Risk factors and outcomes associated with surgical site infections after [PDF]

RISK FACTORS AND OUTCOMES ASSOCIATED WITH SURGICAL SITE INFECTIONS AFTER. CRANIOTOMY AND CRANIECTOMY by. Hsiu-Yin Chiang

0 downloads 4 Views 9MB Size

Recommend Stories


Reducing the Risk of Surgical Site Infections
Why complain about yesterday, when you can make a better tomorrow by making the most of today? Anon

Obstetric Surgical Site Infections
Life is not meant to be easy, my child; but take courage: it can be delightful. George Bernard Shaw

Prevalence and risk factors associated with sexually transmitted infections
If you feel beautiful, then you are. Even if you don't, you still are. Terri Guillemets

Risk Factors for Surgical Site Infection (SSI)
You miss 100% of the shots you don’t take. Wayne Gretzky

Patient safety: Surgical Site Infections
If you want to go quickly, go alone. If you want to go far, go together. African proverb

Risk factors and etiology of surgical site infection after radical neck dissection in patients with head
If you feel beautiful, then you are. Even if you don't, you still are. Terri Guillemets

Post Caesarean Surgical Site Infections
Why complain about yesterday, when you can make a better tomorrow by making the most of today? Anon

Epidemiology, Bacteriology and Risk Factors of Surgical Wound Infections
Never let your sense of morals prevent you from doing what is right. Isaac Asimov

Factors associated with surgical site infection after lower extremity bypass in the Society for
When you talk, you are only repeating what you already know. But if you listen, you may learn something

Prevalence of gastrointestinal helminthic infections and associated risk factors among
Happiness doesn't result from what we get, but from what we give. Ben Carson

Idea Transcript


University of Iowa

Iowa Research Online Theses and Dissertations

Summer 2012

Risk factors and outcomes associated with surgical site infections after craniotomy and craniectomy Hsiu-Yin Chiang University of Iowa

Copyright 2012 Hsiu-Yin Chiang This dissertation is available at Iowa Research Online: https://ir.uiowa.edu/etd/3277 Recommended Citation Chiang, Hsiu-Yin. "Risk factors and outcomes associated with surgical site infections after craniotomy and craniectomy." PhD (Doctor of Philosophy) thesis, University of Iowa, 2012. https://doi.org/10.17077/etd.8vyb3qga

Follow this and additional works at: https://ir.uiowa.edu/etd Part of the Clinical Epidemiology Commons

RISK FACTORS AND OUTCOMES ASSOCIATED WITH SURGICAL SITE INFECTIONS AFTER CRANIOTOMY AND CRANIECTOMY

by Hsiu-Yin Chiang

An Abstract Of a thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Epidemiology in the Graduate College of The University of Iowa

July 2012

Thesis Supervisor: Professor Loreen A. Herwaldt

1 ABSTRACT Few investigators have used robust analytic methods to assess risk factors and outcomes for surgical site infections (SSIs) after craniotomy and craniectomy (CRANI) procedures. We performed a retrospective study among patients undergoing CRANI procedures between 2006 and 2010 at the University of Iowa Hospitals and Clinics (UIHC) to assess the effect of an intervention (e.g., limiting Gliadel wafer use among patients with malignant brain tumors) on the trend of SSI rates, to identify independent risk factors for SSIs, and to evaluate one-year postoperative patient outcomes associated with these SSIs. We abstracted demographic data and clinical data from medical records or from the UIHC’s Health information Management System. We identified 104 patients with SSIs and selected 312 controls. Of SSIs, 88% were deep incisional or organ space infections, 70% were identified after patients were discharged from their initial hospitalizations, 32% were caused by Staphylococcus aureus alone or in combination with other organisms, and 27% were caused by Gram-negative organisms alone or in combination with other organisms. Significant independent risk factors for SSIs were: previous chemotherapy (odds ratio [OR], 10.0; 95% confidence interval [CI] 1.1, 92.1), preoperative length of stay ≥ 1 day (OR, 2.1; 95% CI 1.3, 3.5), preoperative serum glucose ≥ 100 mg/dL (OR, 1.7; 95% CI, 1.0, 3.0), Gliadel wafer use (OR, 8.6; 95% CI 3.2, 23.1), and postoperative cerebrospinal fluid leak (OR, 4.0; 95% CI, 1.6, 10.3). Gliadel wafer use was the strongest risk factor; however, limiting Gliadel wafer use did not decrease SSI rate significantly among patients with brain tumors. Perioperative ventricular drains or lumbar drains were not independently associated

2 with an increased risk of SSIs, but drains may have clinical significance. An SSI risk index that included the significant preoperative patient-related risk factors had a better predictive power than the National Healthcare Safety Network (NHSN) risk index. After adjusting for preoperative length of stay, age, comorbidity score, severity of illness score, the reason for the procedure, and procedure month, patients with SSIs were hospitalized longer postoperatively than were controls during their readmissions (2.3 days; P < 0.0001). After controlling for the same covariates and treating SSI as a timevarying factor, patients with SSIs were more likely than controls to: die (hazard ratio [HR], 3.3; 95% CI, 1.8, 5.8), be readmitted (HR, 4.1; 95% CI, 2.9, 5.8), and have reoperations (HR, 56.6; 95% CI, 38.1, 84.0). In conclusion, surgeons could predict patients’ risk of SSIs based on their preoperative risk factors and surgeons could modify some processes of care to lower the SSI risk. Preventing SSIs after CRANI procedures could improve patient outcomes and decrease healthcare utilization.

Abstract Approved: ________________________________ Thesis Supervisor ________________________________ Title and Department ________________________________ Date

RISK FACTORS AND OUTCOMES ASSOCIATED WITH SURGICAL SITE INFECTIONS AFTER CRANIOTOMY AND CRANIECTOMY

by Hsiu-Yin Chiang

A thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Epidemiology in the Graduate College of The University of Iowa

July 2012

Thesis Supervisor: Professor Loreen A. Herwaldt

Copyright by HSIU-YIN CHIANG 2012 All Rights Reserved

Graduate College The University of Iowa Iowa City, Iowa

CERTIFICATE OF APPROVAL _______________________

PH. D. THESIS ____________

This is to certify that the Ph. D. thesis of Hsiu-Yin Chiang has been approved by the Examining Committee for the thesis requirement for the Doctor of Philosophy degree in Epidemiology at the July 2012 graduation.

Thesis Committee: ________________________________ Loreen A. Herwaldt, Thesis Supervisor ________________________________ James C. Torner ________________________________ Elizabeth A. Chrischilles ________________________________ Eli N. Perencevich ________________________________ Marin L. Schweizer ________________________________ Joseph E. Cavanaugh

To my family and friends who support me all the time, especially my dearest family – Ming-I Chiang, Shu-Chen Yang, Wei-Chung Chiang, Hung-Lin Chen, and Ryan Chen – I could not finish this without your love.

ii

ACKNOWLEDGMENTS

These five years have been very important in my life. I came to this friendly city alone without knowing much about epidemiology or Iowa. Fortunately, I have met many wonderful people who helped me and guided me through all the difficulties. This dissertation would not be possible without the guidance and the help of several individuals. First and foremost, I would like to thank my advisor, Dr. Herwaldt, for her support and mentorship. Dr. Herwaldt taught me a lot about conducting research and writing manuscripts. She always encouraged me to learn more and think deep. I also appreciate the valuable advice and expert knowledge from my committee members: Dr. Cavanaugh, Dr. Chrischilles, Dr. Perencevich, Dr. Schweizer, and Dr. Torner. I am grateful to Dr. Kamath, who devoted her time to reading medical records and assigning the severity of illness score for each patient. My appreciation goes to Dr. Yang, who assisted me patiently with some data analyses. I would like to sincerely thank the Department of Neurosurgery, especially Dr. Howard and Dr. Greenlee, for providing much clinical inputs to the current study. I would like to acknowledge the staff of the Clinical Quality, Safety, and Performance Improvement (CQSPI) office, especially Jean Pottinger and Martha Freeman. Jean taught me everything about surgical site infections and responded to my questions promptly with her professional knowledge and experience. Martha and her husband Jim have been very supportive spiritually.

iii

Last but not the least, my family have been giving me the strength to finish my Ph.D. degree. My parents, Ming-I Chiang and Shu-Chen Yang, took care of my baby and everything at home when I was working on my dissertation. My brother Wei-Chung Chiang always cheered me up when I felt down. My husband Hung-Lin Chen stayed awake for countless nights to take care of our newborn so that I could rest, and my cute little baby Ryan was a healthy boy, who seldom let me worry about his conditions. It’s their support and love that allow me to keep going.

iv

ABSTRACT Few investigators have used robust analytic methods to assess risk factors and outcomes for surgical site infections (SSIs) after craniotomy and craniectomy (CRANI) procedures. We performed a retrospective study among patients undergoing CRANI procedures between 2006 and 2010 at the University of Iowa Hospitals and Clinics (UIHC) to assess the effect of an intervention (e.g., limiting Gliadel wafer use among patients with malignant brain tumors) on the trend of SSI rates, to identify independent risk factors for SSIs, and to evaluate one-year postoperative patient outcomes associated with these SSIs. We abstracted demographic data and clinical data from medical records or from the UIHC’s Health information Management System. We identified 104 patients with SSIs and selected 312 controls. Of SSIs, 88% were deep incisional or organ space infections, 70% were identified after patients were discharged from their initial hospitalizations, 32% were caused by Staphylococcus aureus alone or in combination with other organisms, and 27% were caused by Gram-negative organisms alone or in combination with other organisms. Significant independent risk factors for SSIs were: previous chemotherapy (odds ratio [OR], 10.0; 95% confidence interval [CI] 1.1, 92.1), preoperative length of stay ≥ 1 day (OR, 2.1; 95% CI 1.3, 3.5), preoperative serum glucose ≥ 100 mg/dL (OR, 1.7; 95% CI, 1.0, 3.0), Gliadel wafer use (OR, 8.6; 95% CI 3.2, 23.1), and postoperative cerebrospinal fluid leak (OR, 4.0; 95% CI, 1.6, 10.3). Gliadel wafer use was the strongest risk factor; however, limiting Gliadel wafer use did not decrease SSI rate significantly among patients with brain tumors. Perioperative ventricular drains or lumbar drains were not independently associated v

with an increased risk of SSIs, but drains may have clinical significance. An SSI risk index that included the significant preoperative patient-related risk factors had a better predictive power than the National Healthcare Safety Network (NHSN) risk index. After adjusting for preoperative length of stay, age, comorbidity score, severity of illness score, the reason for the procedure, and procedure month, patients with SSIs were hospitalized longer postoperatively than were controls during their readmissions (2.3 days; P < 0.0001). After controlling for the same covariates and treating SSI as a timevarying factor, patients with SSIs were more likely than controls to: die (hazard ratio [HR], 3.3; 95% CI, 1.8, 5.8), be readmitted (HR, 4.1; 95% CI, 2.9, 5.8), and have reoperations (HR, 56.6; 95% CI, 38.1, 84.0). In conclusion, surgeons could predict patients’ risk of SSIs based on their preoperative risk factors and surgeons could modify some processes of care to lower the SSI risk. Preventing SSIs after CRANI procedures could improve patient outcomes and decrease healthcare utilization.

vi

TABLE OF CONTENTS LIST OF TABLES……………………………………………………………………………………………………………xi LIST OF FIGURES………………………………………………………………………………………….……..………xiii CHAPTER 1. 1.1 Introduction…………………………………………………………………………………………….…...1 1.2 Specific Aims…………………………………………………………………………………..…………….3 1.3 Conceptual Framework…………..…………………………………………………………………….6 2. 2.1 Evaluation of the Surgical Site Infection Rates Pre- and Post-Intervention…...8 2.1.1 Quasi-experimental Study Design………..…………………………………………8 2.1.2 Analytic Methods…………………………………………………………………………..9 2.1.3 Approaches in This Study………………………………………..……………………11 2.2 Surgical Site Infections after Craniotomy/Craniectomy…………………………..…..12 2.2.1 Surgical Site Infections in General……………………….…………………….…12 2.2.2 Post-Craniotomy/Craniectomy Surgical Site Infections……..………….14 2.3 Risk Factors and Risk Indices for Surgical Site Infections after Craniotomy/ Craniectomy………….…………………………………………………………….……………………………15 2.3.1 Potential Risk Factors………..………………………………………………………….15 2.3.1.1 Comorbidities…………………………………………………………………19 2.3.1.2 Reason for Procedure…………………………………………………….19 2.3.1.3 Wound Classification……………………………………………………20 2.3.1.4 Antimicrobial Prophylaxis………………………………………………20 2.3.1.5 Operation Duration………………………………………………………..21 2.3.1.6 Gliadel® Wafer Implantation………………………………………….21 2.3.1.7 Cerebrospinal Fluid Drainage…………………………………………22 2.3.1.8 Cerebrospinal Fluid Leakage…………………………………………..23 2.3.2 Surgical Site Infection Risk Index……………………..……………………………23 2.3.2.1 The National Healthcare Safety Network Risk Index………24 2.3.2.2 New NHSN Procedure-Specific Risk Model…………………….26 2.3.2.3 Procedure-Specific Surgical Site Infection Risk Indices……27 2.3.2.4 Development of a Surgical Site Infection Risk Index……….29 2.3.3 Approaches in This Study………………………………………..……………………29 2.4 Outcomes Associated with Surgical Site Infections after Craniotomy/ Craniectomy……………………………………………………….…………………………………………….29 2.4.1 Outcomes Associated with Surgical Site Infections……………………….30 2.4.2 Approaches to Assess Outcomes…….……………………………………………32

vii

2.4.2.1 Controlling for Length of Hospitalization before the Procedure………………………………………………………………………………….32 2.4.2.2 Adjusting for Underlying Severity of Illness and Comorbidities…………………………………………………………………………….32 2.4.2.3 Surgical Site Infection as a Time-Varying Factor……………..34 2.4.3 Approaches in This Study…………..…………………………………………………35 2.5 Pilot Study: Clinical Significance of Positive Bone Flap Cultures And Associated Risk of Surgical Site Infection after Craniotomies or Craniectomies…36 2.5.1 Introduction……..………………………………………………………………………….37 2.5.2 Methods……………………………………………………………………………………….38 2.5.3 Results………………………………………………………………………………………....39 2.5.4 Summary and Discussion……………..……………………………………………….45 2.6 Significance………..……………………………………………………………………………………….47 3. 3.1 Overview of Study Design…………………………..……………………………………………….49 3.1.1 Study Population…………………………….……………………………………………49 3.1.1.1 Specific Aim 1…………………………………………………………………50 3.1.1.2 Specific Aim 2…………………………………………………………………52 3.1.1.3 Specific Aim 3…………………………………………………………………53 3.1.2 Study Design………………………………………………………………………………...55 3.1.2.1 Specific Aim 1…………………………………………………………………55 3.1.2.2 Specific Aim 2…………………………………………………………………55 3.1.2.3 Specific Aim 3…………………………………………………………………56 3.2 Data Collection and Measurement……………..………………………………………………57 3.2.1 Specific Aim 1……………………………………………………………………………….57 3.2.2 Specific Aim 2……………………………………………………………………………….58 3.2.2.1 Exposure Variables…………………………………………………………58 3.2.2.2 Outcome Variable………………………………………………………….63 3.2.3 Specific Aim 3……………………………………………………………………………….64 3.2.3.1 Exposure Variable………………………………………………………….64 3.2.3.2 Confounding Variables…………………………………………………..64 3.2.3.3 Outcome Variables…………………………………………………………66 3.3 Power Calculation………..……………………………………………………………………….…….67 3.4 Statistical Analyses……………………….…………………………………………………………….68 3.4.1 Specific Aim 1……………………………………………………………………………....68 3.4.1.1 Describe the Overall Surgical Site Infection Rate and the Trend in Surgical Site Infection Rates over Time………………..69 3.4.1.2 Compare Surgical Site Infection Rates Before and After Use of Gliadel® Wafers Was Limited in April, 2009 among Patients with Malignant Brain Tumors (Intervention Group) and Patients without Malignant Brain Tumors (Control Group)...71 3.4.2 Specific Aim 2………………………………………………………………………….…...81 viii

3.4.2.1 Identify Patient-Related and Procedure-Related Factors Associated with SSIs after CRANI by Bivariable Analyses……………72 3.4.2.2 Identify Significant Independent Patient- and Procedure-Related Risk Factors by Multivariable Logistic Regression Analyses…………………………………………………………….…….72 3.4.2.3 Develop a Preoperative Risk Index Consisting of Independent Patient-Related Factors Associated with SSIs After CRANI…………………………………………………………………………………….....74 3.4.3 Specific Aim 3……………..………………………………………………………………..76 3.4.3.1 Identify Postoperative Outcomes Associated with SSIs after CRANI by Bivariable Analyses………………………………………….…76 3.4.3.2 Compare the Kaplan-Meier Estimates of the Survival Functions between Patients with SSIs and Controls for Postoperative Outcomes……………………………………………………………76 3.4.3.3 Assess the Association between SSIs and Postoperative Outcomes by Multivariable Regression Analyses……………………….77 3.4.3.4 Assess the Association between SSIs and Postoperative Outcomes by Cox Proportional Hazard Analyses……………………….78 3.5 Study Implementation and Quality Control……………..………………………………….80 3.5.1 Study Implementation……………..…………………………………………………..80 3.5.2 Data Collection…………..…………………………………………………………………80 4. 4.1 Specific Aim 1…………..…………………………………………………………………………………82 4.1.1 Gliadel Wafer Use………..……………………………………………………………….82 4.1.2 Overall Surgical Site Infection Rate and Trend…………..………………….82 4.1.2.1 Regression Analysis………………………………………………………..83 4.1.2.2 Time-Series Analysis………………………………………………………83 4.1.3 Pre- and Post-Intervention……..…………………………………………………….86 4.1.3.1 Two-Group Test……………………………………………………………..86 4.1.3.2 Regression Analyses……………………………………………………….87 4.1.4 Conclusions………..………………………………………………………………………..91 4.2 Specific Aim 2……………………..……………………………………………………….……………..91 4.2.1 Descriptive Epidemiology of Surgical Site Infections…….…..………….91 4.2.2 Risk Factors for Surgical Site Infections…………….…………………………93 4.2.2.1 Bivariable Analyses………………………………………………………..93 4.2.2.2 Multivariable Analyses…………………………………………………..95 4.2.3 Surgical Site Infection Risk Index……………….………………………………..99 4.2.4 Conclusions………………………………..………………………………………………101 4.3 Specific Aim 3………..………………………………………………………………………………….101 4.3.1 Bivariable Associations between Surgical Site Infections and Outcomes……………………………………………………………………………………………101 4.3.2 Postoperative Length of Stays……………………………..……………………..103 ix

4.3.2.1 Kaplan-Meier Survival Analyses……………………………………103 4.3.2.2 Multivariable Analyses…………………………………………………103 4.3.3 Postoperative Deaths……………………..………………………………………….105 4.3.3.1 Kaplan-Meier Survival Analyses……………………………………105 4.3.3.2 Multivariable Analyses…………………………………………………106 4.3.4 Postoperative Readmissions………………………….……………………………106 4.3.4.1 Kaplan-Meier Survival Analyses……………………………………106 4.3.4.2 Multivariable Analyses…………………………………………………107 4.3.5 Postoperative Reoperations……………..………………………………………107 4.3.5.1 Kaplan-Meier Survival Analyses……………………………………107 4.3.5.2 Multivariable Analyses…………………………………………………107 4.3.6 Conclusions…………………………..……………………………………………………108 5. 5.1 Summary of Findings………..……………………………………………………………………….145 5.2 Discussion…..……………………………….……………………………………………………………147 5.2.1 Gliadel® Wafer Implants……………………………………………………………..147 5.2.2 Gram-Negative Surgical Site Infections……………………………………….148 5.2.3 Hyperglycemia……………………………………………………………………………149 5.2.4 Ventricular and Lumbar Drains……………………………………………………151 5.2.5 Cerebrospinal Fluid Leakage……………………………………………………….153 5.2.6 Potential Interventions……………………………………………………………….154 5.2.7 Postoperative Outcomes…………………………………………………………….159 5.3 Limitations……………..…………………………………………………………………………………160 5.4 Future Directions……………..……………………………………………………………………….162 5.5 Overall Summary and Conclusions…………….……………………………………………..164 APPENDIX........................................................................................................................167 REFERENCES……………………….………………………………………………………………………………….…..214

x

LIST OF TABLES TABLE 1.

2.

3. 4. 5.

6. 7.

-

8. 9.

10.

11.

12.

13.

-

-

14.

15.

xi

16.

-

17.

-

-

18.

19.

20.

-

21. 22.

23.

24.

25.

A1. Criteria for defining a surgical site infection (SSI)….………………………………….……….….168 A2. American Society of Anesthesiologists (ASA) score…………………..………………………….170 A3. ICD-9-CM procedure codes for craniotomy/craniectomy…………..…………………………171 A4. ICD-9-CM diagnosis codes for malignant brain tumors………………..……………………….173 A5. Data abstraction form……………….…………………………………………………………………………174 A6. Variable library……………..…………..…………………………………………………………………………188 A7. Comparison of Charlson and Elixhauser comorbidity measures……………………..…….212 xii

LIST OF FIGURES FIGURE 1.

2.

-

3. 4. 5.

6. 7.

8. 9. 10.

11.

12.

13.

14.

xiii

1 CHAPTER 1 – INTRODUCTION AND SPECIFIC AIMS

1.1 Introduction A craniotomy is an incision through the skull done to excise, repair, or explore the brain, and a craniectomy is a procedure in which a portion of the skull is excised (i.e., removed) surgically to access the brain [Stedman, 1995]. These procedures may be necessary to treat medical conditions including: brain tumors, bleeding (hemorrhage) or blood clots (hematomas) from injuries, defects in blood vessels (cerebral aneurysms), abnormal blood vessels (arteriovenous malformations), damage to tissues covering the brain (dura), localized infections (brain abscesses), severe nerve or facial pain (trigeminal neuralgia or tic douloureux), trauma to the skull (skull fractures), and some seizure disorders (epilepsy) [Craniotomy mortality measure, 2009]. The number of craniotomy/craniectomy (CRANI) procedures done each year in the United States (US) has increased steadily since the 1990s, according to data from the Healthcare Cost and Utilization Project Nationwide Inpatient Sample (HCUP-NIS) of the Agency for Healthcare Research and Quality (AHRQ), which is the largest all payer database of hospital discharges in the US. AHRQ’s HCUP-NIS indicated that in 1997, 89,852 national discharges listed one of the CRANI procedure codes as a principle procedure and 123,004 listed one of these procedures codes as a secondary procedure. By 2009, these numbers had increased to 125,117 and 165,353, respectively, representing increases of 39% and 34% (NIS Overview, 2009; HCUPnet, 2012).

2 In 2006-2008, the surgical site infection (SSI) rate after CRANI reported by the National Healthcare Safety Network (NHSN) surveillance system of the Centers for Disease Control and Prevention (CDC) ranged from 2.15% (patients with a NHSN risk index score of 0 or 1) to 4.66% (patients with a NHSN risk index score of 2 or 3) (Edwards et al., 2009). The NHSN risk index has been used commonly to stratify surgical patients’ risk of acquiring SSIs and to adjust for a priori risk of SSIs when comparing SSI rates. The NHSN risk index score is based on the wound classification (clean, clean-contaminated, contaminated, or dirty), the American Society of Anesthesiologists (ASA) score (1 to 5) (Appendix 2), and the duration of the operation. However, the NHSN risk index score stratifies patients undergoing CRANI procedures poorly, because most CRANI procedures are clean procedures and ASA scores for patients undergoing CRANI procedures do not vary substantially. A risk index, which utilizes risk factors specific for SSIs after CRANI procedures, would help neurosurgeons predict a patient’s risk of SSI before an operation, and would help infection preventionists to adjust for differences in the patients’ intrinsic risk when they compare SSI rates between surgeons or hospitals. Few studies specifically evaluated infections after CRANI, and most of these studies did not use robust epidemiological methods to either identify risk factors for SSIs after CRANI or assess the effect of SSIs on patients’ postoperative outcomes. For example, some studies did not assess patients’ comorbidities and underlying severity of illness that may be associated with SSIs. In addition, none of the published studies adjusted for preoperative length of stay (LOS), age, comorbidities, and severity of underlying illness

3 when evaluating patient outcomes. Moreover, only one study included all CRANI procedures and all types of SSIs, and only two studies were done in the US. Thus, there are gaps in the literature and available data may not be generalizable to patients undergoing CRANI procedures in the US. In 2008-2009 at the University of Iowa Hospitals and Clinics (UIHC), the mean SSI rate for CRANI procedures was 5.15% for high-risk patients (patients with NHSN risk score of 2 or 3), which was higher than the NHSN’s mean SSI rate of 4.66% for high-risk patients. Previously, we performed a prospective cohort study that included all patients who underwent CRANI procedures between 2007 and 2008 at the UIHC, and we identified several risk factors associated with increased risk of SSIs, including having Gliadel® wafer implants among patients with brain tumors (Chapter 2.5) (Chiang et al., 2011). We showed the neurosurgeons the results of our study in April 2009, and they

subsequently limited the use of Gliadel® wafer implants. The pilot study gave us useful clinical information about SSIs after CRANI procedures; however, we did not evaluate some potential risk factors associated with SSIs and we did not evaluate patient outcomes in the pilot study. Therefore, we performed the current study to address the research gaps.

1.2 Specific aims The primary objectives of the present study were to describe the epidemiology of SSIs after CRANI and to provide information about risk factors and patient outcomes associated with SSIs after CRANI procedures. The specific aims include:

4 Aim 1: To assess the overall trend of the rate of surgical site infections after craniotomy and craniectomy between January 1, 2006 and December 31, 2010 at the University of Iowa Hospitals and Clinics and to evaluate the effect of limiting use of Gliadel® wafers on surgical site infection rates. We hypothesized that the SSI rates would decrease after limiting use of Gliadel® wafers, and the decreasing trend would be specific for patients with malignant brain tumors because neurosurgeons used Gliadel® wafers to treat only patients with brain tumors. Aim 2: To identify risk factors for surgical site infections and to develop a preoperative surgical site infection risk index for patients undergoing craniotomy and craniectomy procedures between January 1, 2006 and December 31, 2010 at the University of Iowa Hospitals and Clinics. We hypothesized that certain patient-related and procedure-related factors were associated with an increased risk of SSIs, and some of the factors were modifiable. We also hypothesized that the preoperative SSI risk index consisting of patient-related risk factors specific for SSIs after CRANI would better predict SSIs after these procedures than would the NHSN risk index. Aim 3: To evaluate patient outcomes associated with surgical site infections among patients who underwent craniotomy and craniectomy between January 1, 2006 and June 30, 2010 at the University of Iowa Hospitals and Clinics.

5 We hypothesized that, compared with patients who did not acquire SSIs, patients who acquired SSIs had worse postoperative outcomes during the first year after their CRANI procedures. To address these specific aims, we did a literature review and assessed most of the potential risk factors for SSIs after CRANI in the literature. We conducted a retrospective study of patients who underwent CRANI procedures done by the neurosurgeons between January 1, 2006 and December 31, 2010 at the UIHC, and we included patients who survived for at least two days after their CRANI procedures. We retrospectively collected data on demographic and clinical information from patients’ medical records or from the UIHC’s Health Information Management. Few investigators in the infection control field have analyzed time-series data (e.g., infection rates) using appropriate statistical methods that took into account of autocorrelation between consecutive observations (i.e., infection rates by month) and the distribution of the observations. In addition, few studies have assessed risk factors for SSIs after CRANI procedures and investigators have not created a valid SSI risk index specific for all CRANI procedures. Furthermore, investigators have not controlled for patients’ preexisting conditions (e.g., comorbidities, severity of underlying diseases) and the time to the event of interest when assessing the postoperative outcomes associated with SSIs after CRANI. The current study will inform neurosurgeons and infection preventionists about the epidemiology of SSIs after CRANI procedures and the impact of SSIs on patients’

6 outcomes. Thus, the study could contribute to the long-term goal of reducing SSIs after CRANI procedures.

1.3 Conceptual framework The etiology of SSI after CRANI procedures is complex. Patient-related factors and procedure-related factors both affect the likelihood that a patient will acquire an SSI. The conceptual framework below delineates the relationships of some key potential risk factors, SSIs post-CRANI procedures, and patient outcomes (Figure 1).

7 -

-

Surgical Site Infections

-

-

Figure 1. Conceptual framework of the relationships between potential risk factors, surgical site infections, and postoperative outcomes.

8 CHAPTER 2 – BACKGROUND AND SIGNIFICANCE

2.1 Evaluation of the surgical site infection rates pre- and post-intervention 2.1.1 Quasi-experimental study design Investigators studying the epidemiology of infectious diseases, especially healthcareassociated infections (HAIs), often evaluate the effect of an intervention by assessing whether the rates of infection changed after the intervention. Harris et al. reviewed studies published in two infectious disease journals from January 1, 2002 and June 30, 2003 and identified 36 quasi-experimental studies (Harris et al., 2004). The infection rates are often collected at equally spaced time intervals (e.g., monthly) before and after the intervention. These nonrandomized pre- and post- studies are one type of the quasi-experimental study. Quasi-experimental studies encompass a broad range of non-randomized intervention studies. Investigators frequently use these designs when it is not feasible or not ethical to conduct a randomized controlled trial. In addition, as the capacity to collect routine clinical data (e.g., surgical site infection [SSI] rates by month) has increased, the use of quasi-experimental study designs has also increased for studying the epidemiology of infectious diseases. However, results of such studies could be adversely affected by confounding from unmeasured factors, regression to the mean, or maturation effect (Harris et al., 2004). For example, Harris et al. determined that the results of studies they evaluated may have been affected by confounding factors such as the patients’ severity of illness, the quality of medical care, and clinical practice which

9 might have differed between the pre- and post-intervention periods. Regression to the mean is a statistical phenomenon that if a variable is extreme on its first measurement, it will tend to be closer to the average on a second measurement, and so on. Regression to the mean could cause investigators wrongly conclude that a change in infection rates is due to an intervention when it is actually due to chance. Maturation occurs when a variable changes due to natural changes (e.g., seasonal trends) during the course of time. For example, the etiological agents causing SSIs or patient clinical conditions could change with the season, which affect SSI rates and could cause an investigator to conclude incorrectly that an intervention caused the observed change. Nevertheless, investigators can limit the likelihood that regression to the mean and maturation effects explain their results if they collect multiple pre- and postintervention data points. Similarly, investigators can limit the likelihood that unmeasured confounding factors explain their results if they include a control group that does not undergo the intervention. Investigators can also limit the likelihood that they will mistakenly attribute a change in rates, which occurs after an intervention, to the intervention by using proper analytic methods (See Chapter 2.1.2). 2.1.2 Analytic methods Methods used to analyze pre-and post-intervention data can be summarized into two-group tests, regression analysis, and time series analysis. Each of the methods is based on specific assumptions and each has strengths and limitations (Shardell et al., 2007). Two-group tests, such as two-sample t-test and two-rate Chi square test, assume that observations (e.g., monthly infection rates) are independent to each other, which is

10 implausible because quasi-experimental data are typically correlated observations. The two-sample t-test requires data to be normally distributed and the Chi square test assumes data follow a Poisson distribution. Neither of these tests is the best method for analyzing SSI rates because monthly rates are time-series data that could be autocorrelated and because the distribution of SSI rates is often skewed to the right. Autocorrelation is the similarity between observations as a function of the time separation between them. For example, autocorrelation occurs if infection rates from successive months are more similar than infection rates collected two months apart, and skewness occurs if the counts of SSIs are usually low. These two-group tests can allow us to assess the data quickly but they should not be used for the definitive assessment.

-

-

11

In the present study, the intervention of interest was decreasing Gliadel® wafer use among patients with brain tumors. Gliadel® wafer implants, chemotherapy implants used to treat patients with malignant brain tumors, were significantly associated with SSIs among patients with brain tumors in our previous study (Chiang et al., 2011). In April, 2009, the surgeon (surgeon A) who used Gliadel® wafers most frequently left the University of Iowa Hospitals and Clinics (UIHC). During the same month, infection preventionists met with the neurosurgeons to report the results of the previous study, including the observation that the risk of SSIs was higher among patients with brain tumors who had Gliadel® wafers implanted than for those who did not. Subsequently, surgeon B, the other surgeon who used this product, reduced his use of Gliadel® wafer. Therefore, we evaluated the effect of decreasing Gliadel® wafer use among patient with malignant brain tumors (i.e., intervention group) on the SSI rates. To control for the potential problems in this pre- and post-intervention study, we included the non-tumor group as a non-equivalent control group, and we collected SSI rates for 39 months before the intervention and for 20 months after the intervention for both groups. Moreover, we utilized three analytic approaches, two-group tests,

12 regression tests, and a time-series method, to evaluate the influence of decreasing Gliadel® wafer use on the mean and on the trend in SSI rates by month.

2.2 Surgical site infections after craniotomy/craniectomy 2.2.1 Surgical site infections in general In the United States (US), the Centers for Disease Control and Prevention (CDC)’s National Nosocomial Infections Surveillance (NNIS; now known as the National Healthcare Safety Network [NHSN]) system, established in 1970, monitors trends in HAIs and estimates the magnitude of HAIs among hospitals participating in the NNIS system. On the basis of data from the NNIS system (1990-2002), the National Hospital Discharge survey (2002) and the American Hospital Association Survey (2000), approximately 1.7 million HAIs occurred in US hospitals, nearly 99,000 patients died from these infections. SSIs were the third most common HAIs, accounting 20% of all HAIs and for 8% of the deaths (Klevens et al., 2007). By CDC’s definition, SSIs are infections that affect the operative sites, which occur up to 30 days after surgical procedures, or up to one year after procedures during which implants were placed (The NHSN Manual, 2009). Appendix 1 lists the CDC’s criteria for defining an SSI. SSIs are classified by their depth: superficial incisional SSIs involve skin or subcutaneous tissue, deep incisional SSIs involve fascia or muscle layers, and organ/space infections involve an organ or a space (e.g., a body cavity) (Figure 2). Deeper infections are usually more severe than superficial infections, with deeper

13 infections being more likely to cause death or to require re-operations (Astagneau et al., 2001).

(a)

(b)

Superficial Incisional SSI Deep Incisional SSI

Organ/Space SSI

Figure 2. Cross-section of (a) abdominal wall (Mangram et al., 1999) and (b) scalp layers (adapted from Gray’s Anatomy Scalp Diagram [Public Domain]) depicting National Healthcare Safety Network (NHSN)’s classifications of surgical site infections.

14 2.2.2 Post-craniotomy/craniectomy surgical site infections SSIs after craniotomy or craniectomy (CRANI) procedures comprise a spectrum of infectious processes, including superficial scalp infections, deep incisional infections such as subgaleal infections, and organ/space infections such as osteomyelitis, bacterial meningitis, cranial epidural abscess, subdural empyema, and cerebral abscess (Osenbach et al., 2005). Figure 2 depicts the tissue layers of scalp through which CRANI incisions are made. Most SSIs after CRANI procedures are deep incisional or organ/space infections, of which meningitis is the most common infection. For example, meningitis accounted for 47.9% of SSIs Korinek et al. identified after 2,944 craniotomies done in France (Korinek et al., 1997). These deep incisional or organ/space infections are severe conditions that require further treatment and that increase morbidity and mortality. The incidence of SSIs post-CRANI varies substantially. McClelland conducted a metaanalysis of published studies done in North America, each of which included more than 500 intracranial procedures. McClelland found an SSI rate of 2.2% (McClelland, 2008). The CDC’s NNIS system reported that the SSI rate from 1992 through 2004 ranged from 0.91% (low-risk group) to 2.4% (high-risk group), and the SSI rate from 2006 through 2008 ranged from 2.15% (low-risk group) to 4.66% (high-risk group) (Edwards et al., 2009; NNIS Report, 2004). Rates of SSI after CRANI in individual health centers have ranged from 4% to 8.9% in studies performed from 1997 to 2007 (Korinek et al., 1997; Korinek et al., 2005; Kourbeti et al., 2007; Reichert et al., 2002). During our pilot study (Chiang et al., 2011), the SSI rate among patients undergoing CRANI between November

2007 and November 2008 at the UIHC was 5.8%. The variation in incidence may reflect

15 differences in the patient populations, the types of procedures, and the types of infections included in the studies.

2.3 Risk factors and risk indices for surgical site infections after craniotomy/ craniectomy 2.3.1 Potential risk factors Both patient-related factors and procedure-related factors can affect the risk of SSIs. Patient-related factors, such as age and comorbidities, affect the patient’s intrinsic risk of acquiring a SSI. Procedure-related factors, such as surgical technique, preoperative skin preparation, and antimicrobial prophylaxis, affect the extrinsic risk of SSIs. Surgeons cannot eliminate or ameliorate most patient-related factors to reduce the patients’ risk of SSIs, but surgeons can use these factors to predict a patient’s risk of SSI and infection preventionists can adjust for these intrinsic factors when comparing SSI rates among different surgeons and different hospitals. Some procedure-related factors are modifiable. Thus, surgeons could implement interventions to eliminate or ameliorate the modifiable risk factors and, thereby, reduce SSIs after CRANI. Numerous studies and reviews have been published on risk factors for SSIs after neurosurgical procedures, including CRANI procedures, spinal operations, cerebral spinal fluid (CSF) drainage procedures, and peripheral nerve system procedures (Federico et al., 2001; Lietard et al., 2008; Valentini et al., 2007; Broekman et al., 2011; Schuster et al., 2010; Erman et al., 2005). Examples of risk factors for SSIs after these neurosurgical procedures include: older age (Valentini et al., 2007; Schuster et al., 2010; Erman et al., 2005), comorbidities such as obesity (Schuster et al., 2010), diabetes

16 (Schuster et al., 2010; Erman et al., 2005), and malnutrition (Schuster et al., 2010), higher the American Society of Anesthesiologists (ASA) score (Schuster et al., 2010), higher Acute Physiology and Chronic Health Evaluation (APACHE) III score (Federico et al., 2001), longer operation duration (Valentini et al., 2007), intraoperative blood transfusions (Schuster et al., 2010), presence of foreign body (McClelland, 2008; Erman et al., 2005), presence or duration of drains, such as shunts (Erman et al., 2005; Lietard et al., 2008) or ventriculostomies (Federico et al., 2001), intracranial pressure monitoring (Erman et al., 2005), CSF leakage (Lietard et al., 2008), and early reoperation (Lietard et al., 2008). However, few of these studies focused specifically on risk factors for SSIs after CRANI. CRANI procedures are intrinsically different from other neurosurgical procedures because the incision is through the skull rather than through the spine or soft tissue. We identified nine published studies written in English that specifically evaluated infections after CRANI procedures; only two were performed in the US (Kourbeti et al., 2007; Hardy et al., 2010) and the remaining studies were done in Brazil (Reichert et al., 2002), France (Gaberel et al., 2011; Korinek et al., 1997; Korinek et al., 2005; Korinek et al., 2006), Japan (Shinoura et al., 2004), and Mexico (Sanchez-Arenas et al., 2010). The results of studies done in other countries may not be generalizable to the population in the US. In addition, the inclusion criteria for these nine studies varied. Most studies focused only on specific procedures (e.g., elective) or on specific infections (e.g., meningitis). Two studies included craniotomies done as treatment for tumors (Shinoura et al., 2004; Hardy et al., 2010), and one study included craniotomy procedures that did

17 not involve placing implants (Sanchez-Arenas et al., 2010). Three studies evaluated only meningitis (Korinek et al., 2006; Kourbeti et al., 2007; Reichert et al., 2002), and one study evaluated patients with SSIs that needed a second operation to cleanse the wound (Gaberel et al., 2011). Two studies included patients with any type of SSI (Korinek et al., 1997; Sanchez-Arenas et al., 2010). Overall, only one study included all CRANI procedures and all patients with SSIs (Korinek et al., 2005). The findings of studies evaluated specific infections or specific CRANI procedures may not be applicable to the patient population for all CRANI procedures. The risk factors evaluated in these nine studies varied as well. Table 1 lists the significant risk factors that were evaluated by multivariable analyses in at least three studies. Male gender, age greater than 70 years, reason for procedure, ASA score greater than two were patient-related factors associated with increased risk of SSIs. Procedure-related risk factors included emergent procedure, no antimicrobial prophylaxis, longer operation duration, postoperative CSF leakage, and early subsequent procedure. In addition, the significant risk factors that were evaluated in fewer than three studies included: lower Glasgow Coma Score (GCS) (Korinek et al., 1997), underlying chronic disease (Reichert et al., 2002), infection before craniotomy (Shinoura et al., 2004), wound classified as clean-contaminated or dirty (Korinek et al., 1997), surgical approach through a sinus (Kourbeti et al., 2007), use of an external ventricular drain (Korinek et al., 1997; Kourbeti et al., 2007) or intracranial pressure monitoring (Kourbeti et al., 2007), use of BioGlue® (Gaberel et al., 2011), and procedure performed during the late shift (Sanchez-Arenas et al., 2010).

18 These studies did not evaluate thoroughly some potential risk factors for SSIs: comorbidities, smoking, obesity, hair removal, skin preparation, external drains, infection at a remote site and antimicrobial prophylaxis (timing, dosage and antimicrobial agent).

Table 1. Summary of risk factors independently associated with surgical site infections after craniotomy/craniectomy procedures. Risk factors

Independent risk factors in multivariable analyses -Male gender (Korinek et al., 2005; Korinek et al., 2006; Shinoura et al., 2004)

Patient-related

-Age ≥ 70 years (Shinoura et al., 2004) -Reason for procedure is meningioma (Korinek et al., 2005)or nontrauma (Sanchez-Arenas et al., 2010) -ASA score > 2 (Kourbeti et al., 2007) -Emergent procedure (Korinek et al., 1997) -No antimicrobial prophylaxis (Korinek et al., 2005) -Longer operation duration (Korinek et al., 2005; Korinek et al., 2006;

Procedure-related

Hardy et al., 2010; Gaberel et al., 2011) -Postoperative CSF leakage (Korinek et al., 1997; Korinek et al., 2005; Korinek et al., 2006; Shinoura et al., 2004) -Early subsequent procedure at the same site (Korinek et al., 1997; Korinek et al., 2005; Reichert et al., 2002; Sanchez-Arenas et al., 2010)

Abbreviation: ASA score = American Society of Anesthesiologists score; CSF = cerebral spinal fluid. Note: Risk factors listed in the table were all evaluated by multivariable analysis in at least three studies.

19 In the current study, we evaluated a patient population in the US and included all CRANI procedures and all SSIs. Thus, data from our study would be more generalizable to patients undergoing CRANI in the US. In addition, we evaluated not only the wellknown risk factors, but also many potential risk factors that have not been studied carefully. The following paragraphs describe some potential risk factors of interest. 2.3.1.1 Comorbidities Numerous studies have found that comorbidities, such as diabetes and obesity, are strongly associated with SSIs after general surgical procedures. We found two studies that assessed whether chronic diseases increased the risk of SSIs post-CRANI (Reichert et al., 2002; Sanchez-Arenas et al., 2010). Sanchez-Arenas et al. found that chronic diseases (diabetes, neoplasm, arterial hypertension, hepatic cirrhosis, pulmonary disease and other diseases) at the time of admission increased the risk of SSIs by threefold, when controlling for antimicrobial usage, operations due to chronic disease, subsequent operations at the same surgical site, and work shift when the procedure was done (Sanchez-Arenas et al., 2010). 2.3.1.2 Reason for procedure Sanchez-Arenas et al. found that patients with non-traumatic reasons for their procedures had a higher risk of acquiring SSIs (Odds ratio [OR] = 1.87) compared with patients who underwent CRANI for traumatic reasons, after controlling for presence of chronic diseases, ASA score, operation duration, wound contamination class, and other risk factors (Sanchez-Arenas et al., 2010). Korinek et al. found that reason for procedure was an independent risk factor for SSIs after adjusting for surgeon, no antimicrobial

20 prophylaxis, re-operation, operation duration, and CSF leakage. Patients who underwent craniotomy for meningioma or brain metastasis appeared to be at higher risk of SSIs than patients who underwent craniotomy for other reasons, but Korinek et al. did not assess the difference statistically (Korinek et al., 2005). 2.3.1.3 Wound classification The surgical team grades the extent of microbial contamination at the operative site by assigning a wound classification. The wound classification schema divides surgical wounds into four classes: clean, clean-contaminated, contaminated, and dirty/infected wound (Mangram et al., 1999). Wounds classified as clean-contaminated to dirty are more likely to become infected than clean wounds. This classification helps the surgeon determine whether or not to prescribe perioperative antimicrobial prophylaxis or treatment. Wound classification is one of the three components of NHSN risk index, which is used to stratify patients’ risk of SSIs for surgical procedures. A wound classification of contaminated or dirty-infected was a significant risk factor for SSIs after craniotomy in one study (Korinek et al., 1997), but it was not associated with SSIs in three other studies (Korinek et al., 2005; Korinek et al., 2006; Sanchez-Arenas et al., 2010). 2.3.1.4 Antimicrobial prophylaxis Antimicrobial prophylaxis is widely used to reduce intraoperative contamination, and thus, reduce SSIs risk. The first dose should be timed to ensure that appropriate antibacterial concentrations are achieved in serum and tissue at the time of the incision and these concentrations should then be maintained for up to a few hours after wound

21 closure (Owens et al., 2008). The first dose should be given at least 30 minutes before the incision is made. A meta-analysis, including six randomized clinical trials, demonstrated that prophylactic antimicrobials reduced the rate of meningitis by approximately one-half after craniotomy (OR = 0.43; P = 0.03) (Barker, 2007). Absence of prophylaxis was an independent risk factor for SSIs after craniotomy in one study done by Korinek et al. (Korinek et al., 2005), but it was not associated with meningitis in another study done by that group (Korinek et al., 2006). However, the efficacy of different types of antimicrobial agents and the timing of administration have not been thoroughly evaluated for CRANI procedures. 2.3.1.5 Operation duration Patients whose operations last more than “T” hours, where T represents the 75th percentile of the distribution of the operation durations from the national NHSN database, have a higher risk of SSIs (Culver et al., 1991). Each operation type has its own operation duration cut point. For CRANI procedures, the cut point was 225 minutes according to the NHSN report in 2009 (Edwards et al., 2009). Operation duration more than T hours is one of the three components of the NHSN risk index. However, operation duration was not a significant risk factor in four studies of SSIs after CRANI procedures (Lietard et al., 2008; Reichert et al., 2002; Sanchez-Arenas et al., 2010; Shinoura et al., 2004). 2.3.1.6 Gliadel® wafer implantation Gliadel® wafers are biodegradable, carmustine impregnated polymers (1.45 cm in diameter and 1 mm thick) that are implanted in tumor beds after the tumors have been

22 resected. Neurosurgeons usually place six to eight wafers in the tumor cavity to treat patients with brain tumors, primarily high-grade gliomas. These wafers release the chemotherapeutic agent locally for approximately three weeks, thereby, treating tumors in the brain with a high concentration of carmustine at the tumor site and minimizing adverse effects of systemic chemotherapy (Attenello et al., 2008). The rate at which the wafers biodegrade varies from patient to patient. Computer tomography scans obtained seven weeks after implantation demonstrated that wafer remnants were still present in of 11 of 18 patients evaluated (GLIADEL® Wafer, 2012). Results of two studies that assessed the association of Gliadel® wafers and SSIs were contradictory (Attenello et al., 2008; McGovern et al., 2003). Attenello et al. retrospectively reviewed 1,013 patients undergoing CRANI for resection of malignant brain astrocytoma, and they found that patients receiving Gliadel® wafers and patients without Gliadel® wafers had similar incidence of SSIs (2.8% vs. 1.8%; P = 0.33). In contrast, McGovern et al. followed 32 patients who received wafers and 28% of these patients acquired SSIs. Similarly, we previously found that Gliadel® wafer implants were associated with a significantly increased risk of SSIs (OR = 8.38) among patients who underwent CRANI at the UIHC for brain tumors (Chapter 2.5) (Chiang et al., 2011). 2.3.1.7 Cerebrospinal fluid drainage CSF drains, such as ventricular drains and lumbar drains, could increase patients’ risk to infections because microorganisms may enter the central nervous system through the drain catheters. External drains are used to monitor intracranial pressure, temporarily divert CSF from an obstructed ventricular system, or to help treat infected

23 internal catheters. In contrast, internal drains (e.g., ventriculo-peritoneal shunts) are used to treat hydrocephalus long term (Van de Beek et al., 2010). Three studies found that in patients undergoing CRANI procedures, patients with external CSF drainage devices, such as intracranial pressure monitors, had a higher risk of acquiring SSIs (Korinek et al., 1997) or meningitis (Reichert et al., 2002; Kourbeti et al., 2007) after CRANI. 2.3.1.8 Cerebrospinal fluid leakage

2.3.2 Surgical site infection risk Index A risk index consisting of patient-related factors can help surgeons identify patients at high risk for SSIs. Surgeons may be able to lower SSI rates by focusing specific interventions on patients with high risk scores. Additionally, if one wants to compare SSI rates in a valid manner, the rates must first be corrected as much as possible for the differences in the patients’ intrinsic risk of SSIs. The remaining differences may reflect differences in the quality of care. Thus, the preoperative risk index may: (1) allow surgeons to identify patients that are at high risk of acquiring SSIs; (2) allow various groups (e.g., infection prevention programs, surgeons, CDC, state health departments,

24 insurance companies) to adjust SSI rates for patient-related risk factors so that they can compare SSI rates between hospitals, between surgical teams, or over time for the same surgical team. 2.3.2.1 The National Healthcare Safety Network risk index The National Healthcare Safety Network (NHSN) risk index (or NNIS risk index), the most widely used SSI risk index score, was developed by CDC in 1991 to stratify the risk of SSIs among surgical patients undergoing a broad range of surgical procedures. CDC first developed the SENIC (Study on the Efficacy of Nosocomial Infection Control) risk index, which evaluated 10 potential risk factors, including the total number of operations among patients undergoing various operations (Haley et al., 1985). These 10 potential risk factors included the total number of operations, the maximum number of surgical procedures performed during any operation, the maximum duration of the patient’s operations, the highest category of any of the patient’s wounds, an indicator of the anatomic location of the operation, the number of days of preoperative stay before the patient’s first operation, an indicator for the use of corticosteroid or immunosuppressive therapy, the total number of discharge diagnoses excluding diagnoses for SSIs and their complications, the presence of an infection remote from the operative site before surgery, and the patient’s age and sex. They subsequently developed the simpler NHSN risk index. The NHSN risk index score, which categorizes patients into four risk groups (0, 1, 2, 3), is calculated by adding one point for each criterion the patient meets: (1) a patient with an ASA score of 3, 4, or 5, (2) a wound classification of contaminated or dirty-infected at the time of the procedure, (3) an

25 operation lasting over T hours, where T represents the 75th percentile of the distribution of the operation durations (Culver et al., 1991). The ASA score is a physical status classification system anesthesiologists use to evaluate the patient’s preoperative physical status before a procedure. The higher the ASA score, the more severe the patient’s underlying illnesses and the lower the likelihood the patient will survive the operation (Appendix 2). The NHSN risk index predicts SSIs risk significantly better than the traditional wound classification and it performs well across many operative procedures (Culver et al., 1991). However, the NHSN risk index had low sensitivity and specificity in predicting whether individual patients would acquire an SSI after some procedures, including digestive procedures, cardiac procedures, total hip or total knee replacement procedures, abdominal hysterectomy procedures, and CRANI procedures (Gaynes et al., 2000; de Oliveira et al., 2006; Roy et al., 2000; Russo et al., 2002; Clement et al., 2007). In a prospective multicenter study of 2,944 patients undergoing craniotomy in France, the NHSN risk index had a poor sensitivity (49%) and specificity (66%) in predicting an individual patient’s risk of acquiring an SSI (Korinek et al., 1997). In a recent study of risk factors associated with craniotomy SSIs done in Mexico, the predictive power (defined by the c-statistic) of the NHSN risk index was only 0.558, indicating that this index had a low predictive power (Sanchez-Arenas et al., 2010). The NHSN risk index has three major limitations. First, surgeons cannot utilize the NHSN risk index to predict a patient’s risk of SSIs before an operation, because the operation duration is known after the procedure. Second, the NHSN risk index does not

26 stratify patients’ risk of SSIs after some procedures, including CRANI (Korinek et al., 1997; Ercole et al., 2007), because the three factors in the NHSN risk index often are not significantly associated with SSIs after these procedures or if they are risk factors, they are not equally important. For CRANI procedures, an ASA score of more than two and a wound classification of contaminated or dirty were not significant risk factors for SSIs in three studies (Korinek et al., 1997; Korinek et al., 2005; Sanchez-Arenas et al., 2010), and operation duration was not a significant risk factor in four studies (Lietard et al., 2008; Reichert et al., 2002; Sanchez-Arenas et al., 2010; Shinoura et al., 2004). None of the three factors were associated with SSIs post-CRANI in our previous study (Chiang et al., 2011). The third limitation is related to the second limitation; if most patients undergoing a clean procedure have ASA scores of less than three, the NHSN risk index will classify the patients into two risk groups based solely on the procedure duration. Thus, the NHSN report for 2006-2008 stratified the patients undergoing CRANI into two risk groups (0 and 1, 2 and 3) (Edwards et al., 2009). 2.3.2.2 New NHSN procedure-specific risk model Because of the limitations of traditional NHSN risk index, many investigators have developed procedure-specific multivariable risk models that incorporated additional weighted patient factors to better stratify patients’ risk of SSIs. Recently, the NHSN investigators analyzed 849,659 procedures from the current NHSN database and created 39 procedure-specific, multivariable risk models, including a model for SSIs after CRANI (younger age, ASA > 2, operation duration, hospital bed size > 500, and trauma as the reason for the procedure) (Mu et al., 2011). Although 33 of 39 NHSN procedure-

27 specific models had better predictive power than the traditional NHSN risk index, the improvement was limited. For example, the CRANI-specific model had a c-statistic of 0.65, just slightly better than the c-statistic of 0.56 for the NHSN risk index. The new risk models were not strongly predictive because Mu et al. utilized the variables that existed in the NHSN database to develop the models and they did not have access to data on important risk factors, such as preoperative chemotherapy. Second, the investigators included only statistically significant variables in the models and, thus, they could have missed clinically important risk factors that did not reach statistical significance (Moehring et al., 2011). To create better risk indices, investigators should examine more risk factors that are specific to the procedure of interest, and they should include variables that are statistically significant and variables that are clinically significant but did not reach statistical significance in when modeling the risk of SSIs. 2.3.2.3 Procedure-specific surgical site infection risk indices One research group in Australia has done two studies to identify risk factors for SSIs after CABG in multivariable models and to create CABG-specific SSI risk indices (Friedman et al., 2007; Russo et al., 2002). They used a different combination of the independent risk factors to construct three risk indices: Index 1 included only preoperative factors (i.e., obesity, peripheral or cerebrovascular disease, and insulindependent diabetes mellitus) for all SSIs. Index 2 included these preoperative factors and operation duration for all SSIs (Russo et al., 2002), and index 3 included the preoperative factors body mass index (BMI) and diabetes for sternal SSIs (Friedman et al., 2007). The authors used the Goodman-Kruskal nonparametric correlation coefficient

28 (G; G < 0.3 poor; G = 0.3 - 0.5 moderate; G > 0.6 high predictive power) to evaluate the predictive power of indices 1 and 2 and they used the c-statistic to evaluate the predictive power of index 3. All these three CABG-specific SSI risk indices predicted SSIs better than the NHSN risk index (index 1 had G = 0.33; index 2 had G = 0.34; index 3 had c = 0.64; NHSN risk index had G = 0.31 and c = 0.52). On the basis of data from a recent study in Mexico, Sanchez-Arenas et al. built two risk indices. One risk index included patient-related risk factors as indicators of the patients’ underlying conditions (i.e., presence of chronic disease and non-trauma as the reason for the procedure). The other index included procedure-related risk factors as indicators of the healthcare quality (i.e., repeated procedure at the same surgical site and operation performed during the late shift). The c-statistic was 0.625 for the patientrelated risk index and 0.731 for the procedure-related risk index compared with 0.558 for the NHSN risk index (Sanchez-Arenas et al., 2010). Although the two risk indices developed by Sanchez-Arenas et al. could predict SSIs better than NHSN risk index, the external validity of this study is limited because the investigators excluded patients who received implants during their procedures, which are common among CRANI procedures. In addition, Sanchez-Arenas et al. used the same patient cohort to develop their risk indices and to assess the predictive power of these indices, which might optimized the predictive power. One could better validate a risk index with bootstrap or cross-validation resampling methods in the same cohort (Schumacher et al., 1997) or by applying the risk index in another patient cohort.

29 2.3.2.4 Development of a surgical site infection risk index A risk index should be developed through multivariable analysis of a pool of variables representing all the important underlying risk. Haley discussed the principles of developing a risk index that measured the patient’s intrinsic risk of SSIs. The principles included: (1) all potential risk factors for SSIs should be collected and analyzed; (2) variables in the index should be feasibly and consistently obtained; (3) variables in the index should be parsimonious but representative for the risk of SSIs; (4) a simple scale for each variable should be used to build the index; (5) the predictive ability of the index should be validated (Haley, 1991). 2.3.3 Approaches in this study We reviewed the literature comprehensively to identify as many potential risk factors for CRANI SSIs as possible. We included all types of CRANI procedures and all SSIs in our case-control study and we used multivariable analysis to identify important risk factors for SSIs. Subsequently, we developed a preoperative SSI risk index using patient-related risk factors, and we validated the risk index using a cross-validation method.

2.4 Outcomes associated with surgical site infections after craniotomy/craniectomy Some SSIs are preventable which in turn could prevent adverse outcomes associated with SSIs. Patients’ clinical outcomes, such as postoperative death, intensive care unit (ICU) admission, reoperation, readmission, and postoperative length of stay (LOS), are particularly important now given the financial recession and increased public interest in patient safety.

30 2.4.1 Outcomes associated with surgical site infections SSIs are associated with substantial morbidity and mortality and they increase length

of hospitalizations, hospital readmissions, and costs (Herwaldt et al., 2006; Hollenbeak et al., 2000; Kasatpibal et al., 2005; Kirkland et al., 1999; Scott, 2009). A study done in a community hospital with 255 patient pairs matched by age, procedure, the NHSN risk index, procedure date, and surgeon found that patients who acquired SSIs were twice as likely to die during the postoperative hospitalization and to stay in the hospital for an additional 6.5 days. Patients with SSIs were greater than five times more likely to be readmitted to the hospitals than were patients who did not acquire SSIs (Kirkland et al., 1999). A study of 3,864 surgical patients at a large tertiary care medical center and an affiliated Veterans Affairs Medical Center found that SSIs were significantly associated with longer postoperative LOS (adjusting for patients’ functional impairment, preoperative LOS, and age). Among patients who underwent neurosurgical procedures and survived to be discharged, patients with SSIs were more likely to be readmitted to the hospital than those without HAIs (OR = 5.62; P = 0.002), after controlling for patients’ underlying diseases, comorbid conditions, and the NHSN risk index (Herwaldt et al., 2006). Many investigators have assessed the influence of SSIs on patient outcomes for different types of procedures. However, few investigators have specifically studied outcomes associated with SSIs after CRANI. Kasatpibal et al. conducted a matched case-control study of patients undergoing surgical procedures (i.e., CRANI, appendectomy, herniorrhaphy, cholecystectomy, mastectomy, and colectomy) in Thailand; 62 pairs of cases and controls had CRANI

31 procedures. Cases and controls were matched by the types of operations, the diagnosis, the operative year, and the ASA score. They identified a mean extra charge of 67,116 Baht (2,177 USD) and a median extra charge of 49,637 Baht (1,610 USD) that were attributable to SSIs after CRANI procedures. The mean excess postoperative LOS attributable to SSIs after CRANI procedures was 31 days (median = 25 days) (Kasatpibal et al., 2005). Reichert et al. conducted a matched case-control study with 50 pairs of patients who underwent CRANI procedures in a tertiary university hospital. Each patient who acquired postoperative meningitis (case) was matched to a patient who did not acquire meningitis (control) by age, diagnosis, procedure, and procedure date. The investigators identified 15 in-hospital deaths (30%) among cases and none among controls. The mean total LOS (preoperative and postoperative LOS) in the initial stay was significantly different among cases and controls (43 days vs. 19 days) (Reichert et al., 2002). However, it is inappropriate to include preoperative LOS in the analysis of outcomes because preoperative stay could not be contributed by SSIs. Our previous study showed that 85% of patients with SSIs required additional surgical procedures to treat these infections (Chiang et al., 2011). The previous studies of outcomes associated with SSIs after CRANI might have overestimated outcomes because these studies did not control for covariates, such as patients’ comorbidities, severity of underlying illnesses, or preoperative LOS, that may also contribute to the patients’ outcomes. Moreover, these studies did not evaluate other important outcomes such as hospital readmissions or reoperations.

32 2.4.2 Approaches to assess outcomes When determining the attributable patient outcomes related to HAIs, one must adjust for preoperative LOS, severity of illness, and underlying comorbid conditions. In addition, one must include SSI as a time-varying factor in the model because patients would not have SSIs at the time when they have their procedures. Investigators who do not adjust for these factors could bias the estimates of attributable outcomes. 2.4.2.1 Controlling for length of hospitalization before the procedure Patients’ clinical conditions could be reflected in prolonged preoperative LOS, and thus, preoperative LOS could contribute to postoperative outcomes. For example, patients undergoing CRANI to treat head trauma tend to have short preoperative LOS and to have long postoperative LOS. In addition, as the preoperative and postoperative LOS increases, the patients’ risk of acquiring SSIs would increase. Studies that do not adjust for the time at risk of acquiring HAIs could overestimate the LOS attributable to HAIs by up to 2-fold because prolonged LOS may itself be an important risk factor for HAIs (Schulgen et al., 2000). 2.4.2.2 Adjusting for underlying severity of illness and comorbidities

33

McCabe and Jackson used a simple three-category score to predict mortality in patients with bacteremia caused by Gram-negative organisms. Higher MJ scores represent a lower severity of illness (McCabe et al., 1962). Studies evaluating outcomes associated with HAIs (Herwaldt et al., 2006) or methicillin-resistant Staphylococcus aureus (MRSA) bacteremia (Cosgrove et al., 2005) have used the MJ scoring system to adjust for its confounding effect. The MJ score is subjective and is based completely on the judgment of person reviewing the patient record. This score does not include objective physiological data, which limits its generalizability from study to study. Cosgrove et al. indicated that the MJ score worked well as predictor of mortality but not as a predictor of morbidity and cost (Cosgrove et al., 2003). The Charlson comorbidity index was developed for classifying comorbid conditions that might alter the risk of mortality (Charlson et al., 1987). It included 19 diseases that

34 have been shown to be associated with increased mortality. Each disease was assigned a weight varying from 0 to 6 (Appendix 7). Investigators have included the Charlson comorbidity index in their analyses of risk factors for antimicrobial-resistant infections (McGregor et al., 2005) and in a study examining mortality associated with S. aureus bacteremia (Lesens et al., 2003) to adjust for patients’ underlying comorbidities. However, no published study has validated the Charlson comorbidity index in a study assessing outcomes associated with SSIs. In addition, the associations between the Charlson comorbidity index and other outcomes such as costs, readmission, or reoperation have not been evaluated. Infections affect patients’ illness severity and comorbid conditions. If an investigator assesses patient’s severity or comorbidities when the patient is actively infected, the severity of illness and the comorbidities may represent an intermediate variable in the chain of events between infection and the outcome of interest. If investigators adjust for an intermediate variable, they are likely to underestimate the effect of the exposure of interest (i.e., SSIs) on the outcome. Therefore, one must assess the severity of illness before the first signs of infection (Cosgrove et al., 2003). 2.4.2.3 Surgical site infection as a time-varying factor

-

-

35

- -

-

-

-

In this study, we evaluated the patient outcomes of postoperative hospital stay during the initial admission and readmission, postoperative death, hospital readmission, and reoperation that were associated with the initial CRANI procedure, controlling for patient’s age, comorbidities, severity of illness, preoperative LOS, the reason for the procedure, and procedure month period. The reason for the procedure can be associated with patient outcomes such as postoperative LOS. For example, a patient

36 who undergoes CRANI to lower intracranial pressure resulting from an automobile accident often has other injuries that must be treated. Therefore, the postoperative LOS for such patients tends to be longer than that for patients who undergo CRANI to remove brain tumors. This excess postoperative LOS is attributed to the reason for the procedure, but not to the exposure of interest (i.e., SSIs). Thus, investigators must adjust for the reason for the procedure to obtain the unbiased estimates of outcomes. In addition, we controlled for procedure month period in the analyses because residual confounding could result from incomplete frequency matching. Moreover, we used the Cox proportional hazards model to include important covariates and to treat SSIs as a time-dependent covariate in evaluating these outcomes. Thus, the study results should provide a good understanding and precise estimate about the outcomes attributable to SSIs after CRANI.

2.5 Pilot study: Clinical significance of positive bone flap cultures and associated risk of surgical site infection after craniotomies or craniectomies Previously, we conducted a prospective cohort study to investigate the influence of positive bone flap culture on the risk of SSIs after CRANI at the UIHC. We did this pilot study because surgeons were concerned that reimplanting contaminated cranial bone flaps increased the risk of SSIs. The results of the pilot study were published in June, 2011 in the Journal of Neurosurgery (Chiang et al., 2011).

37 2.5.1 Introduction Between January 2006 and December 2006, nine (12.9%) of 70 autograft cranial bone flaps sent to the UIHC’s Tissue Bank for storage and later reimplantation had positive cultures when they were removed; six bone flaps (66.7%) grew small numbers of Propionibacterium acnes (P. acnes). Between January and October 2007 at the UIHC, 23 (38%) of 60 of the cranial bone flap grew positive cultures; 15 bone flaps (25%) grew P. acnes and eight (13%) grew other organisms. Bone flaps that grew P. acnes were reimplanted in two patients, both of whom subsequently acquired P. acnes SSIs. One patient acquired a P. acnes SSI after a culture-negative bone flap was reimplanted. P. acnes, a common member of the skin flora, can cause SSIs with delayed onset after neurosurgical procedures (Esteban et al., 1995; Nisbet et al., 2007). Thus, staff questioned the safety of reimplanting cranial flaps contaminated with small numbers of P. acnes or with other organisms and they placed a moratorium on reimplanting these cranial flaps. We conducted a literature search but did not find any studies that evaluated the SSI risk for patients whose bone flaps were contaminated. Therefore, we conducted a prospective cohort study that included all patients who underwent CRANI procedures in the UIHC between November 2007 and November 2008 to: (1) determine the frequency of positive cranial flap cultures, (2) assess the relationship between positive bone flap cultures obtained during the original CRANI and SSIs after reimplantation during either the initial operation or a subsequent cranioplasty, and (3) identify risk factors for SSIs following the initial CRANI.

38 2.5.2 Methods All neurosurgery patients undergoing CRANI procedures at the UIHC between November 1, 2007 and November 30, 2008 whose cranial flaps were reimplanted during the initial CRANI or were sent to the Tissue Bank were evaluated for inclusion in the cohort study. Patients with cranial infections at the time of their initial procedures were excluded. If a patient had more than one CRANI within 30 days, only the first procedure was included. Nursing personnel swabbed all surfaces of the cranial flaps for aerobic and anaerobic cultures before disinfecting the flap in Bacitiracin for 10 minutes. The cranial flaps were then either reimplanted during the same procedure or packaged for storage. Circulating nurses collected data about the surgical team, hair removal, skin preparation, and use of a sealant or an incise drape and recorded these data on a data collection tool developed for the study. We collected data on demographics, the procedure (for example, operation date, wound classification, reason for the operation, the urgency of the operation, and duration of the operation), whether a Gliadel® wafer (Carmustine, Guilford Pharmaceuticals) was implanted in the tumor bed (only patients with tumors), and microbiology results from the patients’ medical records. We also determined the indication for the procedure by reviewing the procedure notes and then categorized the indications as tumor, bleed, trauma, or other. An infection preventionist independently identified SSIs occurring after the initial craniotomies/craniectomies or after delayed cranioplasties through routine surveillance using the NHSN’s definitions (The NHSN Manual, 2009) (Appendix 1). We entered all data into Excel.

39 We evaluated patients whose bone flaps were replaced during their initial CRANI procedures and those whose flaps were replaced during delayed cranioplasties to assess the effect of reimplanting contaminated bone flaps. We included only the initial CRANI procedures in analyses assessing SSI risk factors and we based the analyses of risk factors on the number of procedures, not on the number of patients. We used the Fisher’s exact test or logistic regression analysis for categorical variables and the twosample t-test for continuous variables to do the bivariable analyses. We used exact logistic regression for the bivariable stratified analyses. We set significance at P = 0.05, and we calculated 95% confidence intervals (CIs). We used the SAS software program (SAS Institute, Inc.) to conduct the statistical analyses. 2.5.3 Results Three hundred seventy-three patients underwent 393 CRANI procedures, of which 377 procedures met the study criteria (with four patients underwent two procedures). (Figure 3) The mean age of the study population was 48 years and 195 (52.3%) patients were male. The mean procedure duration was 215 minutes, which approximates the NHSN risk index cut point of 219 minutes (NNIS Report, 2004). After the 377 initial procedures, 21 patients acquired 22 SSIs (5.8%) (one patient acquired an SSI after each of two procedures). Twenty patients (5.4% of all 373 patients) acquired SSIs after their initial CRANI. Two patients acquired SSIs after delayed cranioplasties; one of whom had an infection after the initial CRANI as well. Most SSIs were deep (nine; 40.9%) or organ space (nine; 40.9%) infections. The median time to

40 onset of the infections was 20.5 days (range 2-117 days). Eighteen (85.7%) of 21 patients with SSIs required a second surgical procedure to treat their infections. One hundred eighty-six (50%) of the 372 cranial bone flaps for which culture results were known had positive cultures, of which 147 (79%) grew P. acnes alone or in combination with other flora. Most (184) of the positive cultures had rare (range 1-18) colonies, and only two cultures had few colonies; no cultures had moderate or many colonies. In contrast to contamination, only three (13.6%) of 22 SSIs were caused by P. acnes. S.aureus and coagulase-negative staphylococci (CoNS) were common etiologic agents. Only two patients were infected with the organisms contaminating their bone flaps. Three hundred thirty-three patients underwent 336 procedures to reimplant bone flaps. Reimplantation was performed either during the initial procedure or during a separate cranioplasty. Nine (6.0%) of 151 procedures associated with positive bone flap cultures and eight (4.8%) of 165 procedures associated with negative cultures were complicated by SSIs. Reimplantation, during either the initial procedure or a subsequent cranioplasty, of bone flaps with positive cultures was not significantly associated with SSIs (p = 0.80). Positive bone flap cultures (P = 0.64; OR = 1.39), having P. acnes in bone flap cultures, age, duration of the procedure, hair removal, emergency procedures, and the indication for the procedure were not risk factors for SSIs after the initial procedures (Table 2). Use of 10% povidone-iodine gel followed by povidone-iodine gel solution to prepare the skin and allowing the skin antiseptic to dry before the procedures were significantly

41 associated with decreased risk of SSIs. The difference between povidone iodine gel and solution and other prep solutions remained significant when the data were stratified by the scheduling of the case (emergency vs. nonemergency; P = 0.015, OR = 0.29), indication for the procedure (trauma vs. other; P = 0.013, OR = 0.28), and duration of the procedure (≥ 219 minutes vs. < 219 minutes [the duration cut point for the NHSN risk index]; P = 0.016, OR = 0.29). The difference between allowing the skin antiseptic to dry and not allowing it to dry also remained significant when the data were stratified by the scheduling of the case (P = 0.026, OR = 0.25), the indication for the procedure (P = 0.026, OR = 0.26), and the duration of the procedure (P = 0.029, OR = 0.26). Female sex was associated with an increased risk of SSIs; Gliadel® wafers were significantly associated with an increased risk of SSIs after procedures to treat tumors (Table 2).

42

Figure 3. Flow chart describing information obtained in the study population. ______________________________________________________________________________ Abbreviation: SSI = Surgical site infection. *The risk of SSIs after the initial procedures when bone flaps were reimplanted immediately (16; 5.4%] of 297) was not significantly different from the risk of SSI when bone flaps were banked (4; 5%] of 80) (P = 1.00, Fisher’s exact test). †Twenty SSIs occurred in 20 patients after the initial craniotomy/craniectomy procedures. ‡Follow-up information through 11/30/2009 is provided for the 80 procedures (79 patients) for which bone flaps were banked (beneath the broken line). However, risk factors for SSIs were assessed only for the initial craniotomy/craniectomy procedures (above the broken line), not for delayed cranioplasty procedures.

43 Figure 3. Continued §These two SSIs were not included in the study of risk factors for SSIs. One of the two patients also acquired an SSI after the craniotomy/craniectomy procedures; this initial SSI was included in the risk factor study.

44 Table 2. Potential risk factors and their associations with surgical site infections after craniotomy/craniectomy. SSI (N = 20)

No SSI (N = 357)

Pvalue

OR (95% CI)

Age, Mean (range)

51.2 (8-71)

47.6 (0-95)

0.49

N/A

Female Gender

15 (75.0%)

165 (46.2%)

0.02

3.49 (1.24, 9.81)

3 hr & 43 min

3 hr & 34 min

0.74

N/A

Case Classification Emergent Non-emergenta

7 (35.0%) 13 (65.0%)

108 (30.2%) 249 (69.8%)

0.63

1.24 (0.48, 3.20) 1.00

Indication for Procedure Tumor Bleed Trauma Others

13 (65.0%) 5 (25.0%) 2 (10.0%) 0

194 49 72 42

(54.3%) (13.7%) (20.2%) (11.8%)

0.14

N/A

8 (42.1%) 9 (47.4%) 2 (10.5%)

178 (50.4%) 138 (39.1%) 37 (10.5%)

0.70

N/A

2 (11.1%)

30

(8.6%)

0.66

0.75 (0.17, 3.44)

6 (33.3%) 9 (50.0%) 3 (16.7%)

224 (63.1%) 107 (30.1%) 24 (6.8%)

0.04

0.21 (0.05, 0.91) 0.67 (0.17, 2.67) 1.00

Skin Preparation Dried (N = 330)

12 (75.0%)

289 (92.0%)

0.04

0.26 (0.08, 0.86)

Gliadel® Wafer in Patients With Tumors (N = 207)

6 (46.2%)

18 (9.3%)

0.001

8.38 (2.54, 27.6)

Variable Name

Operation Duration (mean)

Organism Contaminating Bone Flap (N = 372) None P. acnes and others Others No Hair Removal (N = 366) b

Skin Preparation (N = 373) 10% PVP gel & solution 10% PVP gel CHG

Abbreviation: CHG = Chlorhexidine gluconate; CI = Confidence interval; N/A = Not applicable; OR = Odds ratio; PVP = Povidone-iodine.

45 Table 2. Continued Note: If the total number is not specified, the total number of craniectomy/craniotomy procedures evaluated was 377. a. Non-emergent cases included urgent and elective cases. b. Calculated by logistic regression analysis.

2.5.4 Summary and discussion This pilot study evaluated the effect of contaminated bone flaps and identified other operative factors that may be associated with SSIs after CRANI procedures. A high proportion (50%) of bone flaps were contaminated, but contaminated bone flaps, including those that grew P. acnes, were not associated with the risk of SSIs. However, SSIs were more likely to occur when the skin prep was not allowed to dry, which is biologically plausible because antiseptics need time to kill the bacteria on the skin. We also found that SSIs were more likely to occur when the skin preparation was done with either PVP solution alone or with chlorhexidine gluconate (CHG). However, our data should not be interpreted as suggesting that PVP is superior to CHG because staff members did not follow the manufacturer’s instructions when they used CHG to prepare a patient’s skin. For patients with brain tumors, Gliadel® wafer implants significantly increased the risk of SSIs. Thus, operative factors may be more important than low numbers of skin flora contaminating the bone flap in the pathogenesis of SSIs after CRANI. However, this study had a few limitations. First, very few cultures either grew even moderate numbers of bacteria or grew highly virulent organisms. Thus, we could not

46 evaluate the effect of these factors on the risk of SSIs. Second, unmeasured patient factors (e.g., comorbidities and smoking history) or operative factors (e.g., antimicrobial prophylaxis) may have confounded our results. We do not think that most of these factors would have been associated with whether the skin prep was allowed to dry or with the choice to use Gliadel® wafers. However, factors such as prior operations, radiation therapy, and immunosuppression may have been associated with recurrent tumors and, thus, with the choice to use Gliadel® wafers. Third, the small sample size limited our ability to detect risk factors that had small ORs or that increased the risk of SSIs slightly at the significance level of 0.05. For example, given that 30% of the procedures were emergent, the study had a power of 80% to detect an OR of 3.75 in the risk of SSIs for emergency cases compared with nonemergency cases, but it had only a 33% power to detect an OR of 2.0. Thus, we may have missed some important risk factors that had smaller odds ratios or had less pronounced effects on the risk of SSIs. The pilot study gave us useful clinical information that caused the neurosurgeons to change their practice (e.g., decreasing the use of Gliadel® wafers and implanting bone flaps contaminated with small numbers of commensal organism). However, we did not evaluate many potential risk factors associated with SSIs in the pilot study. To better understand the epidemiology of SSIs after CRANI, we proposed to do a case-control study with a larger sample size and a longer study period that would assess more potential risk factors and confounders related to SSIs after CRANI.

47 2.6 Significance Investigators assessing interventions to prevent HAIs, such as SSIs, often use a quasiexperimental study design. However, investigators often use suboptimal analytic approaches that might lead them to false inferences. For example, numerous investigators have used two-group tests to assess data that violated the assumptions that the data are normally distributed and independent and tests that cannot determine whether the trend of HAI rates changed over time. In addition, the investigators rarely include a comparison group to control for unmeasured confounders, maturation effect, or regression to the mean, which common affect quasi-experimental studies. A number of published studies have evaluated risk factors for SSIs after CRANI procedures. However, most studies included only certain CRANI procedures (e.g., elective) or certain infections (e.g., meningitis) and only two published studies were done in the US. Thus, the results of the prior studies might have limited generalizability. In addition, studies examining patient outcomes associated with SSIs after CRANI procedures did not control for patients’ underlying severity of illness or comorbid conditions and they did not use an appropriate analytic method (e.g., survival analysis) to assess the outcomes. We designed the current study to address the gaps in the literature and the methodological errors. Thus, the current study has several important features. First, we included all CRANI procedures and all SSIs during the study period and we evaluated as many potential risk factors for SSIs as possible so that we could better understand the epidemiology of SSIs after CRANI. Second, we used appropriate analytic methods and

48 included a control group to evaluate the effect of reducing the use of Gliadel® wafers on SSI rates. Third, we created a simple SSI risk index that included the preoperative risk factors specific for CRANI. Surgeons and infection preventionists can use this risk index to identify patients at high-risk of SSIs and to adjust for heterogeneity among patient populations when comparing SSI rates among surgeons or hospitals. Fourth, we utilized appropriate analytic methods to control for covariates such as age, preoperative LOS, severity of illness, or comorbidities, and to take into account the time-to-event when assessing the postoperative patient outcomes. These methods should provide better estimates of the outcomes attributable to SSIs after CRANI than the methods used in prior studies. Thus, the results of this study will contribute substantially to the literature on SSIs after CRANI.

49 CHAPTER 3 – RESEARCH DESIGN AND METHODS

3.1 Overview of study design 3.1.1 Study population The study population comprised all the craniotomy or craniectomy (CRANI) procedures done in the University of Iowa Hospitals and Clinics (UIHC)’s Department of Neurosurgery between 01/01/2006 and 12/31/2010 that met the inclusion criteria for this study. CRANI procedures were defined by the ICD-9-CM procedure codes listed in Appendix 3. We identified patients by obtaining a list of procedures from the UIHC’s Health Information Management. Patients could have more than one CRANI procedures during the study period. CRANI procedures done in the Department of Otolaryngology were excluded because the surgical techniques and approaches of otolaryngologic surgeons would be different from those of neurosurgeons, and most CRANI procedures were performed by neurosurgeons. Otolaryngologic surgeons and neurosurgeons would cooperate during a procedure. Procedures were included if the neurosurgeons was the primary surgeon. In addition, patients who died within two days after their procedures were excluded because these patients were unlikely to acquire surgical site infections (SSIs) within that period. We based this criterion on data from our previous cohort study; the time to onset of SSIs after CRANI ranged from two to 117 days (Chiang et al., 2011). In addition, two studies of SSIs after CRANI included patients who survived at least seven days after

50 their procedures (Korinek et al., 1997; Kourbeti et al., 2007). Figure 4 illustrates how the study population was generated. 3.1.1.1 Specific aim 1 For aim 1, we further excluded any subsequent procedures for patients who acquired SSIs, because the subsequent CRANI procedures were no longer “at-risk” of SSIs. We used the ICD-9-CM diagnosis codes for malignant brain tumors (Appendix 4) to identify the intervention group (patients with ICD-9-CM diagnosis codes for malignant brain tumors) and control group (patients with other ICD-9-CM diagnosis codes). The intervention involved reporting the results of the previous cohort study (Chapter 2.5) to the neurosurgeons and their decision to significantly limit their use of Gliadel® wafers (Figure 5). Because Gliadel® wafers were used exclusively to treat patients with malignant brain tumors, limiting the use of Gliadel® wafer should only affect the intervention group.

51

Figure 4. Flow chart describing the study population for each specific aim. ________________________________________________________________________ Abbreviation: CRANI = Craniotomy and craniectomy; SSI = Surgical site infection.

52 3.1.1.2 Specific aim 2 For aim 2, we only included one CRANI procedure if the patients had more than one procedure during the study period. If a patient had more than one procedure during the same admission and second procedure was performed through the same incision within 24 hours of the original operative procedure, we counted these procedures as one procedure and combined information from the two procedures (e.g., the operation durations was the sum of the duration for the first procedure and for the second procedure). If the patient’s status changed between the procedures, we recorded the worst values. For example, if a patient had an American Society of Anesthesiologists (ASA) score of two before the first procedure and ASA score of three before the second procedure, we recorded an ASA score of three for that patient (The NHSN manual, 2009). We excluded patients who had infections at the time of their procedures. We defined cases as any patient who underwent CRANI between January 1, 2006 and December 31, 2010 and who was identified by the University of Iowa Hospitals and Clinics (UIHC)’s Program of Hospital Epidemiology meeting National Healthcare Safety Network (NHSN)’s criteria for SSIs (The NHSN manual, 2009). See Chapter 2.2 and Appendix 1 for the definition of SSIs. Only the first episode of SSI was included for patients who acquired more than one SSI during the study period. A control was any patient in the study population who had at least one follow-up visit or follow-up phone call and who did not acquire a SSI, either from the initial CRANI procedure or any subsequent procedure(s). We used an on-line random number generator (http://stattrek.com/tables/random.aspx) to select three controls per case by

53 frequency matching them to cases by procedure month period so that cases and controls were distributed equally by procedure month. We divided the 60-month study period into 10 procedure month periods (e.g., 1/2006 to 6/2006, 7/2006 to 12/2006, 1/2007 to 6/2007, and so on, until 7/2010 to 12/2010). We chose controls in this manner to ensure that case and control patients would have received similar clinical care and would have been exposed to the same infection prevention interventions. This nested case-control study had 104 patients with SSIs and 312 controls (Table 3). 3.1.1.3 Specific aim 3 For aim 3, the inclusion and exclusion criterion were the same as those for aim 2, except that the study period was from January 1, 2006 to June 30, 2010, which was six months shorter than that for aim 2. This is because we did not evaluate the McCabe and Jackson severity of illness score for patients having CRANI during July 1, 2010 and December 31, 2010. Finally, this outcome study included 93 patients with SSIs and 279 controls.

54 Table 3. Number of cases and controls by procedure month period. Procedure Month Period

No. of SSIs

% of SSIs

No. of Controls

01/2006 – 06/2006

9

8.7%

27

07/2006 – 12/2006

5

4.8%

15

01/2007 – 06/2007

5

4.8%

15

07/2007 – 12/2007

17

16.4%

51

01/2008 – 06/2008

19

18.3%

57

07/2008 – 12/2008

7

6.7%

21

01/2009 – 06/2009

13

12.5%

39

07/2009 – 12/2009

12

11.5%

36

01/2010 – 06/2010

6

5.8%

18

07/2010 – 12/2010

11

10.6%

33

Total

104

Abbreviation: SSI = Surgical site infection.

312

55 3.1.2 Study design Although the study population for each aim was from the same study cohort (i.e., CRANI procedures done during January, 2006 and December, 2010), the study design for each aim was different. 3.1.2.1 Specific aim 1 The design for aim 1, the SSI rate study, is a quasi-experimental study, or a pre- and post-intervention study, with a control group. We compared the SSI rates by month before and after an intervention — limiting use of Gliadel® wafers in April, 2009.

-

-

-

3.1.2.2 Specific aim 2 -

-

56

-

Recall bias is one weakness of a case-control study. In addition, in some instances, the temporal relationship between the exposure and the outcome cannot be established. Recall bias would not affect this study because we did not question patients about exposures and we were able to determine whether exposures occurred before the SSIs because information about the exposures was recorded in the patients’ medical records before the SSIs occurred. Furthermore, in a nested case-control study, cases and controls come from the same study population, which makes controls comparable to cases because controls had the same chance of becoming classified as cases as did the cases if they would have acquired SSIs. However, information bias could occur in this retrospective study because we obtained data from patients’ medical records, and data could be missing or documented inaccurately. Information bias could cause misclassification of exposures (risk factors) status in this study. However, the misclassification should be independent of the status of SSIs, and therefore, it would be a non-differential misclassification. Non-differential misclassification tends to bias the association toward the null hypothesis. 3.1.2.3 Specific aim 3 The study design for aim 3, the outcome study, is a matched cohort study because cases (exposed to SSIs) and controls (not exposed to SSIs) were frequency matched by the procedure month period and they were followed for one-year to identify

57 postoperative outcomes (i.e., postoperative length of stay [LOS], death, readmission, and reoperation). We compared the outcomes of patients who had SSIs after their initial CRANI procedures with the outcomes of patients who did not have SSIs, while controlling for variables that might affect the outcomes of interest. The goal of controlling for these variables was to ensure that the observed differences in outcomes would be contributed by SSIs, not by those confounding variables.

3.2 Data collection and measurement The list of CRANI procedures done from January, 2006 through December, 2010 was generated electronically from the UIHC’s Health Information Management in an Excel worksheet. CRANI procedures were defined by ICD-9-CM procedure codes listed in Appendix 3. These procedure codes should identify only patients who had CRANI procedures. However, we found that some procedures were miscoded and were not CRANI procedures. The problematic codes included 01.22 (removal of intracranial neurostimulator leads), 02.93 (implantation or replacement of intracranial neurostimulator leads), and 38.81 (Other surgical occlusion of vessels, intracranial vessels). We reviewed the procedure notes describing each procedure with these codes and excluded the ones that were not CRANI procedures. 3.2.1 Specific aim 1 For aim 1, the UIHC’s Health Information Management provided a list of patients with malignant brain tumors and a list of patients with Gliadel® wafer implants. Malignant brain tumors were defined by ICD-9-CM diagnosis codes (Appendix 4), and

58 Gliadel® wafer implants were defined by ICD-9-CM procedure code 00.10 (Implantation of chemotherapeutic agent: brain wafer chemotherapy, interstitial/intracavitary). 3.2.2 Specific aim 2 On the basis of the literature review described in Chapter 2, we identified several potential risk factors for SSIs after CRANI. We retrospectively collected data regarding potential risk factors (patient-related and procedure-related factors) from the patients’ electronic and paper medical records. The UIHC implemented Epic, a proprietary electronic medical record, in May, 2009. Most of the data needed for this study were available in Epic; however, some data remained in the previous electronic medical record (Inform Patient Record or Inform) or on paper records. We entered the data directly into a database in Microsoft Access 2010. 3.2.2.1 Exposure variables The exposure variables were the potential risk factors for SSIs after CRANI, which were identified by reviewing the literature (Chapter 2). We created a variable library (Appendix 6) that defined all the variables in the data abstraction form (Appendix 5) and in the analyses. Table 4 lists key potential risk factors in this study, some of which are described in the following paragraphs.

59 Table 4. Key potential risk factors for surgical site infections after craniotomy/ craniectomy. Patient-related factors

Procedure-related factors

Older age

Skin preparation

Gender

No antimicrobial prophylaxis

Higher body mass index

Improper timing of antimicrobial prophylaxis

Smoking

Surgeon

Higher comorbidity score

Operative path traverses a sinus

ASA score > 2

Deep brain neurostimulator implant

Wound classified as contaminated or

Intracranial pressure monitoring

dirty Reason for procedure

Use of external drains

Emergent procedure

Longer procedure duration

Previous brain operation

Gliadel® wafer implant for patients with tumors

No private insurance

Postoperative CSF leakage

Infection at another site

Postoperative hematoma at incision site

Longer preoperative hospital stay

Subsequent procedure at the same site

Abbreviation: ASA score = American Society of Anesthesiologists score; CSF = cerebrospinal fluid.

Age: Older age has been identified as a significant risk factor in one study of SSIs after CRANI (Shinoura et al., 2004). Aging is related to weaken immunity, decreased physical function, and it is a risk factor for SSIs after many procedures, such as appendectomy,

60 colorectal surgery, cholecystectomy, and hip prosthesis procedures (Brandt et al., 2004; Mangram et al., 1999). Therefore, it is important to evaluate for the effect of age on the SSI risk. Comorbidity: We identified patients’ comorbid conditions by their discharge ICD-9CM diagnosis codes from electronic medical records. We used two commonly used comorbidity indices, the Elixhauser method (Elixhauser et al., 1998) and the Charlson comorbidity index (Charlson et al., 1987), to summarize important comorbidities at the time of the procedure (Appendix 7). To assess the Elixhauser comorbidities, we followed the instructions from the Comorbidity Software v3.7 provided by Agency for Healthcare Research and Quality (AHRQ)’s Healthcare Cost and Utilization Project (HCUP) (HCUP Comorbidity Software, 2011). This comorbidity software consists of two SAS computer programs. One is the format program that defines a format library that contains the diagnosis and diagnosisrelated group (DRG) screens necessary for the comorbidity analysis. Another is the analysis program that uses the data of ICD-9-CM procedure codes and the format library to create the Elixhauser comorbidity variables. Elixhauser method does not provide a weighted score, but we summarized the number of comorbidities for each patient as a score. The SAS program for calculating Charlson comorbidity score was kindly provided by Dr. Marcia Ward in the Department of Health Management and Policy at the UIHC. The program creates Charlson comorbidity variables, assigns weight score to each comorbidity, and calculates the Charlson comorbidity scores by summing up the weight scores.

61 We assessed the risk of SSIs for individual comorbidity included in the two comorbidity methods. We also included the two comorbidity scores separately and compared their performance in predicting SSIs. The Elixhauser method generally performs better in predictive models but can sometimes be challenging to interpret. The Charlson method is less robust but yields a single composite, weighted score. The major difference between the Elixhauser method and the Charlson method is that the former is used to examine the independent effect of each comorbidity, while the latter is commonly used as a composite score to account for the effect of the number and severity of all comorbidities. Reason for procedure: We determined the reason for the procedure by reviewing the procedure notes describing the primary purpose for the procedure and using Korinek et al.’s categories — tumor, bleed, trauma, biopsy, infection, and others – to classify the reasons for the procedures (Korinek et al., 1997). “Tumor” was defined as a procedure involving exploration or removal of a brain tumor of any type. “Bleed” was defined as a procedure done to treat an aneurysm, a hemorrhagic stroke, an infarct, or a hematoma. “Trauma” was defined as a procedure treating traumatic brain injuries caused by motor vehicle accidents, falls, or other traumatic accidents. “Biopsy” was defined as a procedure to perform biopsy. “Infection” was defined as a procedure treating an existing brain infection. Patients in the “infection” category were excluded from this study. Reasons for the procedures that could not be classified into these five classes were classified as “other.” For example, a procedure done as treatment for epilepsy or Parkinson’s disease would be classified as “other.”

62 Private insurance: No private insurance is a proxy indicator for low social economic status. A study of SSIs after hysterectomy found that no private insurance was associated with 1.7-fold increase in the risk of SSIs (Olsen et al., 2009). The association between low social economic status and SSIs after neurosurgery has not been studied. We categorized patients’ insurance status into no insurance, Iowa Care, Medicaid, Medicare, Medicaid with supplemental private insurance, Medicare with supplement private insurance, and private insurance alone. We defined no private insurance as coverage by Medicaid or Medicare or as no health insurance. Wound classification: The wound classification schema divides surgical wounds into four classes: clean, clean-contaminated, contaminated, and dirty/infected wound (Mangram et al., 1999). If a patient had more than one procedure during the same admission and within 24 hours of the original procedure, we recorded the higher wound class if the wound class changed from one procedure to the next (The NHSN manual, 2009). In this study population, most of the patients had clean wounds. For patient who underwent CRANI through a sinus, the wound would be clean-contaminated. Antimicrobial prophylaxis: We collected data about whether prophylactic antimicrobial agents were given, the types of antimicrobial agents used, and the timing of prophylactic treatment (e.g., whether the antimicrobial agents were given before the incision was made). The antimicrobial dosage per body weight (mg/kg) was calculated for each type of antibiotics. Gliadel® wafers implantation: Gliadel® wafers are used as chemotherapy for patients with brain tumors, primarily for malignant gliomas. We previously found that Gliadel®

63 wafer implants were associated with a significantly increased risk of SSIs (Chapter 2.5) (Chiang et al., 2011). We reviewed patients’ medical records or their ICD-9-CM procedure

codes to determine whether they had Gliadel® wafer implants. Operation duration: Operation duration was collected routinely and entered into the procedure log in the electronic medical record. If a patient has more than one procedure during the same admission and another procedure is performed through the same incision within 24 hours of the original incision, we summarized the operation duration for both procedures (The NHSN manual, 2009). Subsequent procedure: Several groups have found that an early subsequent procedure at the same operative site increased the risk of SSIs after CRANI (Korinek et al., 1997; Korinek et al., 2005; Reichert et al., 2002; Sanchez-Arenas et al., 2010). A subsequent procedure was defined as another surgical procedure done at the same operative site within 30 days after the initial CRANI procedure. 3.2.2.2 Outcome variable The outcome variable is SSIs after CRANI. SSI is defined in Chapter 2.2. An SSI must occur within 30 days after the operation if there is no implant or within 365 days if there is an implant and the infection must appear to be related to the operative procedure. In addition, we reviewed patients’ medical records and collected the date when the SSIs occurred, the organism(s) causing the SSIs, the depth of the SSIs, and the setting where the SSIs was identified. The SSI depth is categorized as superficial incisional, deep incisional, and organ/space SSIs. Superficial incisional SSIs involve only the skin and subcutaneous

64 tissue of the incision (e.g., scalp infections); deep incisional SSIs involve deep soft tissues (fascial or muscle layers) of the incision (e.g., subgaleal infections); organ/space SSIs involve any part of the anatomy (organs or spaces) other than the incision, that is opened or manipulated during the operative procedure (e.g., osteomyelitis, meningitis, abscess, and empyema). (Korinek et al., 1997; The NHSN manual, 2009) The setting for the SSI is defined as the geographic site where the SSI was diagnosed, including inpatient, readmission, and outpatient. SSIs identified during the patients’ initial admissions were defined as inpatient, those identified during patients’ readmissions were defined as readmission, and those identified during outpatient visit, such as a clinic visit or an emergency room visit, were defined as outpatient. 3.2.3 Specific aim 3 3.2.3.1 Exposure variable The exposure variable in the outcome study is SSIs after CRANI. The definition of SSIs and the data collection are described above in the “outcome variable” section in aim 2. 3.2.3.2 Confounding variables We controlled for confounding variables of age, severity of illness, comorbidities, preoperative LOS, reason for procedure, and procedure month period. Older age, severe underlying illness, preexisting comorbidities, and reason for procedure may confound the association between SSIs and outcomes, such as postoperative LOS, death, readmission, or reoperation. In addition, we controlled for procedure month period because incomplete frequency matching could cause residual confounding. We

65 collected the data for confounding variables recorded during patients’ initial hospitalizations from medical records. We used the McCabe and Jackson (MJ) score to grade severity of illness (McCabe et al., 1962). The MJ score categorizes patients into three classes according to their admitting diagnoses and the clinical assessment of the severity of underlying illnesses: “rapidly fatal” (score 1) includes patients who are expected to die within two weeks; “ultimately fatal” (score 2) includes patients who are expected to die within five years; and “nonfatal” (score 3) includes patients who are expected to live for more than five years (Cosgrove et al., 2005). Because assigning MJ score requires clinical knowledge, we asked Dr. Aparna Kamath to review each patient’s preoperative notes and assign the score. We printed out each patient’s preoperative notes (e.g., admission notes, H&P, preoperative vital signs and lab values) from medical records, and Dr. Kamath read the notes and assigned the MJ score for each patient. Because Dr. Kamath was blinded to the patient’s outcomes after the procedure, the MJ score was not be biased by the patient’s actual outcomes (e.g., SSIs or deaths). The Charlson comorbidity index was used as an aggregate comorbidity measure as described above in the “exposure variable” section in aim 2. Some investigators have used the Charlson comorbidity index as an indicator of the patients’ comorbidities in their studies of risk factors for SSIs and outcomes of SSIs (Lee et al., 2006; McGarry et al., 2004). The reason for the procedures was also defined in the “exposure variable” section in aim 2.

66 3.2.3.3 Outcome variables The outcome variables in the outcome study were postoperative LOS, total postoperative LOS, death, readmission, and reoperation. Initially, we also evaluated postoperative intensive care unit (ICU) admission. However, we found that almost every patient was admitted to surgical ICU immediately after their CRANI procedures. Thus, we decided not to evaluate ICU admission as an outcome. The follow-up time for each patient was one year after the initial procedure because, according to the NHSN’s definition of SSIs, patients who have implants must be followed for one year after the index procedure to determine whether they acquire SSIs. The actual follow-up time may be different from patient to patient because neurosurgeons at the UIHC do not actively follow each patient for the same time period; however, neurosurgeons usually follow patients with brain tumors for years or until they die and they follow patients with head trauma for at least three months after their procedures. Data of postoperative LOS, death, readmissions, and reoperation was collected from the patients’ medical records. Postoperative LOS was defined as the number of days the patient stayed in the hospital from the date of procedure to the date of discharge during the patient’s initial admission. In addition, if patient was readmitted for treatment of a complication within the first year after the initial procedure, the hospital days after the procedure during the initial admission and all hospital days during the subsequent admission were added together to get the total postoperative LOS. Postoperative death was defined as any death that occurred within one year after the patient’s initial procedure. Readmission was defined as any readmission related to a complication of the

67 initial procedure within one year after the procedure. Reoperation was defined as any subsequent operation, either during the initial hospitalization or any readmission related to a complication of the initial procedure within the first year after the initial procedure. We did not have access to information about admissions to other hospitals, and thus, this study may underestimate the incidence of some outcomes.

3.3 Power calculation We estimated that 2,500 CRANI procedures were done between 01/01/2006 and 12/31/2010 at the UIHC and that about 95 patients acquired SSIs during this time period. We based all power estimates on a two-sided hypothesis, and a 0.05 Type I error. For a case:control ratio of 1:3 (95 cases and 285 controls), the risk factor study had an 80% power to identify factors of interest that have ORs of at least 1.95. (Table 5) In addition, for the outcome of postoperative LOS, the outcome study had an 80% power to detect a minimal mean difference of 1.9 days when comparing patients with SSIs and patients without SSIs, given that the standard deviation is six days within the control population. The study population included 104 patients with SSIs and 312 controls. Given that the sample size was larger than that used for the power calculations, the current study should have a sufficient power to do the analyses.

68 Table 5. Power estimation for potential risk factors for surgical site infections after craniotomy/ craniectomy.

0.80

Minimal detectable OR 2.09

0.40b

0.80

1.95

Gliadel® wafer use

0.10b

0.80

2.49

Postoperative CSF leakage

0.015a

0.80

5.67

Risk factor No antimicrobial

Prevalence of risk factor in controls 0.21a

Power

prophylaxis Operation duration ≥ 4 hours

Abbreviation: CSF = cerebrospinal fluid; OR = odds ratio. a. The prevalence is based on data from the study by Korinek et al. (Korinek et al., 1997). b. The prevalence is based on data from our preliminary study (Chiang et al., 2011).

3.4 Statistical analyses 3.4.1 Specific Aim 1 The specific aim 1 is to assess the overall trend of the rate of surgical site infections

after craniotomy and craniectomy between January 1, 2006 and December 31, 2010 at the University of Iowa Hospitals and Clinics and to evaluate the effect of limiting use of Gliadel® wafers on surgical site infection rates. -

-

-

69

-

-

-

3.4.1.1 Describe the overall surgical site infection rate and the trend in surgical site infection rates over time

70 -

-

-

-

71 3.4.1.2 Compare surgical site infection rates before and after use of Gliadel® wafers was limited in April, 2009 among patients with malignant brain tumors (intervention group) and patients without malignant brain tumors (control group)

-

-



72

The specific aim 2 is to identify risk factors for surgical site infections and to develop a preoperative surgical site infection risk index for patients undergoing craniotomy and craniectomy procedures between January 1, 2006 and December 31, 2010 at the University of Iowa Hospitals and Clinics. 3.4.2.1 Identify patient-related and procedure-related factors associated with SSIs after CRANI by bivariable analyses

-

-

-

-

-

3.4.2.2 Identify significant independent patient- and procedurerelated risk factors by multivariable logistic regression analyses We built three different multivariable models: a pre-operative model that included only the patient-related factors, an operative model that included only the procedure-

73 related factors, and an overall model that included both patient- and procedure-related factors. To build a multivariable model, first, we put all the factors that were associated with SSIs by the bivariable analyses (P < 0.15) or that had clinical significance into a multivariable logistic regression model. Next, we used three different variable selection methods (i.e., forward, backward, and stepwise) to select variables that had P < 0.15 given that other variables are in the model. We compared the variables that remained in the model after the different methods, and we used the variables in common to build the final model. To control for potential residual confounding effect resulting from inefficient matching, we included the “procedure month period” in the model. Ordinal logistic regression analysis can handle cases and controls that are frequency matched (Szklo et al., 2007). If the P-value of “procedure month period” was more than 0.15, we discarded that variable from the model.

-

Subsequently, we assessed the collinearity among the risk factors in the multivariable model by examining the correlation between predictors. Collinearity is a feature of regression analysis that results when two or more predictor variables (exposure variables) are “near linear” functions of each other. When the predictor variables are

74 collinear, the standard errors of the regression parameter estimates can be very large (inflated) and the P-values for the test of hypothesis can be inaccurate. We computed the Variance Inflation Factor (VIF), a measure for multiple collinearity, for each factor in the multivariable model. VIF for the variable Xj is defined by VIFj =

(j = 1, 2, …, j). Rj2

represents the squared multiple correlation based on regressing Xj on the remaining (j-1) predictors. A rule of thumb for evaluating VIFs is to be concerned with any value larger than 10. If any variable has VIF > 10, we should remove it from the model because that variable does not provide addition information about the dependent variable. We used PROC REG in SAS to compute the VIF for each multivariable model. Let π(SSIi) denote the probability of SSIs for the ith subject. The logistic regression model is:

= β0 + β1(Factor 1)i + β2 (Factor 2)i + β3 (Factor 3)i + … + βk (Factor n)i

3.4.2.3 Develop a preoperative risk index consisting of independent patient-related factors associated with SSIs after CRANI On the basis of the preoperative multivariable model, we simplified the model by omitting the predictor variables that were not statistically or clinically significant and by multiplying regression coefficients by 10 and rounding to create the risk score. This simplified risk index model was the preoperative risk index model. We assessed the predictive power of the preoperative risk index by the c-statistic, which was computed as the area under the receiver-operating characteristic (ROC) curve. The ROC curve is the true positive rate (sensitivity) plotted against the false positive rate (1-specificity), which shows the tradeoff between sensitivity and specificity. The closer the curve follows the left-hand border and the top border of the ROC space, the better the model

75 is. Thus, the closer the area under curve (c-statistic or AUC) is to one, the better the model is, and the closer the c-statistic is to 0.5, the worse the model is. The general guideline to determine the predictive power is: c = 0.97 to 1.00 is excellent, c = 0.92 to 0.97 is very good, c = 0.75 to 0.92 is good, and c = 0.50 to 0.75 is fair. In addition, because the assessment of a model can be optimistically biased if the data used to fit the model are also used to assess the model, we used the crossvalidation method to compute the unbiased c-statistic. Assessment via crossvalidation is done by fitting the model to the complete dataset and using the crossvalidated predicted probabilities to provide an ROC analysis. The crossvalidated predicted probability for an observation simulates the process of fitting the model ignoring the observation and then using the model fit to the remaining observations to compute the predicted probability for the observation that is ignored (i.e., leave-one-out crossvalidation) (SAS Usage Note 39724, 2012). In addition, we assessed the model fit with the Hosmer-Lemeshow goodness-of-fit test. The null hypothesis for Hosmer-Lemeshow test is that the model provides a good fit. Therefore, a P-value more than 0.05 indicates reasonable model fit. Finally, we compared the predictability and model fit of the preoperative risk index with that of the NHSN risk index. Because the new risk index included only risk factors significantly associated with SSIs after CRANI, we hypothesized that this preoperative risk index would predict SSIs better and would fit the data better than the NHSN risk index did.

76 3.4.3 Specific aim 3 The specific aim 3 is to evaluate patient outcomes associated with surgical site infections among patients who underwent craniotomy and craniectomy between

January 1, 2006 and June 30, 2010 at the University of Iowa Hospitals and Clinics. 3.4.3.1 Identify postoperative outcomes associated with SSIs after CRANI by bivariable analyses For the outcomes of postoperative LOS during the initial admission and the total postoperative LOS during both the initial hospitalization and any readmission, we analyzed their association with SSIs after CRANI using the t-test or the Wilcoxon rank sum test. For the outcomes of postoperative death, readmission, and reoperation, we tested their association with SSIs after CRANI by the Chi square test. 3.4.3.2 Compare the Kaplan-Meier estimates of the survival functions between patients with SSIs and controls for postoperative outcomes

77

-

3.4.3.3 Assess the association between SSIs and postoperative outcomes by multivariable regression analyses

-

For postoperative death, readmission, and reoperation, we used logistic regression to analyze their association with SSIs. For readmission, we evaluated patients who survived to be discharged at the end of their initial hospitalization, because only patients who survived their initial hospitalization can be readmitted. For death and reoperation, we evaluated all patients. We created separate multivariable regression models for each outcome (Table 6). If π(Outcomei) denotes the probability of the postoperative outcome for the ith subject, then the logistic model is:

78

3.4.3.4 Assess the association between SSIs and postoperative outcomes by Cox proportional hazard analyses We used Cox proportional hazard regression models to assess the association between SSIs and postoperative LOS, death, readmission, and reoperation. A Cox proportional hazard model is a type of survival analysis that can take into account the time to the event of interest (e.g., death, reoperation, discharge, and readmission), censoring, and the time-varying variable. Using this method, we can compute a hazard ratio (HR) which is similar to a relative risk ratio for the outcomes of interest. We evaluated all patients in the study for each outcome except readmission. For readmission, we evaluated only patients who survived long enough to be discharged at the end of their initial admission. For postoperative death, we treated patients who survived one year after the initial procedure as censored observations. For other postoperative outcomes, we treated patients who died before the occurrence of other outcomes or patients who did not have any outcomes of interest within one year after the initial procedure as censored observations. In addition, we made the binary variable “SSI” a time-varying variable in the Cox proportional hazard model because patient’s status of “SSI” changed over time. Patients did not have SSI at the time of their procedures, but they acquired SSIs after their procedures. For patients with SSIs, the status of time-varying SSI equals to 0 before the

79 onset of SSIs and it equals to 1 after the onset of SSIs. We created separate multivariable Cox regression models for each outcome of interest (Table 6). Let h(t)i be the hazard function of a postoperative outcome for the ith subject, and h0(t) represents the baseline hazard. The Cox proportional hazard model was: h(t)i = h0(t) exp {β1 (Time-varying SSI)i + β2 (Preoperative LOS)i + β3 (Age)i + β4 (Severity of illness)i + β5 (Comorbidities)i + β6 (Reason for procedure)i + β7 (Procedure period)i i} We handled the tied event times (i.e., multiple events that occurred at same event time) with Efron’s method. Hazard ratio (HR) and its 95% CI were calculated for each variable. The proportional hazards model assumes that the effect of a variable is constant over time. We tested the proportional hazard assumption for each variable in the model by assessing the interaction of the variable with a function of time. Table 6 describes the models we created to analyze the postoperative outcomes. Different statistical methods may generate different results for the same outcome. If the results of different models conflicted (e.g., OR > 1 in the logistic analysis but HR < 1 in the Cox regression analysis), we chose the most plausible result.

80

Method

Linear regression

Logistic regression

Survived the initial stay

Survived the initial stay

Population

Outcome

All

Postop LOS

Model 1

Total Postop LOS

Model 3

Death

All

All

Survived the initial stay

Model 2

Model 4

Readmission Reoperation

Cox proportional hazard

Model 5 Model 6

Model 8

Model 7 Model 9

Abbreviation: LOS = length of stay.

3.5 Study implementation and quality control 3.5.1 Study Implementation We assembled a research team with expertise in hospital epidemiology, SSI surveillance, study design, and statistical analysis and we collaborated closely with the neurosurgeons who provided their clinical expertise. The Institutional Review Board in the University of Iowa approved the study protocol and the data abstraction form. 3.5.2 Data Collection Hsiu-yin Chiang, who gained considerable experience in data abstraction while reviewing medical records for the previous study, abstracted all of the data for the

81 current study. Before data collection began, the research team reviewed the data abstraction form, and Ms. Chiang tested the form by abstracting data from the medical records of 10 patients. Thus, we knew that data collection using the data abstraction form was feasible. Ms. Chiang entered the data directly into the database constructed in Microsoft Access. She setup data format validation in Access to ensure that all data would be entered in the correct format. After she completed data entry, she validated the data for a 10% random sample of the study population to ensure that the data were abstracted and entered correctly. After validating the data abstraction and entry, she cleaned the data before doing the data analysis. She corrected incomplete, missing, or erroneous data after reviewing the patient medical record again. To protect confidentiality, the database was stored in a password protected computer in a locked room. We assigned a unique study identification (ID) number, which has no meaning external to the study database, for each case and control, and assigned a unique surgeon ID number for each surgeon. We did not collect any personal identifiers.

82 CHAPTER 4 – RESULTS

4.1 Specific aim 1 4.1.1 Gliadel® wafer use During the study period, Gliadel® wafer implants were implanted during 86 procedures; of which only three procedures were implanted after April, 2009. (Figure 5) Among the tumor group, Gliadel wafers were implanted in 20.2% (83/411) of the procedures before April, 2009 and 1.6% (3/186) of the procedures after that (Fisher’s exact test, P = 0.0001). Therefore, neurosurgeons decreased their use of Gliadel® wafers significantly after the intervention in April, 2009. 4.1.2 Overall surgical site infection rate and trend From January 1, 2006 through December 31, 2010, the total number of craniotomy and craniectomy (CRANI) procedures included in the study was 2,865 procedures; 597 procedures were done to treat patients with malignant brain tumor (i.e., tumor group) and 2,268 procedures were done to treat patients with other medical conditions (i.e., non-tumor group). Among 104 surgical site infections (SSIs) detected during the study period, 36 occurred in the tumor group and 68 occurred in the non-tumor group. For all procedures, the SSI rates by month ranged from 0% to 13 %, and the overall SSI rate was 3.6% (104/2865). The overall SSI rate was 6.0% (36/597) for tumor group and 3.0% (68/2268) for non-tumor group. Figure 6a shows the SSI rates by month for all procedures, and Figure 6B shows the rates for tumor and non-tumor group.

83 4.1.2.1 Regression analysis We used the two regression models described below to assess the trends in SSI rates for all procedures.

-

-

-

4.1.2.2 Time-series analysis SSI rates by month for 60 months are time-series data. Thus, we first performed a time-series analysis to check if autocorrelation existed between the series of SSI rates sequenced in time. The unit of analysis was the number of SSIs per 100 CRANI procedures (i.e., SSI rate, %) by month. Because the time series analysis assumed that the data were normally distributed, we added 1% to each monthly rate and

84 logarithmically (log) transformed the data to make it non-zero and approximately normally distributed. We utilized the computer programming language R v2.10.1 and the time series analysis (TSA) package to perform time series analyses. We tried to identify any potential autoregressive moving average (ARMA) autocorrelation model among the SSI rates. An autocorrelation function (ACF) described how a specific month’s SSI rate was related to the previous month’s SSI rate. Serial dependence was assessed by calculating the autocorrelations between observations (SSI rate) separated by different time interval (one month) or lags in the series. A lag-1 autocorrelation was computed by paring the initial observation with the second observation, the second with the third observation, and so on until the second from the last is paired with the last observation. Another function, partial ACF (PACF) described the memory of the data series by measuring correlations between observations that are k time periods apart after controlling for correlations at intermediate lags. The results of model identification for autocorrelation function were shown in Figure 7. We identified moving-average (MA) model by the ACFs of the series, and identified autoregressive (AR) model by the PACFs of the series. ACFs after lag q (q, q+1, and higher) that are not significantly different from zero would indicate a MA(q) model, and PACFs after lag p (p, p+1, and higher) that are not significantly different from zero would indicate a AR(p) model. Figure 7 shows the results of different autocorrelation functions used to identity ARMA model. The blue dash line in the graph of ACF (or PACF) indicates

85 the 95% confidence interval (CI) of ACF (or PACF). ACFs (or PACFs) for lag k that exceed the 95% CI are considered to be significantly different from zero. We could not identify any MA or AR model because no ACF or PACF was significant (i.e., none exceeded the 95% CI). Moreover, to identify potential mixed ARMA models, we used both the best subset method and the extended autocorrelation function (EACF) method. We used the best subset method to calculate the Bayesian information criterion (BIC) for different ARMA model. (Figure 7 Best subset) Each row in the best subset table corresponds to a subset ARMA model. The best model (lowest BIC) is in the top row and the selected variables are indicated with darker shading (Cryer et al., 2010). Thus, the best subset method indicated that a MA(5) model that included only lag 5 of the error process would be the best, because it had the smallest BIC. The EACF table summarizes the EACF for different ARMA models using zero and “X”, where “X” represents EACFs that are significantly different from zero. The upper left-hand vertex of the triangle of zeros indicates the ARMA model. In this study, the vertex of the triangle of zeros was located at ARMA (0, 0), which did not support a significant ARMA model. Since MA(5) was the only model that was identified, we fit the data series with the MA(5) model that included only lag5 of the error process. However, the parameter estimate for lag 5 error was 0.229 (95% CI [-0.065, 0.523]), which was not significantly different from zero. On the basis of these results, we concluded that the data series of SSI rates did not correlate with time because a significant ARMA autocorrelation model could not be specified. Although the SSI rate by month was essentially time-series data, the SSI rate did not necessarily correlate with time. Thus, this data series could be treated as an

86 independent and identically distributed data, in which time-series analysis method was not necessary. Thus, regression analysis such as a linear regression or a Poisson regression would be appropriate analytic approaches, and we did not use time-series analysis for the subsequent analyses for SSI rates. 4.1.3 Pre- and post-intervention

-

-

-

4.1.3.1 Two-group test -

-

-

-

-

-

-

-

87 -

-

-

4.1.3.2 Regression analyses

-

88

-

-

-

-

-

89

-

-

-

-

-

Standard Poisson regression: Poisson regression models should be more appropriate than linear regression models when estimating the association between the intervention and monthly SSI rates, because SSI rates, which are counts of SSIs divided by the number of procedure, are not normally distributed. Poisson regression does not require the distribution to be normally distributed. The model for standard Poisson regression was:

-

90

-

Before the intervention, the increasing SSI rate for all procedures and the decreasing SSI rate for the tumor group were not significant, but the increasing rate for the nontumor group was significant (0.03; P = 0.04; Table 10). The SSI rates for all procedures, the tumor group, and the non-tumor group decreased non-significantly after the intervention. The decrease in the SSI rate for the non-tumor group was close to significance level (P = 0.06; Table 10). Figure 9 depicts the fitted model with dashed red lines and the monthly observed SSI rates with black lines. Poisson regression analysis is commonly used in public health to model the number of events (e.g., SSIs) or rate (e.g., number of SSIs per procedure). However, the Poisson model assumes the variance is equal to the mean, an assumption that is often violated. The standard error estimates would be affected when the assumption is violated, which may result in false inferences. Overdispersion occurs when the variance is greater than the mean and it is commonly due to excess zeros (Rose et al., 2006). In this study, the count of SSI for all procedures was zero for 13 of 60 months. This assumption can be relaxed by fitting a zero-inflated Poisson model. To control for the potential overdispersion problem, we also fit the data with zero-inflated Poisson. However, the parameter estimates and P-values were the same as those for the Poisson regression model (data not shown).

91 4.1.4 Conclusions During the entire study period, SSI rates after all procedures, in the tumor group, and in non-tumor group did not change significantly. Similarly, the SSI rates did not change significantly from the pre-intervention period to the post-intervention period, which suggested that limiting Gliadel® wafer use did not decrease SSI rates in these groups. However, the SSI rate decreased somewhat (P = 0.06) in the non-tumor group after Gliadel® wafer use was limited. These results are the opposite of our hypothesis that the SSI rate would decrease only in the tumor group.

4.2 Specific aim 2 4.2.1 Descriptive epidemiology of surgical site infections Between January 1, 2006 and December 31, 2010, 104 patients who underwent CRANI procedures acquired SSIs (cases) at the University of Iowa Hospitals and Clinics (UIHC), and 312 patients without SSIs were selected as controls. The common symptoms of SSIs after CRANI included: purulent drainage from the incisional site (52.9%), temperature higher than 38°C (41.4%), localized swelling, redness, or heat (29.8%), and incisional pain or tenderness (21.1%). Cultures were positive for 83.6% of these SSIs (Table 11). The most common organisms causing SSIs were Staphylococcus aureus (S. aureus) alone or in combination of other organisms (33; 31.7%) including methicillin-resistant S. aureus (MRSA; 10; 9.6%), coagulase-negative staphylococci (CoNS) alone or in combination of other organisms (20; 19.2%), and Propionibacterium acnes (P. acnes) alone or in combination of other organisms (14;

92 13.5%). Fifty-five (52.9%) SSIs were caused by Gram-positive bacteria alone, seventeen (16.3%) were caused by Gram-negative bacteria alone, eleven (10.6%) were caused by mixed flora of Gram-positive and Gram-negative organisms, and two were caused by mixed flora of unknown species. Cultures were negative for 14 (13.5%) of the SSIs. The infected wounds of three patients were not cultured. Among the 104 patients with SSIs, fifty-four patients (51.9%) had organ/space infections (30 meningitis, 13 epidural infections, five osteomyelitis, and six unspecified infections), thirty-eight patients (36.5%) had deep incisional infections, and 12 patients (11.5%) had superficial incisional infections. Among 38 deep incisional infections, seventeen (44.7%) were caused by S. aureus alone or in combination of other organisms, twenty-seven (71.1%) were caused by at least one Gram-positive organism, and 13 (34.2%) were caused by at least one Gram-negative organism; among 54 organ/space infections, fourteen (25.9%) were caused by S. aureus or in combination of other organisms, thirty-five (64.8%) were caused by Gram-positive organisms, and 14 (25.9%) were caused by at least one Gram-negative organisms. Seventy percent of infections were identified after the patients were discharged; 60% were identified during readmissions and 10% during postoperative visits (e.g., clinic visits or emergency room visits). The time to the onset of SSIs ranged from 2 to 221 days after CRANI, with a mean of 37.1 ± 46.6 days. Of 32 cases without implants, half acquired SSIs within 20 days after their CRANI procedures; of 72 cases with implants, half acquired SSIs within 30 days after their procedures (Figure 10). However, by NHSN’s criteria, surveillance for SSIs among patients without implants is terminated at 30 days

93 after the procedures but surveillance for SSIs among patients with implants is terminated at one year after the procedures. The mean time to the onset was 43.2 ± 50.4 days for SSIs caused by Gram-positive organisms alone, 26.7 ± 35.6 days for SSIs caused by Gram-negative organisms alone, and 54.6 ± 62.1 days for the infections caused by both Gram-positive and Gram-negative organisms. 4.2.2 Risk Factors for Surgical Site Infections 4.2.2.1 Bivariable analyses Bivariable associations between patient-related factors (e.g., patients’ demographic and clinical characteristics) and SSIs after CRANI are shown in Table 12. Patients with SSIs and controls were similar with respect to: age, gender, body mass index (BMI), smoking history, insurance status, diabetes, number of Elixhauser comorbidities, American Society of Anesthesiologists (ASA) score > 2, and emergent or urgent procedures (i.e., procedures within four to 24 hours after admission). Compared with controls, cases were more likely to: stay in the hospital longer before their procedures, have higher preoperative glucose levels, have a Charlson comorbidity score ≥ 2, have had a previous brain operation, have had previous brain in radiation within 30 days before procedures, and received chemotherapy within 30 days before their CRANI. Patients who underwent CRANI to treat brain tumors were more likely to acquire SSIs than were patients who underwent CRANI for other medical conditions (e.g., aneurysm, trauma, or epilepsy). Bivariable associations between procedure-related factors (i.e., surgical techniques and other operative variable such as skin preparation, antimicrobial prophylaxis, or

94 operating room) and SSIs after CRANI are shown in Table 13. Naficillin (55%) and vancomycin (31%) were the antimicrobial agents used most commonly for perioperative prophylaxis. Cases and controls were similar with respect to: type of skin preparation (most patients’ skin was prepared with either povidone-iodine gel or solution), using nafcillin or vancomycin as antimicrobial prophylaxis, the operation path (i.e., crossed a sinus or did not cross a sinus), operation duration, perioperative transfusions, subsequent brain operations, postoperative infection at other site, and postoperative glucose levels. Compared with controls, patients with SSIs were more likely to have a National Healthcare Safety Network (NHSN) risk index ≥ 2, to have ventricular or lumbar drains placed perioperatively (P = 0.07), to have their bone flaps rigid-plated (e.g., using plates and screws to fix their bone flaps), to receive Gliadel® wafer implants, and to have postoperative cerebrospinal fluid (CSF) leakage. Among patients with SSIs in our study, the mean of days to the onset of SSIs was 25.6 ± 40.5 days for patients who had ventricular or lumbar drains, 36.6 ± 58.5 days for patients who had other drains, and 43.0 ± 43.9 days for patients without drains (ANOVA test, P = 0.28). We suspected that patients who acquired SSIs long after the initial CRANI were unlikely to be affected by procedure-related factors. Thus, we did a subgroup analysis for which we excluded patients who acquired SSIs more than 90 days after their CRANI and then assessed the bivariable association between SSIs and procedure-related factors among the 93 patients who acquired SSIs within 90 days after their CRANI. The results from the subgroup analysis were similar to the results from the whole group analysis. Compared with controls, patients with SSIs were more likely to have NHSN risk

95 index ≥ 2 (OR = 1.8; P = 0.04), ventricular or lumbar drains placed (P = 0.09), rigid-plated bone flaps (OR = 1.6; P = 0.06), Gliadel® wafer implants (OR = 7.1; P < 0.0001), and postoperative CSF leakage (OR = 3.7; P = 0.002). The bivariable analyses revealed that having a brain tumor was strongly associated with SSIs after CRANI. Because many risk factors were related to brain tumors, such as preoperative brain radiation, previous chemotherapy, and having Gliadel® wafer implants, we assessed the bivariable associations between some risk factors for SSIs and brain tumor as the reason for procedure (Table 14). Preoperative length of stay (LOS) ≥ 1 day, Charlson comorbidity score ≥ 2, longer operation duration, NHSN risk index score ≥ 2, having bone flaps rigid-plated, and postoperative CSF leakage were associated with brain tumor as the reason for the procedure. Therefore, the reason for the procedure was a potential effect modifier for the association between SSIs and other risk factors. We tried to control for the effect of the reason for the procedure by including this variable in the multivariable models and we also tested its interaction with other risk factors (discussed in the following section). 4.2.2.2 Multivariable analyses Independent risk factors for SSIs were identified by using multivariable logistic regression. First, we built a preoperative model that included only patient-related factors that identified preoperative steroid use (protective), preoperative LOS ≥ 1 day, preoperative glucose level ≥ 100 mg/dL, preoperative chemotherapy, and the reason for the procedure as independent patient-related risk factors for SSIs. However, preoperative steroid use and preoperative glucose level were correlated with each

96 other (Pearson correlation coefficient 0.116; P = 0.02) and, thus, only one of these variables should be included in the model. In the final preoperative model, we included the preoperative glucose level rather than the steroid use because this model had a better c-statistic (Table 15). We included the cut-point of 100 mg/dL for preoperative glucose level because it had the highest odds ratio (1.7; 95% CI [1.0, 2.9]) and the lowest P-value (P = 0.03) among the three cut-points of 100, 120, and 150 mg/dL we tested in the bivariable analysis. The final preoperative model had a c-statistic of 0.665 and a cross-validated c-statistic of 0.608, which indicated fair predictive power. The model provided a good fit according to the Goodness-of-fit test with P-value of 0.91. Second, we built an operative model that included only the procedure-related factors. After the initial variable selection, the variable “Surgeon No. 10” remained in the model because it was associated with a decreased risk of SSIs. Compared with other neurosurgeons, surgeon No. 10 was less likely to operate on patients with older age (51.6 ± 21.0 vs 20.9 ± 16.1 years old; P < 0.0001), patients with brain tumors (47.5% vs 30.8%; P = 0.05), patients undergoing emergent or urgent procedures (43.5% vs 0%; P < 0.0001), or patients whose procedures went through a sinus (7.4% vs 0%; P = 0.09). However, we decided not to include “Surgeon No. 10” in the final operative model for the following reasons. First, different surgeons could operate on patients with different clinical conditions. Therefore, the variable “surgeon” could be a surrogate indicator of patients’ underlying conditions and not an indicator of surgeon’s technique. Second, the variable “surgeon” is not equivalent for patients admitted to different hospitals because different surgeons practice at different hospitals. This variable is also not equivalent for

97 patients admitted to a single hospital at different times because surgeons may leave and new surgeons may join the staff during the study period. If we want to include the variable “surgeon” in the model, we should treat “surgeon” as a random effect so that we can model the correlation among a cluster of patients whose operations are performed by the same surgeon. However, the modeling becomes complicated because more than one surgeon (e.g., attending surgeon, resident surgeon) operates on a single patient in an academic center. Therefore, we decided to exclude “Surgeon No. 10” when building the final operative model and the overall model. In addition, we included ventricular or lumbar drains in the model because the drains could be clinically significant. In the final operative model (Table 16), independent procedure-related risk factors included Gliadel® wafer implants and postoperative CSF leakage. Ventricular or lumbar drains and postoperative glucose levels ≥ 100 mg/dL were associated with higher risk of SSIs, but the associations were not statistically significant. We included the cut-point of 100 mg/dL for postoperative glucose level because it had the highest odds ratio (OR = 2.9; 95% CI [0.9, 9.9]) and the lowest P-value (P = 0.07) among the three cut-points of 100, 120, and 150 mg/dL that we tested in the bivariable analysis. The final operative model had a c-statistic of 0.675 and a cross-validated c-statistic of 0.612, which were similar to those for the preoperative model. The model provided a good fit according to the Goodness-of-fit test, with P-value of 0.96. Gliadel wafer was such a strong risk factor (OR = 8.6; P < 0.0001), we wondered if we had ignored other important risk factors whose association with SSIs were masked by

98 including the variable “Gliadel wafer” in the model. When we excluded Gliadel wafer during the variable selection, a NHSN risk index score ≥ 2, , ventricular or lumbar drains, and postoperative CSF leakage remained in the model and the variable “rigid-plated bone flap” became significant with an OR of 1.9 (95% CI [1.2, 3.0]; P = 0.01). Finally, we included both patient-related factors and procedure-related factors in an overall model. The selection method initially included preoperative steroid use and did not include the reason for the procedure in the model. However, we removed the preoperative steroid use to avoid its collinearity with preoperative glucose level, and we added the reason for the procedure in the model to control for the potential confounding effect caused by tumor as the reason for the procedure. Independent risk factors included preoperative glucose level ≥ 100 mg/dL, preoperative length of stay (LOS) ≥ 1 day, Gliadel® wafer implants, and postoperative CSF leakage (Table 17). BMI ≥ 30 kg/m2, peripheral vascular disease, preoperative chemotherapy, and operation duration more than 225 minutes marginally increased the risk of SSIs. Preoperative chemotherapy and Gliadel® wafer implants increased the risk of SSIs even when the analysis was controlled for the reason for the procedure. The overall model had the highest predictive power of three multivariable models, with a c-statistic of 0.732 and a cross-validated c-statistic of 0.681. The model provided a good fit according to the Goodness-of-fit test, with P-value of 0.49. For each model, we tested for interaction between the reason for the procedure and other variables, such as BMI ≥ 30, preoperative LOS, preoperative glucose level, operation duration, and postoperative CSF leakage, in separate logistic regression

99 models. However, none of the interaction terms were significant. Thus, we did not include interaction terms in the final multivariable models. We also checked for multicollinearity among variables by assessing the Variance Inflation Factor (VIF) for each model. None of the variables had a VIF > 10, suggesting that none of the models were affected by multicollinearity. Therefore, we did not delete any variables from the three multivariable models. 4.2.3 Surgical site infection risk index We developed a SSI risk index that included the variables in the final preoperative model: previous chemotherapy, preoperative glucose level ≥ 100 mg/dL, preoperative LOS ≥ 1 day, and the reason for the procedure. We simplified the model by replacing the reason for the procedure with a dichotomous variable - tumor reason or non-tumor reason. The simplified preoperative model had a c-statistic of 0.641 (Table 18a). We assigned risk points to each factor by multiplying the parameter estimates from the logistic regression model by 10 and rounding to the nearest integer. The SSI risk index score was the sum of the risk points from each factor in the preoperative model. We categorized patients into three risk classes (preoperative risk classes) based on the c-statistic associated with different SSI risk index scores (Table 18b). We calculated predicted probability of SSIs based on the parameter estimates from the logistic regression model that included the preoperative risk class as a predictor. Preoperative risk class 1 included patients with risk index scores < 8, class 2 included patients with scores 8-13, and class 3 included patients with scores ≥ 14. When this preoperative risk class was included as a predictor in the logistic regression model, the c-statistic was

100 0.637, which is slightly lower than the c-statistic from the simplified preoperative model. In the current study, 62% of patients with SSIs and 39% of controls were in preoperative risk class 2 or 3 (high risk group) (Table 18b). We could use the preoperative risk index score to categorize a preoperative patient into a specific risk class and then determine that patient’s priori risk of an SSI based on the risk class. For example, if a patient had a preoperative glucose level of 150 mg/dL and he/she stayed in the hospital for more than one day before the CRANI procedure, the patient would have risk index score of 11 and would be in preoperative risk class 2. Thus, this patient would have a probability of acquiring an SSI of at least a 41.5%. The NHSN risk index is used to stratify surgical patients into four risk groups (0, 1, 2, 3) by their risk of acquiring SSIs. The NHSN risk index includes three factors: ASA score > 2, operation duration > T (T represents the 75th percentile of the distribution based on hospitals reporting to NHSN; 225 minutes for CRANI), and a wound classified as contaminated or dirty/infected (Edwards et al., 2009). NHSN considered the low risk group to be patients in risk group 0 or 1 and the high risk group to be patients in risk group 2 or 3. In the current study, the proportion of cases and controls with ASA scores > 2 (61.5% vs 57.4%; P = 0.46) and operation durations > 225 minutes (40.4% vs 32.7%; P = 0.15) were similar, and none of the patients had contaminated or dirty/infected wounds. Therefore, we expected that the NHSN risk index would not stratify this patient population well. Indeed, the logistic regression model that included the NHSN risk index as a predictor had a low c-statistic (0.549), indicating that the NHSN risk index score had no predictive power (Table 19). Moreover, in our study, seventy-six percent of patients

101 with SSIs and 85% of controls were in NHSN’s low risk group, while only 24% of the patients with SSIs and 15% of controls were in NHSN’s high risk group (Table 19). Thus, the preoperative SSI risk index that included patient-related factors specific for SSIs after CRANI could predict and stratify the risk of SSIs for patients undergoing CRANI procedures better than the NHSN risk index. 4.2.4 Conclusions Most patient-related risk factors for SSIs after CRANI (e.g., tumor as the reason for the procedure, preoperative chemotherapy) were non-modifiable, but surgeons could use them to identify patients at high-risk of acquiring SSIs. Surgical staff might be able to modify or improve a patient-related factor like high glucose levels before CRANI and some procedure-related risk factors (e.g., ventricular drains or lumbar drains management, CSF leakage) to lower the risk of SSIs. Moreover, surgeons could use the preoperative SSI risk index score, which we based on the preoperative model, to predict a patient’s risk of SSIs after CRANI and infection preventionists could use risk index scores to stratify SSI rates after CRANI procedures by the patients’ intrinsic risk of SSIs.

4.3 Specific aim 3 4.3.1 Bivariable associations between surgical site infections and outcomes The study population included 93 patients with SSIs and 279 controls who underwent CRANI procedures during January 1, 2006 and June 30, 2010. The time period was six months shorter than that for the SSI risk factor study because we have had enough sample size to evaluate the postoperative outcomes and we evaluate the severity of

102 illness score only for the patients undergoing CRANI during January 1, 2006 and June 30, 2010. Patients with SSIs stayed in the hospital for a mean of 14.4 ± 12.9 days after they acquired SSIs during their initial admission or during readmissions. The bivariable associations between SSIs and postoperative outcomes are shown in Table 20. Patients with SSIs were more likely than controls to have antimicrobial agents on more days after CRANI during their initial hospitalizations (excluded the first two days after procedure), longer hospital stay during either the initial hospitalizations or during readmissions, to die within one year after their procedures, to be readmitted to the UIHC within one year after their procedures, and to have reoperations to treat complications within one year after their procedures. To determine the postoperative outcomes attributed by SSIs, one must adjust for patients’ preoperative conditions because these conditions could also contribute to patients’ postoperative outcomes. The prior bivariable analyses unveil that, compared with controls, patients who had SSIs stayed in the hospital longer before their CRANI procedures (1.0 ± 2.7 days vs. 2.1 ± 4.2 days; P = 0.02), had higher Charlson comorbidity scores (41.6% vs. 54.8% had score ≥ 2; P = 0.03), and were more likely to have brain tumors (41.9% vs. 57.0%; P = 0.01). Older age (48.4 ± 23.1 years old for controls vs. 50.7 ± 20.4 years old for cases; P = 0.39) and worse McCabe and Jackson (MJ) severity of illness scores (47.0% of controls vs. 47.3% of cases were ultimately fatal or rapidly fatal; P = 0.95) were not significantly associated with SSIs, these variables could be clinically important. Thus, we included all of these covariates in the multivariable regression models for each outcome variable. In addition, to control for unmeasured factors (e.g.,

103 changes in clinical practice) during each procedure period, we included procedure month period in the models. 4.3.2 Postoperative length of stays 4.3.2.1 Kaplan-Meier survival analyses The mean postoperative LOS during the initial hospitalization was significantly longer for cases than for controls (12.2 days vs. 7.9 days), and the mean postoperative LOS for readmissions was also significantly longer for cases than for controls (9.0 days vs. 2.6 days). To do a survival analysis for outcomes, one must define both the time to the event of interest and the patients whose data are censored. For postoperative LOS, the time to the event of interest was the time from the CRANI procedure to discharge. Patients who died during the initial admission and patients who were not discharged within one year after their procedures (no patients in this study met the second censored criterion) were treated as censored observations. The Kaplan-Meier survival curves demonstrated that about 25% of patients with SSIs and 15% of controls remained in the hospital 20 days after their CRANI procedures during the initial admissions (Figure 11). The two survival curves were significantly different (log-rank test, P = 0.04). 4.3.2.2 Multivariable analyses

-

104

-

-

105 Cox proportional hazard analysis: Cox proportional hazard model is a multivariable model that can take into account of time to the event of interest, censoring, and variables that vary over time (i.e., time-varying). We also analyzed the postoperative LOS during the initial hospitalizations with a Cox proportional hazard model. SSI was treated as a time-varying variable in the Cox model for postoperative LOS because patients would not have infections at the time of their procedures. Therefore, patients’ SSI status changed over time. When we treated SSI as a time-varying variable and controlled for covariates with the Cox model, SSI was no longer associated with postoperative LOS (Table 22). 4.3.3 Postoperative deaths 4.3.3.1 Kaplan-Meier survival analyses Twenty-six percent of patients with SSIs and 14% of controls died within one year after their CRANI procedures. The survival curves for cases and controls were significantly different (P = 0.02) (Figure 12). The two survival curves crossed at about the 100th day after the procedures. Before the 100th day, patients with SSIs died later than controls; after the 100th day, cases died earlier than controls. Controls were more likely than cases to have acute medical conditions (e.g., 34% of controls and 24% of cases had head trauma or intracranial bleeding). These acute medical conditions may have caused the early deaths among the controls.

106 4.3.3.2 Multivariable analyses We used logistic regression and Cox proportional hazard method to analyze the association between SSIs and postoperative death. The results of risk ratio (OR and HR) from both methods were similar. SSI was associated with a significant 3-fold increase in the risk of postoperative death when assessed with logistic regression, when considered SSI to be a time-varying variable and assessed by Cox proportional hazard, SSI was associated with a significant 3.3-fold increase in the risk of death (Table 23). In addition, for every 10-year increase in age, patients’ risk of dying within one year of the CRANI increased 50% (logistic regression) or 40% (Cox proportional analysis). Patients with worse comorbid conditions (higher Charlson comorbidity scores) or more severe underlying disease (lower MJ scores) were more likely to die within one year after their CRANI procedures than patients who had better scores. 4.3.4 Postoperative readmissions 4.3.4.1 Kaplan-Meier survival analyses Patients who died during their initial hospitalization could not be readmitted, thus, we included 93 patients with SSIs and 261 controls who survived their initial hospitalizations in survival analyses for readmissions. Seventy-six percent of cases and 30% of controls were readmitted within one year postoperatively (Table 20). The survival curves revealed that about 50% of patients with SSIs and 15% of controls were readmitted to the UIHC within 50 days after their procedures (Figure 13). The two survival curves were significantly different (P

Smile Life

When life gives you a hundred reasons to cry, show life that you have a thousand reasons to smile

Get in touch

© Copyright 2015 - 2024 PDFFOX.COM - All rights reserved.