Introduction

Emergency abdominal surgery is followed by a substantial risk of postoperative complications which influence the risk of death of 15–25%.1,2,3 Complications develop in over 30% of patients and vary according to patient characteristics, hospital characteristics, and the underlying pathology.4,5,6 Adverse events often prolong the hospital stay, and postoperative complications are more strongly associated with mortality than are pre- and intra-operative variables.7,8 A recent study found major complications in 47% of patients following emergency laparotomy for gastrointestinal obstruction, perforation, or bleeding and a mortality risk of 26%.9 The risk of major cardiac and pulmonary complications was high the first days after surgery as was the risk of death. However, little is known about which postoperative complications are most strongly associated with death following emergency gastrointestinal (GI) surgery.

The association between postoperative complications and death has been addressed in several cohorts within elective surgery. The risk of death in patients with major postoperative complications is referred to as failure-to-rescue (FTR). FTR has proven to be effective for evaluating the postoperative course at an institutional level because it allows for the development of post hoc protocols to optimize care. Although the FTR metric was originally developed for planned surgical procedures with a low risk of complications or death,10 it has recently been gaining ground in emergency surgery,11,12,13,14 and thus, its use is extending to patient with other diagnoses. Moreover, the incidence, type, and severity of complications following emergency surgery may differ considerably from those of elective cohorts and is likely to affect the association between specific complications and death. Importantly, specific postoperative complications may serve as valuable in situ markers of when to escalate care to prevent a fatal outcome.

Different complications appear at different times throughout the postoperative course, as does death. As such, the association between a complication and death needs to take into account the time without a complication (unexposed) and the time from a debuting complication to death (exposed). Not taking this into account may lead to what is known as “immortal time bias.” Taking into consideration the “unexposed” and the “exposed” time may add important understanding of the association between individual complications and death following an eventful course of emergency GI surgery.

We hypothesized that complications evolve in continuums and that certain complications are more strongly associated with 90-day mortality than others following emergency GI surgery. Further to this, some complications may serve as markers of an evolving adverse course independently of the underlying pathology. Identifying these index complications may render an in situ marker of patients needing escalation of care. Thus, the aim of this study was to identify which postoperative complications are most strongly associated with death following emergency surgery for GI obstruction or perforation.

Methods

This study was approved by the ethical committee (J.nr. 16-000014) of Zealand Region, Denmark. The requirement for written informed consent was waived by the committee. Approval by the Danish Data Protection Agency (REG-149-2016) and the Danish Patient Safety Authority (3-3013-1999/1) was granted. The manuscript adheres to the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) statement.15

All patients scheduled for emergency gastrointestinal surgery between July 1, 2014, and July 31, 2015, at three hospitals in Zealand Region, Denmark, were retrospectively included. The hospitals treat all emergency cases among 800,000 citizens. Emergency treatment is offered free of charge at public hospitals in Denmark, with no private alternative. Emergency surgery is performed day and night without barriers, and extra staff are on call if needed. At the hospitals in our study, general chief surgeons or surgeons in their final year of specialization are on duty around the clock. If the surgeon is still in training, a chief surgeon is on call.

We included patients aged 18 years or older who underwent emergency surgery for GI obstruction or perforation diagnosed by radiological examination. Thus, minor procedures as appendectomies and cholecystectomies were excluded. Emergency surgery was defined as the need for laparoscopy or laparotomy without planned delay. We excluded patients who had undergone intraabdominal surgery up to 30 days prior to the index procedure, patients with an iatrogenic or traumatic perforation, patients pregnant at the time of surgery, or patients receiving chronic dialysis. Patients eligible for inclusion more than once were only included at the first procedure. Only Danish residents were included.

We manually screened all patients planned for GI surgery in the electronic booking system used at the participating hospitals. All emergency procedures due to obstruction or perforation, or undescribed cases were identified through the patients’ personal identification number, which allowed data collection from the electronic patient files. Data from the pre-, intra-, and postoperative courses were collected. The postoperative follow-up was 90 days on complications and mortality. Mortality data completeness was achieved through the Danish Civil Registration System. Data were collected between June 15, 2017, and March 31, 2018.

All patient files were accessed by two independent, clinically experienced researchers, and the data collected in two separate case report forms. Case report forms were regularly assessed by the project leader (AWV) to settle disagreements. Minor variations were solved by the project leader. Major inconsistencies were settled through dialogue between the project leader and the senior consultant responsible for the study (BB). Double data entry was performed, and irregularities corrected according to the case report form. Finally, a range check was performed for all data.

The preoperative data collected were age, sex, smoking and alcohol habits, height, weight, comorbidity (hypertension, cardiac, pulmonary, renal, diabetes, or presence of active cancer), American Society of Anesthesiologists physical status classification (ASA class), Sepsis-2 score, description of the radiological examinations, and time-to-surgery defined as time from decision of surgery to surgery. The intraoperative variables collected were time of surgery, the procedure performed, the intraabdominal pathology, intravenous fluid administration, and blood loss. The postoperative variables collected were Sepsis-2 score, re-admissions, and in-hospital complications as defined in Table 1. Only complications occurring postoperatively were registered and graded according to the Clavien-Dindo Classification.16 Preoperative conditions were evaluated and only in case of substantial postoperative worsening was the condition registered as a complication (increase in CDC class)17, e.g., medically treated pneumonia preoperatively was only registered as a complication if it warranted mechanical respiratory support postoperatively. The date of appearance was used for complications.

Table 1 Definition of postoperative complications
Full size table

The primary exposure variables comprised 16 predefined complications (Table 1). We combined some individual complications that are considered to evolve in continuum or had similar treatment profiles. The follow-up on complications was contemplated as 90 days for all complications as most of the complications demanded hospital admission. Planned operations such as a “second-look” or change of vacuum-assisted coverings were not deemed complications. The primary outcome was 90-day all-cause mortality.

Statistics

Parametric and non-parametric statistics were used as appropriate. We presented events of complications as numbers and absolute risks. All-cause mortality was presented as a risk for individual groups of complications (FTR). The primary outcome was analyzed using a multivariable Cox proportional hazards model. To evaluate the influence of the first complication on mortality, we delayed the entry time to the date of the complication; i.e., the patient was included as unexposed (no complication) before that date, thereby avoiding immortal time bias.18 Any patient who did not die within the predefined 90 days was censored. The model presumes a progressive time span between a complication and death. Some complications appeared on the same day as death. Thus, half a day was added to the day of death or censoring. If a date was missing on a complication, the median time from surgery to the same complication in the cohort was used, or the time to death if it appeared first.

We created a multivariable model adjusting for variables significantly (p < 0.05) associated with death in a univariable Cox regression model with delayed entry. All variables demonstrating a significant association were included in the model. Several significant variables were found (Supplementary Tables 1–11), and we decided post hoc to limit the adjusted analysis to complications occurring in over 40 patients to avoid overparameterization. Independent variables in the model were as follows: hospital, age, ASA class (categorized in class 1–2 or 3–5), pre-operative Sepsis-2 score (categorized as group 0–2 or 3–4), cardiac comorbidity (yes or no), hypertension (yes or no), renal comorbidity (yes or no), active cancer (yes or no), the diagnosis (bowel obstruction or perforation), and the type of surgery (bowel resection and stoma formation or other procedures). Three preoperative Sepsis-2 scores were missing. Data on all other independent variables were complete. Post hoc, we decided to replace the missing preoperative Sepsis-2 scores. We used the postoperative Sepsis-2 score and subtracted the median increase in Sepsis-2 score (1.0 (IQR 1.0–2.0)) between the pre- and postoperative course. A test for linearity demonstrated a better fit for age in the potency. Proportionality was tested using Schoenfeld residuals. The proportionality assumption was violated by the variable “active cancer”; hence, the baseline hazard was stratified by the “active cancer” group. We performed subgroup analyses for patients with GI obstruction or perforation. Bonferroni correction was used based on 16 outcomes, and a 2-sided p value of < 0.003 was considered significant. We used R version 3.5.0 GUI 1.70 El Capitan©R Foundation for Statistical Computing, 2016, and RStudio version 1.1.453 for the statistical analysis.

Results

A total of 349 patients were included in the analysis (Fig. 1). The follow-up on complications and death was complete (31,410 patient days) due to the link between the patient files system and the Danish Civil Registration System.19

Fig. 1
figure 1

Trial profile

Full size image

We registered 832 complications during the 90-day follow-up in 281 (80.5%) patients. Patients with a complication were more likely to have a higher ASA class and gastrointestinal perforation (Table 2). Dates were missing for 35 (4.2%) complications (Table 3). The median time to the first appearing complication was 3.0 days (IQR 1–4).

Table 2 Background characteristics according to the incidence of complications or death
Full size table
Table 3 Number of patients with a complication according to the postoperative day (POD)
Full size table

On the day of surgery and the first postoperative day (POD), 105 (12.6%) complications and 19 (20.9%) deaths were registered (Table 3). A total of 420 (50.5%) complications appeared between POD 2 and 7; 211 (25.4%) complications between POD 8 and 30; and 61 (7.3%) complications between POD 31 and 90 (Supplementary Fig. 1). The incidence of complications was evenly distributed between POD 0 and 7 and POD 8 and 90 for deep wound complications (24 vs. 21), renal impairment (12 vs. 11), and re-operations (41 vs. 38). The majority of the following complications appeared late in the postoperative course (POD 0–7 vs. POD 8–90): superficial wound complications (25 vs. 51), peritonitis (8 vs. 16), urinary tract infection (10 vs. 30), pleural exudation (22 vs. 34), and venous thrombo-embolic events (1 vs. 5).

The overall risk of death was 26.1% (91) at 90-day follow-up. The patients who died tended to be older, had a higher ASA class or Sepsis-2 score preoperatively, and more often had known active cancer than surviving patients (Table 2). Further to this, these patients more often had gastrointestinal perforation and anastomosis or stoma formation. Time to surgery was not significantly associated with the risk of complications or death.

Complications and Death

The risk of death was 20.6% (14) for patients with none of the registered complications, and 27.4% (77) for patients with complications. In the group with no registered complications, 13 of the 14 deceased patients increased in Sepsis-2 score after surgery and 10 had septic shock and died within postoperative day one. The risk of death, according to the 16 individual complications (FTR), ranged from 21% in patients with prolonged paralysis to more than 50% for patients with renal impairment or atrial fibrillation (Table 4).

Table 4 Risk of a complication or all-cause mortality at 90 days and their association
Full size table

The crude Cox regression analysis with delayed entry showed that out of ten significant associations, renal impairment, arterial thromboembolic events, and atrial fibrillation were the complications most strongly associated with death (Table 4). The adjusted multivariable model showed 7 significant associations out of 11 analyzed complications. Atrial fibrillation (HR 3.3 (95%CI 2.1–5.2), p < 0.001), deep wound complication (HR 3.2 (1.7–5.8), p < 0.001), and respiratory failure (HR 2.9 (1.6–5.1), p < 0.001) were most strongly associated with 90-day mortality (Table 4).

Out of all patients, 87 (24.9%) had only 1 complication with a mortality risk of 14.9%; 2 complications appeared in 57 (16.3%) patients and the mortality risk was 17.5%; and 3 or more complications appeared in 137 (39.3%) patients with a mortality risk of 39.4% (Supplementary Fig. 1).

Gastrointestinal Obstruction or Perforation

In total, 261 patients had GI obstruction, of whom 204 (78.2%) had 1 or more of 547 registered complications. The overall 90-day mortality risk was 21.8% (57). The risk of death was 15.8% (9) for patients with no complications and 23.5% (48) for patients with complications. Of the nine deceased patients with no registered complications, five had septic shock and died within postoperative day one. The crude Cox regression model demonstrated 10 complications significantly associated with 90-day mortality, of which renal impairment, pulmonary edema, and respiratory failure dominated (Table 5).

Table 5 The association between complications and 90-day mortality was stratified on subgroups with gastrointestinal obstruction or perforation
Full size table

Eighty-eight patients had GI perforation, of whom 77 (87.5%) had 1 or more of 285 registered complications. The overall 90-day mortality risk was 38.6% (34). The risk of death was 45% (5) for patients with none of the registered complications and 37.7% (29) for patients with complications. Of the five deceased patients with no registered complications, all had septic shock and died within postoperative day one. Atrial fibrillation was the only complication significantly associated with death in this subgroup of patients. In both subgroups, the number of patients with complications was small, so no adjusted analysis was performed.

Discussion

In this observational retrospective study of patients undergoing emergency surgery for gastrointestinal obstruction or perforation, we found that 81% of the patients had complications, 27% of whom were deceased at 90-day follow-up. One-third of the complications debuted after the first week of surgery, and the majority of patients had two or more complications. Renal impairment and arterial thromboembolic events showed the strongest association with death, but were rare. In the adjusted analysis, the complications showing the strongest association with death at 90-day follow-up were atrial fibrillation and deep wound complications.

We found a significant association in 10 of 16 complications and death with varying hazard rates of 2.4–6.8 and risk of death ranging from 32 to 57%. The variability supports our hypothesis that different complications correlate unevenly with death and emphasizes the importance of both minor (e.g., atrial fibrillation) and major (e.g., renal impairment) complications in the postoperative course in the urgent care setting.

The risk of death in patients with complications is known as failure-to-rescue (FTR). When initially introduced by Silber in 1992, the FTR metric included arrhythmia.10 Alternative definitions followed, some of which focused on surgical complications (e.g., wound infection or re-operations), while others primarily included medical complications (e.g., pulmonary, cardiac, or infectious).11,20 In contrast to most other studies in the field, which register the most severe complications, our focus in this study was on the first appearance of a specific complication and its association with death. Further to this, we considered the unexposed time (without a complication) and the exposed time (time after the appearance of a complication) to avoid immortal time bias. Interestingly, we found that atrial fibrillation and deep wound complication (fascia dehiscence and deep wound infection) demonstrated the highest hazard ratio for death. Fascia dehiscence was the primary reason for a re-operation in our cohort.

Re-operation was performed in 23% of the patients in our cohort, with mortality risk of 32%, and was strongly associated with death. The association has previously been documented following emergency laparotomy, with a risk of re-operations ranging from 20 to 36%.4,21,22 However, the mortality risk varies widely, falling between 20 and 72%. Re-operations have been associated with an increased risk of medical complications and additional re-operations, with each additional re-operation increasing the risk of death.23 Such findings have several possible explanations: the surgical stress response is repeated, an inflammatory response amplified, and the side effects of intravenous fluid therapy and the anaesthesia accumulate and may accelerate an adverse outcome.

Re-operations are generally not optional. Not operating could have vital consequences, and re-operations may be the only chance to rescue the patient. In our study, unplanned re-operations were dominated by fascia dehiscence. We found 11% with fascia dehiscence. The risk of fascia dehiscence varies between 3.8 and 28% following emergency laparotomy21,24 and is associated with morbidity and death.25,26 Moreover, the risk of fascia dehiscence is associated with patient- and doctor-related factors. The dominant patient-related factors are obesity, smoking and alcohol habits, or the presence of peritonitis. The iatrogenic factors are choice of suture and sewing technique. One study found that the risk of fascia dehiscence and subsequent death significantly decreased, compared to a historical cohort, following use of a new suture and sewing technique in patients undergoing emergency laparotomy.25

A striking finding in our study was the marked association between atrial fibrillation and death. Although we found that different complications dominate in patients with GI obstruction or perforation, atrial fibrillation uniformly demonstrated one of the strongest associations with death in both subgroups of patients. Atrial fibrillation is the most common postoperative arrhythmia. The incidence varies according to the type of surgery, ranging from 1.4% in non-cardiac surgery to over 30% following cardiac surgery. 27,28 The risk of atrial fibrillation in our cohort was 19%.

The association between atrial fibrillation and a postoperative adverse course has been documented following esophagostomies and cardiac surgery, but studies within gastrointestinal surgery are scarce.29,30,31 Postoperative atrial fibrillation has previously been associated with sepsis, a leaking bowel anastomosis, or death,31,32,33,34 which supports our finding. The pathophysiological relation is, however, not well understood, since it is unlikely that atrial fibrillation alone is the mediator of various complications or death. The inflammatory response and release of catecholamines following surgery have been argued to prompt postoperative atrial fibrillation. The association between atrial fibrillation and stroke or myocardial infarction has been documented and is a rational relation.35 However, the association between atrial fibrillation and subsequent surgical complications is more difficult to explain. It has been argued that atrial fibrillation alters circulation and may compromise blood flow at the surgical site. Another possible explanation is that perioperative intravenous fluid administration combined with a hormonal stress response that prompts fluid retention causes tissue edema and also induces atrial fibrillation; this further accelerates the risk of pulmonary congestion and edema of the surgical site and thus poor wound and anastomosis healing.36 Both mechanisms may explain why atrial fibrillation appears early in the postoperative course, whereas surgical complications evolve days to weeks later. Our results suggest that postoperative onset of atrial fibrillation should prompt a thorough assessment of the patient to detect underlying pathology and may serve as an early marker of patients needing escalation of care.

The strengths of our study are the double data extraction and registration, clear definitions of study variables, and the analytical adjustment for delayed entry to avoid immortal time bias.18 The contribution from multiple sites increases the external validity and generalizability of the study results.

The limitations of our study are inherent in the retrospective design that relies on patient files. We compensated for this by using clear definitions of complications and conducting double registration of the prospectively collected data in a public health system ensuring 100% follow-up on mortality of all Danish residents.19 Despite a clear definition of the cohort, different intraabdominal pathologies were disclosed and could have unevenly influenced the risk of complications and death. However, we corrected for important confounders in the adjusted analysis. No matter the adjustment, the results of this study are merely hypothesis-generating.

Our study discloses a particularly vulnerable group of patients and an eventful postoperative course following emergency gastrointestinal surgery. The intraabdominal pathology involved every segment of the GI tract, and a variety of surgical interventions were performed. Future studies are encouraged to address how these variables may influence the association between specific complications and death. Importantly, the associations we found between renal impairment, arterial thrombosis, or atrial fibrillation and death were strong and should be addressed in future trials to test for causality. Another area to address in future studies is the marked risk of death (21%) within the first day after surgery.

Conclusion

In this observational study of patients undergoing emergency surgery for gastrointestinal obstruction or perforation, we found that 80% of the patients had a complication and that two-thirds of the complications appeared within the first postoperative week. Renal impairment and arterial thromboembolic events, though rare, showed the strongest association with death. Of the more frequent complications, atrial fibrillation and deep wound complications were most strongly associated with death. Atrial fibrillation was uniformly associated with death in both subgroups of patients with gastrointestinal obstruction or perforation and may serve as an important early marker of patients needing escalation of care.