Variability in Rates of Use of Antibacterials Among 130 US Hospitals and Risk‐Adjustment Models for Interhospital Comparison
Objective. To describe variability in rates of antibacterial use in a large sample of US hospitals and to create risk‐adjusted models for interhospital comparison.
Methods. We retrospectively surveyed the use of 87 antibacterial agents on the basis of electronic claims data from 130 medical‐surgical hospitals in the United States for the period August 2002 to July 2003; these records represented 1,798,084 adult inpatients. Hospitals were assigned randomly to the derivation data set (65 hospitals) or the validation data set (65 hospitals). Multivariable models predicting rates of antibacterial use were created using the derivation data set. These models were then used to predict rates of antibacterial use in the validation data set, which was compared with observed rates of antibacterial use. Rates of antibacterial use was measured in days of therapy per 1,000 patient‐days.
Results. Across the surveyed hospitals, a mean of 59.3% of patients received at least 1 dose of an antimicrobial agent during hospitalization (range for individual hospitals, 44.4%‐73.6%). The mean total rate of antibacterial use was 789.8 days of therapy per 1,000 patient‐days (range, 454.4‐1,153.4). The best model for the total rate of antibacterial use explained 31% of the variance in rates of antibacterial use and included the number of hospital beds, the number of days in the intensive care unit per 1,000 patient‐days, the number of surgeries per 1,000 discharges, and the number of cases of pneumonia, bacteremia, and urinary tract infection per 1,000 discharges. Five hospitals in the validation data set were identified as having outlier rates on the basis of observed antibacterial use greater than the upper bound of the 90% prediction interval for predicted antibacterial use in that hospital.
Conclusion. Most adult inpatients receive antimicrobial agents during their hospitalization, but there is substantial variability between hospitals in the volume of antibacterials used. Risk‐adjusted models can explain a significant proportion of this variation and allow for comparisons between hospitals for benchmarking purposes.
Received August 14, 2007; accepted December 5, 2007; electronically published February 1, 2008.
Antibacterial use in healthcare institutions is one of the primary forces driving the spread of drug resistance among human pathogens.1‐4 Strategies advocated by government and professional organizations to curb the spread of antibacterial resistance emphasize the need to accurately measure and monitor use of antibacterials.3,5,6 Because a greater aggregate rate of antibacterial use has been associated with increased rates of resistance at the national level and the hospital level,7‐11 the Centers for Disease Control and Prevention has included the goal of improvement in the monitoring of antimicrobial usage in healthcare systems as part of its action plan to combat bacterial resistance.2 Few recent studies are available that describe antimicrobial use in US healthcare institutions.12‐14 The growth in digitalization of healthcare information will increase the opportunities for healthcare institutions to measure their performance in outcomes and processes of care, including antibacterial use.15 To put their data on antibacterial use in context, healthcare institutions will need comparative data, either by monitoring their own usage over time or by comparing their own usage with that of other institutions. The latter approach is known as “benchmarking” and is an increasingly important tool for quality improvement in health care.16,17 Benchmarking aims to quantify variability in practices and to identify institutions whose performance substantially departs from some absolute or normative standard. Data that are available from National Nosocomial Infections Surveillance System and other multicenter surveys of antibacterial use have demonstrated wide variations in rates of antibacterial use between different hospitals.12‐14 Thus, the potential exists to make benchmark comparisons between the rates of use of antibacterials among institutions as an avenue to improve antibacterial utilization.
Although there is no absolute standard for the “correct” aggregate rate of antibacterial use, several studies suggest that antibacterials are used excessively.18‐25 Thus, healthcare institutions that use greater quantities of antibacterials relative to their peer institutions are likely to be of more concern, from a quality standpoint, than are institutions in which the overall rate of antibacterial use is low. A key challenge in benchmarking is to provide valid comparators for institutions, because institutions may vary substantially in the types of patients for whom they care.26 The Infectious Diseases Society of America and Society for Healthcare Epidemiology of America guidelines for antimicrobial stewardship state “…hospitals may compare their antimicrobial use with that of other similar hospitals, recognizing the challenges of interhospital comparisons and the potential need for ‘risk adjustment’.”27(p171) Although institutions can be compared on the basis of superficial criteria, such as number of beds and teaching status, a more sophisticated approach is to identify hospital‐level factors associated with the variable of interest and build risk‐adjustment models to correct for these factors. Comparisons are then made among the model‐predicted values of the variable and the observed values, and those institutions whose observed values fall outside an acceptable range (“outliers”) are flagged as having potential quality concerns.26 This approach has been used to benchmark hospitals for such indicators as mortality and length of stay for coronary artery bypass surgery, myocardial infarction, and pneumonia, with the results made publicly available.28,29 Institutions identified as having high‐end outlying rates of antibacterial use may be motivated to investigate the reasons for their excess usage and take steps, where appropriate, to better manage antibacterial use. This could result in benefits in reduction of selective pressure for the development of antibacterial resistance, as well as potentially reducing costs of health care.
We performed an analysis of antibacterial use among 130 US hospitals to describe the current patterns of use, to create risk‐adjusted models for prediction of aggregate rates of antibacterial use, and to assess the performance of these models in benchmark comparison of antibacterial use across institutions.
Methods
Study Design and Data Source
The study was an observational, retrospective analysis of antimicrobial use among 130 hospitals that contribute to the AC Tracker database of Solucient (now part of Thomson Healthcare), an information‐products company serving the healthcare industry. The AC Tracker database contains information on drug utilization and demographic information derived from patient‐level billing data. Data were aggregated at the hospital level, preventing the identification of individual patients. The investigators were blinded to the identity of the study hospitals. The study was approved by the Institutional Review Board of Virginia Commonwealth University.
Data on Antibacterial Use
Data on total amounts of antimicrobial drugs used were obtained from hospital billing records for patients discharged during the 12‐month period from August 1, 2002, through July 31, 2003. Antimicrobial agents were identified by Solucient’s proprietary Standard Transaction Coding, which incorporates standard definitions, such as American Hospital Formulary Service classification codes, to identify unique drugs. A total of 87 systemic antibacterial preparations were included in the database (a complete list is available from the authors). Use of antiviral, antimycobacterial, and antiparasitic drugs was not analyzed. Antifungal use was included in the determination of the percentage of patients who received any antimicrobial agent but was not further investigated. Patient records were screened by age, and the records of patients younger than 18 years were excluded; the results represent antibacterial use in adult patients only. The number of patients who received each drug and the total number of days of therapy with that drug were recorded for each hospital. Measuring the number of days of therapy allows the quantification of drug use without the need to use a standardized defined daily dose to estimate drug use. We have shown previously that use of the number of days of therapy is a viable alternative to defined daily doses in describing aggregate rates of antimicrobial use and overcomes some of the shortcomings of methods that use the defined daily dose.30 One day of therapy represents the administration of a single agent on a given day, regardless of the number of doses or the size of the doses administered. For example, administration of cefazolin as a single dose or as 3 doses given 8 hours apart would both represent 1 day of therapy. A single patient who received both vancomycin and ceftazidime would be recorded as having received 2 days of therapy (1 of vancomycin and 1 of ceftazidime). The rate of antimicrobial use at each hospital was expressed as the number of days of therapy per 1,000 patient‐days.
Hospital Characteristics
Data on the characteristics of each hospital were obtained from the Solucient database. Hospital size is reported as the number of licensed beds. Hospitals that had an accredited residency program were considered teaching hospitals; hospitals could also be considered teaching hospitals on the basis of a formula that takes into account membership in the Council of Teaching Hospitals of the American Medical Association, accreditation from the Joint Commission on Accreditation of Health Care Organizations, medical school affiliation, and nursing school affiliation. Geographic regions reflect US Census regions: Northeast, North‐Central, South, and West. Data on patient demographic characteristics reflect adult patients only. The case‐mix index was calculated as the average diagnosis‐related group weights for discharged Medicare patients.31 The total number of patient discharges and days of hospital stay for all adult patients discharged were obtained from the database. Patients could be counted more than once if they were discharged multiple times during the study period. The total number of days patients spent in the hospital’s intensive care unit (ICU) were aggregated and expressed as ICU‐days per 1,000 patient‐days. The number of surgical patients at each hospital was determined on the basis of diagnosis‐related group codes for surgical patients and was expressed as the number of surgical patients per 1,000 discharges. The total number of International Classification of Diseases, Ninth Edition (Clinical Modification) (ICD‐9‐CM) codes for each of the following infections were recorded: bacteremia (codes 038.0‐038.9), pneumonia (codes 481‐486.99), and urinary tract infection (codes 590.1‐590.3, 590.8, 595.0, and 599.0). These infection‐related ICD‐9‐CM codes were normalized by expressing them as the count per 1,000 discharges.
Statistical Analysis
Descriptive statistics were compiled for all the hospitals in the data set to allow assessment of the generalizability of the study. The 130 hospitals in the study were then randomly divided into 2 equally sized subgroups: a derivation data set and a validation data set. This approach attempts to reduce the problem of “overfitting” a model on the basis of a limited data set and provides a basis for assessment of model performance.32 The derivation data set was used to create multivariable models to predict rates of antibacterial use. Although our outcome variable (rate of antimicrobial use expressed as days of therapy per 1,000 patient‐days) is derived from count data, we used standard linear regression rather than Poisson or negative binomial models based on the approximately normal distribution of the outcome. Initially, univariable linear regression was used to characterize the unadjusted relationships between the variables for hospital characteristics and the rate of antibacterial use. These variables were then used to construct multivariable models to explain the observed variance in rates of antibacterial use among hospitals. Separate models were constructed for the total rate of antibacterial use and for the rates of use of selected groups of antibacterial agents: fluoroquinolones, third‐ and fourth‐generation cephalosporins, carbapenems, antipseudomonal penicillins (piperacillin and ticarcillin, with and without β‐lactamase inhibitors), and vancomycin. All predictor variables were initially entered into the models. Variables with the weakest association were sequentially removed from the model until the most parsimonious model with the greatest adjusted 
value was achieved. The best model for each antibacterial group was then applied to the validation data set to produce a predicted rate of antimicrobial use (in days of therapy per 1,000 patient‐days) for each hospital in the validation set. Standard errors for each individual prediction were used to create 90% prediction intervals around the predicted value. Prediction intervals reflect the uncertainty around an individual outcome predicted by the model and are wider than confidence intervals around the estimated mean of a parameter. The 90% intervals were chosen to improve sensitivity to outlying values. Observed values for the rate of antimicrobial use (in days of therapy per 1,000 patient‐days) were compared with predicted values, and hospitals whose observed rate of use was outside the bounds of the 90% prediction interval were considered to have outlying rates. Statistical analysis was performed using Stata SE (StataCorp).
Results
Hospital Characteristics
Table 1 displays the characteristics of the study hospitals. Fifteen hospitals (11.5%) were classified as teaching hospitals. The study hospitals were primarily located in the South (68 hospitals) and North‐Central (51 hospitals) regions; 10 hospitals were located in the West, and a single hospital was located in the Northeast.
Antimicrobial Use
Across all study hospitals, a mean of 59.3% of all patients (range, 44.4%‐73.6%) received at least 1 dose of an antimicrobial agent (antibacterial or antifungal) during their hospital stay. The mean total rate of antibacterial use in adults (±SD) across all hospitals was 789.8 ± 123.5 days of therapy per 1,000 patient‐days (range, 454.4‐1153.4). Table 2 displays the distribution of total rates of antibacterial use and the 10 most commonly used antibacterials across all study hospitals. These 10 antibacterials accounted for 62.7% of the total rate of antibacterial use. The 20 most commonly used antibacterials accounted for 87.5% of the total rate of antibacterial use (data not shown).
Unadjusted Relationships Between Hospital Factors and Total Rate of Antibacterial Use
For the 65 hospitals in the derivation data set, the strongest association with the total rate of antibacterial use was seen for the factors number of ICU‐days and number of cases of bacteremia per 1,000 discharges (Table 3). Number of beds and teaching status were not significantly associated with total antibacterial use on unadjusted analysis.
Modeling and Prediction of Rates of Antibacterial Use
The model that incorporated the factors number of beds, number of ICU‐days per 1,000 patient‐days, and number of surgeries per 1,000 discharges, and number of cases of pneumonia, bacteremia, and urinary tract infection per 1,000 discharges provided the best model fit (
) with the fewest number of variables (Table 4). This model was applied to the derivation data set. When the values for the validation data set were entered into the model, the mean predicted rate of antibacterial use was 794.2 days of therapy per 1,000 patient‐days, compared with an actual mean observed rate of 784.4 days of therapy per 1,000 patient‐days in the validation data set. The predicted values ranged from 609.4 to 919.5 days of therapy per 1,000 patient‐days; the lowest boundary of the 90% prediction interval was 438.1 and the highest boundary was 1102.2 days of therapy per 1,000 patient‐days. The Figure displays the comparison of actual with predicted values and ranges. Five hospitals had observed rates of antibacterial use greater than the upper bound of the 90% prediction interval of the predicted rate, 5 had an observed rate below the lower bound of the 90% prediction interval of the predicted rate, and a number of hospitals had observed rates that bordered very closely on the limits of the 90% prediction intervals of the predicted rate.
Figure. Comparison of observed and predicted total rates of antibacterial use among hospitals in the validation data set. The model used to calculate the predicted rate was as follows: predicted rate of antibacterial use = 0.09(no. of beds) + 0.74(no. of intensive care unit days per 1,000 patient‐days) + 0.43(no. of surgeries per 1,000 discharges) + 3.30(no. of cases of pneumonia per 1,000 discharges) + 6.89(no. of cases of bacteremia per 1,000 discharges) + 1.68(no. of cases of urinary tract infection per 1,000 discharges) + 237.08.
Separate multivariable models were constructed to predict use of fluoroquinolones, third‐ and fourth‐generation cephalosporins, carbapenems, antipseudomonal penicillins (piperacillin and ticarcillin with and without β‐lactamase inhibitors), and vancomycin (Table 4). Model fit values (
values) varied from 0.063 for antipseudomonal penicillins to 0.584 for vancomycin. Outlying rates were identified for each drug class.
Discussion
In this survey of antibacterial use in US hospitals, a mean of almost 6 of 10 adult patients received at least 1 dose of an antimicrobial during their hospitalization, although there was substantial interhospital variation. The rate of antibacterial use normalized for the number of patient‐days for each hospital varied over an almost 3‐fold range, from 454 to 1153 days of therapy per 1,000 patient‐days. Only a handful of published studies have described the pharmacoepidemiology of antimicrobial use across multiple hospitals. Table 5 summarizes multihospital studies that presented either an estimation of the proportion of patients who received an antimicrobial or some measure of the rate of antimicrobial use. Although the hospital populations and measurements varied by study, compared with previous studies in the United States, the volume of antibacterial use appears to be greater in the current study and also higher than that reported in comparable studies from outside the United States (with the exception of a study from Taiwan35). Future studies using standardized measurements would aid in understanding rates of antimicrobial use across regions and the changes in usage patterns over time.
Approximately 31% of the variability in rates of antibacterial use could be accounted for by the hospital‐specific factors we measured, including number of beds, surgical volume, case‐mix index, number of ICU‐days, and number of infections. The purpose of such risk adjustment is to determine what fraction of antimicrobial use is accounted for by “unmodifiable” factors, such as patient mix, allowing scrutiny of the remaining variability. Some of this remaining variability may represent a quality deficit—in the case of antibacterial use, excessive and inappropriate use of antibacterials. Allowing for random variation by constructing 90% prediction intervals around the predicted values permitted the identification of hospitals with outlier rates of total antibacterial use or use of specific antibacterial classes that may warrant examination. Variables that have been previously recommended for risk adjustment in benchmarking rates of antimicrobial use among hospitals are number of beds and teaching status.38 The presumption is that these 2 factors are markers for hospital characteristics that affect antimicrobial use; thus, matching hospitals by these 2 factors allows hospitals to make comparisons with their “peers.” We found that these factors were weak predictors of rates of antibacterial use after normalization for total number of patient‐days. Other factors, such as case‐mix index, number of ICU‐days, surgical volume, and number of infection‐related ICD‐9‐CM codes contributed more strongly to the predictive value of the model. These administrative data are frequently captured by hospitals from electronic billing data and are generally standardized to allow for interhospital comparisons. Healthcare institutions participating in group purchasing and/or benchmarking networks would be especially well positioned to incorporate benchmarking of rates of antibacterial use into their quality improvement programs.
This study has a number of limitations. Most importantly, our sample is not a random sample of all US hospitals. Although the percentage of hospitals in our sample that were community (as opposed to teaching) hospitals approximates that seen in the United States as a whole (85%),39 we cannot be sure that hospitals in our sample are representative of US hospitals in terms of the volume of antibacterial use or patient demographic characteristics. We obtained data on antibacterial use and patient demographic characteristics from hospital billing records. The exact protocols and definitions used in electronic data capture and data cleaning are unique to this hospital network. Hospitals interested in benchmarking should ensure that data are processed comparably at their comparator institutions. Drug identification from electronic billing data requires that the hospital have a unique billing code for each agent. Some hospitals may not have a unique code for uncommonly used agents; thus, these agents may be underrepresented. However, it is unlikely that this would significantly bias estimates of total rates of antibacterial use, because the 20 most commonly used antibacterials accounted for 88% of the total rate of antibacterial use.
We quantified antibacterial use by using days of therapy rather than defined daily doses, which has been used as a standard to express aggregate rates of drug use.40 However, estimates based on defined daily doses may be inaccurate because the World Health Organization standards for dosages do not accurately reflect the dosages used in clinical practice for many antimicrobials in the United States, as we have shown in a previous study.30 This is especially problematic when the total rate of antibacterial use is measured, because the differential use of agents for which the defined daily dose is not reflective of the dose used in practice across hospitals will bias the total rate of antibacterial use.41 Whether use of days of therapy or use of defined daily dose values is a better predictor of the ecologic effect of antimicrobials on resistance rates in pathogens remains to be determined. We determined the number of infections that occurred in the hospital during the study period by the number of infection‐related ICD‐9‐CM codes. There may be substantial variability in coding practices among hospitals, such that the actual number of infections may be over‐ or underestimated.42
The predictor variables we selected for this study were chosen because data on these variables are readily availabile from administrative databases and they have face validity as potential predictors of the aggregate rate of antibacterial use. Other unmeasured hospital‐specific characteristics, such as presence of an antimicrobial stewardship program, may also be important predictors of the total rate of antimicrobial use. Future studies that incorporate a wide range of potential predictor variables are warranted. Because antimicrobial use is necessary for patient care, it is not possible to create an absolute standard for volume of antimicrobial use (eg, in contrast to a goal of zero for the surgical‐site infection rate). However, given that studies have documented that antimicrobial use is generally excessive, if a hospital’s level of antimicrobial use is substantially greater than that of its peers, it seems likely that this represents overuse on that hospital’s part rather than underuse by its peers.
Our study suggests that the pharmacoepidemiology of antimicrobial use in US hospitals is changing, with a greater proportion of patients receiving antimicrobials and an increasing use of broad‐spectrum agents (such as fluoroquinolones and cephalosporins) and vancomycin, compared with historical data. This scenario (ie, more patients receiving broader‐spectrum drugs) raises fears that the rate of development of antimicrobial resistance among pathogens in hospitals may accelerate in the coming years. Characteristics of the hospital’s patient population, obtainable from administrative data, can be used to create risk‐adjustment models that allow institutions to make benchmark comparisons of their rate of drug use with that of comparable institutions. Hospitals that have excessive, outlying rates of antibacterial drug use may need to take especially aggressive steps to ensure that their prescribing of antibacterials meets standards for quality. Given the dearth of new antibacterial agents in advanced development, proper monitoring and management of the use of the currently available antibacterial agents may be the most important strategy in attempting to curtail antibacterial resistance in the near future.
Acknowledgments
We thank Anne Mahoney and Jim Letcavage, formerly of Solucient, for their assistance in obtaining the study data. We thank Solucient (now part of Thomson Healthcare) for providing the study data. We acknowledge the financial support of C.M.’s fellowship from Merck and Schering Plough.
Potential conflicts of interest. R.E.P. reports that he has been a consultant for Schering‐Plough and Forest Pharmaceuticals and has been on the speakers’ bureau for Astra‐Zeneca. C.M. reports no conflicts of interest relevant to this article.
References
- 1. McGowan JE Jr. Antimicrobial resistance in hospital organisms and its relation to antibiotic use. Rev Infect Dis 1983; 5:1033‐1048.
- 2. Interagency Task Force on Antimicrobial Resistance. A public health action plan to combat antimicrobial resistance. Part I. Domestic issues. Atlanta, GA: Centers for Disease Control and Prevention, 2001.
- 3. Avorn J, Barrett JF, Davey PG, McEwen SA, O’Brien TF, Levy SB. Antibiotic resistance: synthesis of recommendations by expert policy groups: a background document for the global strategy for containment of antimicrobial resistance. Boston, MA: Alliance for the Prudent Use of Antibiotics, 2001.
- 4. Guillemot D. Antibiotic use in humans and bacterial resistance. Curr Opin Microbiol 1999; 2:494‐498.
- 5. Shlaes DM, Gerding DN, John JF Jr, et al. Society for Healthcare Epidemiology of America and Infectious Diseases Society of America Joint Committee on the Prevention of Antimicrobial Resistance: guidelines for the prevention of antimicrobial resistance in hospitals. Clin Infect Dis 1997; 25:584‐599.
- 6. Goldmann DA, Weinstein RA, Wenzel RP, et al. Strategies to prevent and control the emergence and spread of antimicrobial‐resistant microorganisms in hospitals: a challenge to hospital leadership. JAMA 1996; 275:234‐240.
- 7. Zervos MJ, Hershberger E, Nicolau DP, et al. Relationship between fluoroquinolone use and changes in susceptibility to fluoroquinolones of selected pathogens in 10 United States teaching hospitals, 1991‐2000. Clin Infect Dis 2003; 37:1643‐1648.
- 8. Polk RE, Johnson CK, McClish D, Wenzel RP, Edmond MB. Predicting hospital rates of fluoroquinolone‐resistant Pseudomonas aeruginosa from fluoroquinolone use in US hospitals and their surrounding communities. Clin Infect Dis 2004; 39:497‐503.
- 9. Monnet DL, MacKenzie FM, Lopez‐Lozano JM, et al. Antimicrobial drug use and methicillin‐resistant Staphylococcus aureus, Aberdeen, 1996‐2000. Emerg Infect Dis 2004; 10:1432‐1441.
- 10. Bhavnani SM, Callen WA, Forrest A, et al. Effect of fluoroquinolone expenditures on susceptibility of Pseudomonas aeruginosa to ciprofloxacin in U.S. hospitals. Am J Health Syst Pharm 2003; 60:1962‐1970.
- 11. Bhavnani SM, Hammel JP, Jones RN, Ambrose PG. Relationship between increased levofloxacin use and decreased susceptibility of Streptococcus pneumoniae in the United States. Diagn Microbiol Infect Dis 2005; 51:31‐37.
- 12. National Nosocomial Infections Surveillance (NNIS) System Report, data summary from January 1992 through June 2004, issued October 2004. Am J Infect Control 2004; 32:470‐485.
- 13. Carling PC, Fung T, Coldiron JS. Parenteral antibiotic use in acute‐care hospitals: a standardized analysis of fourteen institutions. Clin Infect Dis 1999; 29:1189‐1196.
- 14. Lesch CA, Itokazu GS, Danziger LH, Weinstein RA. Multi‐hospital analysis of antimicrobial usage and resistance trends. Diagn Microbiol Infect Dis 2001; 41:149‐154.
- 15. Niland JC, Rouse L, Stahl DC. An informatics blueprint for healthcare quality information systems. J Am Med Inform Assoc 2006; 13:402‐417.
- 16. Platt R. Toward better benchmarking. Infect Control Hosp Epidemiol 2005; 26:433‐434.
- 17. Ellis J. All inclusive benchmarking. J Nurs Manag 2006; 14:377‐383.
- 18. Kislak JW, Eickhoff TC, Finland M. Hospital‐acquired infections and antibiotic usage in the Boston city hospital—January, 1964. N Engl J Med 1964; 271:834‐835.
- 19. Scheckler WE, Bennett JV. Antibiotic usage in seven community hospitals. JAMA 1970; 213:264‐267.
- 20. Roberts AW, Visconti JA. The rational and irrational use of systemic antimicrobial drugs. Am J Hosp Pharm 1972; 29:828‐834.
- 21. Maki DG, Schuna AA. A study of antimicrobial misuse in a university hospital. Am J Med Sci 1978; 275:271‐282.
- 22. Gonzales R, Malone DC, Maselli JH, Sande MA. Excessive antibiotic use for acute respiratory infections in the United States. Clin Infect Dis 2001; 33:757‐762.
- 23. Fakih MG, Hilu RC, Savoy‐Moore RT, Saravolatz LD. Do resident physicians use antibiotics appropriately in treating upper respiratory infections? A survey of 11 programs. Clin Infect Dis 2003; 37:853‐856.
- 24. Hecker MT, Aron DC, Patel NP, Lehmann MK, Donskey CJ. Unnecessary use of antimicrobials in hospitalized patients: current patterns of misuse with an emphasis on the antianaerobic spectrum of activity. Arch Intern Med 2003; 163:972‐978.
- 25. Avorn J, Solomon DH. Cultural and economic factors that (mis)shape antibiotic use: the nonpharmacologic basis of therapeutics. Ann Intern Med 2000; 133:128‐135.
- 26. Iezzoni LI. Risk adjustment for measuring health care outcomes. 3rd ed. Chicago, IL: Health Administration Press, 2003.
- 27. Dellit TH, Owens RC, McGowan JE, Jr, et al. Infectious Diseases Society of America and the Society for Healthcare Epidemiology of America guidelines for developing an institutional program to enhance antimicrobial stewardship. Clin Infect Dis 2007; 44:159‐177.
- 28. Pennsylvania Healthcare Cost Containment Council. Public reports. Available at: http://www.phc4.org/reports/. Accessed July 30, 2007.
- 29. California Office of Statewide Planning and Development, Healthcare Quality and Analysis Division. Healthcare outcomes reports. Available at: http://www.oshpd.ca.gov/HQAD/index.htm. Accessed July 30, 2007.
- 30. Polk RE, Fox C, Mahoney A, Letcavage J, MacDougall C. Measurement of adult antibacterial drug use in 130 US hospitals: comparison of defined daily dose and days of therapy. Clin Infect Dis 2007; 44:664‐670.
- 31. Hornbrook MC. Hospital case mix: its definition, measurement and use. Part I. The conceptual framework. Med Care Rev 1982; 39:1‐43.
- 32. Concato J, Feinstein AR, Holford TR. The risk of determining risk with multivariable models. Ann Intern Med 1993; 118:201‐210.
- 33. Shapiro M, Townsend TR, Rosner B, Kass EH. Use of antimicrobial drugs in general hospitals. II. Analysis of patterns of use. J Infect Dis 1979; 139:698‐706.
- 34. Janknegt R, Wijnands WJ, Caprasse M, Brandenburg W, Schuitenmaker MG, Stobberingh E. Antimicrobial drug use in hospitals in The Netherlands, Germany and Belgium. Eur J Clin Microbiol Infect Dis 1993; 12:832‐838.
- 35. McDonald LC, Yu HT, Yin HC, Hsiung CA, Hung CC, Ho M. Correlates of antibiotic use in Taiwan hospitals. Infect Control Hosp Epidemiol 2001; 22:565‐571.
- 36. Liem TB, Filius FM, van der Linden PD, Janknegt R, Natsch S, Vulto AG. Changes in antibiotic use in Dutch hospitals over a six‐year period: 1997 to 2002. Neth J Med 2005; 63:354‐360.
- 37. Blix HS, Hartug S. Hospital usage of antibacterial agents in relation to size and type of hospital and geographical situation. Pharmacoepidemiol Drug Saf 2005; 14:647‐649.
- 38. Bhavnani SM. Benchmarking in health‐system pharmacy: current research and practical applications. Am J Health Syst Pharm 2000;57(Suppl 2):S13‐S20.
- 39. American Hospital Association. Available at: http://www.aha.org/aha/resource‐center/Statistics‐and‐Studies/fast‐facts.html. Accessed November 7, 2007.
- 40. Madaras‐Kelly K. Optimizing antibiotic use in hospitals: the role of population‐based antibiotic surveillance in limiting antibiotic resistance. Insights from the society of infectious diseases pharmacists. Pharmacotherapy 2003; 23:1627‐1633.
- 41. Dalton B, Sabuda D, Conly J. Trends in antimicrobial consumption may be affected by units of measure. Clin Infect Dis 2007; 45:399‐400.
- 42. Campbell SE, Campbell MK, Grimshaw JM, Walker AE. A systematic review of discharge coding accuracy. J Public Health Med 2001; 23:205‐211.