Evaluation of antimicrobial use in a pediatric intensive care unit

Transcription

1 University of Iowa Iowa Research Online Theses and Dissertations Summer 2009 Evaluation of antimicrobial use in a pediatric intensive care unit Josiah Olusegun Alamu University of Iowa Copyright 2009 Josiah Olusegun Alamu This dissertation is available at Iowa Research Online: Recommended Citation Alamu, Josiah Olusegun. "Evaluation of antimicrobial use in a pediatric intensive care unit." PhD (Doctor of Philosophy) thesis, University of Iowa, Follow this and additional works at: Part of the Clinical Epidemiology Commons

2 EVALUATION OF ANTIMICROBIAL USE IN A PEDIATRIC INTENSIVE CARE UNIT by Josiah Olusegun Alamu 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 2009 Thesis Supervisor: Professor Loreen Adele Herwaldt

3 1 ABSTRACT A pediatric intensivist in the University of Iowa Hospitals and Clinic s (UIHC) Pediatric Intensive Care Unit (PICU) was concerned about antimicrobial use in the unit. However, no one had quantified antimicrobial use in the UIHC s PICU or described the patterns of antimicrobial use in this unit. To address the intensivist s concern, the principal investigator (PI) conducted a retrospective study to determine the percentage of patients who received antimicrobial treatments, to determine the indications for antimicrobial use, and to identify antimicrobial agents used most frequently in the unit. On basis of our data, we hypothesized that empiric antimicrobial use, particularly the duration of therapy, could be decreased. We implemented a six-month intervention during which we asked the pediatric intensivists to complete an antimicrobial assessment form (AA) to document their rationale for starting antimicrobial treatments. We postulated that this documentation process might remind physicians to review antimicrobial therapies, especially empiric therapies, when the microbiologic data became available. In addition, we utilized the AA form to identify factors pediatric intensivists considered when deciding to prescribe empiric antimicrobial treatments. Data from the AA forms suggested that pediatric intensivists in the UIHC s PICU often considered elevated C-reactive protein, elevated white blood cell counts, and elevated temperatures when deciding to start empiric antimicrobial therapy. Data from the three nested periods showed that the median duration of empiric and targeted treatments decreased during the intervention and remained stable during the postintervention period. The PI estimated that 193 days of empiric antimicrobial therapy and 59 days of targeted antimicrobial therapy, respectively, may have been saved by the decreased durations of therapy. Time series analysis assessing the trend in use of piperacillin-tazobactam, cefepime, and ceftriaxone (measured in mg/wk) did not reveal a significant change over time.

4 2 On the basis of our results, an intervention strategy using an AA form alone may not be an effective strategy for antimicrobial stewardship in PICUs. Additional measures such as automatic stop orders and computer decision support may be useful for reducing the duration of empiric therapy in PICUs. Abstract Approved: Thesis Supervisor Title and Department Date

5 EVALUATION OF ANTIMICROBIAL USE IN A PEDIATRIC INTENSIVE CARE UNIT by Josiah Olusegun Alamu 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 2009 Thesis Supervisor: Professor Loreen Adele Herwaldt

6 Graduate College The University of Iowa Iowa City, Iowa CERTIFICATE OF APPROVAL PH.D. THESIS This to certify that the Ph.D. thesis of Josiah Olusegun Alamu has been approved by the Examining Committee for the thesis requirement for the Doctor of Philosophy degree in Epidemiology at the July 2009 graduation. Thesis Committee: Loreen Herwaldt, Thesis Supervisor Stanley Perlman Elizabeth Chrischilles Michael Jones Kung-Sik Chan James Torner

7 I dedicate this project to Dr. Loreen Herwaldt, without her this project could not have been possible. ii

8 ACKNOWLEDGEMENTS I wish to thank Drs. Herwaldt, Loreen A; Torner, James C; Chrischilles, Elizabeth A; Jones, Michael P; Smith, Tara; Perlman, Stanley; and Chan, Kung-Sik for their unflinching support during the course of this project. I appreciate all the advice and words of encouragement this incredible committee have provided me while carrying out this research. I would like to acknowledge the tremendous support of Dr. (Davis) Volk, Paige who coordinated the intervention part of this project from the Pediatric Intensive Care Unit (PICU). In addition, my appreciation goes to the attending physicians, residents, fellows, nurse practitioners, pharmacists, and auxiliary staff of the PICU for supporting this project. I would like to thank the infectious disease specialists Drs. Ziebold, Christine S; and Gomez, Oscar G for their valuable contributions. My gratitude also goes to Dr. Diekema, Daniel for his expert advice on culture results, Ms. Leder, Laurie who helped with data entry, Ms. Seidel Jennifer who helped with literature searches; Ms. Von Behren, Sandra and Dr. Helms, Charles M who provided much needed letters of support when I applied for a grant from the Association for Professionals in Infection Control and Epidemiology (APIC); and to all the staff of the Program of Hospital Epidemiology for their support and encouragement. My sincere appreciation also goes to Ms. Anderson, Marilyn and Ms. Colbert, Kathy who helped with scheduling meetings with my committee. I am highly indebted to my lovely wife Ms. (Sykes) Alamu, Chenoa Annette for her spiritual support. She cared for our two small children and made the home front a iii

9 congenial atmosphere for me while I was busy with this project. I am indebted to my two jolly children, Victoria and Jonathan who were born after I started this project and who never complained about not seeing their daddy. They are a blessing to me. I will not forget the support of my in-laws, Mrs. (also known as Mama) Sykes, Lue; Ms. Sykes, Lori; and Mr. Sykes, Brian for their persistent prayers for my success. I would like to thank Dr. Lafollette, Sharron, Head, Department of Public Health University of Illinois at Springfield, for supporting my PhD candidacy. She helped with proofreading and editing. Finally, I would like to thank APIC s scientific council for providing grant for the intervention stage of this project. iv

10 LIST OF TABLES LIST OF FIGURES TABLE OF CONTENTS vii ix CHAPTER 1 ANTIMICROBIAL USE Introduction Infections in Intensive Care Units (ICUs) Antimicrobial use in ICUs Antimicrobial Stewardship Strategies Significance of the Problem Statement of the Problem 16 CHAPTER 2 EVALUATION OF ANTIMICROBIAL USE IN A PEDIATRIC INTENSIVE CARE UNIT Introduction Methods Results Discussion 28 CHAPTER 3 FACTORS THAT PEDIATRIC INTENSIVISTS CONSIDER WHEN DECIDING TO START ANTIMICROBIAL AGENTS Introduction Methods Results Discussion 58 CHAPTER 4 INTERVENTION TO IMPROVE ANTIMICROBIAL STEWARDSHIP IN A PEDIATRIC INTENSIVE CARE UNIT Introduction Methods Results Discussion 96 v

11 CHAPTER 5 CONCLUSION Introduction Summary of Findings Strengths and Limitations Future Directions Overall Summary and Conclusion 131 APPENDIX A CHART ABSTRACTION FORM 133 APPENDIX B ISOGRAPHS FOR SAMPLE SIZE CALCULATION FOR MATCHED CASE-CONTROL STUDY/ INTERVENTION 139 APPENDIX C TIME SERIES DIAGNOSTICS AND TECHNICAL DETAILS 143 REFERENCES 151 vi

12 LIST OF TABLES Table 1. Table 2. Table 3. Studies of Nosocomial Infections in Adult Intensive Care Units 17 Studies of Nosocomial Infections in Pediatric Intensive Care Units 18 Sepsis Criteria: Age-specific Vital Signs and Laboratory Variables 35 Table 4 Descriptive Characteristics of the Patient Population 36 Table 5. Comparison of Antimicrobial Use by Age Category 37 Table 6. Exposure to the 10 most Commonly Prescribed Antimicrobial Agents: All Age Categories 38 Table 7. Indications for Antimicrobials Therapy 39 Table 8. Table 9. Characteristics of Long-stay and Short-stay Patients in the PICU 40 Patients Who Met Sepsis Criteria and Those Who Did Not Meet Sepsis Criteria among the Subcategory of Patients Whose First Course of Antimicrobials was Empiric 43 Table 10. Compliance with the Antimicrobial Assessment Form 68 Table 11. Table 12. Factors Pediatric Intensivists Considered While Prescribing Prophylactic Antimicrobial Agents (Data from Antimicrobial Assessment Form) 69 Factors Pediatric Intensivists Considered While Prescribing Empiric or Targeted Antimicrobial Agents (Data from Antimicrobial Assessment Form) 70 Table 13. Descriptive Characteristics of Patient Population 72 Table 14. Univariate Analysis: The Association of Empiric Antimicrobial Treatments and Infection Parameters 73 vii

13 Table 15. Adjusted Odds Ratio for the Association of Antimicrobial Empiric Antimicrobial Treatments and Infection Parameters 74 Table 16. Descriptive Characteristics of the Patient Populations 105 Table 17. Table 18. Treatment Days for Empiric, Prophylactic, and Targeted Antimicrobial Use 107 Description of Organisms and Category of Antimicrobial Treatment for Patients who Had Positive Blood Cultures 109 Table 19. Susceptibility Results: Staphylococcus aureus 111 Table 20. Susceptibility Results: Pseudomonas aeruginosa 112 Table 21. Susceptibility Results: Enterobacter cloacae 113 viii

14 LIST OF FIGURES Figure 1. Duration of Stay in the PICU 44 Figure 2. Figure 3. Indications for the 10 most Commonly Prescribed Antimicrobial Agents 45 Antimicrobial Agents Reported on the Antimicrobial Assessment Form 75 Figure 4. Study Design and Nested Periods 114 Figure 5. Piperacillin-tazobactam, Cefepime, and Ceftriaxone Use in the PICU 115 Figure 6. Cefazolin Use in the NICU 116 Figure 7. Figure 8. Pre-Intervention Period: Weekly Proportion of Patients Exposed to Empiric Antimicrobial Agents in the PICU 117 Intervention and Post-Intervention Periods: Transformed Data of Patients Exposed to Empiric Therapy and Fitted Values 118 ix

15 1 CHAPTER 1: ANTIMICROBIAL USE 1.1 Introduction An antimicrobial agent or an antibiotic is defined as any therapeutic agent produced by an organism or made synthetically that selectively destroys or inhibits the growth of micro-organisms, such as bacteria, fungi, or protozoa (1). Antimicrobial agents differ markedly in physical, chemical, and pharmacological properties (2). Broadspectrum antimicrobial agents, such as amoxicillin and levofloxacin, are effective against a wide range of disease-causing bacteria. In contrast, narrow-spectrum antimicrobial agents, such as penicillin G and vancomycin, are effective against only specific groups of bacteria. Antimicrobial agents are not effective against viral infections. More than 5,000 antimicrobial agents have been identified (3) and about 100 of these are used clinically to treat infections. Current classes of antimicrobial agents include penicillins (e.g., penicillin and amoxicillin), aminoglycosides (e.g., gentamicin and streptomycin), tetracylines (e.g., tetracycline and doxycycline), macrolides (e.g., erythromycin and clarithromycin), glycopeptides (e.g., vancomycin and teicoplanin), cephalosporins (e.g., cephalexin and cefuroxime), and fluoroquinolones (e.g., ciprofloxacin and levofloxacin) (4-7). 1.2 Infections in Intensive Care Units (ICUs) A nosocomial infection is a localized or systemic condition that: (1) results from an adverse reaction to the presence of an infectious agent(s) or its toxin(s) and (2) was not present or incubating at the time of admission to the hospital (8). Thus, nosocomial infections are generally defined as infections occurring 48 hours or more after hospital

16 2 admission (e.g., bloodstream infections) or within 30 days after a surgical procedure (i.e., surgical site infections) (9). Nosocomial infections cause morbidity, and mortality, and increase the costs of health care. The attributable mortality rate is about 15% for nosocomial bloodstream infections, making these infections the eighth leading cause of death in the United States (US) (10). The crude and attributable mortality rates for nosocomial pneumonias are approximately 30% and 10%, respectively (11-14). Although ICUs account for fewer than 10 percent of the total beds in most hospitals, more than 20 percent of all nosocomial infections are acquired in intensive care units (ICUs) (15). Several factors contribute to the high incidence of nosocomial infections in ICUs. Patients in ICUs have more comorbid illnesses compared with patients in the general hospital population and are often exposed to intravascular catheters that provide microorganisms a portal of entry into the bloodstream. Healthcare workers frequently access or manipulate these devices and other invasive catheters and can transmit pathogenic organisms from patient to patient if they do not follow appropriate infection prevention practices. Thus, patients can become colonized or infected with nosocomial pathogens. Moreover, patients in ICUs are at risk of exposure to antimicrobial-resistant organisms, such as methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant enterococci (VRE), and infections caused by these resistant pathogens are difficult and costly to treat effectively. Several studies have evaluated the prevalence and incidence of infections among children in pediatric intensive care units (PICUs). Given the possible association of nosocomial infection rates and antimicrobial use, the principal investigator reviewed the

17 3 literature on nosocomial infection rates in adult ICUs and in PICUs. Tables 1 and 2 summarize data from multi-center studies of nosocomial infections. From the studies reviewed, nosocomial infection rates in PIUs ranged from 6.1 to 11.9 per 100 patients (16, 17) or 0.15 to 31.4 per 1000 patient days (18, 19), compared with 6.1 to 21.0 per 100 patients (17, 20, 21) in adult ICUs. However, the types of nosocomial infections most frequently identified in PICUs are different from those identified in adult ICUs. Catheter-associated bloodstream infections (CABSIs) (6.5 per 1000 catheter days) (18) are the most nosocomial infections among children hospitalized in PICUs, while ventilator-associated pneumonia (VAP) (24.1 per 1000 ventilator days) (20) are the most nosocomial infections among adults hospitalized in ICUs. The report from Richards and colleagues suggested that these variations are not due to differences in device utilization between PICUs and adult ICUs because utilization of central venous catheters and endotracheal tubes in PICUs was comparable to that in adult medical ICUs (17). Moreover, the distribution of nosocomial infection sites differs by age among patients hospitalized in the PICUs. Primary bloodstream infections and surgical site infections were the nosocomial infections most frequently identified among infants aged 2 months or less (16). Whereas urinary tract infections were more frequently identified among children older than 5 years than among younger children (16). The types of organisms that cause nosocomial bloodstream infections and urinary tract infections among patients hospitalized in PICUs and adult ICUs are similar. Richards and colleagues (16, 17) reported that coagulase-negative staphylococci were the most common cause of primary bloodstream infections in PICUs and adult ICUs and Escherichia coli was the most common cause of urinary tract infections (16, 17).

18 4 Pseudomonas aeruginosa, often caused pneumonia among patients hospitalized in PICUs, whereas Staphylococcus aureus often caused pneumonia among adult patients in ICUs (16, 17). 1.3 Antimicrobial Use in ICUs Published reports have documented widespread antimicrobial use among patients both in communities and hospitals (22-38). These reports have suggested that widespread antimicrobial use might be associated with the emergence of antimicrobial-resistant bacteria (24, 39-41). Use of antimicrobials for conditions or infections that are not caused by bacteria is one reason why many pathogens have become resistant to commonly used antimicrobial agents. Infections caused by resistant organisms such as MRSA are associated with prolonged length of stay in the hospital, and higher mortality than infections caused by susceptible organisms (42, 43). Thus, antimicrobial stewardship has become an important public health issue. Although antimicrobial agents have saved lives, their misuse has caused significant problems and patients have died from complications related to antimicrobial misuse, such as severe Clostridium difficile-associated diarrhea Antimicrobial Use in Adult ICUs Several investigators evaluated use of antimicrobial agents in adult ICUs 44-47). Unfortunately, different investigators used different measurement units when (22-25, 27, reporting data on antimicrobial use. Several groups reported antimicrobial use expressed as the number of World Health Organization (WHO) defined daily doses (DDD) per 1,000 patient-days in individual ICUs or in groups of ICUs (24, 26, 48, 49). In these ICUs, antimicrobial use ranged from 490 to 3,456 WHO DDD per 1,000 patient-days (25, 50, 51). Other investigators have collected data at the patient level. For example, Petersen and

19 5 colleagues reported that antimicrobial use in a group of four Danish ICUs, ranged from 1,390 to 2,510 antimicrobial treatments per 1,000 patient-days (25). The authors reported that antimicrobial use was highest in an ICU that routinely used selective decontamination of the digestive tract. Other investigators have assessed the percentage of ICU patients who received at least one antimicrobial agent. Investigators who have performed studies in individual adult ICUs reported that 68 80% of patients received antimicrobial agents (52-55). The European Prevalence of Infection in Intensive Care (EPIC) point prevalence, a study of 1,047 ICUs in 17 European countries, found that 62% of patients received antimicrobial agents (56). The Nosokomiale Infektionen in Deutschland: Erfassung und Pravention (NIDEP study), a German point prevalence survey, found that 53% of patients hospitalized in ICUs in 72 hospitals received antimicrobial agents (57). Erlandsson and colleagues studied antimicrobial use from patients admitted to 23 Swedish ICUs during a 2-week period. They found that the percent of patients exposed to antimicrobial agents increased as the level of care increased. Eighty-four percent (median; range, 58% to 87%) of patients hospitalized in tertiary care centers were prescribed antimicrobials compared with 67% (median; range, 35% to 93%) of patients hospitalized in ICUs in secondary hospitals, and 38% (median; range, 24% to 80%) of patients hospitalized in ICUs of primary hospitals (26). Erbay and colleagues evaluated the appropriateness of antimicrobial use in relation to the patients diagnoses and to the bacteriological findings in the adult ICUs of a tertiary care hospital with an antimicrobial restriction policy (44). The authors found that 223 (60.6%) of 368 patients admitted to the study ICUs received antimicrobial agents and 47.3% of all antimicrobial prescriptions were inappropriate. The most frequently

20 6 prescribed antimicrobials were first-generation cephalosporins (16.1%), third-generation cephalosporins (15.2%), aminoglycosides (12.1%), carbapenems (10.5%), and ampicillin-sulbactam (8.7%) (44). Investigators have also evaluated indications for empiric antimicrobial therapy in adult ICUs. A cohort study conducted at a teaching hospital in the US found that 42% of the patients were treated for suspected infection, suspected sepsis, severe sepsis, or septic shock. Only 41% of these patients subsequently had infections documented by positive cultures (46). Similarly, a multicenter prospective study of patients receiving antimicrobials in adult ICUs in Australia and New Zealand, found that suspected infection (i.e., empiric treatment) was second only to prophylactic treatment as an indication for antimicrobials. Of 183 patients who received empiric therapy, only 25% were subsequently confirmed to have infections (47) Antimicrobial use in Neonatal Intensive Care Units (NICU) The following articles described antimicrobial use in NICUs. Lesko and colleagues found that 71% of the infants hospitalized in two NICUs for more than 24 hours between 1978 and 1979 were exposed to gentamicin, 44% to ampicillin, and 34% to penicillin. The authors observed that vancomycin and clindamycin were not used during 1978 and 1979, but were each used in 1% of infants during 1985 and 1986; the use of cephalosporins did not change during the study period (30). Fonseca and colleagues found that 75% of infants born between February and March 1991 and who stayed at least 24 hours in Yale-New Haven Hospital s NICU were treated with antimicrobials during their first 48 hours of life. The authors noted that the highest rate of antimicrobial therapy was among premature infants with birth weights less

21 7 than 1,500 g. Ninety-two percent of these infants received antimicrobials in the first 48 hours and ampicillin and gentamicin were the most frequently prescribed antimicrobials (29). Warrier and colleagues conducted a retrospective study that assessed antimicrobial use among newborns in an NICU between January 1997 and June Their study showed that the mean number of antimicrobial prescriptions per infant was highest among the 24- to 27-week infants (11.8 antimicrobial prescriptions per infant) followed by the < 23-week infants (9.9 antimicrobial prescriptions per infant). Ampicillin and cefotaxime were the antimicrobials used most frequently; all infants of gestational age < 23 weeks were treated with these agents. The authors concluded that the high rate of antimicrobial exposure in their study unit was due to the empiric therapy protocol mandating that physicians start antimicrobial therapy for all sick neonates while waiting for the results of bacterial cultures and thus, did not reflect the incidence of bacterial infection (28) Antimicrobial use in PICUs Six studies have evaluated antimicrobial use in PICUs (31-33, 37, 38, 58), two of which were conducted in the US (33, 37). Van Houten and colleagues found that 36% of patients hospitalized in a combined PICU and NICU were exposed to antimicrobial agents during an 8-week study period. Only 12.3% of patients who received antimicrobial agents had proven bacterial infections (38). Fischer and colleagues evaluated antimicrobial use in a PICU of a university teaching hospital (32). Two hundred and fifty-eight (57%) of 456 patients received

22 8 systemic antimicrobial agents. Thirty-three percent of patients who were treated for a suspected VAP did not need antimicrobial treatments. Grohskopf and colleagues conducted two multicenter point prevalence surveys of infections in NICUs and PICUs reporting to the Pediatric Prevention Network (PPN) (37). On two days when the survey was conducted, 756 (71%) of 1067 patients in the PICUs were receiving antimicrobial agents. Briassoulis and colleagues, who evaluated antimicrobial use in a cohort study of all patients admitted to a PICU, reported that 67.2% of patients admitted to this PICU received antimicrobial agents on their admission days and 80.5% of patients received at least one antimicrobial during their PICU stays (35). All patients who presented with shock on admission (32.8%) were treated with empiric antimicrobial agents (35). Ding and colleagues conducted a study of antimicrobial use in PICUs at three tertiary children s hospitals in China (31). The authors reported that 524 (95%) of the 540 children admitted during the study period received at least one antimicrobial agent, and that antimicrobial therapies were started empirically for 72% of these patients. The overall proportion of patients who were exposed to antimicrobial agents in Ding s study was higher than that reported in the other studies described previously. Toltizs and colleagues assessed antimicrobial use among children who were admitted to a PICU and who had fever (axillary temperature > 38.3 C) (33). Only seven (3.3%) of 211 patients in the study did not receive parenteral antimicrobial agents during their febrile episodes, suggesting that pediatric intensivists almost always start antimicrobial agents for febrile patients. However, this study was not designed to assess factors intensivists consider when starting antimicrobial agents.

23 9 The most commonly prescribed antimicrobial agents varied by study. Fischer and colleagues reported that aminopenicillin and aminoglycosides accounted for nearly onehalf of all drugs prescribed during the first 3 days of hospitalization (32), whereas Grohskopf and colleagues reported that cefazolin, vancomycin and cefotaxime were the most frequently prescribed antimicrobial agents. In addition, half of all vancomycin use was empiric and all cefazolin use was prophylactic (37). Ding and colleagues reported that third generation cephalosporins, which accounted for 31% of all antimicrobials prescribed, were the major antimicrobials used in all PICUs participating in their study (31). In summary, a high proportion of patients in the PICUs studied were treated with antimicrobial agents. Few of the patients who received antimicrobial agents had proven bacterial infections. 1.4 Antimicrobial Stewardship Strategies Researchers have evaluated various strategies for improving antimicrobial stewardship in ICUs. These strategies include but are not limited to formulary systems, antimicrobial cycling, and computer-assisted programs. Only four studies have evaluated strategies for improving antimicrobial use in PICUs (31, 59-61). Strategies assessed in these studies include antimicrobial cycling (59), restricted use of specific antimicrobial agents (60), educational interventions (31), and computer-assisted programs (61). Moss and colleagues conducted a pilot study to determine whether antimicrobial cycling would reduce colonization and infection by resistant bacteria. The investigators introduced different types of antimicrobial agents during each 3 month cycle over the 18 month study period. They used imipenem/cilastatin during cycle 1,

24 10 piperacillin/tazobactam during cycle 2, and ceftazidime and clindamycin-cefepime during cycle 3. The investigators found that cycling broad-spectrum antimicrobials for empirical treatment was safe for patients in their PICU. However, the prevalence of children colonized with resistant bacteria did not change (29% vs. 24%; P = 0.41) during the study (59). Toltzis and colleagues attempted to reduce the rate of colonization and infection with resistant organisms by controlling antimicrobial use (60). The authors restricted use of ceftazidime, a third-generation cephalosporin because Bush et al. previously documented that this antimicrobial agent can induce the class 1 (AmpC) chromosomal beta-lactamases of Enterobacter, Serratia, Citrobacter, and Pseudomonas (62). Use of ceftazidime decreased 96%, but the incidence density of ceftazidime-resistant organisms increased during the study from 1.57 to 2.16 isolates/100 patient-days. The authors observed that in many instances children were colonized with the resistant Gram-negative bacteria on admission to the unit. The investigators concluded that an antimicrobial restriction policy would not diminish the size of the endemic reservoir for antimicrobialresistant Gram-negative rods in a PICU and that antimicrobial stewardship programs implemented in the absence of infection control interventions would have little effect on the incidence of antimicrobial resistant organisms (60). Ding and colleagues showed that a combination of antimicrobial stewardship strategies could help minimize the use of broad-spectrum antimicrobial agents in a PICU (31). The authors introduced an educational program, which focused primarily on pediatricians, and antimicrobial guidelines that compelled physicians to obtain prior approval from a senior pediatrician before ordering restricted antimicrobials. The

25 11 following measures decreased significantly: prescription rates for third-generation cephalosporins and macrolides (P < 0.01); empiric treatment (defined as a treatment based on the clinical evidence of infection and not dependent on culture results) (P < 0.01); the incidence of P. aeruginosa resistant to imipenem, cefepime, and ceftazidime (P < 0.05); and the incidence of E. coli and K. pneumoniae resistant to cefepime (P < 0.01) (31). Mullett and colleagues introduced bedside computer terminals, which ran the Health Evaluations through Logical Processing (HELP software) (61), to provide decision support for physicians prescribing antimicrobial agents. Through HELP, physicians were able to retrieve patients vital signs, laboratory test results, radiology test results, and pathology results. The HELP software also helped physicians chose antimicrobials and doses based on logic incorporated into the software. After the intervention, the rate of erroneous antimicrobial doses decreased by 59%, the rate of subtherapeutic (defined as antiinfective therapy that fell below the minimum recommendation) risk days decreased by 36%, and the rate of excessive-doses (defined as antiinfective therapy that excluded the maximum recommendation) risk days declined by 28%. Furthermore, the number of orders placed per anti-infective course decreased 11.5% from an average of 1.56 to 1.38 orders/patient-anti-infective (P < 0.01) (61). However, the costs of antimicrobial agents for the PICU and for the hospital did not decrease. In summary, the principal investigator identified only four studies that evaluated antimicrobial stewardship programs in PICUs (31, 59-61), only two of which were successful (31, 61). The intervention strategy that utilized computer decision support improved antimicrobial prescribing (61) but such software is expensive and requires support from

26 12 information technology staff. The intervention strategy that combined educational sessions and antimicrobial guidelines (31) is also desirable. However, implementing this strategy would also be complex. For example, a committee of experts would need to design the guidelines and educational programs must be repeated frequently. 1.5 Significance of the Problem Problems Associated with Overuse of Antimicrobials Antimicrobial Resistance The cause-and-effect relationship between antimicrobial use and the emergence of resistance has been difficult to prove. Nevertheless, several studies have demonstrated a correlation between antimicrobial use and antimicrobial resistance at the hospital level (63, 64). This evidence includes: (1) antimicrobial resistance is more prevalent among bacterial isolates causing nosocomial infection than among organisms from community-acquired infections (65, 66) ; (2) antimicrobial resistant isolates are more prevalent among patients who have received prior antimicrobial therapy than among controls (67) ; (3) the prevalence of antimicrobial resistance has changed in parallel with changes in antimicrobial use (68, 69) ; (4) the prevalence of antimicrobial-resistant bacteria is highest in areas within the hospital having the highest antimicrobial usage (69). Infections caused by antimicrobial-resistant bacteria are associated with higher mortality rates (70-74). For example, Lee and colleagues conducted a retrospective matched case-control study among patients infected with multi-drug resistant (MDR) Acinetobacter baumannii bacteremia and patients with non-mdr A. baumannii. The authors reported that the sepsis-related mortality rate for case patients with MDR A. baumannii infections was 34.8%, whereas, it was only 13.8% for patients with non-mdr A. baumannii infections, for an attributable mortality rate of 21.8% (odds ratio [OR] = 4.1, P = 0.006).

27 13 Several studies have demonstrated an association between antimicrobial-resistant bacteria and prolonged length of hospital or ICU stays (70-74). Evans and colleagues conducted a retrospective cohort study to determine the association between infections caused by resistant Gram-negative rods (rgnr) and patients outcomes (74). The authors reported a longer median hospital length of stay among patients with infections caused by rgnr than among patients with infections caused by susceptible strains (sgnr) (29 days vs. 13 days, P < ) (74). Lee and colleagues reported that patients who had infections caused by MDR A. baumannii stayed in the hospital longer than patients with infections caused by non-mdr A. baumannii (54.2 days vs days, P = 0.006) (70). In addition, these infections are very costly. The Office of Technology Assessment (OTA) estimated that the in-hospital cost of nosocomial infections caused by six common antimicrobialresistant bacteria is at least $1.3 billion in the US. This estimate does not include the costs of infections caused by other organisms, the costs of lost work days, and the costs for post-hospital care Adverse Drug Reactions Several studies have evaluated the link between adverse drug reactions and antimicrobial agents among hospitalized patients (75), among patients in emergency departments (EDs) (76), and at the national level (authors utilized nation-wide database for reported adverse drug reactions) (77). For example, Mazzeo and colleagues conducted an observational prospective study for 20 days in six departments of a university hospital (75). Seven (0.04%) of the 171 patients evaluated had adverse events that included leucopenia (trimethoprim/sulfamethozale), nephrotoxicity (netilmicin and teicoplanin; cefotaxime), diarrhea (ceftriaxone), neurotoxicity (isoniazid), angioneurotoxic edema (piperacillin), and skin rashes (ceftriaxone) (75). Shehab and colleagues examined data from the National Electronic Injury Surveillance System-Cooperative Adverse Drug Event Surveillance (NEISS-CADES) system to determine the rates of visits to EDs in the US for adverse events associated

28 14 with systemic antimicrobial agents (76). Nineteen percent of 142,505 ED visits were related to antimicrobial-associated adverse events. Most (78.7%; 95% confidence interval [CI], 3% %) of these visits were for allergic reactions. Among commonly prescribed antimicrobials, sulfonamides and clindamycin were associated with the highest rate of ED visits (18.9 ED visits per 10,000 outpatient prescriptions; 95% CI, ). Salvo and colleagues analyzed an Italian database, Italian Interregional Group of Pharmacovigilance (the GIF database) comprising spontaneous reports of suspected adverse drug reactions, for adverse drug reactions to amoxicillin (AMX) and amoxicillin/clavulanic acid (AMC) (77). The GIF database collected 37,906 reports, of which 1,088 were related to AMC and 1,095 to AMX. Adverse events were reported at a rate of 2.11/1,000,000 DDD for AMC per year compared with 1.52 reports/1,000,000 DDD per year for AMX. The proportion of serious adverse reactions to AMC was higher than that to AMX (39% vs. 34%, P = 0.017). Moreover, liver toxicity was 9 fold more frequent for AMC than for AMX (77) Adverse Events Numerous studies have demonstrated that antimicrobial use is associated with diarrhea (78-80) ; and that Clostridium difficile is responsible for 15% to 25% of cases of antimicrobial-associated diarrhea (AAD) (81). A prospective study by Wistrom and colleagues determined the frequency of AAD and C. difficile-associated diarrhea (CDAD) among patients in five Swedish hospitals (82). The investigators reported that 4.9% of patients treated with antimicrobials developed AAD and that patients treated with antimicrobials for 3 days had a significantly (P = 0.009) lower frequency of AAD than those treated for longer periods. Treatment with cephlosporins, clindamycin, or broad-spectrum penicillins was associated with an increased risk of AAD and 55.4% of fecal samples from patients with AAD were positive for C. difficile cytotoxin B (82).

29 15 Chaudhry and colleagues analyzed 524 stool specimens from patients suspected to have CDAD (67), 95% of whom were on more than one antimicrobial agent. Thirty-seven (7.1%) of stool samples were positive for C. difficile toxin; 41% of the 37 toxin-positive stool samples were also positive for C. difficile by culture. The investigators noted that 8 of these 37 (21.6%) patients died, possibly related to CDAD (83). Polgreen and colleagues reported that CDAD was significantly associated with cephalosporin use (P = 0.009) among patients who were treated for community-acquired pneumonia in a small rural hospital. Only half of these patients had clinical evidence of pneumonia. The crude mortality rate associated with CDAD was 33% (79) Cost of Antimicrobial Treatments Kern and colleagues used pharmacy data to assess antimicrobial use and expenditures at four university-affiliated hospitals in southwestern Germany. Annual expenditures for antimicrobial drugs per university hospital ranged between 4.5 and 8 million Deutsch Marks ($2.97 million to $5.28 million). Study site (P < 0.001), intensive care (P = 0.015), type of infection (P < 0.001), gastrointestinal tract disease (P = 0.037), and pressor therapy (P = 0.002) were significantly associated with antimicrobial expenditures in these hospitals (84). Inan and colleagues estimated the mean daily costs for 1407 episodes of nosocomial infection among patients hospitalized in six adult ICUs to be $89.64 per infection (85). The mean daily antimicrobial cost for pneumonia was $94.32 compared with $31.31 for bloodstream infections and $52.37 for UTIs (85). Vandijck and colleagues studied 446 BSI episodes acquired by 310 adult patients hospitalized in a university hospital s ICU. The mean overall daily antimicrobial cost was Euro ($147.34). The daily antimicrobial cost per patient with BSI caused by a MDR organism was 50% higher than for patients with non-mdr BSI ($ vs. $106.60, P < 0.001) (86). In summary, widespread antimicrobial use can lead to the emergence of antimicrobial-resistant organisms. Infections caused by resistant bacteria may be difficult

30 16 to treat or even untreatable. These infections prolong patients hospitalizations, increase the cost of care, and increase mortality. Additionally, antimicrobial use can also cause adverse reactions (e.g., allergies, renal failure, etc.) and adverse events (e.g., CDAD), which increase mortality among hospitalized patients. Furthermore, antimicrobial use increased cost of patient care and add to the burden of hospital pharmacy budget. 1.6 Statement of the Problem A pediatric intensivist in the University of Iowa Hospitals and Clinics (UIHC) PICU was concerned about antimicrobial use in the unit. However, no one had quantified antimicrobial use in the UIHC s PICU or described the patterns of antimicrobial use in this unit. To address the intensivist s concern, the principal investigator (PI) first reviewed the literature and found only a small number of studies that assessed antimicrobial use in PICUs and fewer studies that evaluated interventions to reduce antimicrobial use in PICUs. Moreover, the PI did not identify any papers describing factors that pediatric intensivists consider when prescribing antimicrobial agents. The PI conducted the quality improvement studies described in the following chapters to fill these gaps Aims The specific aims of this project were to: 1. Describe antimicrobial use in the UIHC s PICU; 2. Identify the factors that pediatric intensivists considered when deciding to start antimicrobial agents; and 3. Determine whether antimicrobial use, particularly empiric use, in the UIHC s PICU would decrease when pediatric intensivists filled out a form to document why they prescribed antimicrobial agents.

31 17 Table 1. Studies of Nosocomial Infections in Adult Intensive Care Units Study/Authors Dates Study Sites Overall Infection Rates EPIIC (21) ICUs in 17 countries medicalsurgical ICUs from 152 hospitals INICC (20) ICUs in 46 hospitals from 8 developing NNIS (87) Sites/Infection Types (% patients infected) 21/100 patients Pneumonia 47% LRTI 18% UTI 18% BSI 12% Wound 7% 6.1/100 patients Pneumonia 30 33% UTI 18 30% BSI 13 16% 14.7/100 patients VAP 24.1/1000 vd 22.5/1000 ICU CABSI 12.5/1000 cd days CAUTI 8.9 /1000 cd Organism (% infections) S. aureus 30% P. aeruginosa 29% CNS 19% E. coli 13% Enterococci 12% Acinetobacter spp. 9% Klebsiella spp. 8% CNS 38.7% S. aureus 17.0% GNR 46.0% varied by site countries Abbreviations: EPIIC = The European Prevalence of Infections in Intensive Care study; NNIS = The National Nosocomial Infections Surveillance System; INICC = International Nosocomial Infection Control Consortium; ICUs = intensive care units; LRTI = other lower respiratory tract infection; UTI = urinary tract infection; BSI = bloodstream infection; VAP = ventilator-associated pneumonia; CABSI = catheter-associated bloodstream infection; CAUTI = catheter-related urinary tract infection; vd = ventilator days; cd = catheter days; CNS = coagulase-negative staphylococci; GNR = Gram-negative rod

32 18 Table 2. Studies of Nosocomial Infections in Pediatric Intensive Care Units Study/Authors Dates Study Sites Overall Infection Rates Sites/Infection Types (% patients infected) PPN (18) PICUs *13.9/1000 pd CABSI 6.5 / 1000 cd (range ) CAUTI 5.4 / 1000 cd VAP 3.7 / 1000 vd PPN (16) PICUs 11.9/100 patients BSI 41.3% LRI 22.7% UTI 13.3% SSTI 8.0% NNIS (17) PICUs 6.1/100 patients BSI 28.0% Pneumonia 21.0% UTI 15.0% SPSNI (19) 1999 PICUs 1.5/100 pd BSI 51.7% Organism (% infections) Varied by site CNS 21.3% Candida spp. 17.3% Enterococcus spp. 13.3% S. aureus 12.0% P. aeruginosa 10.7% varied by site varied by site Pneumonia 19.0% * Median overall nosocomial infection rate Abbreviations: PPN = Pediatric Prevention Network; NNIS = National Nosocomial Infections Surveillance System; ICUs = intensive care units; LRI = lower respiratory tract infections; UTI = urinary tract infection; BSI = bloodstream infection; VAP = ventilatorassociated pneumonia; CVCABSI = central venous catheter-associated bloodstream infection; CAUTI = catheter-associated urinary tract infection; SSTI = skin/soft tissue infection; pd = patient days; vd = ventilator days; cd = catheter days; CNS = coagulase-negative staphylococci

33 19 CHAPTER 2: EVALUATION OF ANTIMICROBIAL USE IN A PEDIATRIC INTENSIVE CARE UNIT 2.1 Introduction Several investigators have reported high rates of antimicrobial use among patients hospitalized in intensive care units (ICUs) (28, 29, 31-33, 35, 36). For example, four studies in adult ICUs found that 68% to 80% of patients hospitalized in adult ICUs received antimicrobial agents (53-55, 88). Similarly, 56% to 97% of patients hospitalized during five studies in neonatal intensive care units (NICUs) or pediatric intensive care units (PICUs) received at least one antimicrobial agent during their ICU hospitalizations (31-33, 37, 89). Antimicrobial therapy is common among patients hospitalized in ICUs for several reasons. Patients in ICUs have more chronic comorbid illness and acute illnesses, injuries, and surgical procedures compared with patients in the general hospital population. In addition, patients in ICUs are often exposed to invasive devices or procedures that provide a portal of entry for microorganisms (90). These patients are also susceptible to colonization and infection with nosocomial pathogens. At times these nosocomial pathogens may be antimicrobial-resistant organisms, such as methicillinresistant Staphylococcus aureus (MRSA) and vancomycin-resistant enterococci (VRE) and thus, may be difficult to treat (90). The principal investigator identified six studies on antimicrobial use in PICUs (35, 37, 91), only two of which were conducted in the United States (US) (33, 37). These studies evaluated the classes of antimicrobials prescribed, duration of treatment, and the indications for treatment in PICUs (35, 37, 38). However, these studies have not evaluated the pediatric intensivists antimicrobial prescribing behavior and the factors that determine empiric use of antimicrobial agents in PICUs. Additionally, results from studies in adult ICUs may not apply to pediatric populations, because children have different underlying illnesses and require different processes of care. To fill this gap, we conducted a quality improvement study to determine the percentage of patients who

34 20 received antimicrobial agents, to determine the indications for antimicrobial use and to identify antimicrobial agents used most frequently in the University of Iowa Hospitals and Clinic s (UIHC) PICU. 2.2 Methods Patients This retrospective quality improvement project was done in the UIHC s multidisciplinary, tertiary-care PICU, which has 16 beds. PICU staff care for approximately 700 medical and surgical patients per year. The five pediatric intensivists who attend on the unit have week-long rotations and residents and fellows have month-long rotations Data collection The principal investigator (PI) retrospectively abstracted the paper and electronic medical records of all patients who stayed longer than 24 hours in the PICU from January 1, 2005 to June 30, Each admission was considered a patient encounter; patients who were hospitalized in the PICU more than one time during the six-month study period were included for each separate encounter. The PI developed a form (Appendix A) on which he recorded information about the antimicrobials prescribed, duration of use, route of administration, rationale(s) for use, and important patient characteristics. The PI also collected information about surgical procedures, invasive devices, mechanical ventilation, vital signs, and results of microbiologic and laboratory tests. A hospital epidemiologist who is an adult infectious disease specialist reviewed the abstracted patient records and the patients medical records to determine whether antimicrobial therapy was empiric, prophylactic, or targeted. In addition, a pediatric intensivist reviewed some of the medical records to assess reasons for initiating antimicrobial therapy or to determine whether chest radiographs were consistent with pneumonia.

35 Definitions Empiric antimicrobial use was defined as antimicrobial therapy begun when a physician suspected infection but (1) microbiologic, clinical, and laboratory data were pending or did not identify a site of infection or (2) the patient did not meet the Centers for Disease Control and Prevention s (CDC) criteria for nosocomial infection. Targeted antimicrobial use was defined as antimicrobial therapy when (1) the patient had an infection that was documented by a positive microbiologic or serological result and the patient was treated with an antimicrobial agent to which the etiologic agent was susceptible or (2) the patient was admitted on antimicrobial therapy as treatment for an infection and the microbiologic results were not available at the UIHC. Prophylactic antimicrobial use was defined as antimicrobial therapy initiated when (1) the patient had no evidence of infection and the patient was immunocompromised, had an anatomical defect, had recurrent infections, or had an indwelling device or (2) the patient was scheduled for a surgical procedure and antimicrobial therapy was given to prevent surgical site infections. The following sub-categories of prophylactic antimicrobial use were defined: prophylactic-urinary tract infections (UTI) was therapy to prevent further UTIs in a patient with a history of UTIs; prophylactic-immunocompromised was antimicrobial therapy initiated for patients who were immunosuppressed due to medications or to their underlying disease; prophylactic-perioperative was antimicrobial therapy initiated to prevent surgical site infections; prophylactic-other was prophylactic antimicrobial therapy that did not meet the definitions for the other subcategories of prophylactic treatments.

36 22 The PI considered a patient exposed if he/she received one or more dose(s) of one or more antimicrobial agents. The principal investigator adapted the definitions of antimicrobial prescriptions and antimicrobial courses used by Bergmans and colleagues (23). The PI defined an antimicrobial prescription as the initiation of one antimicrobial agent. Thus, if a physician wrote an order for gentamicin and ampicillin-sulbactam, the patient would have received two prescriptions. The principal investigator defined an antimicrobial course as an episode in which one or more antimicrobial agents were prescribed, either consecutively or in combination for prophylaxis or to treat a suspected or documented infection. The PI used criteria published by the American College of Chest Physicians and the Society of Critical Care Medicine to identify patients who had sepsis (92). According to the criteria, a child is considered septic if he/she has a systemic inflammatory response syndrome (SIRS) in the presence of or because of suspected or proven infection (92). The criteria defined SIRS as the presence of at least two of the following four criteria (one of which must be either abnormal temperature or abnormal leukocyte count): (1) abnormal temperature defined as > 38.5 C or < 36 C; (2) abnormal heart rate defined as either tachycardia, (either a mean heart rate > 2 standard deviations [SD] above normal for age in the absence of external stimulus, chronic drugs, or painful stimuli; or otherwise unexplained persistent elevation over a 0.5 to 4-hour time period) or as bradycardia (a mean heart rate < 10 th percentile for age in the absence of external vagal stimulus, β-blocker drugs, or congenital heart disease or otherwise unexplained persistent depression over a 0.5-hour time period); (3) abnormal respiratory rate defined as a mean respiratory rate > 2 SD above normal for age or mechanical ventilation for an acute

37 23 process not related either to an underlying neuromuscular disease or to general anesthesia; (4) abnormal leukocyte count defined as elevated or depressed for age (not chemotherapy-induced leucopenia) or < 10% neutrophils. The PI modified these criteria because he did not collect information about factors that could influence tachycardia or bradycardia. Table 3 summarizes age-specific values for vital signs and laboratory variables specified in the sepsis criteria (92). The PI defined length of stay (LOS) as the PICU discharge date minus the PICU admission date if admission and discharge occurred on different calendar days. The PICU admission and discharge dates were never on the same day because the study included only patients who stayed in the PICU greater than 24 hours. The PI defined long-stay patients (LSP) as patients whose length of stay in the PICU was greater than or equal to the 90 th percentile ( 18 days), and short-stay patients (SSP) as patients whose length of stay in the PICU was less than the 90 th percentile (< 18 days) Statistical Analysis The PI classified patients into four age groups for specific analyses and generated frequency counts, percentages, medians, means and standard deviations. The PI used unpaired t tests to compare means, the Wilcoxon rank sum to compare medians, and Chisquare tests of proportion to compare categorical variables. The PI used the programming language R to develop an algorithm that identified patients who met the definition of sepsis and SAS 9.1 for all other analyses. 2.3 Results Table 4 provides a description of patient characteristics. One hundred and fortyfive patients stayed longer than 24 hours during the 25-week study period. Of these

38 24 patients, 132 (91%) were admitted to the PICU only once, 12 (8.3%) had two admissions, and one (0.7%) had three admissions. The four most common reasons for admission to the PICU were: (1) management after cardiac operations (31.7%), (2) management after neurosurgery (8.9%), (3) respiratory distress (7.5%), and (4) suspected pneumonia (6.9%). Most patients were white, nearly 60% were males, and greater than 90% were discharged alive from the PICU. The median length of stay in the PICU was 3 days (Figure 1). While in the PICU, 11 (7.6%) patients did not receive antimicrobial treatment, 48 (33.1%) received one antimicrobial agent, and 86 (59.3%) received two or more antimicrobial agents. Patients who received antimicrobials while in the PICU stayed longer (median: 4 days) than patients who did not receive antimicrobials (median: 2 days) (P = 0.05). The mean age of patients who received antimicrobials (5.7 years) was similar to the mean age (6.1 years) for patients who did not receive antimicrobials (P = 0.98). Exposure to antimicrobials did not differ among age groups (Table 5). However, more neonates than infants (P = 0.002), children (P = 0.007), and adolescents (P = 0.02) received gentamicin while in the PICU (table 6). In addition, more neonates than patients in other age groups received ampicillin-sulbactam (neonates vs. infants, P = 0.01; neonates vs. children, P = 0.001; neonates vs. adolescents, P = 0.004). Patients received a total of 437 antimicrobial prescriptions during the study period, of which 192 (43.9%) were given empirically, 186 (42.6%) were given prophylactically, and 41 (9.4%) were given as targeted treatment. The physician who reviewed the medical records could not determine the indication for 18 (4.1%) prescriptions.

39 25 The 10 most commonly prescribed antimicrobial agents during the study period are listed in Table 6. Overall, gentamicin, cefazolin, and vancomycin were the three most commonly used antimicrobials, accounting for 60.3% of the total antimicrobial prescriptions during the study period. Figure 2 presents the percentages of empiric, prophylactic, or targeted use for the 10 most commonly prescribed antimicrobial agents. Ceftriaxone, piperacillin-tazobactam, azithromycin, and cefepime were the agents most frequently prescribed as empiric treatment; > 70% of the prescriptions of these agents was for empiric treatments. Trimethoprim-sulfamethoxazole, cefazolin, and nafcillin were the agents most frequently prescribed as prophylactic treatment; > 85% of the prescriptions for these agents was for prophylaxis. The two most common antimicrobial combinations used for empiric treatment were gentamicin and ampicillin; and piperacillin-tazobactam, vancomycin, and genatamicin. Gentamicin and vancomycin or gentamicin and cefepime were the two most common antimicrobial combinations used for targeted therapy. Table 7 presents the distribution of indications for antimicrobial therapy. Antimicrobials prescribed for empiric use accounted for 868 (41.7%) of 2084 antimicrobial days. Of the 868 empiric treatment days, 200 (23.0%) were for patients whose only positive cultures were of tracheal aspirates. Only 58 (6.7%) of empiric treatment days were for patients whose cultures and susceptibility results were not available while the patients were still in the PICU. Antimicrobials prescribed for prophylactic use accounted for 815 (39.1%) of 2084 antimicrobial days. Of these treatment days, 519 (63.7%) were for perioperative prophylaxis and 139 (26.8%) were for patients who were immunocompromised. Of the

40 patients admitted during the study period, 46 (31.7%) had cardiac operations. These patients accounted for 348 (42.7%) of the 815 prophylactic antimicrobial days and for 67% of the perioperative prophylaxis days. Cardiac surgery patients received longer perioperative-prophylactic treatments than patients who had neurosurgery (mean treatment days: 5.0 vs. 3.0, P = 0.03). Eleven percent of 145 patients were categorized as LSPs (16/145) and 89% as SSPs (129/145; Table 8). Significantly, more LSPs than SSPs had cardiac operations. In addition, more LSPs than SSPs had at least one culture positive for a bacterial pathogen. Significantly, more LSPs than SSPs had at least one positive blood culture (P < ) and had at least one positive tracheal aspirate. LSPs received more antimicrobial prescriptions per patient (2.7) than the SSPs (1.2). In addition, LSPs were 12 times more likely to have received empiric prescriptions and 4 times more likely to have received targeted antimicrobial prescriptions than the SSPs. In contrast, the intensity of antimicrobial use expressed per 100-days was significantly higher among SSPs than among LSPs. The mean duration of empiric treatments was longer for LSPs (6.5 days) than for SSPs (3.4 days) as was the mean duration of prophylactic treatments (Table 8). Eighty-nine percent of the patients who had tracheal aspirate cultures done had positive cultures compared with 26.7% of those who had blood cultures obtained (P < ). Similarly, 140 of 167 (84.0%) tracheal aspirate cultures were positive compared with 16 of 114 (14.0%) blood cultures (P < ). The most common microorganisms isolated from tracheal aspirate cultures were Escherichia coli, Pseudomonas aeruginosa, and Staphylococcus aureus. Coagulasenegative staphylococci were frequently isolated from blood and urine cultures.

41 27 Patients who had positive tracheal aspirates had a similar average duration of stay in the PICU compared with those who had negative tracheal aspirates (mean: 16.1 days vs days, P = 0.58), but their average length of stay was longer than for those patients who did not have tracheal aspirate cultures done (mean: 16.1 days vs. 6.0 days, P = ). Twenty-four (60%) of 40 patients who had positive tracheal aspirate cultures received empiric therapy only, 10 (25%) had both empiric and targeted therapies, and 6 (0.15%) had targeted treatments only. Patients who had positive blood cultures had significantly higher average duration of stay in the PICU compared with those who had negative blood cultures (mean: 32.0 days vs days, P < ), and those who did not have blood cultures done (mean: 32.0 days vs. 5.5 days, P < ). Four (25%) of 16 patients who had positive blood cultures had targeted treatments only, 10 (62.5%) had both empiric and targeted treatments, and 2 (12.5%) had empiric treatments only. Patients who had positive tracheal aspirate cultures were more likely to receive empiric treatments than patients who had positive blood cultures (P = 0.003), but were as likely to receive targeted treatments (P = 0.18). Pediatric intensivists were more likely to obtain infectious diseases consults when deciding to prescribe antimicrobials that were given empirically (37.5% of all empiric prescriptions) than when deciding to prescribe antimicrobials for specific infections (5.8% of all targeted prescriptions; P < ). Pediatric intensivists were equally likely to document their rationale for therapies prescribed empirically and those prescribed for specific infections (94.8% vs. 97.6%, P = 0.45). Fifteen of 45 (33.3%) patients whose first antimicrobial courses were empiric met sepsis criteria. Four of 15 (27.0%) who met the sepsis criteria had positive blood cultures,

42 28 7 (47%) had positive tracheal aspirates, and 3 (20%) had positive urine cultures. Table 9 compares the patients demographic characteristics, vital signs, and results of selected laboratory tests for patients who met sepsis criteria and those who did not. The median band counts for those who met sepsis criteria were significantly higher than those who did not meet the sepsis criteria (P = 0.04). Other infection parameters and patient characteristics were not significantly different. 2.4 Discussion The goal of this study was to assess antimicrobial use among patients who stayed longer than 24 hours in the UIHC s PICU. Specifically we wanted to determine the percentage of patients who received antimicrobial agents, the antimicrobial agents used most frequently, and how the agents were used. Descriptive data on antimicrobial use is important because it would reveal the extent of use and temporal trends of antimicrobial use in a PICU. To our knowledge, only two studies on antimicrobial use in PICUs in the US have been published (33, 37). The current study had three principal findings. First, a high proportion (92%) of patients who stayed longer than 24 hours in the UIHC s PICU received antimicrobial agents. Second, only 17% of those who were treated with antimicrobial agents received targeted therapy. Third, 45% of those who received antimicrobial therapy were treated empirically. Studies that assess antimicrobial use in PICUs are uncommon. The six studies (31-33, 37, 38, 58) that were available for review found that between 36% and 97% of patients hospitalized in PICUs received antimicrobial treatments. Three of these studies were conducted in Europe and these studies found the lowest proportion of patients (32, 35, 91)

43 29 exposed to antimicrobials (36% to 67%) compared with studies (33, 37) conducted in the US (71% to 97%) and a study (95%) (31) conducted in China. Thus, the proportion of patients in the UIHC s PICU who received at least one antimicrobial agent was among the highest reported. Antimicrobial use in the UIHCs PICU was higher than that reported by Grohskopf and colleagues (37) (one of the studies conducted in the US). The investigators reported that 70.8% of patients in 35 PICUs during two point prevalence surveys conducted in 1999 and 2004 received antimicrobial agents. The difference between Grohskopf s findings and ours might have several explanations. In the current study, the PI retrospectively abstracted information relevant to antimicrobial use from the medical records of all patients who stayed longer than 24 hours in the UIHCs PICU during a sixmonth period. Grohskopf and colleagues, on the other hand, did point-prevalence surveys in multiple PICUs on only two days and, thus, they may have underestimated the proportion of patients exposed to antimicrobial agents. The patient populations could also have been different. The UIHCs PICU is a combined medical-surgical unit; about 60% of the patients were admitted to the PICU for further management after surgical procedures. Therefore, a high proportion of these patients received perioperative antimicrobial treatments. Grohskopf s study included 35 PICUs, 30 of which were general PICUs, 1 of which was a cardiac PICU, and 2 of which were described as other PICUs. Only two of the study units were medical-surgical PICUs. In the current study, antimicrobial utilization differed according to age category. Neonates received more antimicrobial prescriptions per patient (mean: 5.17 antimicrobial prescriptions per patient), in general, and more prescriptions of three specific agents

44 30 ampicillin-sulbactam, gentamicin, and vancomycin than did patients in other age categories. The main indication for these antimicrobial agents was empiric therapy. For example, 11 (50%) of 22 neonates who received gentamicin were treated empirically, which is consistent with the unit s treatment protocols for treating sick neonates. Neonatologists and pediatric intensivists at the UIHC treat neonates who have signs of possible sepsis or risk factors for infection (e.g., premature labor, prolonged rupture of membranes) with ampicillin and gentamicin and these physicians treat neonates suspected of having nosocomial infections with vancomycin, gentamicin, and piperacillin-tazobactam. Warrier and colleagues (28) observed a similar practice. The investigators retrospectively assessed antimicrobial use among newborns in an NICU and found that the neonates received a mean of 9.9 antimicrobial prescriptions. In this NICU, ampicillin and cefotaxime were the antimicrobials used most frequently. The investigators concluded that the high rate of antimicrobial exposure in their study was probably due to the standard practice of administering empiric therapy for all sick neonates while waiting for the results of bacterial cultures and, thus, did not reflect the incidence of bacterial infections (28). In the current study, 45% of the patients who received antimicrobial therapy were treated empirically. These patients also accounted for 42% of the total antimicrobial treatment days. Patients who received empiric antimicrobial therapy had a longer median duration of therapy and a higher number of prescriptions per course of antimicrobial treatment (2.7 empiric antimicrobial prescriptions per course) than patients who received prophylactic antimicrobial treatments (1.7 prophylatic antimicrobial prescriptions per course) and patients who received targeted antimicrobial treatments (1.6 targeted

45 31 antimicrobial prescriptions per course). This observation makes sense because physicians usually prescribe two or three antimicrobials for empiric treatment so that they cover a broad spectrum of organisms until they know the culture results, while guidelines for surgical prophylaxis recommend the use of a single antimicrobial agent, which should be discontinued within 24 hours of the operative procedure (93). Some studies have found even higher proportions of empiric antimicrobial treatment in other PICUs than we found in the current study. For example, Van Houten and colleagues (38) and Ding and colleagues (31), respectively, reported that 58% and 72% of antimicrobial therapy was given empirically. Differences in patients characteristics could account for some of this variation. For example, van Houten and colleagues conducted their study in a PICU that was the only tertiary referral center for a population of approximately 2.5 million people, whereas, there are two other PICUs in Iowa to serve its population of 3 million. Therefore, the PICU described by van Houten et al. may have served a patient population that was different than that served by the UIHC s PICU. Study duration and design could affect a study s results. For example, Ding and colleagues conducted their study over a 5-year period (January 2002 to December 2006). They reviewed the clinical records of the first 15 patients admitted every month, for a total of 180 patients. In contrast, we retrospectively abstracted data on antimicrobial use for all patients who stayed longer than 24 hours between January 2005 and June An alternative explanation of the difference between our findings and those of Ding and colleagues could be that physicians practices vary by the location of their practices. In fact, researchers who have studied small area variation have demonstrated that factors other than the patients characteristics may affect utilization of medical tests

46 32 and treatments. Wennberg (94) reported that heterogeneity in physicians practice styles may be an important cause of differences in medical utilization. To our knowledge there is no study on small area variation with respect to antimicrobial use in PICUs. However, we wonder if physicians practice styles may account for some of the variation in empiric antimicrobial use between our study and those reported in the literature. The frequency of nosocomial infections and the patients length of stay in a PICU could also affect antimicrobial use. Yogaraj and colleagues, who assessed risk factors and outcomes of nosocomial primary bloodstream infections among patients admitted to a PICU, found that patients who acquired nosocomial bloodstream infections had higher pediatric risk of mortality (PRISM III) scores on admission than those who did not acquire these infections. This difference was not statistically significant. However, patients who acquired nosocomial bloodstream infection were in the ICU a mean 11.7 days (median: 10 days; range: 2 33 days) when they were found to have their infections. Those patients who did not acquire nosocomial infection had a mean ICU stay of only 4.1 (± 11 days) days. As one would expect most (98.2%) patients with nosocomial bloodstream infections were treated with antimicrobials while only 75.1% of patients without bloodstream infections were treated with antimicrobials (P = 0.007) (95). Like Yogaraj et al., we noted that nosocomial infections occurred in patients who had long hospitalizations in the PICU. Of the nine documented nosocomial infections during the study period, seven occurred in LSPs: two episodes of coagulase-negative staphylococcal bloodstream infections in two different patients, two episodes of C. difficile infections in one patient, and two episodes of ventilator-associated pneumonia (one caused by Haemophilus influenza and once caused by Escherichia coli and

47 33 Enterobacter cloacae) in two different patients. We also assessed our data to compare antimicrobial consumption among SSPs and LSPs. The median length of stay for LSPs was approximately ten times that of SSPs. In addition, all LSPs were on mechanical ventilators and had central venous catheters compared with 56.6% and 60.5%, respectively, among SSPs. Only 11% of patients in the study were categorized as LSPs but they accounted for a disproportionate (72.2%) proportion of the total antimicrobial treatment days. It is, therefore, not surprising that we observed different patterns of antimicrobial use among these patients. For example, a higher proportion of LSPs than SSPs received empiric or targeted antimicrobial therapies. The median duration of empiric therapy among the LSPs was 5 days compared with 3 days for SSPs. Susceptibility test results, which are available 48 to 72 hours after the cultures are obtained, would not have been available before many of the SSPs were discharged from the unit. Therefore, an antimicrobial intervention strategy targeting empiric antimicrobial use among SSPs would be difficult to implement. In the current study, vancomycin, gentamicin, and cefazolin were the antimicrobial agents prescribed most frequently. Of these agents, cefazolin was used primarily for perioperative prophylaxis, while vancomycin and gentamicin were used primarily for empiric or targeted treatment. Grohskopf and colleagues found similar results. They reported that cefazolin, vancomycin, and cefotaxime were the antimicrobial agents most frequently prescribed in their PICU. In addition, the investigators reported that half of all vancomycin use was empiric and all cefazolin use was prophylactic (37). The antimicrobials used in the UIHCs PICU and the PICU studied by Grohskopf were consistent with the Infectious Diseases Society of America s (IDSA) guidelines on

48 34 surgical prophylaxis and recommendations, such as those in Long s textbook Principles and Practice of Pediatric Infectious Disease, for treatment of neonatal sepsis. The IDSA s guidelines on perioperative antimicrobial prophylaxis recommended that surgeons use cefotetan or cefoxitin for operations involving the distal ileum, appendix, or colon, and to use cefazolin for all other procedures (93) and Long s textbook suggested that physicians prescribe vancomycin plus gentamicin or ceftazidime for neonates with suspected sepsis (96). In summary, most antimicrobial use in the UIHC s PICU during the study period was for prophylactic treatment of patients who underwent surgical procedures and for empiric treatment of patients with suspected infections. Patients received few courses of targeted antimicrobials. Therefore, the prescribing patterns for prophylactic and empiric antimicrobial treatments are likely to play a major role in shaping overall antimicrobial use in this unit (Refer to chapter 5 for strengths and limitations of this study).

49 35 Table 3. Sepsis Criteria: Age-specific Vital Signs and Laboratory Variables Age Group Heart Rate, Beats/Min Respiratory Rate (Br/min) Tachycardia Bradycardia Leukocyte Count x 10 3 /mm 3 Systolic Blood Pressure, mm Hg 0 d - 1 wk > 180 < 100 > 50 > 34 < 65 > 1wk - 1 mo > 180 < 100 > 40 > 19.5 or < 5 < 75 > 1 mo - 1 yr > 180 < 90 > 34 > 17.5 or < 5 < yrs > 140 NA > 22 > 15.5 or < 6 < yrs > 130 NA > 18 > 13.5 or < 4.5 < < 18 yrs > 110 NA > or < 4.5 < 117 (92) Abbreviations: min = minute; Br = breaths; mm Hg = millimeters of mercury; d = day; mo = month; yr = year; NA = not applicable

50 36 Table 4. Descriptive Characteristics of the Patient Population Variables Characteristic Total n = 145 Gender Female 61 (42.0%) Age Median age in years (mean ± std) 3.00 (5.7 ± 6.2) Age Categories Neonates (0-1 mo) 22 (15.2%) Infants (> 1 mo 2 yrs) 44 (30.3%) Children (> 2 yrs 12 yrs) 44 (30.3%) Adolescents (> 12 yrs) 35 (24.1%) Ethnicity White 110 (75.9%) Black 10 (6.9%) Hispanic 6 (4.1%) Asian/Pacific Islander 4 (2.8%) Others 15 (10.4%) Weight Length of Hospitalization Median weight in kg (mean ± std) Median length of stay in hospital in days (mean ± std) Median length of stay in PICU in days (mean ± std) 14 (23.44 ± 23.9) 10 (16.85 ± 19.7) 3.00 (8.98 ± 15.2) Invasive Devices Endotracheal tube 89 (61.4%) Foley catheter 97 (66.9%) Central venous catheter 94 (64.8%) Nutrition Parenteral nutrition 3 (2.1%) Reasons for Admission Cardiac operations 46 (31.7%) Neurosurgery 13 (8.9%) Respiratory distress 11 (7.5%) Suspected pneumonia 10 (6.9%) Survival Discharged from the PICU alive 140 (97.0%) Abbreviations: std = standard deviation; kg = kilograms; PICU = pediatric intensive care unit; mo = month; yrs = years

51 37 Table 5. Comparison of Antimicrobial Use by Age Category Age category Number Who Received Antimicrobial Agents (%) Number of Antimicrobial Prescriptions among Patients Who Received Antimicrobials (mean per patient) Median Duration of Antimicrobial Therapy in Days (mean std) All 134 (92) 437 (3.26) 3.00 (4.77 ± 4.07) Neonates (0 1 mo) Infants (> 1 mo 2 yrs) Children (> 2 yrs 12 yrs) 18 (82) 93 (5.17) 4.00 (5.65 ± 4.04) 43 (97) 139 (3.23) 3.00 (4.63 ± 3.59) 41 (93) 104 (2.54) 3.00 (3.66 ± 3.16) Adolescents (> 12 yrs) 32 (91) 101 (3.16) 4.00 (5.30 ± 5.20) Abbreviations: mo = months; yrs = years; std = standard deviation.

52 38 Table 6. Exposure to the 10 most Commonly Prescribed Antimicrobial Agents: All Age Categories Antimicrobial Agent % of Neonates Exposed (n = 22) % of Infants Exposed (n = 44) % of Children Exposed (n = 44) % of Adolescents Exposed (n = 35) Ampicillin-sulbactam Azithromycin Cefazolin Cefepime Ceftriaxone Gentamicin Nafcillin Piperacillin-tazobactam Trimethoprim-sulfamethoxazole Vancomycin

53 39 Table 7. Indications for Antimicrobials Therapy Indication Indication Sub-categories *Number of Patients (%) (n = 134) Number of Antimicrobial Prescriptions (%) (n = 437) Median Treatment Days (mean ± std) Sum of Treatment Days (%) (n = 2084) Empiric Empiric 60 (44.8) 192 (43.9) 4 (4.5 ± 3.5) 868 (41.7) Prophylactic Prophylactic (total) 101 (75.4) 186 (42.6) 3 (4.4 ± 4.1) 815 (39.1) Prophylactic - UTI 5 (3.5) 7 (1.6) 3 (7.7 ± 7.8) 54 (2.6) Prophylactic - other 14 (10.4) 24 (5.5) 4 (4.6 ± 3.3) 103 (4.9) Prophylactic - 15 (11.2) 31 (7.1) 2 (4.5 ± 6.2) 139 (6.7) immunocompromised Prophylactic - perioperative 77 (57.5) 124 (28.4) 3 (4.2 ± 3.2) 519 (24.9) Targeted Targeted 23 (17.2) 41 (9.4) 8 (8.2 ± 5.3) 336 (16.1) Unknown - 8 (6.0) 18 (4.1) 3 (3.6 ± 2.7) 65 (3.1) Note: * Number of patients in each category is greater than 134 because some patients received antimicrobial agents for more than one indication. The total number of prophylactic treatment days was 917. Abbreviations: std = standard deviation

54 40 Table 8. Characteristics of Long-stay and Short-stay Patients in the PICU Variables Description Short-stay Patients n = 129 Long-stay Patients n = 16 P-value Gender Female 55 (42.6%) 7 (43.8%) 0.93 Age Categories Neonates (0 1 mo) 15 (11.6%) 7 (43.8%) 0.03 Infants (> 1 mo 2 yrs) 42 (32.6%) 2 (12.5%) 0.17 Children (> 2 yrs 12 yrs) 43 (33.3%) 1 (6.3%) 0.05 Adolescents ( > 12 yrs) 29 (22.5%) 6 (37.5%) 0.31 Ethnicity White 97 (75.2%) 14 (87.5%) 0.43 Other 32 (24.8%) 2 (12.5%) Invasive Devices Mechanical ventilation 73 (56.6%) 16 (100%) Foley catheter 83 (64.3%) 14 (87.5%) 0.09 Central venous catheter 78 (60.5%) 16 (100%) Reasons for Admission Cardiac operations 36 (27.9%) 10 (62.5%) 0.01 Other 59 (45.7%) 6 (37.5%) Neurosurgery 13 (10.1%) - Respiratory distress 11 (8.5%) - Pneumonia 10 (7.8) - Length of Stay in Median length of stay in the PICU in 3 (4.6 ± 3.7) 34.5 (43.6 ± 25.9) < * PICU days (mean ± std) Range of stay in the PICU in days Sum of length of stay in the PICU in days Survival Discharged from the PICU alive 126 (97.7%) 14 (87.5%) 0.09

55 41 Table 8 continued Variables Description Short-stay Patient n = 129 Culture by Sites At least one microbiologic culture was positive for a bacterial pathogen Long-stay Patient n = 16 P-value 43 (33.3%) 14 (87.5%) 0.04 Number of patients with 1 positive blood culture 3 (2.3%) 13 (81.3%) < Number of patients with 1 positive tracheal aspirates 18 (14.0%) 12 (75.0%) < Number of patients with 1 positive cerebrospinal 4 (3.1%) - - fluid culture Number of patients with 1 positive urine culture 1 (0.8%) 3 (18.8%) - Number of patient with 1 positive other culture 9 (7.0%) 5 (31.3%) - Antimicrobials Number of Antimicrobial Courses by Indication Number of Antimicrobial Prescriptions by Indication Number of patients who received at least one 119 (92.3%) 15 (93.8%) 0.59 antimicrobial agent Number of patients who received empiric 47 (36.4%) 13 (81.3%) Number of patients who received prophylactic 101 (78.3%) 10 (62.5%) 0.27 Number of patients who received targeted 17 (13.2%) 6 (37.5%) 0.03 Number of empiric courses 51 (31.7) 20 (46.5) 0.10 Number of prophylactic courses 95 (59.0) 13 (30.2) Number of targeted courses 15 (9.3) 10 (23.3) 0.03 Number of empiric prescriptions 122 (39.6) 70 (63.1) 0.03 Number of prophylactic prescriptions 158 (51.3) 28 (25.2) Number of targeted prescriptions 28 (0.09) 13 (0.12) 0.54

56 42 Table 8 continued Variables Description Short-stay Patient n = 129 Rate of Antimicrobial exposure Long-stay Patient P-value n = 16 Empiric courses per 100-patient days Prophylactic courses per 100-patient days < Targeted courses per 100-patient days Treatment days per 100-patient days < Abbreviations: std = standard deviation; n = number

57 43 Table 9. Patients Who Met Sepsis Criteria and Those Who Did Not Meet Sepsis Criteria among the Subcategory of Patients Whose First Course of Antimicrobials Was Empiric Variable Septic Non-Septic P-value n=15 n = 30 Number of Patients 15 (36.6%) 30 (63.4%) - Median age in years (mean ± std) Median weight in kg (mean ± std) Median length of stay in hospital in days (mean ± std) Median length of stay in PICU in days (mean ± std) Median Temperature ( C) Median Heart Rate (beats/min) Median Respiratory Rate (breaths/min) Median White Blood Count (kcells/mm 3 ) Median Neutrophil Count (kcells/mm 3 ) Median Band Count (kcells/mm 3 ) 7.9 (7.6 ± 6.3) 1.5 (5.0 ± 6.5) (32.0 ± 28.8) 9.0 (22.0 ± 24.9) (26.9 ± 58.5) 13.0 (20.8 ± 20.4) (12.9 ± 26.9) 6.0 (12.8 ± 19.1) (37.5 ± 2.0) 37.4 (37.3 ± 1.1) (141.5 ± 40.8) (135.7 ± 30.6) (36.0 ± 20.7) 25 (30.4 ± 13.9) (18.6 ± 11.5) 12 (34.9 ± 10.4) (12.3 ± 9.6) 8.8 (8.7 ± 5.5) (2.7 ± 2.6) 0.9 (1.6 ± 1.6) 0.04 Median C-reactive Protein (mg/dl) 4 (11 ± 1.2) 0.9 (3.9 ± 0.5) 0.47 Abbreviations: std = standard deviation; kg = kilograms; C = degrees Celsius; min = minutes; kcells = 1000 cells; mm 3 = cubic millimeters; mg = milligrams; dl = deciliters; mmhg = millimeters of mercury

58 Figure 1. Duration of Stay in the PICU 44

59 45 Figure 2. Indications for the 10 most Commonly Prescribed Antimicrobial Agents Empiric Prophylactic Targeted Vancomycin TMP/SMZ Pip-tazo Antimicrobial agent Naficillin Gentamicin Ceftriaxone Cefepime Cefazolin Azithromycin Amp-sulb % 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Percent

60 46 CHAPTER 3: FACTORS THAT PEDIATRIC INTENSIVISTS CONSIDER WHEN DECIDING TO START ANTIMICROBIAL AGENTS 3.1 Introduction Patients hospitalized in pediatric intensive care units (PICUs) are usually critically ill. Patients may be admitted to PICUs for treatment of acute medical illnesses or exacerbations of chronic conditions. Patients may also be admitted after surgical procedures or trauma. Patients in PICUs often require invasive devices such as central venous catheters, urinary catheters, and endo-tracheal tubes that increase their risk of hospital-acquired infections. Pediatric intensivists therefore, monitor their patients closely for infection and often initiate empiric antimicrobial treatments for patients who have signs and symptoms that resemble those of bacterial infections because these patients are at high risk of dying if not treated quickly (32, 47, 97). Viral infections, congestive heart failure, and sepsis from other causes can mimic infections caused by bacteria. Therefore, pediatric intensivists may prescribe empiric antibacterial therapy for some patients who do not need these agents. The principal investigator did not identify any studies describing factors that pediatric intensivists consider when deciding to start empiric antimicrobial therapies for patients in PICUs. Toltzis and colleagues found that most patients with fever in their PICU received empiric antimicrobial therapy (33). Hepler and colleagues (98) studied the physicians rationale for starting empiric antimicrobials for adult patients in a Veterans Affairs (VA) medical center. Common rationale included: (1) the patient might become septic; (2) an infection was causing an exacerbation of the patient s underlying disease; (3) an infection was causing the patients clinical status to decline; (4) the patients underlying condition or fragile clinical status predisposed the patient to a severe

61 47 infection; (5) the site of possible infection put the patient at risk of a severe infection. However, these studies were not designed to assess factors pediatric intensivists consider when starting empiric antimicrobial agents. If we understood these factors better, we might be able to improve empiric antimicrobial prescribing in PICUs. The purpose of this quality improvement study was to identify factors that the pediatric intensivists in the University of Iowa Hospitals and Clinic s (UIHC) PICU consider when prescribing antimicrobial agents, especially agents for empiric use. 3.2 Methods Study design: Antimicrobial Assessment Form This quality improvement study was conducted in the UIHC s PICU from February 1, 2007 to July 31, In November 2006, the principal investigator presented the descriptive data on antimicrobial use in the UIHC s PICU (Chapter 2) to the pediatric intensivists who were not surprised that 92% of the patients who stayed longer than 24 hours in the PICU received antimicrobial agents. However, they agreed that empiric use of antimicrobial agents could be decreased and, thus, they agreed to complete antimicrobial assessment forms (AA) when they started patients on antimicrobial therapy. The principal investigator, his academic advisor, and the pediatric intensivists designed and implemented the AA form as part of a quality improvement project. The principal investigator adapted an antimicrobial order form developed by Durbin and colleagues (99) for use in this study. The current study s AA form asked the intensivists for: (1) antimicrobials prescribed, (2) possible diagnoses requiring antimicrobial therapy, (3) vital signs (temperature, heart rate, blood pressure, and

62 48 respiratory rate), (4) laboratory test results (e.g., C-reactive protein [CRP], white blood cell count [WBC], etc.), and findings on physical examination. The pediatric intensivists and infectious disease specialists critiqued the AA form (see description in Data Collection Section 3.2.2) and the principal investigator revised the AA form based on their input. The principal investigator presented the finalized AA form to the pediatric intensivists, the nurse practitioners, and a pharmacist in January Physicians began completing the form on February 1, 2007 and stopped on July 31, The principal investigator encouraged the residents or the fellows on duty to document their rationale for therapy on the AA forms within 24 hours of prescribing antimicrobial agents. A resident or a fellow who completed the AA form gave the form to the attending physician who verified the data and signed the form. To improve adherence with using the AA form, the principal investigator visited the PICU weekly and reminded the residents and fellows to fill out the intervention forms Study Design: Case-Control Study Pediatric intensivists filled out the AA form when initiating empiric antimicrobial therapy but they did not fill out this form for patients who did not receive antimicrobial agents. Thus, the form provided information about case patients but not about control patients. Thus, the principal investigator conducted a case-control study to examine the relationship between clinical infection parameters (such as temperature, heart rate, systolic blood pressure, diastolic blood pressure, etc.) and empiric antimicrobial treatments in patients hospitalized in the PICU. Case patients were hospitalized in the UIHC s PICU for more than 24 hours between January 1, 2005 and June 30, 2005 or

63 49 between February 1, 2007 and January 31, 2008, and received at least one course of empiric antimicrobial treatment during the PICU hospitalization. In addition, patients were eligible to be cases if they: (1) had not received prior prophylactic or targeted antimicrobial treatments during their stay in the PICU and (2) did not have a confirmed infection, at the time they were identified as case patients. The principal investigator matched one control subject to each case patient based on age and time from admission until the case patient s first empiric antimicrobial treatment. Control patients did not receive any antimicrobial agents before the PICU day (of his or her stay) when the case received his or her first day of empiric antimicrobial treatments. For example, suppose a five year old patient received an empiric antimicrobial treatment starting on day three of his or her PICU admission. The principal investigator chose a five year old control subject who had not received any antimicrobial agents during the first three days of his or her PICU stay. Thus, a patient who received prophylactic antimicrobial treatment on day four or after could serve as a control but a patient who received prophylactic treatment on days one, two, or three would not be eligible Data Collection: Antimicrobial Assessment Form The principal investigator collected the completed AA forms every Monday from a designated office within the PICU. The principal investigator entered the data into an ACCESS database designed for this purpose Data Collection: Case-control Study The principal investigator abstracted the medical records of patients who stayed in the PICU more than 24 hours between January 1, 2005 and June 30, 2005 or between February 1, 2007 and January 31, 2008 to obtain information on their demographics,

64 50 antimicrobial treatments, and clinical parameters. The principal investigator obtained this information during the three study periods: period A (January 1, 2005 to June 30, 2005), period B (February 1, 2007 to July 31, 2007), and period C (August 1, 2007 to January 31, 2008). A hospital epidemiologist who, is also an adult infectious disease specialist, assessed all abstracted information and categorized antimicrobial use as either empiric, prophylactic, or targeted. Two research assistants helped abstract additional information on controls from the patients medical records Definitions Empiric antimicrobial use was defined as antimicrobial therapy begun when a physician suspected infection but (1) microbiologic, clinical, and laboratory data were pending or did not identify a site of infection or (2) the patient did not meet the Centers for Disease Control and Prevention s (CDC) criteria for nosocomial infection. Targeted antimicrobial use was defined as antimicrobial therapy when (1) the patient had an infection that was documented by a positive microbiologic or serological result and the patient was treated with an antimicrobial agent to which the etiologic agent was susceptible or (2) the patient was admitted on antimicrobial therapy as treatment for an infection and the microbiologic results were not available at the UIHC. Prophylactic antimicrobial use was defined as antimicrobial therapy initiated when (1) the patient had no evidence of infection and the patient was immunocompromised, had an anatomical defect, had recurrent infections, or had an indwelling device or (2) the patient was scheduled for a surgical procedure and antimicrobial therapy was given to prevent surgical site infections. The following sub-categories of prophylactic antimicrobial use were defined: prophylactic-urinary tract infections (UTI) was therapy to prevent further

65 51 UTIs in a patient with a history of UTIs; prophylactic-immunocompromised was antimicrobial therapy initiated for patients who were immunosuppressed due to medications or to their underlying disease; prophylactic-perioperative was antimicrobial therapy initiated to prevent surgical site infections; prophylactic-other was prophylactic antimicrobial therapy that did not meet the definitions for the other subcategories of prophylactic treatments. The principal investigator considered a patient exposed if he/she received one or more dose(s) of one or more antimicrobial agents. The principal investigator adapted the definitions of antimicrobial prescriptions and antimicrobial courses used by Bergmans and colleagues (23). The principal investigator defined an antimicrobial prescription as the initiation of one antimicrobial agent. Thus, if a physician wrote an order for gentamicin and ampicillin-sulbactam, the patient would have received two prescriptions. The principal investigator defined an antimicrobial course as an episode in which one or more antimicrobial agents were prescribed, either consecutively or in combination for prophylaxis or to treat a suspected or documented infection (23) Study size: Case-control Study To determine the number of cases and controls needed for the study, the principal investigator used the isographs of constant sample size for paired case-control studies by Dupont (100) (see Appendix C for the isographs). The principal investigator assumed that the exposure prevalence for control subjects (pp 0 ) (i.e., the proportion of control subjects who received empiric antimicrobials after they were used as controls) would be 0.3 and the correlation coefficient for exposure between matched subjects (φφ) would be 0.2. If we wish to detect an odds ratio of 2.5 with a power of 0.8 and αα of 0.05, then we should

66 52 select 100 case patients for a study with a paired design. The principal investigator selected 101 case patients and 101 control patients for this study Dependent variable Empiric therapy (see section 3.2, for the definition) was the dependent variable or outcome for this case-control study Independent variables The principal investigator recorded the following independent variables: temperature continuous variable measured in degrees Celsius ( C); heart rate continuous variable measured in beats per minute; respiratory rate continuous variable measured in breaths per minute; systolic and diastolic blood pressure continuous variables measured in millimeters of mercury (mmhg); white blood count (WBC), band count, and the neutrophil count continuous variables measured in a thousand cells per cubic millimeter (kcells/mm 3 ); and C-reactive protein (CRP) continuous variable measured in milligrams per deciliter (mg/dl); positive tracheal aspirate culture (yes or no); and chest radiograph had evidence of pneumonia (yes or no) Statistical Methods The principal investigator used frequency tables and basic combinatorial analysis to describe factors that the pediatric intensivists checked on the AA form. The principal investigator used kappa coefficients to assess whether the information provided by the pediatric intensivists on the AA form matched the abstracted data from the patients medical records. The principal investigator used analysis of variance (ANOVA) to compare the means of continuous variables and the chi-square test of proportions for categorical

67 53 variables abstracted from the medical records. For the case-control study, the principal investigator used conditional logistic regression to quantify possible associations between empiric antimicrobial treatment and infection parameters such as, temperature, systolic and diastolic blood pressure, white blood count, etc. The principal investigator categorized six of the nine continuous variables. Thus, he fitted univariate models for categorical and continuous variables. The principal investigator fitted multivariable models, which included some continuous variables and some categorical variables that were statistically significant at the 95% level in the univariate analyses. The principal investigator analyzed the case-control data with proc logistic procedure in SAS. Proc logistic provides the capability of model-building and performs conditional logistic regression analysis for matched case-control studies and exact conditional logistic regression analysis. He utilized the backward elimination method to select variables that entered the final model and the Akaike Information Criterion (AIC) to choose the best model. 3.3 Results Antimicrobial Assessment Form Residents and fellows completed AA forms for 52 patients and 68 antimicrobial courses. Of the 68 antimicrobial courses, 27 (39.7%) were prophylactic and 41 (60.3%) were empiric or targeted. Twenty-three patients received the 27 courses of prophylactic antimicrobials and 35 patients received the 41 courses of empiric or targeted antimicrobials. We combined the latter two indications because the AA form did not distinguish between empiric and targeted antimicrobial treatments. Compliance with the

68 54 filling of the AA form was 90% in February when the intervention began and only 3% in June and July when the intervention ended (Table 10) Vancomycin, gentamicin, and ceftriaxone were the antimicrobials most commonly reported on the AA form (Figure 3) and accounted for 53.1% of all antimicrobial prescriptions reported. These agents accounted for 42% of the prescriptions documented in the medical records during the study period. The physicians checked a single factor for 17 (63.0%) of 27 antimicrobial courses reported on the AA form as prophylactic and multiple factors for 10 (37.0%). Common single factors were the presence of a chest tube (18.5%) and perioperative (25.9%). Two common combinations of factors were: (1) the presence of a chest tube and a central venous catheter (7.4%); (2) compromised immune status and the presence of a central venous catheter (11.1%) (Table 11). Three of five (60.0%) patients with chest tube as a reason for prophylactic therapy had cardiac operations, one patient had a surgical procedure to repair a diaphragm hernia, and one patient had a chest tube placed to treat a pneumothorax. Thus, these courses were actually perioperative prophylaxis. The physicians checked single factors for 4 (9.8%) of the 41 antimicrobial courses reported on AA forms as empiric or targeted and multiple factors for 37 (90.2%) (Table 12). Overall, the combination of elevated CRP, elevated WBC count, and elevated temperature was chosen 12 times and the combination of elevated heart rate and elevated CRP was chosen 14 times. Pediatric intensivists checked multiple factors more frequently for antimicrobials used for empiric or targeted therapy (90.2%) than for antimicrobials used for prophylactic therapy (37.0%) (P = 0.01).

69 55 Overall, for 68 courses of antimicrobials, the physicians checked the following five factors most frequently: (1) elevated temperature (47.1%), elevated CRP (45.6%), elevated heart rate (33.8%), elevated WBC count (32.4%), and elevated respiratory rate (26.5%). Physicians indicated on 22 AA forms that they suspected the patients had pneumonia. However, physicians checked abnormal chest radiograph on only 7 (32.0%) of these forms. In addition, physicians also checked the following factors on these 22 AA forms: elevated temperature 12 (54.5%) times, elevated WBC 6 (27.3%) times, and the combination of elevated temperature and elevated CRP 7 (31.8%) times. We found evidence that 13 (59.0%) of these 22 patients met the CDC criteria for pneumonia. Three criteria were checked frequently on the AA form: (1) ventilator status, (2) a new infiltrate on chest radiograph, and (3) a Gram stain of sputum reveals polymorphonuclear leukocytes (PMNs) and a predominant organism. Physicians indicated on 14 AA forms that they suspected that patients had bloodstream infections (BSI). The physicians checked elevated temperature and elevated CRP for 9 (64.3%) of these 14 patients, elevated CRP for 3 (21.4%) of these patients, and elevated temperature for 2 (14.3%) of these patients. Physicians indicated on six AA forms that they suspected that patients were septic. The physicians checked abnormal heart rate for 4 (66.6%) of these patients, elevated CRP for three (50%) of these patients, elevated temperature for 2 (33.3%) of these patients, left shift for 2 (33.3%) of these patients, and acidosis for 1 (16.7%) of these patients. To verify the accuracy of information pediatric intensivists reported on the AA forms, the principal investigator compared data on temperature, CRP, and heart rate

70 56 reported in the medical records with the information on the AA forms. The information in the medical records and that provided by the pediatric intensivists were in good agreement as indicated by Cohen s Kappas of 0.72 for temperature, 0.81 for CRP, and 0.93 for heart rate Medical Record Review During the intervention period, 173 patients stayed more than 24 hours in the PICU. The patient population was similar to that of the pre-intervention period (January to June 2005). However, the proportion of neonates (22 of 145 patients) was higher in the pre-intervention period than the intervention period (11 of 173 patients) (P = 0.02). Additionally, patients were more likely to have Foley catheters (66.9% vs. 24.9%, P < ), central venous catheters (64.8% vs. 51.4%, P = 0.02), and parenteral nutrition (2.1% vs. 7.5%, P = 0.01) during the pre-intervention period compared with the intervention period (Table 13). One hundred and sixty-two (93.6%) of the 173 patients during the intervention period received antimicrobial treatments compared with 134 (92.4%) of 145 patients in the pre-intervention period (P = 0.84). The pediatric intensivists prescribed 24 different antimicrobial agents for 436 antimicrobial prescriptions during the intervention period. Of these antimicrobial prescriptions, 193 (44.3%) were empiric, 176 (40.4%) were prophylactic, and 59 (13.5%) were targeted. The physician reviewer could not determine why 8 (2%) prescriptions were given. The proportion of each indication relative to all prescriptions was similar between the pre-intervention period and the intervention period (empiric: 43.9% vs. 44.3%, P = 0.98; prophylactic: 42.6% vs. 40.4%, P = 0.56; and targeted: 9.4% vs. 13.5%, P = 0.07). Pediatric intensivists consulted the pediatric

71 57 infectious disease specialists infrequently during both the pre-intervention period (8.3%) and the intervention period (10.4%) (P = 0.33) Case-control Study One hundred and one case patients who received empiric antimicrobial treatments were matched to 101 control subjects. By univariate analysis, elevated temperature, elevated heart rate, the availability of WBC count test results, the availability of CRP test results, the availability of neutrophil count test results, and whether the chest radiograph showed evidence of pneumonia were significantly associated with receiving empiric antimicrobial treatments (Table 14). The odds of receiving empiric treatments for case patients who had a 1 C increase in body temperature was 1.84 times greater than the odds for control patients. The odds for receiving empiric treatments for case patients who had WBC count results available was 6.67 times greater than the odds for control patients and the odds for receiving empiric treatments for case patients who had CRP test results available was 3.92 times greater than the odds for control patients. Six variables were eligible for inclusion in the multivariable model based on the 0.05 significance level. The principal investigator developed a multivariable model with temperature, heart rate, the availability of WBC count test results, the availability of CRP test results, the availability of neutrophil test results, and whether a chest radiograph showed evidence of pneumonia. When the investigator used the backward elimination method in proc logistic procedure, elevated temperature and the availability of CRP test results were the two variables that met the 0.05 significance level for inclusion in the multivariable model. The principal investigator forced the availability of a WBC count result into the model because the WBC count is an important infection parameter. The

72 58 principal investigator assessed whether age, a matching variable, interacted with any of the variables in the multivariable model. None of the interactions was significant at the 0.05 significance level. Thus, the principal investigator removed the interaction terms from the main effects. Table 15 presents the final multivariable model of the association between empiric treatments and infection parameters. Case patients with CRP test results available were approximately 3 times more likely to receive empiric antimicrobial treatments than their matched controls, controlling for temperature and the availability of WBC count results (and age and length of stay before receiving the first antimicrobial agent, via matching). Case patients with a unit increase in body temperatures (1 C) were about 2 times more likely to receive empiric antimicrobial treatment than their matched controls who had normal temperatures, controlling for the availability of CRP and WBC count results (and age and length of stay before receiving first antimicrobial agent, via matching). 3.4 Discussion The goal of this study was to describe factors that the pediatric intensivists working in the UIHCs PICU consider when deciding to start antimicrobial agents, especially those antimicrobials that are prescribed empirically. Prior studies on physicians antimicrobial prescribing habits have described physicians attitudes, knowledge, and beliefs with respect to antimicrobial use (98, ). In general, physicians prescribing decisions are influenced by a number of factors. Raisch categorized these factors as direct (formularies, prescribing restrictions, required consultation); indirect (advertisements, visits by pharmaceutical sales persons, opinions

73 59 of colleagues, scientific data from randomized controlled trials, and medical training); and individual and practice factors (demographics, case mix, organizational structure, and others) (102). Raisch suggested that individual factors, practice factors, and those factors categorized as indirect, affect prescribing decisions by influencing the physicians thought processes (102). Lambert and colleagues described six factors associated with physicians decisions to prescribe amoxicillin, amoxicillin with clavulanate, clarithromycin, cefaclor, cefuroxime, erythromycin, and trimethoprim-sulfamethoxazole (101). The investigators assumed that prescribers beliefs about possible outcomes of therapy affect their decisions about antimicrobial therapy. To test this hypothesis, they wrote scenarios and listed various possible outcomes of therapy, such as therapeutic effects, side effects, and cost-effective for the medical group. The authors asked physicians to estimate (on a Likert scale: -3 for extremely unlikely to 3 for extremely likely) the probabilities that particular outcomes would result from each antimicrobial agent (101). The two most frequent reasons physicians chose to describe their decisions to prescribe any of these six antimicrobial agents were cures the underlying infectious disease and quickly lessens uncomfortable or painful symptoms (101). While much work has been done to understand physicians prescribing behavior in general, comparatively little work has been done to identify clinical infection parameters, such as vital signs, laboratory test results, and microbiologic test results, that pediatric intensivists consider when prescribing antimicrobial agents. The current study had two principal findings. First, data from the AA forms suggested that pediatric intensivists frequently considered combinations of elevated

74 60 temperature, elevated CRP, and elevated WBC count when initiating empiric or targeted antimicrobial treatment, while perioperative coverage was the major reason that they initiated prophylactic antimicrobial treatment. Second, it appeared from the case-control data that CRP was a test the pediatric intensivists often ordered before they prescribed empiric antimicrobial agents. Several studies have evaluated the usefulness of different markers, such as CRP ( ) and procalcitonin, (107, 109, 110) for diagnosing infections and for identifying patients at risk of infections. Toikka and colleagues (107) assessed serum procalcitonin, CRP, and interleukin-6 (IL-6) for distinguishing bacterial and viral infection among children hospitalized for radiologically confirmed community-acquired pneumonia. They found that children with bacterial pneumonia had significantly higher procalcitonin levels (median 2.09 ng/ml vs ng/ml, P = 0.019) and CRP concentrations (96 mg/l vs. 54 mg/l, P = 0.008) than those with viral pneumonia. The investigators used receiver operating characteristic (ROC) curves to assess the sensitivities and specificities of procalcitonin and CRP as diagnostic tests for bacterial pneumonia. They reported that a cutoff value of 2.0 ng/ml for procalcitonin had a sensitivity of 50% and a specificity of 80%; whereas, a cutoff value of 7.0 ng/ml for procalcitonin had a sensitivity of 19% and a specificity of 98%. A cutoff value of 80 mg/l for CRP had a sensitivity of 59% and a specificity of 68% and a 150 mg/l cutoff value for CRP had a sensitivity of 31% and a specificity of 88%. On the basis of these results, the investigators concluded that measurement of serum procalcitonin and CRP had little value in [differentiating] bacterial and viral pneumonia in children. However, they noted that a very high value for serum procalcitonin or CRP suggested that the etiology was bacterial.

75 61 Flood and colleagues (108) conducted a meta-analysis to assess whether serum CRP differentiated children with bacterial pneumonia from those with nonbacterial pneumonia. Eight studies met their criteria for inclusion in their meta-analysis. The results of these studies indicated that children with bacterial pneumonia were significantly more likely to have serum CRP concentrations exceeding mg/l than were children with nonbacterial infections (odds ratio [OR] = 2.58, 95% confidence interval = ). However, when the investigators used the mean odds ratio from their analysis as the positive likelihood ratio, the predictive value positive of a serum CRP greater than mg/l was only 64%. Paran and colleagues (104) found that the CRP velocity (defined as the ratio between CRP on admission and the number of hours since the onset of fever) improved their ability to differentiate between febrile bacterial infections and non-bacterial febrile illnesses among adult patients (age 18 years) who presented to their emergency department with fever. The investigators reported that patients who had febrile bacterial illness had a mean CRP velocity of 3.61mg/l/hour compared with 0.41 mg/l/hour in patients who had non-bacterial febrile illnesses. They used an ROC curve to assess the efficacy of CRP velocity in differentiating between bacterial and non-bacterial febrile illness. The area under the curve (AUC) for CRP velocity was (95% confidence interval = to 0.924) compared with (confidence interval = to 0.850) for the CRP value alone. The investigators concluded that CRP velocity might be a useful test in patients with a known time at which fever began. Povoa and colleagues (105) conducted a prospective, observational study to assess the value of CRP, temperature, and WBC count measurements for the diagnosis of

76 62 infection in critically ill patients in an adult medical-surgical ICU. The authors monitored CRP, temperature, and WBC count daily for 76 infected and 36 non-infected patients in this ICU. A CRP concentration of > 8.7 mg/dl and a temperature of > 38.2 C were associated with infection, with sensitivities of 93.4% and 54.8% and specificities of 86.1% and 88.9%, respectively (105). The authors found that the combination of CRP and temperature increased the specificity for a diagnosis of infection to 100% (105). In the current study, data from the AA form suggested that the pediatric intensivists frequently considered a combination of an elevated CRP, elevated WBC count, and elevated temperature as their reasons for starting empiric or targeted antimicrobial therapies. However, the factors physicians checked varied substantially among three subsets of patients: those for whom the physicians checked pneumonia, bloodstream infection, or sepsis as reasons for starting empiric or targeted antimicrobial treatments. When physicians suspected pneumonia (22 of 41 patients; 53.7%) the factors checked most frequently were consistent with some but not all of the CDC s criteria for nosocomial pneumonia. For example, the physicians also checked elevated temperature (54.5%), abnormal chest radiograph (32.0%), elevated WBC (27.3%), and polymorphonuclear leukocytes on the sputum Gram stain (13.6%). Whereas, when physicians suspected bloodstream infection (14 patients) the factors they checked most frequently were elevated temperature (2/14; 14.3%), elevated CRP (3/14; 21.4%), and a combination of elevated temperature and elevated CRP (9/14; 64.3%). Sepsis remains a major cause of morbidity and mortality among neonates and infants (96). Neonates and infants with sepsis can deteriorate rapidly. Thus, physicians often obtain diagnostic tests, including cultures and laboratory tests, from patients they

77 63 suspect have sepsis and then they start empiric antimicrobial therapy while waiting for the culture and susceptibility results. In the current study, the pediatric intensivists checked sepsis on 6 (14.6%) of 41 AA forms on which they reported empiric or targeted antimicrobial therapies. On these same forms, they also checked other factors such as, elevated heart rate (4/6; 66.6%), elevated CRP (3/6; 50.0%), elevated temperature (2/6; 33.3%), left shift (2/6; 33.3%), and acidosis (1/6; 16.7%). These factors are consistent with those listed by Guzman-Cottrill and colleagues in their chapter on sepsis in the Sarah Long s textbook Principles and Practice of Pediatric Infectious Disease. They state that fever, tachycardia, and tachypnea are the most common physiologic abnormalities associated with sepsis (96) and that the total peripheral WBC count, procalcitonin, and CRP are biologic markers of sepsis in children (96). None of the studies on CRP and infection were conducted in PICUs and none of them evaluated the factors physicians consider when deciding to initiate empiric antimicrobial therapy. However, our data suggest that the pediatric intensivists in the UIHC s PICU probably were aware of studies assessing the usefulness of CRP and that they ordered this test to help differentiate between acute infections and non-infectious conditions in patients they treated with empiric antimicrobials. In the current study, physicians frequently checked single factors for prophylactic treatments but frequently checked multiple factors for empiric or targeted treatments. The reasons for prophylactic treatments are defined more clearly than the reasons for empiric treatments. For example, the IDSA guideline on perioperative antimicrobial prophylaxis recommends the use of cefotetan or cefoxitin for operations involving the distal ileum, appendix, or colon, and to use cefazolin for all other procedures (93). However, the

78 64 reason a patients clinical condition deteriorated may not be evident. A patient who has fever, tachycardia, and tachypnea may have an infection, acute respiratory distress syndrome (ARDS), or sepsis caused by conditions other than infections. Because, the etiology is often uncertain, physicians consider a broader range of factors when starting empiric treatment than when starting prophylactic treatment. Data from the case-control study indicated that the availability of CRP test results and the patients temperature were independent predictors of empiric antimicrobial treatments among patients in the UIHC s PICU. Cases were significantly more likely than controls to have CRP test results available (65% vs. 31%); suggesting that the UIHC s pediatric intensivists frequently ordered CRPs before initiating empiric antimicrobial agents. As noted previously, other investigators have suggested that an elevated CRP is associated with bacterial infections ( ). In Povoa s study, a temperature of 38.2 C performed well for the diagnosis of infection, with an AUC of 0.75 (105). In our case-control study, temperature values were available for 99% of cases and 95% of controls but only 26% of cases and 16% of controls had temperatures greater than 38 C or less than 35 C. Peres Bota and colleagues developed the Infection Probability Score (IPS) (111) to improve the diagnostic accuracy for infections in adult ICUs. The final IPS model comprised six variables: temperature (> 37.5 C), heart rate (> 140 beats/min), respiratory rate (> 25 breaths/min), WBC count (< 5 x 10 3 /mm 3 or > 12 x 10 3 /mm 3 ), CRP (> 6 mg/dl), and the Sequential Organ Failure Assessment (SOFA) score. The combination of heart rate, CRP, and temperature had the highest predictive values for infection (72.2% predictive value positive and 95.9% predictive value negative). The

79 65 current case-control study was not designed to assess predictors of infection but did assess whether temperature, heart rate, respiratory rate, WBC count, and CRP or the availability of results for these measures were associated with empiric antimicrobial treatments. The availability of CRP test results and elevated temperature were associated with receiving empiric antimicrobial treatments in the UIHC s PICU but heart rate, respiratory rate, and WBC count were not. Hepler and colleagues (98) interviewed physicians at a 370-bed, universityaffiliated VA hospital within 72 hours after they initiated empiric antimicrobial treatments to assess their reasons for initiating therapy. The physicians frequently explained their decisions to initiate empiric antimicrobial therapy based on concerns about possible bad outcomes such as, (1) the patient might become septic; (2) an infection was causing an exacerbation of the patient s underlying disease; (3) an infection was causing the patients clinical status to decline; (4) the patient s underlying condition or fragile clinical status predisposed the patient to a severe infection; (5) the site of possible infection put the patient at risk of a severe infection. The current study was not designed to assess physicians beliefs or attitudes; it was designed to assess objective factors physicians consider when starting antimicrobial treatment. However, the pediatric intensivists commented when we were designing this study that they all had cared for patients who died quickly of unrecognized infections. These experiences caused them to worry that this could happen again. Thus, we suspect, but cannot prove, that other factors that we did not measure influenced their decisions to start empiric antimicrobial therapy.

80 66 Adherence to completing the AA form was high when the study first started and decreased as the study progressed. Overall, five attending physicians signed the AA forms; however, one physician (the coordinator of this study in the PICU) verified and signed approximately 50% of the AA forms. The PI did not see any correlation between the attending physicians schedules and adherence to completing the AA form. In fact, adherence to completing the AA form was low while the pediatric intensivist who coordinated this study was on service late in the intervention period. The principal investigator does not know why fewer forms were completed when other attending physicians were on service than when the aforementioned coordinator was on duty. Perhaps these attending physicians did not encourage the residents or fellows to complete the AA forms. Several factors may explain why we observed low compliance with completing the intervention form. First, physicians in the PICU were not obliged to complete the AA forms because completing the forms was voluntary. Second, pediatric intensivists are very busy caring for critically ill children. Therefore, they already had a heavy workload and they might have considered completing an AA form to be a burden. Similarly, pediatric intensivists primary goal is to save the lives of critically ill children. Thus, they might not have been willing to complete the intervention form because they did not think it would help them achieve their goal. Perhaps, incorporating the AA form into an existing electronic ordering system or into their normal workflow would encourage pediatric intensivists to use AA forms. The current study was a pilot study and further work is needed to assess the use of AA forms for studying objective factors physicians in PICUs consider when deciding to

81 67 start empiric antimicrobial therapy. Future research should incorporate an AA form into the physicians workflow in order to facilitate the use of AA forms. To our knowledge, this is the first study to evaluate objective factors that pediatric intensivists consider when deciding to start empiric antimicrobial therapy and factors associated with empiric therapy. We sought to identify factors pediatric intensivists considered when deciding to prescribe empiric antimicrobial treatments (AA forms) and factors associated with empiric therapy (case-control study). Data from the AA forms suggested that pediatric intensivists in the UIHC s PICU often considered elevated CRPs, elevated WBC counts, and elevated temperatures when deciding to start empiric antimicrobial therapy. Data from the case-control study suggested that temperature and the availability of CRP test results were significant predictors of empiric therapy. Cases were twice as likely to have CRP results available than were controls, suggesting that pediatric intensivists in the current study often ordered CRP test results before starting empiric antimicrobial treatment for patients they suspected of having infections (See chapter 5 for strengths and limitations of this study).

82 68 Table 10. Compliance with the Antimicrobial Assessment Form Month Received antimicrobial agents according to medical record (a) Intervention forms submitted (b) Percent compliance (b/a) February March April May June/July

83 69 Table 11. Factors Pediatric Intensivists Considered while Prescribing Prophylactic Antimicrobial Agents (Data from Antimicrobial Assessment Forms) Number of Courses Factors Percent n = 27 One factor - - Immunocompromised patient Ventricular drain present Chest tube present Perioperative Mulitple factors - - Central venous catheter/abnormal urinary tract Chest tube present/immunocompromised patient Skull fracture/perioperative Skull fracture/perioperative/ventricular drain present Immunocompromised patient/perioperative/central venous catheter present/abnormal urinary tract Chest tube present/central venous catheter present Immunocompromised patient/central venous catheter present Abbreviations: n = number

84 70 Table 12. Factors Pediatric Intensivists Considered while Prescribing Empiric or Targeted Antimicrobial Agents (Data from Antimicrobial Assessment Forms) Factors Number of Courses *n = 41 Percent One factor Surgical site infection/purulent drainage from a surgical site Urinary tract infection Elevated respiratory rate Two factors Ventriculoperitoneal shunt infection/temperature > 38 C or < 35 C Infection at another site/known or suspected immunodeficiency Three factors Pneumonia/WBC count is elevated/a Gram stain of sputum reveals PMNs and a predominant organism Lethargy/Sepsis/Left shift Lethargy/Sepsis/Age days Four factors Pneumonia/Temperature > 38 C or < 35 C /A Gram stain of sputum reveals PMNs and a predominant organism/a new infiltrate on CXR Surgical site infection/erythema and/or tenderness and/or swelling around a surgical site/purulent drainage from a surgical site/chronic illness Five factors Pneumonia/Temperature > 38 C or < 35 C /Age days /Prolonged or complicated ICU stay/chronic illness Pneumonia/Bloodstream infection/surgical site infection/left shift/elevated CRP Meningitis/Encephalitis/Neck stiffness and/or positive Kernig s sign and/or positive Brudzinski s sign/elevated CRP/Elevated CSF WBC count /Known or suspected viral infection Changed mental status/sepsis/elevated heart rate/elevated respiratory rate/elevated CRP 1 2.4

85 71 Table 12. Continued Factors Multiple factors Pneumonia/Urinary tract infection/ventilatory status/temperature > 38 C or < 35 C/Elevated respiratory rate/new onset of purulent sputum Pneumonia/Respiratory distress/temperature > 38 C or < 35 C/Elevated respiratory rate/patient on chronic steroids/chronic illness Pneumonia/Elevated heart rate/elevated respiratory rate/elevated CRP/A new infiltrate on CXR/Age days/known or suspected viral infection Pneumonia/Ventilatory status/elevated heart rate/elevated respiratory rate/a new on infiltrate CXR/Age days Pneumonia/Temperature > 38 C or < 35 C/Elevated heart rate/ Elevated CRP/Known or suspected viral infection/prolonged or complicated ICU stay/chronic illness Sepsis/Elevated heart rate/elevated WBC count/elevated CRP/Age days/prolonged or complicated ICU stay Meningitis or encephalitis /Seizures/Erythema and/or tenderness and/or swelling of a skin site (not surgical site or catheter exit site)/elevated CSF WBC count/known or suspected viral infection Number of Courses *n = 41 Percent Ventriculoperitoneal shunt infection/seizures/temperature > 38 C or < 35 C /Elevated WBC /Elevated CRP/A Gram stain of sputum reveals PMN s and a predominant organism/elevated CSF WBC count * Combinations of more than 7 factors were excluded from the table. Therefore, the number of courses does not equal 41. Abbreviations: CSF = cerebrospinal fluid, CXR = chest radiograph, PMNs = polymorphonuclear cells, WBC = white blood cell, ICU = intensive care unit, C = degrees Celsius

86 72 Table 13. Descriptive Characteristics of Patient Population Variables Description Pre-Intervention Period n = 145 (%) Intervention Period n = 173 (%) Gender Female 61 (42) 85 (49) 0.25 Age Median age in years (mean ± std) 3 (5.7 ± 6.2) 4 (6.6 ± 6.9) 0.42 Age Categories Neonates (0 1 mo) 22 (15.2) 11 (6.4) 0.02 Infants (> 1 mo 2 yrs) 44 (30.3) 64 (37.0) 0.26 Children (> 2 yrs 12 yrs) 44 (30.3) 51 (29.5) 0.96 Adolescents (> 12 yrs) 35 (24.1) 47 (27.2) 0.62 Ethnicity White 110 (75.9) 127 (73.4) 0.71 Black 10 (6.9) 10 (5.8) 0.86 Hispanic 6 (4.14) 9 (5.2) 0.86 Asian/Pacific Islander 4 (2.76) 2 (1.2) 0.53 Others 15 (10.35) 25 (14.5) 0.35 Weight Median weight in kg (mean ± std) 14 (23.4 ± 23.9) 15 (26.6 ± 25.7) 0.50 P-value Length of Hospitalization Median length of stay in hospital in days (mean ± std) Median length of stay in PICU in days (mean ± std) 10 (16.9 ± 19.7) 10 (15.7 ± 20.3) (9.0 ± 15.2) 3 (6.3 ± 7.4) 0.61 Invasive Devices Endotracheal tube 89 (61.4) 110 (63.6) 0.77 Foley catheter 97 (66.9) 43 (24.9) < Central venous catheter 94 (64.8) 89 (51.5) 0.02 Nutrition Parenteral nutrition 3 (2.1) 13 (7.5) 0.01 Abbreviations: std = standard deviation; kg = kilograms; mo = month; yrs = years

87 73 Table 14. Univariate Analysis: The Association of Empiric Antimicrobial Treatments and Infection Parameters Variable ββ OR 95% CI Number of Matched Pairs Temperature Temperature (< 35 C or > 38 C) Systolic Blood Pressure Diastolic Blood Pressure Respiratory Rate Heart Rate Heart Rate (Yes or No) White Blood Cell Count White Blood Cell Count (Yes or No) C-reactive Protein C-reactive Protein (Yes or No) Neutrophil Count (Yes or No) Chest Radiograph Consistent with Pneumonia (Yes or No) Positive Tracheal Aspirates (Yes or No) Abbreviations: ββ = estimate of the regression coefficient; OR = odds ratio; CI = confidence interval; C = degrees Celsius; Yes = test result available; No = test result not available

88 74 Table 15. Adjusted Odds Ratio for the Association of Empiric Antimicrobial Treatments and Infection Parameters* Variable ββ OR 95% CI Temperature C-reactive Protein (Yes or No) White Blood Cell Count (Yes or No) Abbreviations: ββ = estimate of the regression coefficient; OR = odds ratio; CI = confidence interval; C = degrees Celsius; Yes = test result available; No = test result not available *Number of matched pairs = 94

89 75 Figure 3. Antimicrobial Agents Reported on the Antimicrobial Assessment Form Vancomycin Gentamicin Ceftriaxone Piperacillin-tazobactam Nafcillin Metronidazole Ampicillin-sulbactam Clindamycin Cefazolin Cefepime Trimethoprim-sulfamethoxazole Rifampin Meropenem Levofloxacin Cefotaxime Linezolid Cephalexin Azithromycin Number of Prescriptions

90 76 CHAPTER 4: INTERVENTION TO IMPROVE ANTIMICROBIAL STEWARDSHIP IN A PEDIATRIC INTENSIVE CARE UNIT 4.1 Introduction Antimicrobial use is common in patients hospitalized in intensive care units (ICUs) ( ). Patients hospitalized in ICUs are at risk of developing bacterial infections (115) because they have serious underlying diseases, are treated with invasive devices, and have prolonged hospital stays (116, 117). Several reports have shown that use of antimicrobials not only promotes the emergence of bacterial resistance, but also could lead to serious adverse outcomes such as Clostridium difficile-associated diarrhea (79, 81) and adverse drug reactions (76, 77, 118). Consequently, the Centers for Disease Control and Prevention (CDC), the Society for Healthcare Epidemiology of America (SHEA), and the Infectious Diseases Society of America (IDSA) have advocated for the judicious use of antimicrobial agents and recommended guidelines to promote antimicrobial stewardship in hospitals (119). The purpose of these guidelines is to provide evidence based recommendations that hospitals can use when developing a program to enhance antimicrobial stewardship and improve the quality of care (119). Patel s systematic review of antimicrobial control strategies in hospitalized and ambulatory pediatric populations (120) found that only a few studies have evaluated strategies for improving antimicrobial use in PICUs (31, 59-61). Two of these studies described unsuccessful interventions (59, 60), and two described successful interventions (31, 61). Moss and colleagues assessed whether antimicrobial cycling would reduce colonization and infection by resistant bacteria. They found that the prevalence of

91 77 children colonized with resistant bacteria did not change during the study (29% vs. 24%, P = 0.41) (59). Toltizis and colleagues restricted ceftazidime use but the rate of infection and colonization with ceftazidime-resistant organisms increased from 1.57 isolates per 100 patient-days to 2.16 isolates per 100 patient-days (60). Ding and colleagues reported that prescription rates for third-generation cephalosporins and macrolides decreased significantly (P < 0.01) (31) when they introduced an educational program focused primarily on pediatricians, and antimicrobial guidelines that compelled physicians to obtain prior approval from a senior pediatrician before ordering restricted antimicrobials. Mullet and colleagues reported that pediatricians placed fewer antimicrobial orders per antiinfective course after a computerized decision support system was introduced (1.56 orders per antiinfective course pre intervention and 1.38 orders per antiinfective course post intervention; P < 0.01) (61). The principal investigator sought to determine whether a simple inexpensive process improvement intervention could decrease empiric use of antimicrobials in the University of Iowa Hospitals and Clinic s (UIHC) PICU. Durbin and colleagues previously (99) introduced an antimicrobial prescription form on surgical and medical wards. Physicians were required to categorize antimicrobial use as prophylactic, empiric, or therapeutic. The investigators automatically discontinued prophylactic antimicrobials after two days, empiric antimicrobials after three days, and therapeutic antimicrobials after seven days, unless the physicians renewed the orders or specified an alternate duration of treatment. Their strategy reduced the proportion of patients who received prophylactic antimicrobial treatment for elective surgery from 68% to 60% and the duration of prophylactic treatments by 2 days (99).

92 78 The principal investigator of the current study encouraged pediatric intensivists to document their rationale for prescribing antimicrobial agents on an antimicrobial assessment (AA) form. In contrast to the intervention introduced by Durbin and colleagues, the principal investigator did not require the pediatric intensivists to fill out the AA form and the pharmacy did not terminate the orders at specified times. To assess the efficacy of this intervention, the principal investigator evaluated antimicrobial use in the UIHC s PICU during the two years (January 2005 to January 2007) before the AA form was introduced, during the six-month intervention period when physicians completed the AA form (February 2007 to July 2007), and during the six-months after the intervention period when physicians no longer completed the AA form (August 2007 to January 2008). 4.2 Methods Setting This quality improvement intervention was implemented in the UIHC s PICU. The principal investigator chose the neonatal intensive care unit (NICU) to be a nonintervention, control unit because the patient populations of the PICU and the NICU overlap. The NICU generally serves neonates and newborns whereas; the PICU serves newborns to patients who are 20 years old. In addition, changes in hospital-wide policies that affect antimicrobial use will affect both units. For example, in January 2005, the Department of Pediatrics hired two infectious disease consultants who see patients in both the PICU and the NICU. The principal investigator assumed that changes in antimicrobial use related to services provided by the infectious diseases consultants would be similar in both units. The principal investigator also assumed that

93 79 neonatologists and pediatric intensivists would have a similar threshold for starting empiric antimicrobial therapy because they would be concerned that patients might die if antimicrobial therapy was delayed until the etiology of their distress or decompesation was identified. Furthermore, the neonatologists and the pediatric intesivists use specific antimicrobial agents in a similar manner. For example, both units use vancomycin, gentamicin, and piperacillin-tazobactam for empiric therapy for presumed nosocomial infections. However, the NICU staff use ampicillin, gentamicin, and metronidazole frequently for prophylactic therapy while the PICU staff frequently use cefazolin, gentamicin, and trimethoprim-sulfamethoxazole for this purpose Design The principal investigator chose a quasi-experimental, pre-post, time series design for this study because a prospective cohort study or a randomized controlled trial would take years to complete. The principal investigator did a preliminary study, from January 2005 through June 2005, to describe antimicrobial use in the PICU and to identify aspects of antimicrobial use that could be improved (Chapter 2). On November 2006, the principal investigator presented the data from the preliminary study to the pediatric intensivists who were not surprised that 92% of the patients who stayed more than 24 hours in the PICU received antimicrobial agents. However, the pediatric intensivists agreed that empiric use of antimicrobial agents could be improved and they agreed to participate in a simple process improvement intervention. Subsequently, the principal investigator, with the help of his academic advisor, developed a draft intervention form (AA form) and the pediatric intensivists and the pediatric infectious diseases consultants critiqued the form (see final version in Appendix B).

94 Intervention The principal investigator did not recruit patients for this study. Rather, the principal investigator asked pediatric residents and fellows in the PICU to fill out the AA form. The AA form had four sections: (1) the antimicrobial agents prescribed, (2) the possible diagnoses requiring antimicrobial therapy, (3) the patient s vital signs, and (4) the patient s laboratory test results and findings on physical examination. During the intervention period (February 1, 2007 to July 31, 2007), residents or fellows in the PICU were encouraged to complete the AA form within 24 hours of initiating new antimicrobial agents. The attending physicians verified that the information was adequate and both the attending physicians and the residents or fellows signed the forms. Every two weeks (Monday afternoon), the principal investigator collected all forms completed during the prior week from a designated drop-off box located in a pediatric intensivist s office. The principal investigator chose Monday afternoon to allow the residents or fellows and the attending physicians, time to complete the forms. Residents or fellows completed forms voluntarily and the principal investigator went to the PICU biweekly to encourage them to complete the forms. The principal investigator hypothesized that the process of justifying their decision might modify the pediatric intensivists prescribing patterns, particularly for empiric treatment, or might cause the physicians to stop empiric antimicrobials if cultures for bacterial pathogens were negative Definitions Empiric antimicrobial use was defined as antimicrobial therapy begun when a physician suspected infection but (1) microbiologic, clinical, and laboratory data were

95 81 pending or did not identify a site of infection or (2) the patient did not meet the Centers for Disease Control and Prevention (CDC) criteria for nosocomial infection. Targeted antimicrobial use was defined as antimicrobial therapy when (1) the patient had an infection that was documented by a positive microbiologic or serological result and the patient was treated with an antimicrobial agent to which the etiologic agent was susceptible or (2) the patient was admitted on antimicrobial therapy as treatment for an infection and the microbiologic results were not available at the UIHC. Prophylactic antimicrobial use was defined as antimicrobial therapy initiated when (1) the patient had no evidence of infection and the patient was immunocompromised, had an anatomical defect, had recurrent infections, or had an indwelling device or (2) the patient was scheduled for a surgical procedure and antimicrobial therapy was given to prevent surgical site infections. The following sub-categories of prophylactic antimicrobial use were defined: prophylactic-urinary tract infections (UTI) was therapy to prevent further UTIs in a patient with a history of UTIs; prophylactic-immunocompromised was antimicrobial therapy initiated for patients who were immunosuppressed due to medications or to their underlying disease; prophylactic-perioperative was antimicrobial therapy initiated to prevent surgical site infections; prophylactic-other was prophylactic antimicrobial therapy that did not meet the definitions for the other subcategories of prophylactic treatments. The principal investigator considered a patient exposed if he/she received one or more dose(s) of one or more antimicrobial agents. The principal investigator adapted the definitions of antimicrobial prescriptions and antimicrobial courses used by Bergmans and colleagues (23), defining an antimicrobial prescription as the initiation of one

96 82 antimicrobial agent. Thus, if a physician wrote an order for gentamicin and ampicillinsulbactam, the patient would have received two prescriptions. The principal investigator defined an antimicrobial course as an episode in which one or more antimicrobial agents were prescribed, either consecutively or in combination for prophylaxis or to treat a suspected or documented infection (23). The principal investigator defined three study periods: the baseline period from January 1, 2005 to January 31, 2007, the intervention period from February 1, 2007 to July 31, 2007, and the post-intervention period from August 1, 2007 to January 31, The principal investigator also defined three nested periods: period A from January 1, 2005 to June 30, 2005, Period B from February 1, 2007 to July 31, 2007, and Period C from August 1, 2007 to January 31, 2008 (See Figure 4) Data collection The principal investigator obtained data on total milligrams of antimicrobials used in the PICU and in the NICU during the study period, January 1, 2005 to January 31, For the period between January 1, 2005 and January 31, 2006, the principal investigator obtained data on use of all antimicrobials (i.e., names of the antimicrobial agents, start and stop dates) in the PICU, the intervention unit, and in the NICU, the nonintervention comparison unit, from the UIHC s billing records, because a computerized database for antimicrobial use was not available during this period. The UIHC implemented a hospital-wide electronic database for medication administration in February Thus, for the period between February 1, 2006 and January 1, 2008, the principal investigator obtained data on antimicrobial use in both the NICU and the PICU from the UIHC s electronic medical record (IPR).

97 83 The principal investigator abstracted the medical records of patients who stayed in the PICU more than 24 hours during the three nested periods (Period A, Period B, and Period C). He subsequently compared the patients characteristics, the specific antimicrobial agents used, the indications for therapy, and microbiologic and diagnostic test results during the periods A, B, and C Outcome measures The primary outcome measures were defined as the overall rate of antimicrobial use measured as milligrams per week (mg/week) in the PICU and the NICU and the proportion of patients exposed to empiric antimicrobial treatments in the PICU. The principal investigator s academic advisor, who is a hospital epidemiologist and an adult infectious disease consultant, reviewed the chart abstraction forms and the patients medical records to determine whether antimicrobial agents prescribed in the PICU during the three nested periods were given empirically, prophylactically, or therapeutically (i.e., targeted therapy). She did not categorize antimicrobial use in the NICU because the number of medical records was very large. Thus, the principal investigator was able to determine whether empiric antimicrobial use in the PICU changed during the intervention period, while prophylactic use remained stable as he hypothesized. The principal investigator was not able to compare trends in prophylactic or empiric antimicrobial use in the PICU with those in the NICU Statistical Methods Data from Medical Records The principal investigator compared information about patients admitted to the PICU during each of the three nested study periods. He used analysis of variance

98 84 (ANOVA) to compare the means of continuous variables, Wilcoxon rank-sum test to compare medians, the chi-square test of proportions to compare categorical variables, and likelihood ratio test for comparison of antimicrobial rates. The principal investigator considered a P-value of less than or equal to 0.05 as significant. All tests were two-tailed Data on Overall Antimicrobial Use The principal investigator used interrupted time series analysis to determine the effect of the intervention. He used autoregressive integrated moving average (ARIMA) models to identify the pre-intervention serial dependence. The principal investigator checked the series for stationarity using Dickey-Fuller test (ADF) (121) and the Ljung-Box test (121) to evaluate if the residuals were white noise (i.e., were distributed independently and distributed normally with constant variance and zero mean). The units of analysis were overall exposure to cefazolin in milligrams per week, overall exposure to piperacillin-tazobactam, cefepime, and ceftriaxone in milligrams per week and the weekly proportion of patients exposed to empiric antimicrobials. A week was considered from Sunday to Saturday. The principal investigator used the Akaike Information Criterion (AIC) to select between two or more pre-intervention models and selected the model with the minimum AIC. The principal investigator used the computer programming language R to perform time series analyses and the ARIMAX function in the time series analysis (TSA) package to model the transfer functions and to determine the effect of the intervention (121). The principal investigator used logtransformation for the overall antimicrobial series and used sin 1 P transformation for the proportion of patients exposed to empiric antimicrobial series.

99 85 The principal investigator used a procedure developed by Box and Tiao to assess the efficacy of the intervention (122). In their approach, a time series is represented by two components: an underlying disturbance process and the set of interventions on the series (see Appendix C for technical details). The intervention model can be written in the form (123) YY tt = ωω(bb) δδ(bb) II tt + θθ(bb) φφ(bb) εε tt. (1) where {I t }is the intervention function. The polynomials φφ(bb), θθ(bb), δδ(bb), and ωω(bb) are defined as for the transfer function model; εε tt is white noise. From equation 1: ø(b) = 1 ø 1 B ø 2 B 2.. ø p B p (2) θ(b) = 1 θ 1 B θ 2 B 2.. θ q B q.. (3) p and q represent the order of the autoregression and the moving average process, respectively. In an autoregressive process, we assume that the series is made up of a random error component and a linear combination of prior observations. While in the moving average process, we assume the series is made up of random error shocks (124). The coefficient of εε tt in equation 1 allows us to incorporate the autoregressive and moving average process with the nonstationary difference component of the series. Therefore, an ARIMA (p,q,d) involves these three components Hypothesis The principal investigator assumed that the effect of the intervention on piperacillin-tazobactam, cefepime, and ceftriaxone use, on cefazolin use, and on the proportion of patients exposed to empiric antimicrobials would be abrupt and constant in nature after the intervention. The principal investigator established a type I error level of 0.05 as significant.

100 86 An abrupt constant change in Y t was tested using the intervention model at time (t) described in equation 4. The operator in the numerator, ω (B), represents the impact(s) of the intervention and the length of time (delay) it takes for the impact(s) to be reflected in the series. The operator in the denominator, δ (B), represents the way in which an impact dissipates. YY tt = ωω(bb) δδ(bb) II tt + θθ(bb) φφ(bb)(1 BB) dd εε tt. (4) II tt = 1, iiii tt TT 1 aaaaaa tt TT 2 0, ooooheeeeeeeeeeee T 1 =beginning of the intervention period (February 1, 2007); T2 = end of the intervention period (July 31, 2007) Occasionally, data may have atypical observations (outliers). Outliers may arise because of measurement and/or copying errors or because of abrupt, short-term changes in the underlying process (121). An outlying response is an additive outlier (AO), if the underlying process is perturbed additively at time T. An AO is modeled by adding a dummy variable on the right hand side of the equation 4. On the other hand, if the noise is atypical it is an innovative outlier (IO). An IO at T perturbs all observations starting at T and thereafter. The principal investigator used the computer programming language R to perform time series analyses and the ARIMAX function in the time series analysis (TSA) package to detect whether an observation was an AO or IO and to test the hypotheses.

101 Sample size Sample size calculations are difficult for time series studies and the published literature includes only a few examples. The principal investigator used the sample size calculation plot reported by Gottman to estimate the number of data points, in weeks, required in the pre-intervention, intervention, and post-intervention periods for this model using a two-sided type I error level of 0.05 and a power of 0.90 (122). The principal investigator hypothesized that the pre-intervention models for this study would follow second order of autoregressive integrated moving average (ARIMA) (2, 0, 0) with the following parameters: ø 1 = 0.3 and ø 2 = 0.2. Thus, to detect a change in the series mean as large as the standard deviation of the series, at least 20 weeks of data would be needed in each of the three study periods (i.e., the pre-intervention, intervention, and postintervention periods (122). The principal investigator did not have prior information on the parameters cited above. On the basis of Gottman s simulations (122), the principal investigator assumed a sample size of at least 20 weeks would be sufficient for a wide spectrum of ARIMA models. Thus, each of the study periods in the PICU was 24 weeks long. 4.3 Results Antimicrobial Use: Nested Periods A, B, and C One hundred and forty-five patients stayed in the PICU longer than 24 hours in period A, compared with 173 patients in period B, and 206 patients in period C. Most patient characteristics were similar during all three periods (Table 16). However, a higher proportion of patients had Foley catheters while in the PICU during periods A and C

102 88 compared with period B. Additionally, a higher proportion of patients received parenteral nutrition during period C than during periods A and B. One hundred and thirty-four of 145 (92%) patients in period A, 162 of 173 (94%) patients in period B, and 186 of 206 (90%) patients in period C received at least one antimicrobial agent. Of these patients 60 (44.8%) in period A, 72 (44.4%) in period B, and 104 (55.9%) in period C received empiric antimicrobial treatments. A higher proportion of patients in period C than in periods A or B received empiric antimicrobial treatments (C vs. A: P = 0.064; C vs. B = 0.043). Compared with period A, exposure to empiric antimicrobials expressed as the number of courses per 100 patient-days was 1.4 times higher in period B and 1.5 times higher in period C. The proportions of patients in periods A, B, and C who received prophylactic antimicrobial treatments were 111/134 (82.8%), 113/162 (69.8%), and 123/186 (66.1%), respectively (P = 0.003). A higher proportion of patients received prophylactic antimicrobial treatments in period A than in period B (P = 0.01) or period C (P = 0.001). However, the proportions of patients who received prophylactic antimicrobial treatments in periods B and C were similar (P = 0.54). Compared with period A, exposure to prophylactic antimicrobials expressed as the number of courses per 100 patient-days was 1.3 times higher in period B and 1.1 times higher in period C. The proportion of patients who received targeted antimicrobial treatments was similar in period A (23/134; 17.2%), in period B (34/162; 21.0%), and in period C (36/186; 19.4%). Compared with period A, exposure to targeted antimicrobials expressed as the number of courses per 100 patient-days was 1.2 times higher in period B and 1.4 times higher in period C.

103 89 The median duration of empiric treatments decreased slightly from 4 days in period A to 3 days in period B (P = 0.63). The median duration of empiric treatments did not change between periods B and C (P = 0.51). The median duration of targeted treatments decreased from 8 days in period A to 6 days in period B (P < ), but was similar between periods B and C (P = 0.57). The median duration of prophylactic treatment did not change between periods A, B, and C Documentation Results Pediatric intensivists documented their rationale for approximately 90% of all empiric antimicrobial prescriptions and this was consistent across all periods: period A (94.8%), period B (87.6%), period C (99.3%) (P = 0.80). Documentation of the rationale for all antimicrobial prescriptions given during period A (88.6%) and period B (85.6%) were similar (P = 0.22). However, documentation of the rationale for all antimicrobial prescriptions in period C (90.6%) was higher than in period B (P = 0.02) Consultation Results Pediatric intensivists consulted the ID specialists more during period A (38% of all empiric prescriptions) than during period B (17% of all empiric prescriptions) and during period C (16% of all empiric prescriptions) when deciding to prescribe empiric antimicrobial agents (A vs. B, P = 0.004; A vs. C, P = 0.001). The frequency of consultation regarding empiric therapy was similar between periods B and C (P = 0.84). In contrast, pediatric intensivists consulted the infectious disease specialists more frequently during periods B (20% of all targeted prescriptions) and C (34% of all targeted prescriptions) than in period A (7% of all targeted prescriptions) when deciding to prescribe targeted antimicrobial agents (A vs. B, P = 0.03; A vs. C, P = 0.003). The

104 90 frequency of consultation for targeted therapy was similar between periods B and C (P = 0.12) Culture Results During period A, eight patients had positive blood cultures (Table 18); five of these patients were treated with empiric and targeted treatments, two patients received targeted treatment only, and one patient received empiric treatment only. Only three (37.5%) of these patients met the CDC criteria for nosocomial bloodstream infections (BSI). During period B, eight patients had positive blood cultures; one of these patients was treated with targeted antimicrobials and eight patients were treated with empiric and targeted antimicrobials simultaneously. Only two (22.2%) of these patients met the CDC criteria for nosocomial BSI. During period C, twenty patients had positive blood cultures; fourteen of these were treated with empiric and targeted antimicrobials simultaneously, six patients were treated with empiric antimicrobials only and one patient was treated with targeted antimicrobials only. Twelve of (57.1%) these patients met the CDC criteria for nosocomial BSI. Forty patients had positive tracheal aspirates during period A, and 26 (65%) of these patients had only positive tracheal aspirates with no positive cultures from other sites such as blood, urine, and peritoneal fluid. Only five of 26 (19%) patients received targeted antimicrobial treatments and 19 (73.1%) patients received empiric antimicrobial treatments. Fifteen (58%) of 26 patients had chest radiographs that were consistent with pneumonia but only three (12%) of these patients met the CDC criteria for VAP according to the UIHC s infection surveillance program. (Note: the UIHC s Program of Hospital Epidemiology [PHE] did surveillance for ventilator-associated pneumonia

105 91 (VAP) during period A but not during periods B and C). One VAP was caused by Hemophilus influenza, one by H. influenzae, Moraxella catarrhalis, and Streptcococcus pneumonia, and one by Escherichia coli and Enterobacter cloacae. Forty-four patients had positive tracheal aspirates in period B and 33 (75%) of these patients had only positive tracheal aspirates with no positive cultures from other sites such as blood, urine, and peritoneal fluid. Nineteen of 33 (58%) received targeted antimicrobial agents and14 (42%) received empiric antimicrobial treatments only. Thirteen (39%) of 33 patients had chest radiographs that were consistent with pneumonia. In period C, 53 patients had positive tracheal aspirates and 37 (70%) of these patients had only positive tracheal aspirates with no positive cultures from other sites. Thirteen of 37 (35%) received targeted antimicrobial agents, 21 (57%) received empiric antimicrobial treatments; an indication could not be determined for three (8%) patients. Seventeen (46%) of 37 patients had chest radiographs that were consistent with pneumonia Susceptibility Results All Staphylococcus aureus isolates were susceptible to vancomycin and most were susceptible to gentamicin (Table 19). The proportion of isolates susceptible to gentamicin decreased from periods A and B to period C (P = 0.06). The proportion of S. aureus isolates susceptibile to clindamycin was lowest during the period B. As expected, all S. aureus isolates were resistant to penicillin. The proportion of isolates susceptible to oxacillin remained stable during the three periods (P = 0.74). All P. aeruginosa isolates were susceptible to cefepime, ciprofloxacin, gentamicin, piperacillin-tazobactam in period A; 93% of the isolates were susceptible to each of these antimicrobial agents

106 92 during period C (Table 20). During the latter two periods, P. aeruginosa isolates were tested for susceptibility to meropenem and trobramycin; all isolates from periods B and C were susceptible to these agents. All E. cloacae isolates from the three periods were susceptible to ciprofloxacin, gentamicin, and meropenem (Table 21) Interrupted time series Piperacillin-tazobactam, Cefepime, and Ceftriaxone Use in the PICU Figure 5 presents the weekly use of piperacillin-tazobactam, cefepime, and ceftriaxone measured in milligrams per week for the PICU. A gray rectangle identifies the intervention period. The pre-intervention period was stationary for the PICU antimicrobial series; therefore, no differencing was required (See diagnostic materials in Appendix C). The selected pre-intervention model for the PICU series was an ARIMA (1, 0, 0). The principal investigator fitted two models. Model 1 was ARIMA (1, 0, 0) with AIC = ; model 2 was ARIMA (0,0,0) with AIC = The principal investigator chose model 1 as the final model because it has the lower AIC value compared with model 2. Five influential outliers were detected at weeks 39, 74, 78, 94, and 153. These outliers were adjusted for in the final model. The final intervention model and its estimates of abrupt constant change are presented in equation (5). YY tt = CC + ωω 0 1 δδ 0 BB II tt + ωω IIII II IIII,tt + εε tt 1 φφ 1 BB. (5) Where II tt = 1 during the intervention period and 0 otherwise; II IIII,tt = 1 in weeks 39, 74, 78, 94, and 153 and 0 otherwise. φφ 1 = se(φφ 1 ) = CC = se(cc ) = ωω IIII(39) = se(ωω IIII(39) ) =

107 93 ωω IIII(74) = se(ωω IIII(74) ) = ωω IIII(78) = se(ωω IIII(78) ) = ωω IIII(94) = se(ωω IIOO(94) ) = ωω IIII(153) = se(ωω IIII(39) ) = δδ 0 = se(δδ 0) = ωω 0 = se(ωω 0 ) = AIC = Log likelihood = and estimated σσ 2 = The estimate of abrupt constant change (see equation 5) for this series was not statistically significant. ωω 0 = and its standard error was , indicating that the intervention did not reduce use of piperacillin-tazobactam, cefepime, and ceftriaxone use in milligrams per week in the PICU during the intervention period. The principal investigator combined these agents for time series analysis for two reasons: (1) they were used primarily for empiric therapy in the PICU and (2) they are dosed similarly Cefazolin Use in the PICU Figure 6 presents the weekly overall cefazolin use measured in mg/wk for the PICU. A gray rectangle identifies when the intervention was implemented in the PICU. The pre-intervention period was stationary for the cefazolin series; therefore, no differencing was required (See diagnostic materials in Appendix C). The selected preintervention model for the series was an ARIMA (0, 0, 0). The principal investigator fitted three models. Model 1 was ARIMA (1,0,0) with AIC = ; model 2 was ARIMA (1,0,1) with AIC = ; and model 3 was ARIMA (0,0,0) with AIC = The principal investigator chose model 3 as the final model because it has the lowest AIC

108 94 value of all the models. The final intervention model and its estimates of abrupt constant change are presented in equation (6). YY tt = CC + ωω 0 1 δδ 0 BB II tt + εε tt. (6) Where II tt = 1 during the intervention period and 0 otherwise. CC = se(cc ) = δδ 0 = se(δδ 0) = ωω 0 = se(ωω 0 ) = AIC = Log likelihood = and estimated σσ 2 = The estimate of abrupt constant change (see equation 6) for this series was negative but not statistically significant. ωω 0 = and its standard error was indicating that cefazolin series decreased slightly during the intervention in the PICU. Approximately 90% of cefazolin use in the PICU was for perioperative-prophylaxis. Intermittent spikes in the series may be caused by increase numbers of patients being admitted to the PICU after major surgical procedures Data from Nested Periods A, B, and C Proportion of Patients Exposed to Empiric Antimicrobials in the PICU Figure 7 presents the proportion of patients exposed to empiric antimicrobials each week in the PICU. The blank space in the figure represents the time between periods A and B during which the PI did not abstract data from the patients medical records. As a result we did not have information on the indications for antimicrobial use during this period. Figure 8 presents the transformed data of patients exposed to empiric therapy during the intervention and post intervention periods. The pre-intervention series was

109 95 stationary; therefore, no differencing was required (See diagnostic materials in Appendix C). The selected pre-intervention model was ARIMA (0,0,0). The principal investigator fitted eight models. Model 1 was ARIMA (2,0,0) with AIC = ; model 2 was ARIMA (1,0,0) with AIC = ; model 3 was ARIMA (1,0,0) with AIC = ; model 4 was ARIMA (0,0,0) with AIC = ; model 5 was ARIMA (0,0,0) with AIC = ; model 6 was ARIMA (0,0,0) with AIC = ; model 7 was ARIMA (0,0,0) with AIC = and model 8 was ARIMA (0,0,0) with AIC = The principal investigator chose model 8 as the final model because it has the lowest AIC value of all the models. The intervention model and its estimates of abrupt constant change are presented in equation (7). CC = se(cc ) = ωω 0 = se(ωω 0 ) = YY tt = CC + ωω 0 1 BB II tt + εε tt. (7) The series of the proportion of patients exposed to empiric antimicrobial agents in the PICU seems to be uncorrelated over time. However, the estimate of abrupt constant change (see equation 7) for this series was negative and statistically significant. ωω 0 = and its standard error was This indicates that there was a significant reduction on the proportion of patients, on the arc sin scale, who were exposed to empiric antimicrobial agents over the intervention period. But the intervention does not have persistent after-effects as the post-intervention time series reverted back to the preintervention level. This decrease on a normal scale is equivalent to xx = (sin ) 2 = Thus, there appears to be a significant 0.42% reduction on the proportion of patients who were exposed to empiric antimicrobial agents over the intervention period.

110 Discussion In this quality improvement study, the principal investigator encouraged pediatric intensivists to document their rationale for prescribing antimicrobials for patients hospitalized in the UIHC s PICU. The principal investigator used documentation as a strategy to improve antimicrobial stewardship because several investigators previously showed that this approach together with measures, such as automatically discontinuing antimicrobials, could decrease antimicrobial use. Descriptive data on antimicrobial use in the PICU during the baseline period (January to June 2005) showed that empiric (45.8% of all antimicrobial prescriptions) antimicrobial use and prophylactic (44.4% of all antimicrobial prescriptions) antimicrobial use were common. We implemented an intervention (described in chapter 3) and used two methods to assess the effect of the intervention on antimicrobial use, particularly on empiric therapy. First, we used basic statistical techniques, such as the Wilcoxon rank sum test, to compare data from the patients medical records on the indications for antimicrobial use during period A, period B, and period C to see whether empiric antimicrobial use changed during these periods. Second, we used interrupted time series analysis (TSA) to assess trends in antimicrobial use. We first used TSA to assess overall antimicrobial use measured in mg/wk (data not shown). This analysis did not reveal a significant change over time. In addition, this analysis was flawed because the dosing regimens for different antimicrobials were not the same. Therefore, we used TSA to assess whether use of three antimicrobials which have similar dosing regimens and which were used frequently for empiric treatment piperacillin-tazobactam, cefepime, and ceftriaxone, measured in mg/wk changed over time. We also used TSA

111 97 to assess the trend in the cefazolin use measured in mg/wk (an antimicrobial agent that was used primarily for surgical prophylaxis) and the trend in the proportion of patients who were exposed to empiric antimicrobial therapy. In the current study, the median duration of empiric and targeted antimicrobial therapies decreased by one day and two days, respectively. The decreases in the duration of empiric and targeted treatments were not significant statistically. However, the PI estimated that 193 days of empiric treatment and 59 days of targeted treatment would have been saved, which would decrease the cost of antimicrobial treatment and would decrease antimicrobial pressure. We do not know whether the decreases in the median treatment days for empiric and targeted therapies were associated with our intervention. There are other possible explanations for these observations. First, if the length of stay in the PICU decreased during the intervention period then the duration of empiric therapy could decrease if the average length of stay was shorter than the average total duration of empiric treatment course. The mean length of stay in the PICU decreased from 9 days in pre-intervention period (period A) to 6 days in intervention period (period B). However, the median length of stay did not change. Second, if the pediatric intensivists consulted the ID specialists more frequently during the intervention period than during the pre-intervention period then empiric antimicrobial treatment might decrease or the duration of empiric therapy might decrease during the intervention period. We found that pediatric intensivists consulted ID specialists more during the pre-intervention period (38% of all empiric prescriptions) than during the intervention period (17% of all empiric prescriptions). This observation suggests that the decrease in the duration of empiric and targeted antimicrobial therapies

112 98 that we noted during the intervention period was not explained by increased ID consultation. Third, the Hawthorne effect might have affected the duration of empiric and targeted antimicrobial treatments. We met with the physicians in our PICU and we discussed the study design with these physicians. The PI visited the unit during the intervention period and encouraged residents and fellows to complete the AA form. Consequently, physicians knew that we were observing their antimicrobial prescribing behavior. Thus, the Hawthorne effect might have accounted for the decrease in the duration of therapy that we observed during the intervention period. The PI expected that the intervention might influence pediatric intensivists to document their rationale for ordering antimicrobial agents more frequently, especially for those agents prescribed empirically. However, documentation did not change (P = 0.80) during the intervention period. The results of the TSA evaluating the proportion of patients exposed to empiric therapy suggested that this measure decreased significantly during the intervention period. However, the results of the TSA for the three common empiric antimicrobial agents (measured in mg/wk) with similar dosing regimens piperacillin-tazobactam, cefepime, and ceftrixone did not confirm this result. We do not know why these two analyses had different results. However, the trend in the time series assessing the proportion of patients who were exposed to empiric therapy did not decrease when compliance with the AA form was highest; it happened when compliance was low. We considered TSA of piperacillin-tazobactam, cefepime, and ceftrixone use to be the gold standard because these three agents were primarily used for empiric treatment (47% of

113 99 all empiric antimicrobial prescriptions). Therefore, the result of the TSA assessing the proportion of patients exposed to empiric antimicrobials might be due to random variation. In addition, the TSA demonstrated that the trend for cefazolin use did not change appreciably. Cefazolin was used primarily as perioperative prophylactic therapy. Our intervention did not target prophylactic antimicrobial use because surgeons, who were not the focus of this intervention, controlled the decisions regarding the perioperative prophylactic agents used and the duration of treatment. The rates of empiric antimicrobial therapy measured as courses per 100-treatment days increased significantly during the intervention period, but the rates of prophylactic antimicrobial therapy measured as courses per 100-treatment days remained fairly stable. Therefore, the intervention did not discourage physicians from starting empiric antimicrobial therapy. The goal of this study was not to prevent physicians from starting empiric antimicrobial therapy but rather to shorten the length of empiric therapy by having physicians adjust antimicrobial therapy once culture and susceptibility results were available, or stop therapy if cultures were negative. It is possible that pediatric intensivists started more empiric antimicrobial therapy during the intervention period compared with the pre-intervention period because the patients were sicker than during the pre-intervention period. Most patient characteristics during the two study periods were similar, although, Foley catheter and central venous catheter utilization varied. For example, fewer patients had Foley catheters or central venous catheters during the intervention period than during the pre-intervention period. In contrast, more patients received parenteral nutrition during the intervention period than during the pre-

114 100 intervention period. On the basis of these observations, we do not have evidence suggesting that patients were sicker during the pre-intervention period than during the intervention period. However, more patients had nosocomial bloodstream infections during the post-intervention period (5.8% of all patients) than during the pre-intervention period (2.1% of all patients) and during the intervention period (1.2% of all patients). In contrast to our study, Durbin and colleagues mandated that physicians in a surgical intensive care unit categorize antimicrobial use as prophylactic, empiric, or therapeutic (99). Administration of antimicrobial agents was automatically discontinued after two days for prophylactic therapy, three days for empiric therapy, or seven days for therapeutic courses. During the two-month intervention period, 60% of surgical patients received prophylactic antimicrobial agents compared with 68% in the baseline period and the mean duration of prophylactic treatment decreased from 4.9 days to 2.9 days. The investigators did not report whether the decrease in prophylactic use they observed in their study was statistically significant. It is not surprising that our results were different from those of Durbin and colleagues. Their intervention was stronger than our intervention because they mandated that physicians define the category of antimicrobial use and the investigators automatically stopped the treatments after specified durations. In contrast to the study by Durbin and colleagues, we used only the AA form without an additional intervention component, such as introducing automatic stop orders or mandating that physicians complete the AA form. In addition, the physicians did not categorize the antimicrobial use and were not informed of the retrospective categorizations of antimicrobial use.

115 101 In contrast to our study, Kowalsky and colleagues (125) required physicians in a teaching hospital to document clinical indications for all antimicrobial prescriptions. These investigators automatically stopped antimicrobials for surgical prophylaxis after 48 hours. They reported a significant 17% reduction in the percentage of patients receiving sentinel antimicrobials (i.e., first, second, and third-generation cephalosporins, penicillin, ampicillin, aminoglycosides, antistaphylococcal agents, antianaerobic agents, and antipseudomonal agents in their hospital (P = 0.007) (125). While Durbin et al. and Kowalsky et al. reported that their interventions reduced prophylactic antimicrobial use; our study and that of Bolon did not find the expected intervention effects. Bolon and colleagues (58) introduced an AA form that required physicians working in a pediatric, tertiary-care teaching hospital to either select one of the criteria specified by the Hospital Infection Control Practices Advisory Committee (HICPAC) (126) as their reason for prescribing vancomycin or to request an infectious disease consultation. Initially, compliance with completing their AA form was less than 50%. The investigators extended their study period by two months, during which compliance improved to between 70% and 80%. They were surprised that the rate of inappropriate use of vancomycin increased from 35% to 39% during the months of poor adherence and to 51% during the months of good adherence. In the current study, compliance with completing the AA form was 90% at the inception of the study but later dropped below 10%. On the basis of our results and results from Bolon s study, we think that an AA form without additional measures is unlikely improve antimicrobial stewardship.

116 102 Researchers have evaluated various strategies for improving antimicrobial stewardship in ICUs. These strategies include but are not limited to formulary systems, antimicrobial cycling, peer review with feedback, education, and implementation of guidelines. None of these strategies have been clearly effective (127). Burke (128) argued that widespread antimicrobial use and the resulting poor quality of care might be related to physicians having inadequate information about their patients rather than to intentional bad behavior. Pestotnik and colleagues (129) suggested that developing antimicrobial strategies that will augment physicians decisions with information that is relevant to the immediate clinical situation seems logical. These investigators, assessed whether computer-decision support system that incorporated patient-specific information coded into rules, algorithms, and predictive models and provided feedback to physicians in real time would affect the prophylactic, empiric, and therapeutic antimicrobial use among inpatients. These investigators found that overall antimicrobial use (measured by defined daily dose) decreased by 22.8%, that the average number of antimicrobial doses administered for surgical prophylaxis was reduced from 19 doses in the baseline period to 5.3 doses in the intervention period, and that adverse events related to antimicrobial agents decreased by 30%. The investigators concluded that their intervention worked probably because the system used patient-specific information, provided feedback to the clinicians, presented the clinicians with choices, and allowed the clinicians to use their own judgment. To the best of our knowledge, only one study assessed whether a computer decision support system would decrease antimicrobial use in a PICU. Mullett and colleagues introduced bedside computer terminals, which ran the Health Evaluations

117 103 through Logical Processing (HELP software) (61), to provide decision support for physicians prescribing antimicrobial agents. Through HELP, physicians were able to retrieve patients vital signs, laboratory test results, radiology test results, and pathology results. The HELP software also helped physicians chose antimicrobials and doses based on logic incorporated into the software. After the intervention, the rate of pharmacy intervention for erroneous drug orders decreased by 59%, the rate of subtherapeutic (defined as antiinfective therapy that fell below the minimum recommendation) risk days decreased by 36%, and the rate of excessive-doses (defined as antiinfective therapy that exceeded the maximum recommendation) risk days declined by 28%. Furthermore, the number of orders placed per anti-infective course decreased 11.5% from an average of 1.56 to 1.38 orders/patient-anti-infective (P < 0.01) (61). However, the costs of antimicrobial agents for the PICU and for the hospital did not decrease. More studies on the role of computer decision support system as an effective antimicrobial stewardship program might be useful in PICUs. Singh and colleagues (97) calculated a clinical pulmonary infection score (CPIS) to identify patients for whom a shorter duration of empiric therapy could suffice. The investigators randomized patients with CPIS 6 (low likelihood of having pneumonia) to receive either therapy as directed by their physicians (standard therapy) or ciprofloxacin monotherapy with reevaluation of the need for further treatment at 3 days. The investigators reported that patients with CPIS 6 who received the therapy defined by the protocol had a significantly shorter duration of empiric antimicrobial treatment than did patients who received standard therapy (3 days versus 9.8 days, P = ). Patients in both treatment arms had similar outcomes, suggesting that the 6.8 extra antimicrobial

118 104 days received by patients on standard therapy was unnecessary. Of note, investigators terminated the study early because as the study progressed it became obvious to physicians that administration of antimicrobial therapy for a prolonged duration did not favorably affect patients outcomes and did not decrease complications. In the current study, few patients had documented serious invasive bacterial infections yet a substantial proportion of patients in periods A, B and C received empiric antimicrobial agents suggesting that a fairly high percent of the patients could have had antimicrobial treatments stopped when their cultures did not identify bacterial pathogens. A study similar to that of Singh s study might be useful in PICUs. Investigators could assess whether a clinical infection index or score (CII or CIS) designed for children could help clinicians guide the duration of empiric antimicrobial therapy. In summary, the median days of empiric and targeted treatment decreased during the intervention and remained stable during the post-intervention period. The PI estimated that 193 days of empiric antimicrobial therapy and 59 days of targeted antimicrobial therapy respectively may have been saved by the decreased durations of therapy. TSA assessing the trend in use of antimicrobials did not reveal a significant change over time. Therefore, an AA form alone may not be an effective strategy for antimicrobial stewardship in PICUs. Additional measures such as automatic stop orders and computer decision support may be useful for reducing duration of empiric therapy in PICUs (See strengths and limitations in Chapter 5).

119 105 Table 16. Descriptive Characteristics of the Patient Populations Variables Description Period A (n = 145) (%) Period B (n = 173) (%) Period C (n = 206) (%) *P-value Gender Female 61 (42) 85 (49) 92 (45) 0.43 Age Median age in years (mean ± std) 3 (5.7 ± 6.2) 4 (6.7 ± 6.9) 2 (5.6 ± 6.3) 0.46 Age Categories Neonates (0 1month) 22 (15.2) 11 (6.4) 19 (9.2) 0.17 Infants (> 1 month 2 years) 44 (30.3) 64 (37.0) 78 (37.9) Children (> 2 years 12 years) 44 (30.3) 51 (29.5) 64 (31.1) Adolescents (> 12 yrs) 35 (24.1) 47 (27.2) 45 (21.8) Ethnicity White 110 (75.9) 127 (73.4) 132 (64.1) Black 10 (6.9) 10 (5.8) 18 (8.7) Hispanic 6 (4.1) 9 (5.2) 8 (3.9) Asian/Pacific Islander 4 (2.8) 2 (1.2) 3 (1.5) Others 15 (10.4) 25 (14.5) 45 (21.9) Weight Median weight in kg (mean ± std) 14 (23.4 ± 23.9) 15 (26.6 ± 25.7) 13 (22.9 ± 23.5) 0.51

120 106 Table 16 continued Variables Description Period A (n = 145) (%) Period B (n = 173) (%) Period C (n = 206) (%) *P-value Length of Hospitalization Median length of stay in the hospital in days (mean ± std) Median length of stay in the PICU in days (mean ± std) 10 (16.9 ± 19.7) 10 (15.7 ± 20.3) 11 (20.3 ± 53.7) (9.0 ± 15.2) 3 (6.3 ± 7.4) 4 (7.6 ± 9.4) 0.63 Invasive Devices Endotracheal tube 89 (61.4) 110 (63.6) 136 (66.0) 0.67 Foley catheter 97 (66.9) 43 (24.9) 125 (60.7) < Central venous catheter 94 (64.8) 89 (51.5) 123 (59.7) 0.05 Nutrition Parenteral nutrition 3 (2.1) 13 (7.5) 32 (15.5) < Mortality Died in the PICU 5 (3.1) 5 (2.6) 10 (4.6) 0.58 Abbreviations: n = number; std = standard deviation; kg = kilograms; PICU = pediatric intensive care unit; yrs = years *P-value compares all three periods; if p < 0.05 then the value of the variable must be different in at least one of the periods. Period A = pre-intervetion period; period B = intervention period; period C = post-intervention period

121 107 Table 17. Treatment Days for Empiric, Prophylactic, and Targeted Antimicrobial Use Indication Description Period A (n = 134) Period B (n = 162) Period C (n = 186) Empiric Number of prescriptions Number of courses Number of patients Sum of patient days Sum of treatment days Courses per 100 patient-days* Courses per 100 treatment-days** Prescriptions per 100 patient-days Prescriptions per 100 treatment-days Median treatment days 4 (4.5 ± 3.5) 3 (4.5 ± 3.8) 3 (4.1 ± 3.1) Range of treatment days Prophylactic Number of prescriptions Number of courses Number of patients Sum of patient days Sum of treatment days Courses per 100 patient-days* Courses per 100 treatment-days** Prescriptions per 100 patient-days Prescriptions per 100 treatment-days Median treatment days 3 (4.4 ± 4.1) 3 (4.3 ± 4.6) 3 (4.2 ± 3.9) Range of treatment days

122 108 Table 17 continued Indication Description Period A (n = 134) Period B (n = 162) Period C (n = 186) Targeted Number of prescriptions Number of courses Number of patients Sum of patient days Sum of treatment days Courses per 100 patient-days* Courses per 100 treatment-days** Prescriptions per 100 patient-days Prescriptions per 100 treatment-days Median treatment days 4 (4.5 ± 3.5) 3 (4.5 ± 3.8) 3 (4.1 ± 3.1) Range of treatment days Abbreviation: Period A = pre-intervetion period; period B = intervention period; period C = post-intervention period. total patients who received antimicrobial agents. Empiric: *P-value = 0.24; **P-value = 0.01; P-value = 0.56; P-value = 0.02 Prophylactic: *P-value = 0.14; **P-value = 0.16; P-value = 0.11; P-value = 0.95 Targeted: *P-value = 0.39; **P-value = 0.76; P-value = 0.13; P-value = 0.61

123 109 Table 18. Description of Organisms and Category of Antimicrobial Treatment for Patients who Had Positive Blood Cultures Period / Patient Organisms Type of Indications Met Definition of Nosocomial Primary Bloodstream Infection A 1 Coagulase-negative staphylococcus Empiric and targeted No 2 Coagulase-negative staphylococcus Empiric and targeted No 3 Klebsiella pneumoniae Empiric and targeted No 4 Coagulase-negative staphylococcus Empiric and targeted Yes 5 Coagulase-negative staphylococcus Targeted Yes 6 Staphylococcus aureus and K. pneumoniae Targeted Yes 7 Coagulase-negative Staphylococcus Empiric and targeted No 8 Enterobacter cloacae Empiric No B 1 Enterococcus spp. Empiric and targeted No 2 Coagulase-negative staphylococcus Empiric and targeted No 3 E. cloacae and Pseudomonas aeruginosa Empiric and targeted Yes 4 Stenotrophomonas maltophilia Empiric and targeted No 5 Coagulase-negative staphylococcus Empiric and targeted No 6 Coagulase-negative staphylococcus, P. Empiric and targeted No aeruginosa 7 Coagulase-negative staphylococcus Targeted No 8 Klebsiella oxytoca Empiric and targeted Yes

124 Table 18 continued Period / Patient Organisms Type of Indications Met Definition of Nosocomial Primary Bloodstream Infection C 1 Enterococcus spp. Empiric and targeted Yes 2 Coagulase-negative staphylococcus Empiric Yes 3 Coagulase-negative staphylococcus Empiric and targeted Yes 5 Streptococcus pneumoniae Empiric and targeted No 6 Coagulase-negative staphylococcus Empiric No 7 Neisseria spp. Empiric and targeted No 8 Serratia marcescens Empiric and targeted Yes 9 Coagulase-negative staphylococcus Empiric Yes 10 Enterococcus spp. Empiric and targeted Yes 11 Enterobacter cloacae Targeted Yes 12 Stenotrophomonas maltophilia Empiric and targeted Yes 13 Enterococcus spp. Empiric and targeted Yes 14 Coagulase-negative staphylococcus Empiric No 15 Klebsiella pneumoniae Empiric and targeted Yes 16 Coagulase-negative staphylococcus Empiric No 17 Coagulase-negative staphylococcus Empiric and targeted No 18 Coagulase-negative staphylococcus Empiric and targeted Yes 19 Hemophilus influenzae Empiric and targeted No 20 Coagulase-negative staphylococcus Empiric Yes 21 Serratia marcescens Empiric and targeted No 110

125 111 Table 19. Susceptibility Results: Staphylococcus aureus Antimicrobials All Period A Period B Period C n % n % n % n % tested susceptible tested susceptible tested susceptible tested susceptible Clindamycin Erythromycin Gentamicin Oxacillin Penicillin Tetracycline Vancomycin Abbreviation: n = number of isolates tested

126 112 Table 20. Susceptibility Results: Pseudomonas aeruginosa Antimicrobials All Period A Period B Period C n % n % n % n % tested susceptible tested susceptible tested susceptible tested susceptible Cefepime Ciprofloxacin Gentamicin Piperacillin-tazobactam Meropenem Tobramycin Abbreviation: n = number of isolates tested

127 113 Table 21. Susceptibility Results: Enterobacter cloacae Antimicrobials All Period A Period B Period C n % n % n % n % tested susceptible tested susceptible tested susceptible tested susceptible Ceftriaxone Ciprofloxacin Gentamicin Levofloxacin Piperacillin-tazobactam Meropenem Abbreviation: n = number of isolates tested

128 114 Figure 4. Study Design and Nested Periods Period A January 1, 2005 to June 30, 2005 Period B February 1, 2007 to July 31, 2007 Entire Study Period January 1, 2005 to January 31, 2008 Period C August 1, 2007 to January 31, 2008

129 Figure 5. Piperacillin-tazobactam, Cefepime, and Ceftriaxone Use in the PICU 115

130 Figure 6. Cefazolin Use in the PICU 116

131 Figure 7. Pre-Intervention Period: Weekly Proportion of Patients Exposed to Empiric Antimicrobial Agents in the PICU 117

132 118 Figure 8: Intervention and Post-Intervention Periods: Transformed Data of Patients Exposed to Empiric Therapy and Fitted Values

133 119 CHAPTER 5: CONCLUSION 5.1 Introduction We conducted a study to determine the percentage of patients who received antimicrobial treatments, to determine the indications for antimicrobial use, and to identify the antimicrobial agents used most frequently in the University of Iowa Hospitals and Clinic s (UIHC) pediatric intensive care unit (PICU). On the basis of our data, we hypothesized that empiric antimicrobial use, particularly the duration of therapy, could be decreased. We implemented a six-month intervention study during which we asked the residents and the fellows to complete an antimicrobial assessment form (AA) to document their rationale for starting antimicrobial therapy. We postulated that documenting their rationale for starting antimicrobial treatments might remind pediatric intensivists to review antimicrobial therapies, especially empiric therapies, when the microbiologic data became available, thus, reducing the duration of antimicrobial therapy and possibly reducing the quantity (measured in grams) of antimicrobial consumption. The following sections summarize findings, strengths, and limitations of this study and briefly describe possible future studies. 5.2 Summary of findings During periods A, B, and C, most patients (92%, 94%, and 90%) who stayed longer than 24 hours received at least one antimicrobial agent. The goal of this study was to decrease the duration of empiric treatments or possibly decrease the amount (measured in grams) of empiric antimicrobials prescribed during and after the intervention period. The PI did not think that the intervention could affect prophylactic use because we did not focus the intervention on surgeons who decide which antimicrobials are used for

134 120 surgical prophylaxis and how long surgical patients will be treated. The proportion of patients who received prophylactic antimicrobial treatments during the entire study period ranged between 60% and 77%.Therefore, even if patients did not receive antimicrobials for any other reasons, antimicrobial use would still be high. Consequently, the principal investigator was not surprised that overall exposure to antimicrobials remained high in the UIHC s PICU. During the pre-intervention period (period A), 45% of patients who received antimicrobial agents were treated empirically. They received a mean of 3.2 empiric prescriptions per patient for a median duration of 4 days. During the intervention period (period B), 44% of patients who received antimicrobial agents were treated empirically but the number of empiric prescriptions per patient decreased to 2.7 and the duration decreased from 4 to 3 days. During the post-intervention period (period C), 56% of patients who received antimicrobial agents were treated empirically, but the number of prescriptions per patient stayed stable at 2.8 and the median duration of empiric therapy remained 3 days. Another goal of this study was to identify factors that pediatric intensivists considered while deciding to start empiric antimicrobial therapy. To achieve this goal we asked the residents and fellows to complete an AA form when they were deciding to start antimicrobial treatments. The pediatric intensivists (residents and fellows) completed 68 AA forms, 27 (39.7%) for prophylactic treatments and 41 (60.3%) for empiric or targeted treatments. The data from the AA forms suggested that perioperative prophylaxis was the major reason physicians prescribed prophylactic antimicrobial agents. The data from the AA forms also suggested that the factors pediatric intensivists considered most frequently

135 121 before deciding to start empiric or targeted antimicrobial treatments were elevated temperature, elevated CRP, elevated heart rate, elevated WBC count, and elevated respiratory rate. This observation was consistent with the result from the case-control study, which identified elevated temperature and the availability of CRP results to be independent predictors of patients who received empiric therapy, after adjusting for the availability of WBC test results, age, and time to receiving the first antimicrobial treatment. Interrupted time series analysis demonstrated that the proportion of patients who were exposed to empiric antimicrobial treatments per week (interrupted time series) decreased linearly by 0.42% during the intervention period. However, this observation was not supported by the results of the interrupted time series assessing piperacillintazobactam, cefepime, and ceftriaxone use (measured in mg/week), which did not find an intervention effect. Piperacillin-tazobactam, cefepime, and ceftriaxone were used primarily for empiric treatments and they accounted for 47% of all empiric antimicrobial prescriptions. It is therefore logical to assume that if the intervention truly affected empiric use, the results of the two measures should be similar. 5.3 Strengths and Limitations Strengths One of the strengths of the current study is the combination of retrospective and prospective study designs. Retrospectively reviewing patient medical records allowed the principal investigator to assess the patterns of antimicrobial use and to identify a specific indication for antimicrobial use that could possibly be decreased. The PI expected that the medical record would be the best source of information on patients antimicrobial use

136 122 because hospital pharmacists, medical doctors, and nurses are required to record the antimicrobials prescribed and the dosages. We used a prospective study design to assess factors that pediatric intensivists considered when deciding to start antimicrobial treatment. Physicians completed the AA form at the time they decided to start antimicrobial treatments, when the rationale for prescribing antimicrobial agents was still fresh in their minds. Thus, prospectively collecting these data decreased the likelihood that these data would be affected by recall bias. The current study assessed only one intervention strategy compared with other studies that used similar AA forms, which often combined an AA form with an automatic stop order. Therefore, it is difficult to evaluate the true intervention effect attributable to their use of an AA form. In addition, the current study utilized time series analysis to assess change overtime in antimicrobial use. The investigators for the four prior studies evaluating strategies for antimicrobial stewardship in PICUs did not use time series analysis. In fact, they all used the paired t-test to evaluate the effects of their interventions. Two of the studies reported successful interventions; however, they might have overestimated the effect of their intervention because clinical data collected at discreet points are usually correlated. Segmented regression analysis is a better method for assessing time series data because it estimates the stability of the data before an intervention, accounts for trends, shows whether the data shifts, and estimates how much an intervention changed an outcome of interest. Few prior studies assessed the indications for antimicrobial use in PICUs. In contrast, the current study assessed indications for antimicrobial use during three nested periods and this information allowed the PI to determine whether the indications for

137 123 antimicrobial use changed over the study period. A higher proportion of patients in the post-intervention period than in the pre-intervention period or in the intervention period received empiric antimicrobial treatments. We were not sure why empiric antimicrobial treatments increased in the post-intervention period. However, our data showed that more patients had nosocomial bloodstream infections in the post-intervention period (5.8% of all patients) than in the pre-intervention period (2.1% of all patients) and in the intervention period (1.2% of all patients). To the best of our knowledge, this is the first study to assess objective factors that pediatric intensivists consider when deciding to start empiric antimicrobial agents. The PI used an AA form to identify the factors that pediatric intensivists considered when deciding to start empiric antimicrobial agents during the intervention period. He validated the results obtained from this data source with a case-control study during the three nested periods. The PI collected information on exposure variables before patients were assigned to be cases or controls and a hospital epidemiologist/infectious disease expert categorized antimicrobial use as empiric, targeted, or prophylactic before the PI assigned patients to be cases or controls. Therefore, misclassification of cases and controls (if any) would be biased non-differentially. Furthermore, the PI controlled for two important variables that might confound the results of the case-control study age and time from admission until the case patient s first empiric antimicrobial treatment Limitations This study had several limitations. The PI collected the data on patient characteristics from the patients medical records. He could not validate the data; thus, he was dependent on what healthcare workers recorded. In addition, the PI did not collect

138 124 data on surgical procedures. However, if the PI had collected data on surgical procedures he would have been able to assess whether perioperative prophylactic antimicrobial use changed when surgical procedures increased or decreased. The goals of antimicrobial stewardship programs often include decreasing the emergence of antimicrobial resistant-organisms, decreasing the number of complications, and decreasing the number of adverse events that may occur as a result of wide-spread antimicrobial use. We did not assess whether patients had complications, such as liver damage and renal insufficiency, from antimicrobial therapy. We did not find evidence for increasing antimicrobial resistance or for frequent Clostridium difficile infections. In fact, antimicrobial susceptibility patterns remained stable and we indentified only a few nosocomial C. difficile infections during the entire study period despite widespread antimicrobial use. The current study was done in a single PICU, which may limit its external validity. However, PICUs with similar patient populations could compare data on antimicrobial use in their units with the results from this study. 5.4 Future Directions Improving the use of antimicrobials in PICUs is important because widespread antimicrobial use may increase healthcare costs, increase adverse drug events, and encourage the emergence of antimicrobial resistant organisms. Studies have found that physicians implement antimicrobial guidelines infrequently when they are introduced as the sole antimicrobial stewardship strategy. Therefore, effective antimicrobial stewardship strategies that are easy to implement are needed. In addition, clinicians would benefit from a tool that helps them discriminate bacterial infections from

139 125 nonbacterial infections and from systemic inflammatory response syndrome. The following paragraphs will address a conceptual framework, based on Rogers s (130) model of innovation diffusion that could facilitate the adoption and implementation of an antimicrobial stewardship in a PICU How to Improve Antimicrobial Prescribing Behavior Characteristics of the Innovation Rogers (130) described an innovation as an idea or practice that is perceived as new by a person (e.g., a pediatric intensivist) or other unit of adoption (e.g., a PICU). The use of the AA form met this definition of innovation. According to Rogers, the characteristics of an innovation, such as the degree to which the innovation is perceived as better than the idea it supersedes (relative advantage); the degree to which the innovation is perceived as being consistent with the existing values, the past experiences, and the needs of potential adopters (compatibility); the degree to which the innovation is perceived as difficult to understand and use (complexity); and the degree to which the results of the innovation are visible to others (observability) may explain the rate at which different people adopt an innovation. In addition, the communication process, the social system, and the users of an innovation can all affect whether the innovation is adopted. In the current study, low adherence to completing the AA form may be related to the nature of the intervention we used. Initially, nurse practitioners and a pharmacist agreed to complete the AA form but two weeks after the intervention began they said they could no longer fulfill this task. At that point we asked the residents and fellows to complete the forms. The intervention was voluntary; therefore, the residents and fellows were not obliged to complete the AA form. Consequently, the characteristics of the

140 126 intervention probably affected the adherence. Unfortunately, we did not assess the residents and fellows perceptions of the relative advantages of the AA form or the degree to which they felt it was compatible with their values and needs. We also did not assess their views on the complexity of the intervention or the visibility of the results. The PI visited the PICU every two weeks to encourage residents and fellows to complete the AA form but he was not able to improve adherence to the AA form. In the future, investigators could introduce a reward or an incentive system whereby residents and fellows are rewarded for completing an AA form. For example, residents and fellows could receive a gift card to the hospital coffee shop for completing a high percentage of their AA forms. Such a reward might help the residents and fellows feel that they benefited from completing the AA forms, and might increase adherence. Additionally, investigators could integrate an AA form into a computer order entry system. Such a system might reduce the workload for the residents and fellows and, at the same time, incorporate the data collection function of the AA form into the normal work flow, thereby, reducing both the complexity of the residents and fellows overall workload and of the intervention. Furthermore, investigators could introduce a feedback mechanism into their interventions. For example, investigators could report results such as the average duration of empiric antimicrobial treatments, any change in the average duration of empiric antimicrobial treatments, the number of empiric antimicrobial prescriptions per course, and any change in the number of empiric antimicrobial prescriptions per course to the residents, fellows, and the attending physicians every month. This feedback might make the results of the intervention more visible to the

141 127 persons who did the work of the intervention and, thus, might encourage residents and fellows to complete the AA forms Communication Process The methods through which an innovation is communicated to the users are important. Numerous investigators ( ) have found that clinicians that are judged by their colleagues as persons, who can assess whether the evidence and local situations are compatible, can facilitate the adoption of interventions. These clinicians are called opinion leaders and change champions and they can determine whether an intervention is adopted rapidly and whether the required behavior change is sustained (130). Opinion leaders and change champions can help other staff adopt an intervention (such as an AA form) because the staff members trust them and because they possess interpersonal communication skills that help promote behavior change (often referred to as peer influence). In the current study, our communication process may have been inadequate. First, we did not educate residents and fellows on the goals and objectives of the study and we did not train them on how to complete the AA form. We met with the attending physicians (faculty), a pharmacist, and two nurse practitioners when we were designing the intervention and we provided them with the preliminary data on antimicrobial use from the baseline period but we did not meet with the residents and fellows and we did not provide them with the baseline data. As noted previously, the pharmacist and nurse practitioners initially completed the AA forms but quickly decided that they could not do this function. Unfortunately, we did not meet with the residents and fellows at that point and explain the intervention to them, ask for their input, and ask for their help in

142 128 completing the project. Instead, we relied on attending physicians to explain the use of the AA form to the residents and fellows. Thus, adherence to completing the AA form might have been dependent on whether the attending physicians encouraged the residents and the fellows to do this extra task. In addition, we did the studies described in this thesis because a pediatric intensivist was concerned about wide-spread antimicrobial use in the UIHC s PICU. We had hoped that he would be an opinion leader who could facilitate the intervention. However, he left the UIHC before the study started. Subsequently, we identified a newly hired pediatric intensivist who expressed interest in our study. We hoped she would serve as both an opinion leader and a change champion. Initially it appeared that she was successful because residents and fellows completed the AA forms for 90% of the antimicrobial courses during February and March 2007 when the intervention started. However, as the intervention progressed it appeared that her influence on the residents and fellows waned. The rate of completing the AA form may have been low because she did not have enough support from her colleagues or because she had limited influence within the social system to ensure that the residents and the fellows completed the AA form. In the future, investigators wanting to utilize an AA form as an antimicrobial stewardship program could meet with the residents and the fellows before intervention starts to: (1) explain the details of their interventions, (2) review existing literature, and (3) discuss preliminary data. Investigators could also ask residents and fellows what types of rewards or incentives would help them adopt the intervention.

143 Social System and Users of an Innovation The social system may influence whether an innovation (such as an AA form) is adopted. A social system is a set of interrelated units that are engaged in joint problemsolving to accomplish a common goal (130). On the basis of this definition, the UIHCs PICU could be described as a social system in which the primary goal of the physicians, nurses, and other staff members is to save the lives of their patients. The structure of a social system can facilitate or impede the diffusion of an innovation (130). In the PICU, the attending physicians supervised the residents and the fellows. Thus, if the attending physicians strongly encouraged the residents and fellows to complete the AA form, they might have been more likely to complete the AA forms. The pediatric intensivist who coordinated the intervention part of this study signed approximately half of the 68 AA forms that were completed and compliance was very high when we initiated the intervention and she was on service. It may be possible that she encouraged residents and fellows to complete the AA form whenever she was on service, particularly at the beginning of the study. However, the PI cannot explain why the adherence became low towards the end of the intervention period even when she was on service. Individual members of a system (such as a resident or a fellow) could decide to adopt or to reject an innovation if implementing the innovation is optional. In contrast, an entire social system (the physician in a PICU) could adopt or reject an innovation. In this case, all members of a social system could collectively agree to adopt an innovation or a person (or persons) in authority could decide that all members of the social system will adopt the innovation (130). The individual members of the social system usually conform to the system s decision (130) because adherence is mandatory, because the members all

144 130 agree with the innovation, or because peer pressure influences their behavior. In the current study, the attending physicians including the medical director of the unit agreed to support an intervention that could decrease empiric antimicrobial use. However, the entire social system did not make completing the AA form a priority and they did not make completing the AA form mandatory. Thus, individual residents or fellows could decide whether or not to complete an AA form, and they often chose not to complete the AA form. In the future, investigators could seek the support of opinion leaders or change champions within their PICUs to facilitate the adoption of an AA form or investigators could seek the support of all attending physicians for a collective decision to adopt the intervention Biomarkers of Infections as a Possible Guide for Antimicrobial Use Data from the current study suggested that pediatric intensivists ordered CRP tests when they suspected bacterial infections. In the future, investigators could measure CRP serially from PICU patients in whom physicians suspect bacterial infections and from a comparison group within the same unit in whom physicians did not suspect bacterial infections. Investigators could plot the receiver operating characteristics (ROC) curves to determine the predictive value positive of CRP. If the predictive value positive is high, investigators could use the CRP cut off value to develop a predictive model that might guide empiric antimicrobial treatments. Data from the AA form suggested that pediatric intensivists most frequently checked elevated temperature, elevated CRP, elevated heart rate, and elevated WBC counts when deciding to start empiric or targeted antimicrobial treatments. In the future,

145 131 investigators could conduct studies, which include all PICU patients to determine whether those factors suggested by the current study could be used to discriminate bacterial infections from non-bacterial infections. Investigators could then develop predictive models that could be used to determine the probability of bacterial infections for those patients who appear ill or who have clinical evidence of bacterial infections. Researchers could use factors that performed well in their predictive models to develop a clinical infection index or score (CII or CIS) to guide empiric antimicrobial use. For example, researchers could do a study similar to that done by Singh and colleagues (97). They could randomize patients who have CIS 6 (hypothetical value) to either receive standard therapy (i.e., choice and duration of antimicrobial therapy at the physicians discretion) or therapy with a specific antimicrobial agent (e.g., ceftriaxone). On the third day, the clinicians could reevaluate the CIS score and determine whether or not the patient s condition has deteriorated to determine whether to continue or to stop the empiric therapy. If the use of the CIS decreases empirical use of antimicrobial in patients who do not have documented infections, the cost for antimicrobial agents could decrease, the risk of selecting resistant organisms might decrease, and complications or adverse events due to antimicrobials might also decrease. 5.5 Overall Summary and Conclusion In the pre-intervention period, most antimicrobial use in the UIHC s PICU was for prophylactic treatment of patients who underwent surgical procedures and for empiric treatment of patients with suspected infections. Patients received few courses of targeted antimicrobials. Furthermore, the current study demonstrated that pediatric intensivists in

146 132 the UIHC s PICU often considered elevated CRPs, elevated WBC counts, and elevated temperatures when deciding to start empiric antimicrobial therapy. Data from the three nested periods showed that the median duration of empiric and targeted treatments decreased during the intervention and remained stable during the post-intervention period. We estimated that 193 days of empiric antimicrobial therapy and 59 days of targeted antimicrobial therapy, respectively, may have been saved by the decreased durations of therapy. The time series analysis assessing the trend in use of antimicrobials did not reveal a significant change over time. Therefore, an AA form alone may not be an effective strategy for antimicrobial stewardship in PICUs. Additional measures such as automatic stop orders and computer decision support may be useful for reducing widespread antimicrobial use in PICUs.

147 133 APPENDIX A CHART ABSTRACTION FORM

148 134

149 135

150 136

151 137

152 138

153 139 APPENDIX B ISOGRAPHS FOR SAMPLE SIZE CALCULATION FOR MATCHED CASE- CONTROL STUDY / INTERVENTION FORM

154 Isographs for Constant Sample Size for Paired Case-control studies 140

155 141

156 142