American Journal of Infection Control 42 (2014) 1291-5
Contents lists available at ScienceDirect
American Journal of Infection Control
American Journal of Infection Control
journal homepage: www.ajicjournal.org
Major article
Using clinical variables to guide surgical site infection detection: A novel surveillance strategy Westyn Branch-Elliman MD, MMSc a, b, c, *, Judith Strymish MD a, c, Kamal M.F. Itani MD c, d, e, Kalpana Gupta MD, MPH a, f a
Department of Medicine, Boston VA Healthcare System, Boston, MA Department of Healthcare Quality, Division of Infection Control, Beth Israel Deaconess Medical Center, Boston, MA Department of Medicine, Harvard University Medical School, Boston, MA d Department of Surgery, Boston VA Healthcare System, Boston, MA e Department of Surgery, Boston University School of Medicine, Boston, MA f Department of Medicine, Boston University School of Medicine, Boston, MA b c
Key Words: Post-operative care Quality improvement Electronic tool
Background: Surgical site infections (SSIs) are a common and expensive health careeassociated infection, and are used as a health care quality benchmark. As such, SSI detection is a major focus of infection prevention programs. In an effort to improve on conventional surveillance methods, a simple algorithm for SSI detection was developed using clinical variables not traditionally included in National Healthcare Safety Network definitions. Methods: A case-control study was conducted among surgeries performed at the Veterans Affairs Boston Healthcare System between January 2008 and December 2009. SSI cases were matched to controls without SSI. Clinical variables (administrative, microbiological, pharmacy, radiology) were compared between the groups to determine those that best identified SSI. Results: A total of 70 SSIs were matched to 70 controls. On multivariable analysis, variables significantly associated with SSI identification were wound culture order, computed tomography scan/magnetic resonance imaging order, antibiotic order within 30 days after surgery, and application of a relevant International Classification of Disease, Ninth Revision code. Among patients with no SSI identifiers, 98% were correctly classified as having no SSI. Among patients with multiple SSI identifiers, 97.1% were correctly identified as having SSI. The area under the curve for this model was 0.87. Conclusion: We have derived a novel surveillance algorithm for SSI detection with excellent operating characteristics. This algorithm could be automated to streamline infection control efforts. Published by Elsevier Inc. on behalf of the Association for Professionals in Infection Control and Epidemiology, Inc.
Surgical site infection (SSI) is a common and expensive health careeassociated infection (HAI), costing up to $10 billion per year in the United States alone.1 SSIs account for almost 20% of all HAIs and complicate approximately 2% of all surgeries.2,3 Given their significant impact on patient quality of life and overall health care costs,4,5 SSIs serve as a quality benchmark6 and are a strong focus of national infection prevention initiatives.7,8 Recent public reporting
* Address correspondence to Westyn Branch-Elliman, MD, MMSc, ECHCS, 1055 Clermont St, Mailstop 111L, Denver, CO 80220. E-mail address: [email protected] (W. Branch-Elliman). Funded in part by the VA Patient Safety Center for Inquiry, VA Boston Healthcare System (grant no. XVA 68-023). Conflicts of interest: None to report.
requirements have prompted near-universal surveillance of postoperative patients to detect SSIs.9,10 Personnel in infection control departments spend valuable time on manual chart review11 to collect mandatory data for public reporting. Limitations to manual review include lack of interinstitutional and interrater reliability,12 as well as an increasing burden on already limited infection prevention resources.11 An automated platform for SSI detection using clinical variables has several potential advantages compared with the gold standard manual review process. First, an electronic monitoring system would provide an objective and consistent reporting platform across multiple institutions.13 Second, an electronic monitoring system would allow for expansion of current surveillance programs to include all surgeries conducted nationwide, rather than simply
0196-6553/$00.00 - Published by Elsevier Inc. on behalf of the Association for Professionals in Infection Control and Epidemiology, Inc. http://dx.doi.org/10.1016/j.ajic.2014.08.013
1292
W. Branch-Elliman et al. / American Journal of Infection Control 42 (2014) 1291-5
the subset currently reviewed.14 Third, in many practice settings, manual review of postsurgical cases is driven by positive bacterial culture results; the use of a positive bacterial culture to guide detection is known to miss a high proportion of cases,15 and an electronic algorithm with additional variables might identify many of the cases missed by standard surveillance protocols. Using clinical variables available as administrative data in the Veterans Affairs (VA) electronic medical record, we sought to derive a surveillance algorithm based on clinical factors that can be extracted electronically to identify SSIs, thereby reducing the need for detailed manual chart review. This approach expands previous SSI electronic surveillance studies by using variables guided by clinical practice, rather than basing SSI identification solely on National Healthcare Safety Network (NHSN) criteria. METHODS Setting The VA Boston Healthcare System is a 245-bed facility that performs approximately 5000 surgeries annually. Surgeries are performed at both the acute care hospital (West Roxbury campus) and the ambulatory care hospital (Jamaica Plain campus). All patients who underwent a clean or clean contaminated surgical procedure within the VA Boston Healthcare System between January 1, 2008, and December 2009 were included in the present analysis. Dental, ophthalmologic, and contaminated surgeries were excluded from the analysis. If a patient underwent several surgical procedures during the study period, only the first was eligible for inclusion.
Table 1 Baseline characteristics of matched patients Characteristic Age, y, mean (95% CI)* Male sex, n (%) Diabetes, n (%) Vancomycin prophylaxis, n (%) Chlorhexidine prep, n (%) Inpatient status, n (%) Surgery type, n (%) Cardiac surgery General surgeryy Neurosurgery Orthopedic surgeryz Otorhinolaryngologyx Thoracic surgery Urologyjj Vascular surgery American Surgical Association class, n (%) 1 2 3 4 Unknown Wound class, n (%) 0 1
SSI (n ¼ 70) 65.8 67 8 12 22 63
No SSI (n ¼ 70)
P value
(62.9-68.6) (95.7) (19.5) (17.1) (31.4) (90)
64.6 66/70 2/70 8/70 27/70 63/70
(61.2-68.0) (94.) (6.9) (11.4) (38.6) (90)
8 30 4 7 2 2 6 11
(11.4) (42.9) (5.7) (10) (2.9) (2.9) (8.6) (15.7)
8/70 30/70 4/70 7/70 2/70 2/70 6/70 11/70
(11.4) (42.9) (5.7) (10) (2.9) (2.9) (8.6) (15.7)
0 11 41 16 5
(0) (16.2) (60.3) (23.5) (7.14)
1/70 10/70 39/70 15/70 2/70
(1.54) (15.4) (60.0) (23.1) (2.9)
.69
47/70 (68.2) 31/53 (44.3)
.16
39 (55.7) 22 (31.9)
1.0 .18 .47 .48 1.0 1.0
Mean age of the overall cohort was 65 years. *Missing age in 2 patients, 1 patient in each category. y Of the 30 general surgery cases included in each group, 1 was performed on an outpatient basis. z Of the 7 orthopedic cases included in each group, 2 were performed on an outpatient basis. x All otorhinolargynology cases were performed on an outpatient basis. jj Of the 6 urology cases included in each group, 2 were performed on an outpatient basis.
Evaluation for SSI Matching Standard NHSN definitions of SSI were used. SSIs were ascertained from the VA Boston infection control and VA National Surgical Quality Improvement Program (VASQIP) databases. Methicillin-resistant Staphylococcus aureus (MRSA), methicillinsensitive S aureus (MSSA), and other organisms were tracked through the infection control database for 30 days postsurgery, or for 1 year postsurgery in cases of implants. Data collection Data were extracted from the VA Health Information Systems and Technology Architecture. Demographic (ie, age, sex, race), clinical (ie, preexisting conditions, smoking status, MRSA colonization status), and surgical data (ie, date of surgery, type of surgery, surgical risk score) were extracted electronically. Relevant international Classification of Diseases, Ninth Revision (ICD-9) codes (998.x) for all patients included in the study were collected as well. Clinical data were abstracted manually, including administrative data (ie, emergency room visits, readmissions, surgery follow-up visits, surgery reconsultations, deaths), laboratory data (ie, wound, tissue, or sterile space culture ordered, wound, tissue, or sterile space culture positive; if bacterial culture positive, then bacterial organisms isolated, blood culture ordered and results, urine culture ordered, and results), pharmacy data (ie, oral vs intravenous antibiotic use, antibiotic type), and radiographic data (ie, X-ray, computed tomography [CT] scan, or magnetic resonance imaging [MRI] ordered). Urine culture and X-ray ordering data were collected to capture variables that may reflect a workup for evaluation of postoperative fever. All antibiotics ordered for patients during the 30-day postoperative period were collected.
All patients with an SSI were identified and then matched 1:1 to another patient without an SSI. Controls were matched based on surgical type and then on the date of surgery. In cases where more than 1 potential control was available, a random number generator was used, and the potential control with the lowest number was chosen. Because the goal was to identify SSI for surveillance purposes, rather than to predict the risk of SSI, matching for known SSI risk factors was not undertaken. Statistical analysis All data were analyzed using JMP version 10.0 (SAS Institute, Cary, NC). Baseline demographic and clinical variables were compared using the Student t test and Fisher’s exact test as appropriate. Clinical data potentially associated with SSI also were compared between the 2 groups using the Student t test and Fisher’s exact test as appropriate. Variables possibly associated with appropriate identification of SSI on univariate analysis (P < .20) were entered into a multivariable logistic regression model. Variables significantly associated with SSI detection (P < .05) were retained. The HosmerLemmeshow test was used to determine model goodness of fit. Two composite antibiotic variables were eligible for inclusion in the mode: an order for any antibiotic and an order for an SSIidentifying antibiotic, which included an order for any antibiotic commonly prescribed in the postoperative setting. After determining the optimal variables for model inclusion, we developed a points system for classifying patients as low risk, medium risk, or high risk for SSI. Points were assigned based on the
W. Branch-Elliman et al. / American Journal of Infection Control 42 (2014) 1291-5
1293
Table 2 Multivariable logistic regression model of administrative variables associated with SSI Variable Wound culture ordered CT or MRI ordered Antibiotic ordered* ICD-9 code for SSIy
OR
95% CI
P value
6.0 6.0 9.5 98
1.7-23.5 1.8-21.7 3.1-32.5 14.8-2023
.006 .003
Contents lists available at ScienceDirect
American Journal of Infection Control
American Journal of Infection Control
journal homepage: www.ajicjournal.org
Major article
Using clinical variables to guide surgical site infection detection: A novel surveillance strategy Westyn Branch-Elliman MD, MMSc a, b, c, *, Judith Strymish MD a, c, Kamal M.F. Itani MD c, d, e, Kalpana Gupta MD, MPH a, f a
Department of Medicine, Boston VA Healthcare System, Boston, MA Department of Healthcare Quality, Division of Infection Control, Beth Israel Deaconess Medical Center, Boston, MA Department of Medicine, Harvard University Medical School, Boston, MA d Department of Surgery, Boston VA Healthcare System, Boston, MA e Department of Surgery, Boston University School of Medicine, Boston, MA f Department of Medicine, Boston University School of Medicine, Boston, MA b c
Key Words: Post-operative care Quality improvement Electronic tool
Background: Surgical site infections (SSIs) are a common and expensive health careeassociated infection, and are used as a health care quality benchmark. As such, SSI detection is a major focus of infection prevention programs. In an effort to improve on conventional surveillance methods, a simple algorithm for SSI detection was developed using clinical variables not traditionally included in National Healthcare Safety Network definitions. Methods: A case-control study was conducted among surgeries performed at the Veterans Affairs Boston Healthcare System between January 2008 and December 2009. SSI cases were matched to controls without SSI. Clinical variables (administrative, microbiological, pharmacy, radiology) were compared between the groups to determine those that best identified SSI. Results: A total of 70 SSIs were matched to 70 controls. On multivariable analysis, variables significantly associated with SSI identification were wound culture order, computed tomography scan/magnetic resonance imaging order, antibiotic order within 30 days after surgery, and application of a relevant International Classification of Disease, Ninth Revision code. Among patients with no SSI identifiers, 98% were correctly classified as having no SSI. Among patients with multiple SSI identifiers, 97.1% were correctly identified as having SSI. The area under the curve for this model was 0.87. Conclusion: We have derived a novel surveillance algorithm for SSI detection with excellent operating characteristics. This algorithm could be automated to streamline infection control efforts. Published by Elsevier Inc. on behalf of the Association for Professionals in Infection Control and Epidemiology, Inc.
Surgical site infection (SSI) is a common and expensive health careeassociated infection (HAI), costing up to $10 billion per year in the United States alone.1 SSIs account for almost 20% of all HAIs and complicate approximately 2% of all surgeries.2,3 Given their significant impact on patient quality of life and overall health care costs,4,5 SSIs serve as a quality benchmark6 and are a strong focus of national infection prevention initiatives.7,8 Recent public reporting
* Address correspondence to Westyn Branch-Elliman, MD, MMSc, ECHCS, 1055 Clermont St, Mailstop 111L, Denver, CO 80220. E-mail address: [email protected] (W. Branch-Elliman). Funded in part by the VA Patient Safety Center for Inquiry, VA Boston Healthcare System (grant no. XVA 68-023). Conflicts of interest: None to report.
requirements have prompted near-universal surveillance of postoperative patients to detect SSIs.9,10 Personnel in infection control departments spend valuable time on manual chart review11 to collect mandatory data for public reporting. Limitations to manual review include lack of interinstitutional and interrater reliability,12 as well as an increasing burden on already limited infection prevention resources.11 An automated platform for SSI detection using clinical variables has several potential advantages compared with the gold standard manual review process. First, an electronic monitoring system would provide an objective and consistent reporting platform across multiple institutions.13 Second, an electronic monitoring system would allow for expansion of current surveillance programs to include all surgeries conducted nationwide, rather than simply
0196-6553/$00.00 - Published by Elsevier Inc. on behalf of the Association for Professionals in Infection Control and Epidemiology, Inc. http://dx.doi.org/10.1016/j.ajic.2014.08.013
1292
W. Branch-Elliman et al. / American Journal of Infection Control 42 (2014) 1291-5
the subset currently reviewed.14 Third, in many practice settings, manual review of postsurgical cases is driven by positive bacterial culture results; the use of a positive bacterial culture to guide detection is known to miss a high proportion of cases,15 and an electronic algorithm with additional variables might identify many of the cases missed by standard surveillance protocols. Using clinical variables available as administrative data in the Veterans Affairs (VA) electronic medical record, we sought to derive a surveillance algorithm based on clinical factors that can be extracted electronically to identify SSIs, thereby reducing the need for detailed manual chart review. This approach expands previous SSI electronic surveillance studies by using variables guided by clinical practice, rather than basing SSI identification solely on National Healthcare Safety Network (NHSN) criteria. METHODS Setting The VA Boston Healthcare System is a 245-bed facility that performs approximately 5000 surgeries annually. Surgeries are performed at both the acute care hospital (West Roxbury campus) and the ambulatory care hospital (Jamaica Plain campus). All patients who underwent a clean or clean contaminated surgical procedure within the VA Boston Healthcare System between January 1, 2008, and December 2009 were included in the present analysis. Dental, ophthalmologic, and contaminated surgeries were excluded from the analysis. If a patient underwent several surgical procedures during the study period, only the first was eligible for inclusion.
Table 1 Baseline characteristics of matched patients Characteristic Age, y, mean (95% CI)* Male sex, n (%) Diabetes, n (%) Vancomycin prophylaxis, n (%) Chlorhexidine prep, n (%) Inpatient status, n (%) Surgery type, n (%) Cardiac surgery General surgeryy Neurosurgery Orthopedic surgeryz Otorhinolaryngologyx Thoracic surgery Urologyjj Vascular surgery American Surgical Association class, n (%) 1 2 3 4 Unknown Wound class, n (%) 0 1
SSI (n ¼ 70) 65.8 67 8 12 22 63
No SSI (n ¼ 70)
P value
(62.9-68.6) (95.7) (19.5) (17.1) (31.4) (90)
64.6 66/70 2/70 8/70 27/70 63/70
(61.2-68.0) (94.) (6.9) (11.4) (38.6) (90)
8 30 4 7 2 2 6 11
(11.4) (42.9) (5.7) (10) (2.9) (2.9) (8.6) (15.7)
8/70 30/70 4/70 7/70 2/70 2/70 6/70 11/70
(11.4) (42.9) (5.7) (10) (2.9) (2.9) (8.6) (15.7)
0 11 41 16 5
(0) (16.2) (60.3) (23.5) (7.14)
1/70 10/70 39/70 15/70 2/70
(1.54) (15.4) (60.0) (23.1) (2.9)
.69
47/70 (68.2) 31/53 (44.3)
.16
39 (55.7) 22 (31.9)
1.0 .18 .47 .48 1.0 1.0
Mean age of the overall cohort was 65 years. *Missing age in 2 patients, 1 patient in each category. y Of the 30 general surgery cases included in each group, 1 was performed on an outpatient basis. z Of the 7 orthopedic cases included in each group, 2 were performed on an outpatient basis. x All otorhinolargynology cases were performed on an outpatient basis. jj Of the 6 urology cases included in each group, 2 were performed on an outpatient basis.
Evaluation for SSI Matching Standard NHSN definitions of SSI were used. SSIs were ascertained from the VA Boston infection control and VA National Surgical Quality Improvement Program (VASQIP) databases. Methicillin-resistant Staphylococcus aureus (MRSA), methicillinsensitive S aureus (MSSA), and other organisms were tracked through the infection control database for 30 days postsurgery, or for 1 year postsurgery in cases of implants. Data collection Data were extracted from the VA Health Information Systems and Technology Architecture. Demographic (ie, age, sex, race), clinical (ie, preexisting conditions, smoking status, MRSA colonization status), and surgical data (ie, date of surgery, type of surgery, surgical risk score) were extracted electronically. Relevant international Classification of Diseases, Ninth Revision (ICD-9) codes (998.x) for all patients included in the study were collected as well. Clinical data were abstracted manually, including administrative data (ie, emergency room visits, readmissions, surgery follow-up visits, surgery reconsultations, deaths), laboratory data (ie, wound, tissue, or sterile space culture ordered, wound, tissue, or sterile space culture positive; if bacterial culture positive, then bacterial organisms isolated, blood culture ordered and results, urine culture ordered, and results), pharmacy data (ie, oral vs intravenous antibiotic use, antibiotic type), and radiographic data (ie, X-ray, computed tomography [CT] scan, or magnetic resonance imaging [MRI] ordered). Urine culture and X-ray ordering data were collected to capture variables that may reflect a workup for evaluation of postoperative fever. All antibiotics ordered for patients during the 30-day postoperative period were collected.
All patients with an SSI were identified and then matched 1:1 to another patient without an SSI. Controls were matched based on surgical type and then on the date of surgery. In cases where more than 1 potential control was available, a random number generator was used, and the potential control with the lowest number was chosen. Because the goal was to identify SSI for surveillance purposes, rather than to predict the risk of SSI, matching for known SSI risk factors was not undertaken. Statistical analysis All data were analyzed using JMP version 10.0 (SAS Institute, Cary, NC). Baseline demographic and clinical variables were compared using the Student t test and Fisher’s exact test as appropriate. Clinical data potentially associated with SSI also were compared between the 2 groups using the Student t test and Fisher’s exact test as appropriate. Variables possibly associated with appropriate identification of SSI on univariate analysis (P < .20) were entered into a multivariable logistic regression model. Variables significantly associated with SSI detection (P < .05) were retained. The HosmerLemmeshow test was used to determine model goodness of fit. Two composite antibiotic variables were eligible for inclusion in the mode: an order for any antibiotic and an order for an SSIidentifying antibiotic, which included an order for any antibiotic commonly prescribed in the postoperative setting. After determining the optimal variables for model inclusion, we developed a points system for classifying patients as low risk, medium risk, or high risk for SSI. Points were assigned based on the
W. Branch-Elliman et al. / American Journal of Infection Control 42 (2014) 1291-5
1293
Table 2 Multivariable logistic regression model of administrative variables associated with SSI Variable Wound culture ordered CT or MRI ordered Antibiotic ordered* ICD-9 code for SSIy
OR
95% CI
P value
6.0 6.0 9.5 98
1.7-23.5 1.8-21.7 3.1-32.5 14.8-2023
.006 .003
