Actionable Risk Model for the Development of Surgical Site Infection after Emergency Surgery

Abstract

Background:

Surgical site infections (SSIs) increase mortality and the economic burden associated with emergency surgery (ES). A reliable and sensitive scoring system to predict SSIs can help guide clinician assessment and patient counseling of post-operative SSI risk. We hypothesized that after quantifying the ES post-operative SSI incidence, readily abstractable parameters can be used to develop an actionable risk stratification scheme.

Patients and Methods:

We reviewed retrospectively all patients who underwent ES operations at an urban academic hospital system (2005–2013). Comorbidities and operative characteristics were abstracted from the electronic health record (EHR) with a primary outcome of post-operative SSIs. Risk of SSI was calculated using logistic regression modeling and validated using bootstrapping techniques. Beta-coefficients were calculated to correlate risk. A simplified clinical risk assessment tool was derived by assigning point values to the rounded β-coefficients.

Results:

A total of 4,783 patients with a 13.2% incidence of post-operative SSIs were identified. The strongest risk factors associated with SSIs included acute intestinal ischemia, weight loss, intestinal perforation, trauma-related laparotomy, radiation exposure, previous gastrointestinal surgery, and peritonitis. The assessment tool defined three patient groups based on SSI risk. Post-operative SSI incidence in high-risk patients (34%; score = 6–10) exceeded that of medium- (11.1%; score = 3–5) and low-risk patients (1.5%; score = 1–2) (C statistic = 0.802). Patients with a risk score ≥10 points evidenced the highest post-operative SSI risk (71.9%).

Conclusion:

Pre-operative identification of ES patient risk for post-operative SSI may inform pre-operative patient counseling and operative planning if the proposed procedure includes medical device implantation. A clinically relevant seven-factor risk stratification model such as this empirically derived one may be suitable to incorporate into the EHR as a decision-support tool.

Surgical site infections (SSIs) are the most common health-care–associated infection and the most common cause of unplanned hospital re-admission [1]. These infections greatly increase the economic burden associated with emergency surgery (ES) [2]. A reliable and accurate system that identifies patients at increased SSI risk may help focus strategies to decrease SSI frequency and cost.

Such as system may help guide clinician assessment and patient counseling regarding post-operative SSI risk. Some scoring systems have been developed and deployed, but appear most reliable in predicting post-operative mortality, sepsis, and in particular, pneumonia [3]. Unfortunately, none appear able to predict SSI precisely, especially in the patient requiring emergency surgery (ES) [4,5]. Therefore, an actionable risk stratification scheme capable of guiding operative management and patient counseling for the risk of post-operative SSI after ES is needed. Accordingly, we hypothesized that by quantifying post-ES SSI risk, we could identify patient parameters and intra-operative findings that would inform a high-reliability model to create an actionable risk stratification scheme.

Patients and Methods

Data source

An Institutional Review Board-approved retrospective cohort study was conducted including consecutive adult patients undergoing emergent exploratory laparotomy (University of Pennsylvania Health System; January 2005 through June 2016). Patients and procedures were identified in the electronic health record (EHR) by International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9 CM) and year-specific Current Procedural Terminology (CPT) codes (Table 1). Patients with missing or incomplete data were excluded. Each patient's index and subsequent encounters were linked to support longitudinal assessment across all inpatient and outpatient care.

Table 1.

Summary of ICD-9 CM Diagnostic Coding Scheme for Medical Comorbidities

Diagnosis	Subgroup	ICD-9 CM diagnostic codes
Diabetes		250, 250.0
Hypertension		401, 401.0, 401.1, 401.9
Smoking		305.1, 305.10, 305.11, 305.13, 989.84, V15.82
Cardiovascular disease	Coronary artery disease	414-414.07, 414.2, 414.8-414.9
	Peripheral vascular disease	443, 443.8, 443.89-443.9
	Hyperlipidemia	272.0, 272.2, 272.4
	Congestive heart failure	428.0, 428.1, 428.22, 428.32-428.33, 428.40-428.43, 428.9
Pulmonary disease	Chronic obstructive pulmonary disease	491.0, 491.1, 491.2-491.22, 491.8, 491.9, 492.0, 492.8, 518.1, 518.2
	Pulmonary hypertension	416
	Respiratory failure	518.83-518.84, V46.1-V46.11
Renal disease	Chronic renal failure	585.0-585.6, 585.9, 586, 587
	Long-term dialysis	V45.10-V45.11
Liver disease	Ascites	789.5-789.51, 789.59
	Cirrhosis	571.0, 571.2, 571.4-571.41, 571.5, 571.6
	Varices	456.0-456.21, 572.3-572.4, 572.8, 573.5
History of radiation		909.2, 990, V15.3, V58.0, V67.1
History of chemotherapy		V07.30, V07.39, V58.1-V58.11, V67.2, V87.41
Coagulopathy		286.0-286.5, 286.9, 289.81
Immunosuppressed	Human immunodeficiency virus	42.0-42.2, 42.9, 43.3, 44.9, V08, V09
	Other	284.1
Malnutrition		261-263.1, 263.8, 263.9
Anemia		285.1, 285.2-285.22, 285.29, 285.3, 285.8, 285.9

ICD-9 CM = International Classification of Diseases, 9th Revision, Clinical Modification.

Demographics, index surgery indications, and medical histories were abstracted from the EHR. Comorbidities were defined using the Elixhauser index; body mass index (kg/m²) was encoded using the World Health Organization classification [1,2]. Composite risk factors were defined as: cardiovascular (coronary artery disease, peripheral vascular disease, congestive heart failure, prior myocardial infarction, or percutaneous intervention); pulmonary (chronic obstructive pulmonary disease, acute or chronic respiratory failure, pulmonary hypertension); hepatic (cirrhosis, ascites, varices); renal (acute kidney failure or chronic kidney disease, long-term dialysis); and prior abdominal surgical interventions (Table 1).

The primary outcome of interest was post-operative SSI, defined by the presence of any of the following ICD-9 CM codes: 682.2, 942.12, 998.5, 998.59, 513.0-513.1. Presence of any SSI was considered a binary event and recorded as present or not present. A secondary outcome included cost of care. Financial data were provided by each institution's finance department for the primary admission and subsequent related readmissions. This data did not include outpatient data.

Analysis

Univariable analysis was performed comparing frequencies of demographic parameters by the study's outcome variable, SSI. The χ² and Fisher exact tests were both utilized. Specifically, Fisher test was utilized when a frequency of n ≤5 was observed on analysis; the χ² test result was used in all other cases. Statistical significance was defined as p ≤ 0.05. All variables with a p value ≤0.10 were defined as model-eligible variables. Covariance of model-eligible parameters was assessed and used in conjunction with clinical judgment to define which model-eligible parameters would be included in the final model. The final set of model-eligible parameters was incorporated into a multivariate logistic regression (MVR), with the outcome parameter of SSI. The model's concordance statistic (C statistics) was calculated to assess predictive capacity.

Risk-scoring transformation

To generate a more clinically useful tool, the model was transformed into a risk-scoring-tool. Beta-coefficients for all independent parameters that reached statistical significance within the model were divided by their greatest common denominator, and then rounded to the nearest whole number; this number was defined as the rounded risk score. Each subject's rounded risk scores were summed based on comorbidities, resulting in a total risk score. To clarify, a subject with no risk factors present has a risk score of zero points, whereas one with all risk factors present has a risk score of 13. Risk scores were then binned into risk tiers (low, medium, or high) based on average hernia incidence for each given risk score. Average SSI incidence was calculated for each risk tier.

Results

Univariable analysis

A total of 4,738 subjects met inclusion criteria, 15.2% (n = 632) of whom developed SSI after emergent laparotomy. There were no differences in age (49.1 ± 16.8 vs. 50.2 ± 18.6; p = 0.19) or body mass index (28.6 ± 11.0 vs. 28.2 ± 7.8; p = 0.8205). The SSI cohort was noted to have a substantially longer follow-up (59.3 ± 32.1 months vs. 53.7 ± 34.0 months; p < 0.001) than those who did not develop post-operative SSI. Additionally, the SSI cohort had a larger proportion of Caucasians (63.7% vs. 53.3%; p < 0.001), and a higher rate of tobacco use (40.7% vs. 32.0%; p < 0.001), anemia (57.6% vs. 43.56%; p < 0.001), recent weight loss (37.8% vs. 19.1%; p < 0.001), hepatic disease (26.7% vs 17.9%; p < 0.001), and prior abdominal surgery (56.9% vs. 19.9%; p < 0.001). Patients in the SSI cohort had a greater frequency of prior oncologic therapy such as chemotherapy (4.4% vs. 4.2%; p < 0.001) or radiation (7.8% vs. 2.5%; p < 0.001). Patients who had SSI were more likely to have peritonitis, which was defined as pus, succus, or stool in the peritoneal space (78.2% vs. 50.4%; p < 0.001), intestinal perforation (18.8% vs 8.9%; p < 0.001), or acute intestinal ischemia (16.1% vs. 8.5%; p < 0.001) and were less likely to have ectopic pregnancy (0.32% vs. 16.2%; p < 0.001) or gastrointestinal hemorrhage (15% vs, 18.8%; p < 0.001) than those who did not develop SSI (Table 2).

Table 2.

Univariable Analysis of Patients Experiencing Post-Operative SSI

	None		SSI
	n	%	n	%	p
N	4,151		632
Age					<0.001
<45	1,692	41%	244	39%
45–55	667	16%	125	20%
55–65	785	19%	144	23%
65–80	788	19%	105	17%
>80	217	5%	14	2%
Obesity	725	17%	142	22%	0.002
Race
Race: Asian	102	2%	13	2%	0.539
Race: Black	1,656	40%	201	32%	<0.001
Race: Caucasian	2,213	53%	403	64%	<0.001
Diabetes	881	21%	152	24%	0.108
Hypertension	2,090	50%	336	53%	0.187
Smoking history	1,399	34%	268	42%	<0.001
Smoking and/or drug history	2,050	49%	305	48%	0.323
Hyperlipidemia	1,248	30%	208	33%	0.147
Anemia	1,808	44%	364	58%	<0.001
Recent weight loss	792	19%	239	38%	<0.001
Cardiovascular disease	884	21%	157	25%	0.044
Liver disease	741	18%	169	27%	<0.001
Renal failure	741	18%	113	18%	0.986
Pulmonary disease	1,145	28%	196	31%	0.074
History of chemotherapy	174	4%	53	8%	<0.001
History of radiation	105	3%	49	8%	<0.001
History of gastrointestinal surgery	829	20%	360	57%	<0.001
0 Elixhauser comorbidities	613	15%	31	5%	<0.001
2+ Elixhauser comorbidities	2,989	72%	553	88%	<0.001
Reason for emergency surgery
Contaminated peritonitis	2,091	50%	494	78%	<0.001
Intestinal ulcer ± perforation	370	9%	119	19%	<0.001
Acute bowel ischemia	354	9%	102	16%	<0.001
Ectopic pregnancy	671	16%	2	0%	<0.001
GI bleed NOS	779	19%	95	15%	0.024
Post-operative hernia	498	12%	193	31%	<0.001
BMI	28.2	7.8	28.6	11.0	0.8205
Age	50.2	18.6	49.1	17.1	0.1948
Follow-up (mo)	53.7	35.0	59.3	34.2	<0.001

SSI = surgical site infection; NOS = not otherwise specified; GI = gastrointestinal; BMI = body mass index.

Multivariable regression analysis

On multivariable regression analysis, those with recent weight loss (β-coefficient: 0.80; p = 0.013), prior gastrointestinal surgery (β-coefficient: 0.1.49; p < 0.001), or radiation therapy (β-coefficient: 1.06; p < 0.001) were at increased risk of developing SSI after ES. Additionally, patients were at increased risk of developing SSI if they had acute intestinal ischemia (β-coefficient: 0.49; p = 0.013), intestinal perforation (β-coefficient: 0.87; p < 0.001), or peritonitis (β-coefficient: 1.37; p < 0.001). Patients undergoing emergency laparotomy after injury did not demonstrate an increased SSI risk (β-coefficient: 1.04; p = 0.144) (Table 3). This model generated a concordance statistic of 0.802, and each parameter included in the model was assigned a rounded risk score (Table 2).

Table 3.

Multivariable Regression Analysis

SSI	β-coefficient	p	Exact risk	Rounded risk score
Acute bowel ischemia	0.49	0.013	1.00	1
History of weight loss	0.80	<0.001	1.63	2
Intestinal ulceration or perforation	0.87	<0.001	1.76	2
Exploratory laparotomy in trauma	1.04	0.144	2.11	2
History of radiation	1.06	<0.001	2.14	2
Contaminated peritonitis	1.37	<0.001	2.79	3
History of GI surgery	1.49	<0.001	3.01	3
Constant	−3.90	<0.001

SSI = surgical site infection; GI = gastrointestinal.

Patients with risk scores of zero to two were considered to be low risk, whereas those with scores of three to five points were medium risk. A score higher than six was defined as high risk. The proportion of patients in each tier and their SSI incidence is presented in Table 4. Patients with a risk score ≥10 points evidenced the highest post-operative SSI risk (71.9%) (Fig. 1).

FIG. 1.

Incidence of surgical site infections (SSI) increased as risk score increased. Risk scores of 0–2 were considered low risk, whereas those with scores of 3–5 points were medium risk. A score >6 was defined as high isk. With increasing score, the risk of SSI increases with those >10 points having >70% risk of SSI.

Table 4.

Surgical Site Infection after Emergency Exploratory Laparotomy Risk Model

Risk tier	Risk score	Percent of cohort	SSI incidence
Low	0–2	36.0%	1.5%
Medium	3–5	48.5%	11.1%
High	6–13	15.5%	34.0%

SSI = surgical site infection.

Cost analysis and secondary outcomes

A secondary outcome was the cost incurred by the healthcare system as a result of having an SSI. Patients with an SSI had a substantial increase of $8,887 ($55,872 vs. 46,985) in their total cost of care (p < 0.001). This resulted in a conglomerate increased care cost of $5,616,546 over the course of the study. Another secondary outcome was the incidence of post-operative hernia. Patients who developed an SSI had a 2.5 times increased incidence of post-operative hernia compared with those without SSI (30.4% vs. 12%, p < 0.001).

Discussion

In this study we discovered and stratified risk factors associated with developing SSI after ES to generate a clinically usable predictive model. Although multiple parameters impact SSI occurrence in ES patients, prior gastrointestinal surgery and the presence of peritonitis unsurprisingly exerted the strongest influence on post-operative SSI. Multiple scoring systems and risk stratification schemes have been developed to predict post-operative infection [6 –9]. Among them, the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) stands out as the largest database for providing risk- and reliability-adjusted data that enables surgeons to compare their outcomes, in particular those related to SSI [10]. The NSQIP risk calculator utilizes multiple parameters to calculate patient morbidity and mortality—and locates that probability within a range—but does not consider the type of operation or whether the surgery was performed emergently [11]. Similar to most models, the ACS NSQIP calculator is incapable of predicting the interaction of a host of elements and therefore may underestimate specific risks including that of SSI. Unsurprisingly, many studies have criticized the ACS NSQIP risk calculator as being inaccurate for the prediction of SSI [12,13].

More recently the Emergency Surgery Score (ESS) has emerged as a more reliable and accurate scheme to predict mortality and post-operative complication in ES patients [9]. Although the ESS has also been shown to predict post-operative complications such as pneumonia or sepsis, its ability to predict SSI is, unfortunately, less precise. Unlike other risk stratifications schemes, our model, and the parameters utilized herein demonstrate great accuracy with regard to SSI in ES patient as noted by a C statistic of 0.802. This is in contrast with previous reports in which the accuracy of SSI risk prediction was lower [9,14].

Our findings have practical implications for the acute care surgeon. Benefits fall into three categories: benchmarking, pre-operative shared decision-making, and intra-operative decision-making. Understanding the likely risk for post-operative SSI after an ES procedure is essential in determining a center's or surgeon's performance. It establishes the norm against which other and subsequent care is assessed. It is that care performance—in patients similar to the one before the surgeon—that informs shared patient decision-making regarding a procedure, be it life-saving or palliative when undertaken as an emergency. Furthermore, if the tool can also inform the surgeon regarding key intra-operative decisions that impact outcome, then it is a tool that has clinical relevance outside of the office or the inpatient room. Such utility also drives one to use such a tool with great frequency because it is not limited to only one segment of patient care. Skin closure versus packing, packing versus negative pressure wound therapy, or a dry dressing compared with overlying negative pressure therapy are all decisions that may be impacted by this tool. Few aspects render a tool more useless than being cumbersome. Ours uses a limited number of elements—seven—in a simple scoring system that is suitable for development and deployment using an app for a desktop or smartphone.

Risk calculators have been developed for a wide variety of surgical procedures and scenarios [12,15]. Emergency surgery poses a unique patient population characterized with issues of perfusion, dysoxia, bleeding, coagulopathy, acidosis, temperature regulation, and concomitant organ failure [16 -19]. These factors inherently make this population at increased risk of all complications but particularly those associated with the development of infection [9,14]. If the tool enables decision-making that decreases infection, risk utilization could lead to a health system financial benefit in excess of $500,000 per year.

The development of SSI not only poses an immediate threat to the patient and their recovery but may also impart downstream risks to the implantation of prosthetics. By understanding a patient's increased risk of SSI, one can modify their operative strategy for the placement of a prosthetic, increase their surveillance of prosthetic infection, or simply choose not to use a prosthetic if the risk is excessive. Given our findings of an increased incidence of post-operative hernia in the SSI group, these management strategies become increasingly important. Biofilm formation after the initial SSI will place any subsequent prosthetic implant at risk of colonization or infection. Although much work has been done to demonstrate that the use of certain prosthetics is safe in the contaminated field, understanding that risk will guide selection of the specific prosthetic (because they have different nanofeatures that influence infection risk) or operative planning [20,21].

There are a variety of relevant limitations that influence the interpretation of our data and the generalizability of our results. First, the data for this tool are derived from our local EHR and were not sufficiently specific to extract the level of SSI or therapy for that infection once it was identified. Nonetheless, using SSI as a binary occurrence is useful in model generation to predict the risk of any kind of SSI. Second, although the tool appears to perform well, it has not been tested in a prospective fashion and has not been validated using data other than those derived from our system. Third, because our data assessed only emergency laparotomy, their performance for other kinds of operative interventions such as thoracotomy may be more or less robust. Fourth, the diagnosis of SSI was extracted from the EHR, and as such, was not subject to the rigor one might impose upon doing so within a clinical trial. Nonetheless, such inaccuracies should be evenly distributed across all patients. Moreover, the goal was to assess for any SSI, not just those occurring at a particular depth. Fifth, we were unable to validate that pre-operative antibiotic agents were given in the correct time frame, and that the antibiotic agents addressed all recovered pathogens. Therefore, it is possible that some pathogens were not susceptible to the administered antibiotic agents, and their lack of coverage could influence study data. Sixth, we were also unable to determine the adequacy of source control at the index procedure, or whether a second procedure was needed to achieve adequate control.

Conclusions

In summary, there are no scoring models that predict SSI occurrence after emergent laparotomy accurately and reliably. Accordingly, those requiring emergency abdominal exploration form a unique and somewhat orphan group in terms of benchmarking infection prevention management activities. Pre-operative identification of ES patients at high-risk for post-operative SSI may inform pre-operative patient counseling to support shared decision-making, as well as intra-operative activities. Such knowledge may be essential if the proposed procedure includes medical device implantation such as prosthetic mesh, a feature that appears to increase SSI risk. A clinically relevant seven-factor risk stratification model such as this empirically derived one may be ideal to incorporate into existing EHR's as a decision-support tool for this unique population.

Footnotes

Funding Information

No funding was received for this work.

Author Disclosure Statement

Drs. Fernandez-Moure, Wes, and Kaplan have no conflicts of interest to disclose. Dr. Fischer is a consultant for Becton Dickinson, Gore, Integra, and Allergan.

References

Merkow

, Ju

, Chung

, et al. Underlying reasons associated with hospital readmission following surgery in the United States. JAMA, 2015; 313:483–495.

Zimlichman

, Henderson

, Tamir

, et al. Health care-associated infections: A meta-analysis of costs and financial impact on the US health care system. JAMA Intern Med, 2013; 173:2039–2046.

Bilimoria

, Liu

, Paruch

, et al. Development and evaluation of the universal ACS NSQIP surgical risk calculator: A decision aid and informed consent tool for patients and surgeons. J Am Coll Surg, 2013; 217:833–842.

Nandan

, Bohnen

, Sangji

, et al. The Emergency Surgery Score (ESS) accurately predicts the occurrence of postoperative complications in emergency surgery patients. J Trauma Acute Care Surg, 2017; 83:84–89.

van Walraven

, Musselman

. The Surgical Site Infection Risk Score (SSIRS): A model to predict the risk of surgical site infections. PLoS One, 2013; 8:e67167.

Richards

, Edwards

, Culver

, Gaynes

. Nosocomial infections in medical intensive care units in the United States. National Nosocomial Infections Surveillance System. Crit Care Med, 1999; 27:887-892.

Berger

, Li

, Hicks

, et al. Development and validation of a risk-stratification score for surgical site occurrence and surgical site infection after open ventral hernia repair. J Am Coll Surg, 2013; 217:974–982.

Gervaz

, Bandiera-Clerc

, Buchs

, et al. Scoring system to predict the risk of surgical-site infection after colorectal resection. Br J Surg, 2012; 99:589–895.

Han

, Lee

, Achanta

, et al. Emergency surgery score accurately predicts the risk of post-operative infection in emergency general surgery. Surg Infect, 2019; 20:4–9.

10.

, Ko

, Hall

, et al. A comparison of 2 surgical site infection monitoring systems. JAMA Surg, 2015; 150:51–57.

11.

Cohen

, Liu

, Ko

, Hall

. Improved surgical outcomes for ACS NSQIP hospitals over time: Evaluation of hospital cohorts with up to 8 years of participation. Ann Surg, 2016; 263:267–273.

12.

Ventral Hernia Outcome Collaborative; Mitchell

, Holihan

, et al. Do risk calculators accurately predict surgical site occurrences?. J Surg Res, 2016; 203:56–63.

13.

Levy

, Lally

, Blakely

, et al. Surgical wound misclassification: A multicenter evaluation. J Am Coll Surg, 2015; 220:323–329.

14.

Mueck

, Kao

. Patients at high-risk for surgical site infection. Surg Infect, 2017; 18:440–446.

15.

Gibbons

, Bruce

, Carpenter

, et al. Identification of risk factors by systematic review and development of risk-adjusted models for surgical site infection. Health Technol Assess, 2011; 15:1–156.

16.

Masuda

, Kuramoto

, Tanimoto

, et al. Intraoperative baseline oxygen consumption as a prognostic factor in emergency open abdominal surgery. J Crit Care, 2016; 32:42–47.

17.

Meledeo

, Herzig

, Bynum

, et al. Acute traumatic coagulopathy: The elephant in a room of blind scientists. J Trauma Acute Care Surg, 2017; 82(6S Suppl 1):S33–S40.

18.

Kaplan

, Kellum

. Comparison of acid-base models for prediction of hospital mortality after trauma. Shock, 2008; 29:662–666.

19.

Williams

, Herndon

. Metabolic and endocrine considerations after burn injury. Clin Plast Surg, 2017; 44:541–553.

20.

Carbonell

, Cobb

. Safety of prosthetic mesh hernia repair in contaminated fields. Surg Clin North Am, 2013; 93:1227–1239.

21.

Fernandez-Moure

, Mydlowska

, Shin

, Vella

, Kaplan

. Nanometric considerations in biofilm formation. Surg Infect, 2019; 20:167–173.