSRC ( rfsrc ) software, with default settings [ 20 ]. Briefly, 5000 trees were grown using log-rank division. Each tree was constructed using an independent bootstrap sample, containing, on average, 63% of the patients (bootstrap data in the sample), and the rest duplicates.
The remaining unselected patients (37%), referred to as out-of-bag (OOB ) observations , were used to calculate OOB cross-validated survival for each patient, and the Variable importance measures (VIMP ) for each of the independent variables.
Positive VIMP values indicate variables that are predictive, fitting all other variables and boxplots, and 95% confidence intervals indicate “significance” of VIMP (SDC Appendix 2: Random Forest Analysis Details ) [22].
The direction, magnitude and risk-adjusted form of the relationship of the variables with the outcome are represented by miniature partial dependence diagrams [23]. Finally, Random Forests for Competing Risks (RF-CR) were performed to identify predictors of the competing events: reconstruction and death before reconstruction [24].
| Results |
> Clinical presentation
The mean age of the cohort was 61 ± 16 years, and 89 (54%) were women. The repair group had the greatest number of iatrogenic perforations from instrumentation, few perforations from cancer, and had the shortest interval from perforation to surgery. The repair and resection groups had fewer spontaneous perforations involving the lower esophagus compared with the resection-diversion group.
Resection and repair patients had few failures with initial endoscopic stent placement or clipping compared with resection and resection-bypass patients. 61% of patients who underwent resection-bypass were transferred from another hospital, compared with 46% who underwent repair, and 50% who underwent resection ( P = 0.10). Of the entire study cohort, 62 patients (37%) presented beyond 72 hours.
> Pittsburg Severity Score
The mean PSS of the entire cohort was 5.0 ± 3.3, and 25% had a PSS > 7. Patients with repair and resection had few pleural effusions and uncontained perforations, relative to the resection group. -derivation.
There were no statistically significant differences between the operative groups in terms of vasopressor use and intubation, at the time of presentation. Repair and resection patients had a lower PSS compared with resection-bypass patients (3 vs 3 vs 6; P = 0.002), and fewer patients in the most severe categories (PSS > 5) (29% vs 30 % vs 52%; P = 0.02). As PSS increased, more patients underwent bypass.
> Predictors of the type of advanced intervention
Patients with repair had a higher probability of having a cervical perforation due to iatrogenic causes and therefore a shorter interval from perforation to operation, and a lower PSS.
Patients with resection were more likely to fail initial nonsurgical management of a malignant perforation and therefore had a longer interval from perforation to operation. Patients with resection-diversion were more likely to have spontaneous lower esophageal perforations and high PSS.
> Results
Patients with repair and resection had shorter lengths of hospital stay than those undergoing resection-bypass, and there were significant differences in the occurrence of complications for each surgical intervention.
Twenty patients (27%) who underwent repair developed a leak. Of these, 8 had their initial repair at another institution and were referred to the Cleveland Clinic . Three of the 8 transferred patients finally underwent resection, and 5 underwent resection-bypass. There were no statistically significant differences in recurrent laryngeal nerve or chylothorax injuries.
> Reconstruction after bypass
In five patients who underwent repairs at other institutions, the initial surgical repair failed and they underwent resection-bypass and were included in the analysis. Of the total 71 patients who underwent resection-bypass, 39 had subsequent upper alimentary tract reconstruction, and 32 did not.
Patients who received subsequent reconstruction after resection-bypass were younger (57 ± 16 vs. 66 ± 16; P = 0.02), with a few having Charlson/Deyo scores > 2 (20% vs. 42% ; P = 0.004), or associated malignancies (5.4% vs 35%; P = 0.007), and had lower PSS scores (5 vs 7; P = 0.02).
Reconstruction was performed at a median of 334 days (Q1-Q3, 306-550) from derivation; 25 gastric conduits and 14 substernal colons were used.
The competing risk of death before reconstruction was highest during the first year after bypass; however, the probability of reconstruction was 52% at 2 years after the initial resection-bypass. A low Charlson/Deyo score, a low PSS, and no association with malignancy were the most important predictors of reconstruction, while a higher PSS and older age were the most important predictors of death before reconstruction.
> Mortality and survival
• Early
Mortality at 90 days after repair, resection, and resection-bypass was 11%, 7.7%, and 23%, respectively. There was variable importance of death predictors with and without the inclusion of PSS in the analysis.
Without taking into account the PSS, the most important preoperative predictors of 90-day mortality were: intubation, vasopressor requirement, high Charlson/Deyo score, associated malignancy, advanced age, and leukocytosis.
However, when PSS was included, it became the most important variable associated with 90-day mortality. The 90-day mortality was 2.7% (n = 8) for patients with a PSS < 2, 6.9% (n = 2) for a PSS of 3 to 5, and 32% (n = 15) for a PSS > 5.
• Late
There were no statistically significant differences in survival between the 3 advanced interventions (log-rank P = 0.12).
| Discussion |
The presentation of uncontained esophageal perforation, although uniformly life-threatening, is surprisingly heterogeneous.
Disease magnitude appears to be reasonably stratified by PSS, and the authors now suggest that treatment allocation be made based on disease severity at presentation.
Although primary repair is always the first consideration, with organ preservation as the goal, the severity present may force more advanced interventions, with greater morbidity and mortality. However, even patients at the highest extremes of severity can be offered life-saving treatment.
The authors of this work have reviewed their series of uncontained perforations, to understand when it is best to consider each of these different advanced interventions. Patients at the lower end of severity, with limited mediastinal necrosis and salvageable esophagus, can and should be managed with primary repair and tissue reinforcement. Such patients will likely benefit additionally from limited follow-up and preservation of normal swallowing.
For patients with an uncontained perforation, low severity at presentation, and perhaps other underlying esophageal pathology, primary resection with gastric elevation is reasonable. This requires longer follow-up and probably additional future interventions (eg, anastomotic dilation).
Finally, patients at the extreme end of severity and with high expected mortality, regardless of the type of intervention, seem reasonable to treat with esophagectomy and complete diversion, with subsequent reconstruction (pending initial recovery). Those patients are certainly the most critically ill, but then, unfortunately, they are subject to much more intensive management in the future. That is certainly not the first treatment option but, for these patients, it is probably the only therapeutic chance.
Any treatment algorithm for the management of esophageal perforations must include not only emergency endoscopic palliation, but also traditional advanced interventions, with reliance on institutional expertise to treat these often severely ill patients through courses complicated hospitals.
The present findings suggest that patients with complicated esophageal perforations, even with extreme severity scores, can be successfully treated surgically. Despite being at the most severe end of PSS, patients who underwent resection-bypass and subsequent reconstruction had surprisingly good survival, compared to what other authors have reported [1-4].
In the modern era, the paradigm for the treatment of esophageal perforation has changed, as have outcomes, which have improved [5]. Endoscopic innovation has allowed many perforations to be managed with minimal morbidity and excellent results. Even in perforations accompanied by pleural effusion and mediastinal contamination, hybrid endoscopic control and drainage using video-assisted thoracic surgery have proven to be very effective [6,7].
The wide range of perforation severity between advanced intervention groups could be explained by the cause of perforation, which sets the course of sepsis and subsequent treatment [7-15].
Patients with primary repair were more likely to have iatrogenic perforations. Patients with resection-bypass more frequently had spontaneous perforations. Iatrogenic perforations are more likely to be recognized immediately and therefore treated before mediastinitis and sepsis can develop. Spontaneous perforations are more likely to be recognized late, allowing mediastinitis and sepsis to progress unrestricted.
In addition to the causes of perforation, there were also differences in initial failed management strategies. Those who underwent resection-bypass were more likely to have had failed endoscopic treatments, further delaying adequate control of sepsis and mediastinal contamination [9-11,15].
Other surgical series treating esophageal perforation focus on primary repair [7]. The authors of the present study agree that the initial surgical choice should be primary repair with a tissue flap; However, extensive mediastinal and pleural contamination, in addition to hemodynamic instability, often make it infeasible.
Sudarshan et al. [7] described their management of 48 patients with esophageal perforation, 30 of whom were treated surgically (20 repairs, 4 drainages, and 6 esophagectomies with immediate reconstruction). These authors noted that this strategy aimed to avoid cervical bypass [7].
However, only 6 (12%) of the 48 patients were diagnosed and treated after 72 hours, while more than a third of the patients in this study presented beyond 72 hours. Delays in definitive treatment often make safe salvage of the esophagus difficult, in which case esophageal resection is the only viable option.
Although this is a larger operation, it can be carried out safely. Seo et al., showed in their study comparing emergency esophagectomy with elective esophagectomy, that morbidity and mortality, in the short and long term, were similar [9].
There are other factors that affect long-term survival in this cohort. Perforation that occurs in the setting of cancer portends a more ominous late mortality. Conditional survival of perforation and its treatment confounds the interpretation of late dangers [10,11].
Despite the severity of the disease at presentation, patients who required esophageal bypass appeared to have benefited from surgical intervention. Based on the PSS data of Schweigeret et al. [4], the expected in-hospital mortality was 37.5% for the group with greater severity (PSS > 5), over a median length of stay of 38 days.
The median PSS for the shunt group in the present study placed them in the most severe category. However, patients with bypass had a median length of stay of 26 days, and a 90-day mortality of 23%.
Despite these encouraging numbers, a third of patients still die before reconstruction. They tend to be older patients, with more chronic diseases, who probably could not survive the initial septic shock. However, a little more than half of the resection-bypass patients were eventually able to be reconstructed. Furthermore, once they overcame the initial septic shock, the reconstructed patients had excellent long-term survival.
> Limitations
Although this is the largest series of advanced interventions for esophageal perforations from a single institution, it is a retrospective review of a rare and heterogeneous population. This creates inherent difficulties in categorizing and comparing perforation etiologies with previously published studies.
A few patients remained difficult to categorize and therefore assumptions were made about how best to include them in the analyses. Patients who had their primary operations performed in other healthcare centers, but who continued to have leaks, were included in this study. Because they highlight patients with extreme illnesses in the presentation, the authors believe their inclusion is informative.
The PSS may have been underestimated, because there were variables (e.g., fever, tachycardia, and respiratory compromise), which were difficult to determine from the medical records of elderly patients; were scored as not occurring in the PSS calculation. Additional details, which likely affected the choice of advanced intervention, such as perforation size, were not readily available in a retrospective review.
Furthermore, endoscopic management with covered esophageal stents, which are more widely used today, was not studied. Previous studies describing management with covered stents focused primarily on iatrogenic perforations that were immediately recognized [5].
| Conclusions |
The authors’ experience of 20 years of advanced surgical interventions for esophageal perforations suggests that repair and resection with immediate reconstruction is quite feasible when patients with a moderate to low PSS score are encountered.
Some patients at the more severe end of disease after perforation have been shown to be still salvageable by esophagectomy and complete alimentary tract diversion , although they are subject to further reconstruction when (and if) sufficient recovery occurs. There appears to still be a role for resection-bypass in the management of this disease.
