ENTEROCUTANEOUS FISTULA AFTER EMERGENCY GENERAL SURGERY: MORTALITY, READMISSION, AND FINANCIAL BURDEN

Abstract

Background:

The burden of enterocutaneous fistula (ECF) after emergency general surgery (EGS) has not been rigorously characterized. We hypothesized that ECF would be associated with higher rates of post-discharge mortality and readmissions.

Methods:

Using the 2016 National Readmission Database, we conducted a retrospective study of adults presenting for gastrointestinal (GI) surgery. Cases were defined as emergent if they were non-elective admissions with an operation occurring on hospital day 0 or 1. We used ICD-10 code K63.2 (fistula of intestine) to identify postoperative fistula. We measured mortality rates and 30- and 90- day readmission rates censuring discharges occurring in December or from October-December, respectively.

Results:

135,595 patients underwent emergency surgery; 1,470 (1.1%) developed ECF. Mortality was higher in EGS patients with ECF than in those without (10.1% vs. 5.4%; OR 1.99, 95% CI 1.67–2.36) among patients who survived the index admission. Readmission rates were higher for EGS patients with ECF than without at 30 days (31.0% vs. 12.6%; OR 3.12, 95% CI 2.76–3.54) and at 90 days (51.1% vs. 20.1%; OR 4.15, 95% CI 3.67–4.70). Similar increases were shown in elective GI surgery.

Conclusions:

ECF after gastrointestinal EGS is associated with significantly increased odds of mortality and readmission, with rates continuing to climb out to at least 90 days. Processes of care designed to mitigate risk in this high-risk cohort should be developed.

Level of Evidence/Study type:

Level II, prognostic

Background

Enterocutaneous fistula (ECF) is a feared complication after gastrointestinal surgery, particularly in the setting of emergency general surgery (EGS). The 1960 landmark paper by Edmunds et al. described a single center’s experience and proposed strategies for prevention and treatment, (1) but subsequent literature has been sparse. Several centers have described their experiences, (2) some with attempts at mortality risk stratification (3) or description of factors associated with earlier closure. (4) Others have attempted to describe the optimal ways of treating ECFs. There has been some interest in medical treatment with somatostatin and octreotide, (5, 6) though this has not definitively been shown to be helpful. Others have published on the merits and logistical considerations of operative closure of ECF. (7, 8)

To date, there has not been a rigorous description of the epidemiology and the burden of postoperative ECF on patients and the health care system. Data on the incidence of ECF in EGS patients is limited, though trauma studies have found a 1.5–2.0% rate in patients undergoing laparotomy for injury. (9, 10) The overall mortality rate following ECF has been broadly stated as 10–30%. (11) Readmissions with ECFs have only been studied in small numbers as a component of studies designed to answer other questions. (12–15)

Based on our institutional experience, we noted relatively high rates of readmission and mortality in the cohort of EGS patients with ECF, but were unable to find national estimates to contextualize our local findings. We therefore set out to describe 30- and 90- day mortality and readmission rates following gastrointestinal EGS using the National Readmissions Database, a nationally representative dataset. We hypothesized that occurrence of ECF would be associated with higher rates of post-discharge mortality and readmissions.

Methods

We conducted a retrospective cohort study using the 2016 National Readmissions Database (NRD). The NRD is a database of all-payer hospital inpatient stays developed for the Healthcare Cost and Utilization project (HCUP) and contains data from approximately 17 million discharges per year using a stratified sampling design. Weighted, this data represents approximately 36 million discharges each year.

Children (age less than 18 years) were excluded. We defined gastrointestinal surgery operations by selecting records with international classification of diseases, 10^th revision (ICD-10) procedure codes for small bowel resection, colorectal resection, hernia repair, GI laparoscopy, exploratory laparotomy, and lysis of adhesions. We used the Clinical Classifications Software (CCS) tool developed and distributed by the Agency for Healthcare Research and Quality (AHRQ) to group representative codes into these surgical procedures. A summary of the CCS codes used is shown in Table 1. Next, we defined emergency operation by selecting cases that were flagged as non-elective admissions in the NRD and took place on hospital day 0 or 1 (within 48 hours of admission). This definition was used in an attempt to capture patients who presented with acute operative surgical issues. Finally, we defined ECF using the ICD-10 diagnosis code K63.2 for “fistula of intestine,” during any admission, including the index, provided it was not the primary diagnosis code upon admission. Patients with a diagnosis of fistula during their first admission for GI surgery were excluded if they had prior admissions with this diagnosis. Patients for whom this code was listed as the primary diagnosis upon index admission were also excluded, as these also do not represent postoperative fistulae.

Table 1.

Clinical Classification Software (CCS) codes used to define gastrointestinal operations.

Clinical Classification Software (CCS) code	Procedure
75	Small bowel resection
78	Colorectal resection
86	Other hernia repair
87	Laparoscopy (GI only)
89	Exploratory laparotomy
90	Excision; lysis peritoneal adhesions

GI, gastrointestinal.

We tabulated baseline descriptive statistics between those with and without fistulae. Statistical significance was determined using Student’s t or chi-squared tests, as appropriate. Outcomes of interest included 30- and 90-day mortality, defined as in-hospital death within the specified timespan as counted from the day of admission, either during index admission or readmission. We excluded those with an unknown survival status from the index admission and those with an unknown index length of stay (LOS) from these analyses. Furthermore, because the database ends at the conclusion of the calendar year, limiting follow-up ability, we censured December discharges from analysis of 30-day mortality and October-December discharges from analysis of 90-day mortality. While this results in cohorts of slightly different sizes for each of these analyses, using this strategy instead of conducting all analyses on a cohort discharged in the first 9 months of the year allows inclusion of the maximum number of patients and guards against bias introduced by seasonal variation.

We calculated 30- and 90-day readmission rates, defined as any readmission documented in the NRD within the specified time as counted from the day of discharge from the index admission, with the same censuring strategy. In addition to the exclusions detailed above for the mortality analysis, those that died during the index visit were excluded (as these patients could not be readmitted). Readmission rates were calculated separately among those undergoing emergent and elective operations. Differences in mortality and readmission rates between groups were calculated using chi-squared tests. We also examined the total number of readmissions in each of the 4 cohorts of interest (elective vs emergent, with and without ECF) and report medians [Interquartile Range (IQR)]. It should be noted that there are varying lengths of follow up for this portion of the analysis, as this data set only spans the calendar year 2016, and patients were seen throughout this year. Comparisons were made using the Mann-Whitney U test.

Finally, charges incurred by patients with and without fistulae were calculated using the sum total of the charges of all stays as documented in the NRD, and statistical difference was determined using a Mann-Whitney U test.

This study was determined by the University of Pennsylvania’s Institutional Review Board to be exempt from review. Statistical analyses were performed using R 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria), SAS 9.4 (SAS Institute, Cary, NC), and Stata version 15.1 (College Station, TX). A two-tailed p- value of <0.05 was considered significant in the final analysis. All analyses were survey weighted according to the NRD design.

Results

Using the criteria defined above, 285 patients were found to have fistulae that were pre-existing at the time of their index admission for gastrointestinal surgery and were therefore excluded from the unweighted dataset, leaving 288,022 records weighted to represent 548,615 patients. Of these GI surgical cases, 7,835 (1.4%) developed a fistula during the follow-up period. Figure 1 shows a flow diagram of exclusions and breakdown of patients. Table 2 shows baseline differences, using appropriately weighted numbers, between patients who did and did not develop fistulae.

Figure 1.

Inclusions and Exclusions.

Table 2.

Baseline characteristics. Totals are survey weighted and rounded to nearest integer. Age tested using Student’s t-test, remainder analyzed using chi-squared.

		Did not develop fistula (n=540,780)	Developed fistula (n=7,835)	p-value
Age in years, mean (SD)		60.2 (16.1)	60.6 (14.6)	0.224
Number male (%)		221,326 (40.9)	3,589 (45.8)	< 0.001
Elixhauser comorbidity index (%)	0	86,618 (16.0)	756 (9.6)	< 0.001
	1 to 2	229,138 (42.4)	3,306 (42.2)
	> =3	225,024 (41.6)	3,773 (48.2)
Emergent operations (%)		134,125 (24.8)	1,470 (18.8)	< 0.001
Procedure type (%)	Small bowel resection	18,400 (3.4)	612 (7.8)	< 0.001
	Colorectal resection	197,491 (36.5)	3,047 (38.9)
	Hernia repair	99,164 (18.3)	1,190 (15.2)
	Laparoscopy (GI only)	53,616 (9.9)	339 (4.3)
	Exploratory laparotomy	10,418 (1.9)	180 (2.3)
	Lysis of adhesions	161,691 (29.9)	2,467 (31.5)

SD, standard deviation; GI, gastrointestinal.

Overall, the rate of fistula development rate after EGS was 1.1% compared to 1.5% in elective surgery (p<0.001). The majority of patients (88.2%) developed ECF during the index admission (median postoperative LOS 15 days [IQR 7–30]), with the remainder re-presenting with ECF on their first readmission (median postoperative day of presentation: 53 [IQR 25–103]). When parsed by procedure, the highest rate of fistula was among elective small bowel resection, at 3.5%. Table 3 shows rates by procedure and emergent/elective status.

Table 3.

Fistula rates in EGS and elective surgery, by procedure group. All values weighted per National Readmissions Database design and rounded to the nearest integer. Procedure groups defined by CCS code.

	Procedure	Fistula	No fistula	ECF rate
EGS	Small Bowel Resection	184	6,441	2.8%
	Colorectal Resection	488	36,504	1.3%
	Hernia Repair	205	22,763	0.9%
	Laparoscopy (GI only)	66	19,514	0.3%
	Exploratory Laparotomy	44	5,150	0.8%
	Lysis of Adhesions	484	43,753	1.1%
Elective	Small Bowel Resection	428	11,958	3.5%
	Colorectal Resection	2,559	160,987	1.6%
	Hernia Repair	985	76,401	1.3%
	Laparoscopy (GI only)	273	34,102	0.8%
	Exploratory Laparotomy	136	5,269	2.5%
	Lysis of Adhesions	1,983	117,938	1.7%

EGS, emergency general surgery; ECF, enterocutaneous fistula; GI, gastrointestinal; CCS, clinical classification software.

Mortality rate in the overall cohort was 3.3% and was higher in patients who developed post-operative fistula. The overall mortality rate among EGS patients who developed a fistula was 10.4%, compared to 7.4% in elective surgery (p<0.001). Table 4 summarizes overall fistula and mortality rates.

Table 4.

Overall mortality rates in EGS and elective surgery, with and without fistula. All values weighted per National Readmissions Database design and rounded to the nearest integer.

		Survived	Died	Mortality Rate
EGS	Fistula	1,318	153	10.4%
	No fistula	126,715	7,410	5.5%
	Total	128,033	7,562	5.6%
Elective	Fistula	5,890	474	7.4%
	No fistula	393,321	13,335	3.3%
	Total	399,211	13,809	3.3%

EGS, emergency general surgery.

At the 30-day time point, the appropriately weighted cohort for the purposes of mortality analysis represented 503,160 patients. Following exclusion of the deaths as described above, the weighted cohort for the readmission analyses consisted of 488,573 patients. 30-day readmission and mortality rates are summarized in Table 5. In both elective and EGS patients, ECF was associated with a more than twofold increase in 30-d readmission rate (EGS: 31.0% vs 12.6%; elective: 28.6% vs 12.6%; p<0.001 for both). Similarly, mortality rates were 9.0% in those with ECF vs 4.9% in those without ECF in EGS patients, and 6.1% with ECF vs 2.7% without among elective patients (p<0.001 for both).

Table 5.

30-day readmission and mortality rates, by fistula and emergency surgery status. All values weighted per National Readmissions Database design and rounded to the nearest integer.

			Patients (n)	30-day Readmission rate		Patients (n)	30-day Mortality rate
EGS	ECF	Readmitted	375	31.0%	Died	118	9.0%
		Not readmitted	835		Survived	1,195
	No ECF	Readmitted	14,810	12.6%	Died	6,066	4.9%
		Not readmitted	102,968		Survived	117,231
Elective	ECF	Readmitted	1,543	28.6%	Died	344	6.1%
		Not readmitted	3,857		Survived	5,338
	No ECF	Readmitted	45,799	12.6%	Died	10,040	2.7%
		Not readmitted	318,386		Survived	362,828
			488,573			503,160	3.3%

EGS, emergency general surgery; ECF, enterocutaneous fistula.

At 90 days, the overall weighted cohort consisted of 402,125 patients for readmission analysis and 414,228 for mortality analysis. In EGS patients, ECF was associated with an increase in 90-day readmissions to 51.1% from 20.1% (p<0.001). In elective patients with ECF, the 90-day readmission rate was 43.6%, vs 20.1% in those without a fistula (p<0.001). 90-day mortality rates were similarly increase in EGS patients (ECF: 10.5% vs no ECF: 5.2%, p<0.001) and elective patients (ECF: 6.6% vs no ECF: 3.1%, p<0.001). These results are summarized in Table 6.

Table 6.

90-day readmission and mortality rates, by fistula and emergency surgery status. All values weighted per National Readmissions Database design and rounded to the nearest integer.

			Patients (n)	90-Day Readmission rate		Patients (n)	90-Day Mortality rate
EGS	ECF	Readmitted	533	51.1%	Died	120	10.5%
		Not readmitted	510		Survived	1,018
	No ECF	Readmitted	19,620	20.1%	Died	5,359	5.2%
		Not readmitted	77,953		Survived	96,759
Elective	ECF	Readmitted	2,032	43.6%	Died	323	6.6%
		Not readmitted	2,628		Survived	4,566
	No ECF	Readmitted	60,114	20.1%	Died	9,321	3.1%
		Not readmitted	238,755		Survived	296,762
			402,145			414,228

EGS, emergency general surgery; ECF, enterocutaneous fistula.

Overall, there was an increase in total number of readmissions over the course of 2016 in patients who developed ECF, whether they had undergone elective or emergent surgery. Among elective surgery patients who were readmitted at least once, the median total number of admissions in those without an ECF was 2 [IQR: 2–3]. It was also 2 [IQR: 2–3] in those with an ECF, though there was a statistically significant increase in these patients (p<0.001). In EGS patients with at least 1 readmission, the medians were 2 [IQR: 2–3] without an ECF and 3 [IQR: 2–4] with an ECF (p<0.001).

When the total hospital charges – including index admission and any readmissions – of each of the 288,022 patients were summed for all stays in 2016, a significant difference in overall charges was demonstrated for those who developed ECF. Median acute care hospital charges for the 4,075 patients with a fistula were $55,469.83 [interquartile range (IQR): $28,954.84-$114,655.86], vs $19,625.06 [IQR: $12,425.08-$35,064.07] for those without (p<0.001), translating into an inpatient burden of $35,844.77 per patient attributable to the development of a postoperative fistula.

Discussion

We conducted the first national study of the mortality, readmission, and acute inpatient care financial burdens of ECF in EGS patients. While the occurrence of ECF was surprisingly more common in elective surgical patients, overall mortality among these patients was significantly higher in EGS. Additionally, ECFs are indeed associated with significantly increased odds of both readmission and mortality at 30- and 90-day time points. Finally, we provide estimates of the costs of acute inpatient care associated with this increased healthcare utilization.

While most EGS surgeons have anecdotal experience with ECF, minimal previous data has been published thus far quantifying the national morbidity and mortality associated with this complication. Previous studies have been limited to single-institution data, (9, 10) and this is the first study to our knowledge to examine the scope of the problem using a nationally representative database as well as to focus specifically on the EGS population. Our rates of fistula development, at 1–2%, are in line with previous trauma studies on the subject. (9, 10) The finding that rates are generally higher in elective operations is counterintuitive. Our procedure-specific calculations were motivated by the possibility that the overall finding was due to case mix; however, the procedure-level findings were similar. This may be due to competing endpoints. Specifically, EGS patients who may have developed ECF – those that may have been high-risk to begin with – may have died prior to ECF occurrence or prior to ECF detection. If this were the case, rates would appear higher in elective surgery. Alternatively, misclassification bias secondary to our EGS definition, the limitations of which are discussed further below, may have contributed to this result.

Mortality rates after ECF development have only been stated broadly in the past, though our rate of 10% in the highest risk groups (EGS patients with a fistula) is concordant with other reported numbers.(11) Given the already high overall risk of the EGS population and the potentially devastating consequences of this complication, this information is potentially quite informative in that it provides a national context for clinicians who take care of surgical patients.

Readmission risk is essentially unstudied in this population. We found that readmission rates are extremely high – rising as high as 50% in EGS patients with ECF at 90 days – demonstrating the need for a clinical focus on this population. It also raises the question of whether or not such patients should be viewed on par with other surgical patients when readmissions are being used as a quality indicator. If the implicit assumption behind such an indicator is that readmissions represent a suboptimal performance on the part of the hospital, perhaps the inclusion of those that might be expected to have a very high risk of readmission (regardless of the quality of their care) should be reconsidered. As an analogy from a non-surgical population, cancer patients are not included in medical readmission rates due to a known high risk of readmission unrelated to the quality of hospital care (16). Given the complexity of EGS ECF patients, an argument could be made for excluding this population from readmission calculations.

The financial burden of ECF has also not been well characterized prior to this work. The median total hospital charges of over $55,000 for fistula patients in our cohort is much higher than is seen in other similar surgical populations. For instance, the median cost per stay for colorectal surgery patients is $9,000. (17) Since this number represents the median total charges from the time of index operation to the end of 2016, it represents nearly a year for some patients and a much shorter time for others. It is therefore likely that this number underestimates the true annual cost associated with inpatient care for ECF patients. Additionally, since many of these patients may be discharged to skilled nursing facilities or require total parenteral nutrition (TPN), the numbers reported here only represent a small portion of the overall economic burden of ECF.

As a retrospective epidemiologic study using a reliable dataset, this constitutes Level II evidence. (18) This study design is optimal for studying this topic, as the exposure (ECF) is relatively uncommon, and loss to follow up over 90 days may be a problem in a prospectively studied cohort. The NRD provides a unique opportunity to examine this particular problem in this particular population. Using it, we were able to study a large sample, which is one of the greatest strengths of the study.

There are some limitations to this study design, however. Retrospective study of a pre-existing administrative dataset is subject to potential coding errors. In addition to errors in entry, there is the possibility that our definition of “ECF” was overly inclusive. As noted in the methods section, K63.2 represents “fistula of the intestine” without specification as to which organ has been fistulized to. It is possible this may include some internal fistulae, though it does not include vesicointestinal fistulae, enterovaginal fistulae, or fistulae in patients with inflammatory bowel disease (IBD), all of which have separate codes. The fact that these common non-cutaneous fistula varieties have their own codes mitigates against misclassification bias of this type.

Another potential limitation of our definitions involves the definition of EGS – while our method is likely to capture the majority of patients admitted urgently for a surgical problem, it may miss those who are a) electively admitted and then develop a separate EGS problem (i.e. elective surgical patient who requires urgent takeback), b) develop an EGS problem more than 48 hours into their admission (i.e. C. diff patient who progresses to toxic megacolon over the course of the stay), or c) both. Admittedly, our definition sacrifices sensitivity for specificity. We chose not to include a broader range of patients – say, any patient with an emergent admission, regardless of day of surgery – out of a concern that inappropriate patients might be included. For example, a stroke patient might be admitted urgently, and then require a feeding tube 2 weeks later. We deemed it worth the loss of some potentially informative patients in order to count these patients as elective.

Finally, while we have demonstrated an association between ECF and the outcomes described here, causation cannot be inferred. As we have shown here, patients who develop ECF have more comorbid conditions than those who do not, and it is possible that some of the mortality and readmission burden of these patients is related to their underlying medical conditions rather than ECF per se. However, it is not possible to tease out the relative contributions of these factors using administrative data and this work still serves to characterize the outcomes of this complex patient group.

Despite the above limitations, the evidence is strong for a significant burden caused by ECF in EGS patients. The results presented here provide important baseline epidemiologic data and may incentivize providers and administrators to think more closely about the contexts in which readmissions should or should not be counted. Future directions in this field should include longitudinal follow-up of ECF patients over an extended period, as our data surely underestimate their burden. Home nursing, wound care, potential need for total parenteral nutrition, reoperation to take down fistulae, and other factors likely have an important impact on patients’ lives and healthcare costs. Other logical next steps following this study are to investigate risk factors associated not only with the development of ECF, but with the consequences demonstrated here – readmissions and mortality. Doing so will allow those of us who care for these patients to design interventions to reduce these risks and the costs associated with them.

Conclusion

The development of ECF after a gastrointestinal operation is associated with significantly increased odds of both mortality and readmission, rates of which continue to climb out to at least 90 days. Mortality risk with ECF is particularly pronounced in EGS. Processes of care should be designed to mitigate risk in this high-risk cohort.