Prescribing rates for the 4,115 physicians in practice throughout the study period were computed as the number of antimicrobial prescriptions in a given year divided by the number of physicians. Prescribing rates were therefore equivalent to the mean number of prescriptions for antimicrobial drugs per physician each year. Population-based or patient-based prescribing rates could not be calculated because the populations served by the prescribing cohort were undefined. To compare prescribing rates for each year of the intervention period (2000–2003) to baseline (1998), we fit Poisson regression models. Generalized estimating equations (GEE) were employed to account for within-physician correlation (26). An autoregressive working correlation structure was specified in the Poisson-GEE models, and appropriate steps were taken to accommodate the absence of data for calendar year 1999. Statistical comparisons of changes in rates in Wisconsin versus those in Minnesota were derived from these models.

Differences in baseline antimicrobial prescribing were addressed by fitting additional Poisson-GEE models using data from the intervention period with the natural log of the baseline number of antimicrobial drug prescriptions included as an independent variable. Prescribing rates per physician-month were computed when analyzing the 12,790 physicians who wrote such prescriptions in ≥1 month of the study period. Months where a physician wrote at least 1 prescription for an antimicrobial drug were included in the denominators of rates (i.e., a physician was assumed not to be in practice in months when he did not write any such prescriptions). Per capita antimicrobial drug sales were calculated based on annual population estimates (www.census.gov). Analyses were completed by using the SAS software (SAS Institute Inc., Cary, NC, USA).

Figure 1

Figure 1. . Temporal trends in per capita antimicrobial drug sales and the mean number of prescriptions per physician in Minnesota and Wisconsin.

The 4,115 primary care physicians in long-term practice included 2,009 (49%) in Minnesota and 2,106 (51%) in Wisconsin. The proportion of physicians in family practice was higher in Wisconsin, and the proportion in internal medicine was higher in Minnesota (Table 3). During 1998, these physicians generated 1.5 million prescriptions for antimicrobial drugs in each state, and the crude antimicrobial prescribing rate was nearly identical across states. From 2000 through 2003, the prescribing rate for antimicrobial drugs in both states gradually declined (Figure 1). From 1998 to 2003, the antimicrobial prescribing rate was reduced by 19.8% in Minnesota and by 20.4% in Wisconsin. The percentage reduction was 19.5% and 18.6%, respectively, in the secondary analysis, which included the 12,790 primary care physicians who prescribed antimicrobial drugs at any time during the follow-up period.

Retail sales of antimicrobial drugs (grams per capita) declined by 27.4% in Minnesota and 17.3% in Wisconsin from 1999 through 2002 (Figure 1). Sales of amoxicillin and penicillin exceeded those of other product groups in each year and accounted for 37% of all retail antimicrobial distribution. In both states, retail sales of fluoroquinolones remained level, and sales of amoxicillin-clavulanate increased. Sales of most other product groups declined by 15% or more.

Stratification of antimicrobial prescribing rates by specialty showed that the reduction in prescribing differed only for physicians in internal medicine and pediatrics (Table 4). Antimicrobial prescribing declined more in Wisconsin than Minnesota among internal medicine physicians, but the opposite was true for pediatricians. In Minnesota, the reduction in antimicrobial use was similar for physicians practicing in the Minneapolis–St. Paul metropolitan statistical area and those practicing elsewhere in the state (Table 5). In Wisconsin, the reduction in antimicrobial prescribing was much less in the Milwaukee-Waukesha metropolitan statistical area than in the remainder of the state. Antimicrobial prescribing declined significantly more in the Minneapolis-St. Paul metropolitan statistical area than in the corresponding Milwaukee-Waukesha area .

Prescriptions for liquid antimicrobial drugs (a surrogate for pediatric use) declined 29% in Wisconsin and 32% in Minnesota from 1998 to 2003 (p = 0.17). Prescriptions for solid formulations declined 15%–17% in each state. The percentage of prescriptions for broad-spectrum agents was unchanged (within 2%) in each state from 1998 through 2003.

Figure 2

Figure 2. . Antimicrobial prescribing rate ratios by year and practice location, adjusted for specialty and baseline (1998) prescribing rate. The vertical bars show 95% confidence intervals. Ratios <1 indicate lower antimicrobial prescribing...

Regression models were fit to compare prescribing rates in Wisconsin and Minnesota during each year from 2000 to 2003, with adjustment for specialty and baseline prescribing in 1998. Two separate models were generated. The first included physicians practicing in the Minneapolis–St. Paul metropolitan statistical area or the Milwaukee-Waukesha metropolitan statistical area. The second included physicians practicing in other counties of Minnesota or Wisconsin. In the latter model, the prescribing rate ratio was <1 in 2002 and 2003, indicating that, when the 2 largest metropolitan areas were excluded, Wisconsin physicians had significantly lower prescribing rates than those in Minnesota (Figure 2). In contrast, the prescribing rate ratio was >1 in each year within the 2 largest metropolitan areas, indicating that physicians in the Milwaukee-Waukesha MSA had higher prescribing rates than those in the Minneapolis-St. Paul MSA, after adjusting for specialty and baseline prescribing.

WARN represents the largest program on appropriate antimicrobial drug use that has been evaluated to measure the effect on prescribing. Previously published studies demonstrated that interventions at the level of the physician, clinic, or community had a modest effect on prescribing (20,27–32). These focused programs are useful in evaluating specific intervention strategies in a relatively controlled setting. However, adoption of new practices may be slow even when the intervention is proven to be effective, and generalizability to larger populations may be limited. In contrast, large-scale programs can reach physicians and the general public in an entire state or large metropolitan area. The development of WARN reflected a need to implement and evaluate a large-scale demonstration project that could influence antimicrobial prescribing throughout Wisconsin in a relatively short time. At the time WARN was conceived and funded, knowledge of national trends in antimicrobial prescribing was limited, but we now know that prescribing declined substantially for both pediatric and adult populations in the 1990s (22–24,33).

Outpatient antimicrobial use declined substantially in both Wisconsin and Minnesota from 1998 to 2003, and no additional intervention-related effect was apparent in Wisconsin. Secondary analyses by specialty and practice location demonstrated variable reductions in prescribing of antimicrobial drugs, but to what extent these reductions were related to WARN interventions, as opposed to other factors that may be influencing secular trends, is unclear. The potential effect of these other factors is illustrated by the observation that Minnesota pediatricians improved their prescribing practices more than Wisconsin pediatricians, despite the absence of an organized, large-scale program to improve pediatric prescribing in Minnesota.

We found that changes in antimicrobial drug use were less pronounced in the Milwaukee-Waukesha metropolitan area than in other regions of Wisconsin. This finding contrasts with findings in Minnesota, where the decrease in antimicrobial drug use in the Minneapolis-St. Paul metropolitan area was similar to that in the rest of the state. When these large metropolitan areas were excluded from the analysis, prescribing of antimicrobial drugs decreased more in Wisconsin than in Minnesota. Several factors may have contributed to the relatively low impact of WARN in the Milwaukee area. These factors include the absence of paid staff working in Milwaukee, few connections with the Milwaukee medical community, and the large number of clinical practices and health plans.

Funding for WARN exceeded levels for other state-based programs on antimicrobial drug resistance, but it was far lower than the funding level for some other public education campaigns. Annual funding for WARN staffing, programs, and materials (excluding indirect costs and evaluation activities) was ≈$238,000–$342,000. By comparison, Wisconsin receives ≈$10 million in state funding and $1.1 million in federal funding each year for smoking prevention programs (Maureen Busalacchi, pers. comm.). WARN funding for education on antimicrobial resistance was not sufficient to conduct a widespread and sustained media campaign, although whether such a campaign would have led to further reductions in antimicrobial use is not known.

One other published study reported the effect of interventions promoting appropriate antimicrobial drug use in a large, highly populated area. In Knox County, Tennessee, a 12-month multifaceted campaign was conducted in 1997 and early 1998 (34). The clinician intervention included professional presentations, distribution of pediatric principles of judicious antimicrobial drug use (35), and newsletter articles. Patient and public education included distribution of print materials, news media coverage, and public service announcements. In the Medicaid managed care program, antimicrobial drug prescribing for respiratory illness declined 19% in Knox County and 8% in the control counties compared to the previous year (intervention attributable effect of 11%, p<0.001). The Knox County intervention was smaller in scale and shorter in duration than WARN, and the generalizability to non-Medicaid populations is unknown.

Evaluating large-scale, multifaceted educational programs such as WARN has several limitations. The control population in Minnesota was not isolated, and educational materials and messages may have diffused into the control area from a variety of sources. For example, during the WARN follow-up period, national guidelines on appropriate antimicrobial use were published and endorsed by the Centers for Disease Control and Prevention and major professional organizations (35–37). WARN also reached limited populations in Minnesota: radio advertisements included areas of eastern Minnesota, and some Minnesota physicians received the WARN satellite broadcasts on appropriate antimicrobial drug use. In addition, efforts within Minnesota to improve prescribing of antimicrobial drugs increased during the follow-up period, particularly during 2002 and 2003. A single state may not be the optimal comparison population, and Minnesota in particular has a highly educated population and a proactive public health infrastructure with strong ties to the healthcare delivery system. As a result, the improvements in antimicrobial drug prescribing within Minnesota may have been greater than those in many other states. Finally, the commercial prescribing data used for the WARN evaluation did not include any information on visits or diagnoses. We therefore cannot determine if the declines in antimicrobial prescribing in Wisconsin and Minnesota were associated with a declining rate of visits for acute respiratory illness or if the rate of prescribing declined for specific diagnoses.

The centralized development of WARN programs and materials facilitated statewide distribution, but it also limited the level of clinician involvement at the local level. Direct, face-to-face communication with physicians was rarely possible. In contrast, practice-level interventions have shown modest success, and we speculate that these focused, participatory interventions may promote physician behavior change more directly than a mass education campaign such as WARN. However, WARN succeeded in changing physicians' knowledge and attitudes regarding appropriate antimicrobial drug use, and WARN materials were widely used by primary care clinicians throughout the state. In a survey of primary care clinicians, 90% of respondents had heard of WARN, and 70% of those had used WARN materials for patient education (38). Models of behavior change suggest that changes in prescribing behavior are preceded by important cognitive changes that proceed in stepwise fashion (39). Improvements in knowledge and beliefs among both physicians and patients may therefore be markers of progress, which may facilitate the future success of provider-level interventions developed by clinics and managed care organizations in Wisconsin.

Increased funding for state-level educational campaigns to promote appropriate antimicrobial drug use does not appear warranted by the results of this evaluation. However, the combined effect of national guidelines for appropriate use of such drugs, increasing attention by the media and professional organizations, and the Centers for Disease Control and Prevention national campaign may have contributed to the observed trend toward declining antimicrobial use. Progress toward decreasing inappropriate use is being made in many states, although antimicrobial prescribing rates remain excessive for bronchitis, and use of broad-spectrum antimicrobial drugs is increasing nationally (22,27,33). State and local organizations should consider a balanced approach that includes limited statewide educational activities with increasing emphasis on local, provider-level interventions and policy development. These activities might include academic detailing by physician opinion leaders, feedback on antimicrobial prescribing performance (including Health Plan Employer Data and Information Set measures), and economic incentives for careful antimicrobial use. These strategies may have the greatest effect if implemented as quality improvement initiatives in collaboration with the leadership of health plans and clinic organizations. Ongoing assessment of prescribing trends and rates of antimicrobial drug resistance will be needed to measure the ultimate effect of these efforts.