Antibiotic resistance is a major public health problem. Infections caused by drug-resistant bacteria lead to worse clinical outcomes than infections caused by susceptible bacteria.1 1 1–3
Improving antibiotic prescribing practices is an important part of the global strategy to reduce antibiotic resistance. Studies aimed at improving antibiotic prescribing, including encouraging narrower-spectrum antibiotic prescribing, have traditionally focused on persuasive (eg, educational) or restrictive (eg, formulary restriction) interventions.4
Hospital “antibiograms,” bacteria-specific antimicrobial susceptibility profiles, are often used to support the need for improving antibiotic prescribing practices. However, antibiograms may also be used to influence antibiotic prescribing.5 6 6 , 7 Streptococcus pneumoniae , the most common bacterial cause of community-acquired pneumonia (CAP), in 2008 after studies demonstrated that narrow-spectrum antibiotics, such as penicillin and aminopenicillins, were effective in treating non-central nervous system (CNS) pneumococcal infections even when classified as nonsusceptible in vitro.7
The aim of this multicenter study was to determine the association between pneumococcal penicillin susceptibility testing results, as reported by hospital antibiograms, and physicians’ prescribing practices for children hospitalized with CAP. We used antibiograms incorporating 2005 pneumococcal susceptibility patterns, as these antibiograms would be available to physicians when prescribing antibiotics for CAP in 2006.
METHODS
Study Design and Data Sources
This multilevel cross-sectional study used hospital data collected from surveys and patient-level data obtained from an administrative database. The Pediatric Health Information System (PHIS) was used to identify hospitals that contributed group-level data and was used to gather prescribing information for patient-level data. The PHIS is a national administrative database containing resource utilization from 38 freestanding, tertiary care children’s hospitals affiliated with the Child Health Corporation of America (Shawnee Mission, KS). Participating hospitals account for 20% of all tertiary care children’s hospitals. For the purposes of external benchmarking, participating hospitals provide discharge data including patient demographics, diagnoses and procedures. Billing data detail all of the drugs, radiologic imaging studies, laboratory tests and supplies charged to each patient. Data quality and reliability are assured through a joint effort between the Child Health Corporation of America and participating hospitals as previously described.8 , 9
Group-level Data . Hospital antibiotic susceptibility patterns for S. pneumoniae were determined via written surveys sent to the microbiology laboratories of each hospital. The surveys requested information regarding antibiotic susceptibility patterns for pneumococcal isolates tested in 2005 in aggregate and, when available, by specific site (ie, blood isolates, respiratory isolates). The cut points were defined using MICs for S. pneumoniae susceptibility as established by the CLSI for 2005 as follows: ≤0.06 mcg/mL, susceptible; 0.12–1.0 mcg/mL, intermediate and ≥2.0 mcg/mL, resistant.10
Individual-level Data . Patient-level information for the calendar year 2006 was retrieved from the PHIS database. Children, 1–18 years of age, with CAP were eligible if they were discharged from any participating hospital between January 1 and December 31, 2006. Subjects were included if they received antibiotic therapy on the first day of hospitalization and if they satisfied one of the following International Classification of Diseases, 9th revision (ICD-9) discharge diagnosis code criteria: (1) Primary diagnosis of pneumonia (ICD-9 codes 481–483.8, 485–486) or (2) Primary diagnosis of a pneumonia-related symptom (ICD-9 codes 780.6 or 786.00–786.52 [except 786.1]) and a secondary diagnosis of pneumonia. Only patients receiving antibiotics considered conventional treatment for childhood CAP (ie, penicillin, macrolide, cephalosporin, vancomycin and clindamycin) on the first day of hospitalization were included.
We identified children with asthma in 2 ways. Asthma-related hospitalizations were defined by an ICD-9 code for asthma (493.0–493.2) in any discharge diagnosis field during any prior hospitalization in the 24 months before the current hospitalization. Chronic asthma controller medication use was defined by treatment with inhaled corticosteroids (eg, fluticasone) or leukotriene receptor antagonists on the first day of hospitalization for CAP, which suggested that these medications were a continuation of baseline therapy.
Data from 5 of the 38 hospitals were excluded because of incomplete patient-level information (n = 3) or because an incomplete antibiogram was returned (n = 2). Children younger than 1 year of age were excluded because they experience a high rate of viral respiratory infections that are difficult to distinguish clinically from bacterial pneumonia. Patients with empyema or pleurisy were excluded as coverage for Staphylococcus aureus with a broad-spectrum antibiotic is typically required. Patients with comorbid conditions that predisposed them to severe or recurrent pneumonia (eg, cystic fibrosis, malignancy, sickle cell disease) were excluded using a previously reported classification scheme.11
Measured Exposures
The primary exposure of interest was the proportion of penicillin-nonsusceptible pneumococcal isolates reported in 2005 by each hospital. An isolate was considered nonsusceptible if it was classified as either intermediate or resistant. Secondary exposures included using the proportion of penicillin-resistant pneumococcal isolates to determine whether a threshold of susceptibility existed, as well as restricting the exposures to blood or respiratory penicillin-nonsusceptible pneumococcal isolates as these isolates were more likely than aggregated isolates to represent invasive disease.
Measured Outcomes
The primary outcome was the receipt of empiric broad-spectrum antibiotic therapy in 2006 (ie, an antibiotic other than penicillin or aminopenicillins). A subanalysis was performed excluding patients who received macrolide therapy alone or in combination as it was most likely prescribed for suspicion of an atypical bacterium as the cause of their pneumonia and not S. pneumoniae .
Data Analysis
Categorical variables were described using frequencies and percentages. χ 2 12 13 , 14
The first model, considered the “empty” model, contained the random-intercept only and no other predictor variables. This model accounted for the probability of receiving broad-spectrum antibiotic prescribing as a function of which hospital the patient attended. The second model, an extension of the “empty” model, added the proportion of penicillin-nonsusceptible pneumococcal isolates. By including the proportion of penicillin-nonsusceptible pneumococcal isolates as a separate measure in the model we were able to estimate its specific effect on antibiotic prescribing. The proportion of penicillin-nonsusceptible pneumococcal isolates was grand-mean centered at 52% (standard deviation [SD]: 11.4).15 S. pneumoniae , 43% (SD: 15.3) for penicillin-nonsusceptible S. pneumoniae blood isolates and 54% (SD: 13.9) for penicillin-nonsusceptible S. pneumoniae respiratory isolates.
The third model tested individually the inclusion of potential effect modifiers, age and asthma status. These interaction terms, determined a priori, remained in the model if the main effect of nonsusceptible pneumococcal patterns and aminopenicillin prescribing changed by 10% or more.16 17
The median odds ratio (OR) quantified the heterogeneity between different hospitals. The median OR, calculated using the variance of the hospitals in each model, is the median value of the ORs when comparing all possible pairs of patients with similar covariates admitted to different hospitals.19 , 20
RESULTS
Hospital Exposure
Hospitals reported the percentage of pneumococcal isolates tested in 2005 that were susceptible to penicillin, overall and, when available, by specific site (ie, blood isolates, respiratory isolates; Table 1 ).
TABLE 1: Variability in Proportion of Penicillin-nonsusceptibility in Streptococcus pneumoniae Across All Hospitals*
Patient Characteristics
There were 4888 patients from 33 hospitals. Narrow-spectrum antibiotic therapy was prescribed to 348 (7%) of the 4888 children with CAP. Of these children, 295 (85%) were 1–5 years old, 37 (10%) were 6–11 years old and 16 (5%) were 12–18 years old. Patients who received a narrow-spectrum antibiotic were between 1 and 5 years old (85% compared with 65%, P < 0.0001), more likely to have a prior asthma-related hospitalization (18% compared with 13%, P = 0.01) and more likely to have received a chronic asthma medication (26% compared with 21%, P = 0.02) than those receiving empiric broad-spectrum antibiotics.
Variability in Antibiotic Prescribing
Commonly used antibiotics were classified into 7 categories based on their spectrum of antibacterial activity to describe hospital variability in antibiotic prescribing (Fig. 1 ). Overall, 45% (n = 2194) of all the patients received cephalosporins as empiric therapy for CAP; cephalosporins were also the most commonly prescribed antibiotic within each hospital. One exception was a hospital where penicillins or aminopenicillins alone were prescribed at a much higher rate, 57%, compared with other hospitals, of which penicillins or aminopenicillin alone accounted for 7% of the total proportion of antibiotics prescribed for CAP during the study period.
Association of Resistance and Prescribing
In the unadjusted analysis there was no correlation between the proportion of penicillin-nonsusceptible pneumococcal isolates and broad-spectrum antibiotic prescribing (Spearman correlation = 0.070, P = 0.703). In the adjusted analysis there was no association between the proportion of penicillin-nonsusceptible pneumococcal isolates and broad-spectrum antibiotic prescribing, either overall or when restricted to blood or respiratory isolates (Table 2 ). However, the association between the proportion of penicillin-resistant pneumococcal isolates and broad-spectrum antibiotic prescribing was significant; patients were 39% more likely to receive broad-spectrum antibiotics for every 10% increase in penicillin-resistant pneumococcal isolates (Table 2). There was no substantive change when patients who received macrolide therapy alone were excluded from the analysis (adjusted OR: 1.38, 95% CI: 1.08–1.67) nor when patients who received macrolide therapy in combination were excluded from the analysis (adjusted OR: 1.41, 95% CI: 1.08–1.73).
One hospital was unique in its prescribing practices (Fig. 1 ) and its inclusion increased the variance component in the models. There was no association between proportion of penicillin-nonsusceptible S. pneumoniae isolates and broad-spectrum prescribing when this hospital was excluded (adjusted OR: 1.07, 95% CI: 0.79–1.35) but the overall fit of the model improved (AIC with the outlier was 2144.3 versus 1984.8 without the outlier). This finding indicates that some, but not all, of the variability between hospitals is due to this hospital, which had a disproportionate amount of narrow-spectrum antibiotic prescribing. In addition, we found no statistical association between the proportion of penicillin-nonsusceptible pneumococcal isolates and broad-spectrum antibiotic prescribing when limiting the data set to the 20 hospitals that reported information in each of the 3 breakpoints (adjusted OR: 1.31; 95% CI: 0.97–1.66).
The median OR for penicillin-nonsusceptible isolates, overall and site-specific, indicated large variability among hospitals in broad-spectrum prescribing (Table 3 ). In contrast, there was less variability between hospitals in the probability of prescribing broad-spectrum antibiotics when adjusting for penicillin-resistant pneumococcal isolates; on average a patient had 1.62 higher odds of receiving a broad-spectrum antibiotic solely based on which hospital they were admitted. The least amount of variability existed between hospitals in prescribing broad-spectrum antibiotics when only accounting for the proportion of penicillin-resistant S. pneumoniae isolates.
TABLE 2: Random-intercept Multilevel Model Predicting the Probability of Being Prescribed a Broad- spectrum Antibiotic*
TABLE 3: Median Odds Ratio for Unadjusted and Adjusted Models to Quantify the Heterogeneity in Broad-spectrum Antibiotic Prescribing Across Hospitals
FIGURE 1: Variability in antibiotic prescribing across hospitals. Each bar represents data from one hospital.
DISCUSSION
This multicenter study found substantial variability in empiric antibiotic prescribing for CAP among children’s hospitals in the United States. High-levels (ie, resistant) but not modest-levels (ie, intermediate susceptibility) of penicillin resistance were associated with broad-spectrum antibiotic prescribing. As narrow-spectrum antibiotics effectively treat most non-CNS pneumococcal infections, our findings suggest that a physician prescribing for CAP is responsive to hospital pneumococcal antibiotic susceptibility patterns. The broader implication of this finding is that strategies to optimally align antibiotic susceptibility patterns and clinical outcomes can lead to meaningful decreases in broad-spectrum antibiotic prescribing.
The degree of variability in empiric therapy prescribing for CAP in this study is similar to prior studies investigating general antibiotic prescribing.21 22 23 24 25
The Centers for Disease Control and Prevention reported that the number of nonmeningitis pneumococcal isolates categorized as resistant decreased from 10.3% to 1.2% using the 2008 CLSI breakpoints.7 S. pneumoniae . In previous studies, however, clinicians typically used antibiograms to prescribe broader-spectrum empiric therapy and continued broad-spectrum antibiotic therapy even when the bacteria were identified as susceptible to narrower-spectrum antibiotics.26
The CLSI determines breakpoints by reviewing the MICs, the pharmacokinetic and pharmacodynamic information for each antimicrobial/pathogen combination, and the data from clinical trials or well-documented case series.27
The CLSI breakpoints that define the interpretative categories in antibiograms must align with clinical outcomes as they influence the choice of empiric therapy. Hospital antibiograms are known to overestimate community-level drug resistance and may, in the community setting, prompt broad-spectrum prescribing. Urinary tract infections, predominantly caused by Escherichia coli, are examples of how breakpoints determined in vitro may contribute to broad-spectrum antibiotic prescribing.28
This study has several limitations. First, there is no information on the patients from whom these isolates were obtained and reported in the antibiograms. Variability in reported susceptibility patterns across hospitals may be due in part to the differences in the underlying patient populations and in part to the number of isolates used to determine susceptibility patterns at each hospital.5
Second, we assumed that the proportion of resistant or nonsusceptible pneumococcal isolates reported by each hospital was the only measure disseminated to physicians and, consequently, to influence their prescribing practices. The outlier hospital in our study reported 46% of pneumococcal isolates to be nonsusceptible to penicillin. However, this hospital had the highest proportion of narrow-spectrum antibiotic prescribing (69%) across all hospitals. Therefore, there may be other determinants that affect antimicrobial prescribing such as hospital policies to direct prescribing (eg, formulary restriction or prior authorization required), the preference of antibiotic in each subspecialty, and the dynamic and expertise of the team of health professionals providing care.29
Third, the use of ICD-9 codes to identify patients with CAP may result in misclassification of disease. However, the ICD-9 CM codes used in this study are similar to previous studies that have shown a relatively high sensitivity and specificity for identifying CAP compared with medical record review.30 , 31
Last, limitations exist in the use of multilevel analysis. The fixed variables available for this analysis may not have accounted for all the different factors that drive a physician’s prescribing practice (ie, the patient’s medical history). However, these unmeasured variables are by default incorporated into the random intercept in the model and could also explain some of the variability that was seen in the model.14
In conclusion, high levels of resistance reported in an antibiogram were associated with broad-spectrum empiric antibiotic therapy. The decision by the CLSI to modify penicillin breakpoints for S. pneumoniae , the most common bacterial cause of childhood pneumonia, was based on considerable supporting clinical and microbiologic data. However, few empiric data were available to suggest that physicians are responsive to the antibiotic susceptibility patterns and that the breakpoint changes would actually change prescribing behavior. Our study suggests that physicians’ response to the penicillin-resistant pneumococcal antibiotic susceptibility patterns may result in improved antibiotic prescribing in light of the recently modified MIC breakpoints for S. pneumoniae .
ACKNOWLEDGMENTS
The authors thank the following investigators for contributing pneumococcal susceptibility data used in this study: Fariba Asghari, Rady Children’s Hospital, San Diego, CA; Elaine B. Dowell, Children’s Hospital, Aurora, CO; Rangaraj Selvarangan, Children’s Mercy Hospital, Kansas City, MO; Anami Patel, Le Bonheur Children’s Medical Center, Memphis, TN; Joseph Campos, Children’s National Medical Center, Washington, D.C.; Robert Wadowsky, Children’s Hospital of Pittsburgh, Pittsburgh, PA; Cindy Maurer, Akron Children’s Hospital, Akron, OH; Keith Zucker, Children’s Hospital Central California, Madera, CA; Caroline Tyndall, Children’s Hospital of The King’s Daughters, Norfolk, VA; Kathleen Eisenach, Arkansas Children’s Hospital, Little Rock, AR; Dennis Kmetz, Children’s Hospital of Orange County, Orange, CA; Sylvia Cuate, Driscoll Children’s Hospital, Corpus Christi, TX; Maria Staeheli, Miami Children’s Hospital, Miami, FL; Mario Marcon, Nationwide Children’s Hospital, Columbus, OH; Linda Snow, Cook Children’s Medical Center, Fort Worth, TX; Kathy Smith, Children’s Hospital of New Orleans, New Orleans, LA; Jane Schilleci, The Children’s Hospital of Alabama, Birmingham, AL; Jeanette Manley, Children’s Hospital and Medical Center, Omaha, NE; Karen Sue Kehl, Children’s Hospital of Wisconsin, Milwaukee, WI; Vera Concho, Texas Children’s Hospital, Houston, TX; Joel E. Mortensen, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH; Gregory A. Storch, Washington University School of Medicine/St. Louis Children’s Hospital, St. Louis, MO; Robert C. Jerris, Children’s Healthcare of Atlanta, Atlanta, GA; Xuan Qin, Seattle Children’s Hospital, Seattle, WA; Karin L. McGowan, Children’s Hospital of Philadelphia, Philadelphia, PA; Patricia Ackerman, Children’s Hospitals and Clinics of Minnesota, Minneapolis, MN; Hossein Salimnia, Children’s Hospital of Michigan, Detroit, MI; Charles Stratton, Vanderbilt Children’s Hospital, Nashville, TN; Phyllis Della-Latta, Children’s Hospital of New York-Presbyterian, New York, NY; Shari Young, Children’s Medical Center of Dallas, Dallas, TX; Kristie Vetterli, Children’s Hospital and Research Center Oakland, Oakland, CA; Pravin H. Patel, Women and Children’s Hospital of Buffalo, Buffalo, NY; Eileen Gorss, Children’s Hospital Boston, Boston, MA .
REFERENCES
1. Nugent R, Back E, Beith A The Race Against Drug Resistance.. 2010 Washington, DC Center for Global Development
2. Peters NK, Dixon DM, Holland SM, et al. The research agenda of the National Institute of Allergy and Infectious Diseases for antimicrobial resistance. J Infect Dis. 2008;197:1087–1093
3. The Infectious Diseases Society of America.Bad Bugs, No Drugs.. 2004 Alexandria, VA The Infectious Diseases Society of America
4. Davey P, Urquhart L. Will a reduction in community prescribing of quinolones decrease the prevalence of quinolone resistance? Clin Infect Dis. 2009;49:876–877
5. Lacy MK, Klutman NE, Horvat RT, et al. Antibiograms: new NCCLS guidelines, development, and clinical application. Hosp Pharm. 2004;39:542–553
6. Clinical and Laboratory Standards Institute.Performance Standards for Antimicrobial Susceptibility Testing; Twenty-First Informational Supplement.. 2011 Wayne, PA Clinical and Laboratory Standards Institute
7. Centers for Disease Control and Prevention.. Effects of new penicillin susceptibility breakpoints for Streptococcus pneumoniae—United States, 2006–2007. MMWR Morb Mortal Wkly Rep. 2008;57:1353–1355
8. Mongelluzzo J, Mohamad Z, Ten Have TR, et al. Corticosteroids and mortality in children with bacterial meningitis. JAMA. 2008;299:2048–2055
9. Shah SS, Hall M, Srivastava R, et al. Intravenous immunoglobulin in children with streptococcal toxic shock syndrome. Clin Infect Dis. 2009;49:1369–1376
10. Feigin R, Cherry J, Demmler-Harrison G, et al. Feigin and Cherry’s Textbook of Pediatric Infectious Diseases..6th ed. Philadelphia, PA Saunders Elsevier2009
11. Feudtner C, Hays RM, Haynes G, et al. Deaths attributed to pediatric complex chronic conditions: national trends and implications for supportive care services. Pediatrics. 2001;107:E99
12. Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics. 1986;42:121–130
13. Hubbard AE, Ahern J, Fleischer NL, et al. To GEE or not to GEE: comparing population average and mixed models for estimating the associations between neighborhood risk factors and health. Epidemiology. 2010;21:467–474
14. Subramanian SV, O’Malley AJ. Modeling neighborhood effects: the futility of comparing mixed and marginal approaches. Epidemiology. 2010;21:475–8; discussion 479
15. Raudenbush SW, Bryk AS Hierarchical Linear Models: Applications and Data Analysis Methods.. 20022nd ed. Thousand Oaks, CA Sage
16. Maldonado G, Greenland S. Simulation study of confounder-selection strategies. Am J Epidemiol. 1993;138:923–936
17. Verbeke G, Molenberghs G Linear Mixed Models for Longitudinal Data.. 2000 New York, NY Springer
18. Long SS, Pickering LK, Prober CG Principles and Practice of Pediatric Infectious Diseases.. 20083rd ed. New York, NY Churchill Livingstone
19. Merlo J, Chaix B, Ohlsson H, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health. 2006;60:290–297
20. Larsen K, Petersen JH, Budtz-Jørgensen E, et al. Interpreting parameters in the logistic regression model with random effects. Biometrics. 2000;56:909–914
21. Gerber JS, Newland JG, Coffin SE, et al. Variability in antibiotic use at children’s hospitals. Pediatrics. 2010;126:1067–1073
22. Bradley JS, Byington CL, Shah SS, et al. The management of community-acquired pneumonia in infants and children older than 3 months of age: clinical practice guidelines by the Pediatric Infectious Diseases Society and the Infectious Diseases Society of America. Clin Infect Dis. 2011;53:e25–e76
23. Yu VL, Chiou CC, Feldman C, et al.International Pneumococcal Study Group. An international prospective study of pneumococcal bacteremia: correlation with
in vitro resistance, antibiotics administered, and clinical outcome. Clin Infect Dis. 2003;37:230–237
24. Cardoso MR, Nascimento-Carvalho CM, Ferrero F, et al.CARIBE Group. Penicillin-resistant pneumococcus and risk of treatment failure in pneumonia. Arch Dis Child. 2008;93:221–225
25. Weinstein MP, Klugman KP, Jones RN. Rationale for revised penicillin susceptibility breakpoints versus Streptococcus pneumoniae: coping with antimicrobial susceptibility in an era of resistance. Clin Infect Dis. 2009;48:1596–1600
26. Saha SK, Darmstadt GL, Baqui AH, et al. Rapid identification and antibiotic susceptibility testing of Salmonella enterica serovar Typhi isolated from blood: implications for therapy. J Clin Microbiol. 2001;39:3583–3585
27. Jorgensen JH. Who defines resistance? The clinical and economic impact of antimicrobial susceptibility testing breakpoints. Semin Pediatr Infect Dis. 2004;15:105–108
28. Khawcharoenporn T, Vasoo S, Ward E, et al. High rates of quinolone resistance among urinary tract infections in the ED. Am J Emerg Med. 2012;30:68–74
29. Hulscher ME, Grol RP, van der Meer JW. Antibiotic prescribing in hospitals: a social and behavioural scientific approach. Lancet Infect Dis. 2010;10:167–175
30. Guevara RE, Butler JC, Marston BJ, et al. Accuracy of ICD-9-CM codes in detecting community-acquired pneumococcal pneumonia for incidence and vaccine efficacy studies. Am J Epidemiol. 1999;149:282–289
31. Whittle J, Fine MJ, Joyce DZ, et al. Community-acquired pneumonia: can it be defined with claims data? Am J Med Qual. 1997;12:187–193
