The study consisted of serial cross-sectional surveys in 2 states with pre- and postintervention measurements. Minnesota served as a control state to distinguish intervention-related changes from the regional secular trend. Minnesota was selected for geographic proximity and similarity in terms of population size and racial/ethnic distribution. Before 2002, educational activities on appropriate antimicrobial drug use were limited in Minnesota. Approval for this study was obtained from the institutional review board of CDC.

Eligible participants for the survey included physicians, nurse practitioners, and physician assistants. Practice specialties for physicians and physician assistants included family practice, pediatrics, internal medicine, emergency medicine, and general practice. Specialties for nurse practitioners included family practice and pediatrics. In 1999 and 2002, independent random samples were selected from Wisconsin and Minnesota licensing databases. The 1999 sampling frame was 7,113 in Wisconsin and 6,335 in Minnesota; the 2002 sampling frame was 6,218 in Wisconsin and 5,800 in Minnesota. The survey sample included 400 Wisconsin and 400 Minnesota clinicians in 1999, and 600 Wisconsin and 400 Minnesota clinicians in 2002. The baseline sample size was selected to provide >80% power to detect a 15% increase in the proportion of clinicians giving the correct or desired response to a specific survey question (α = 0.05). Wisconsin clinicians were oversampled in 2002 to facilitate a within-state analysis of the impact of a television advertising campaign (not reported here). The probability of the same clinician being sampled in both 1999 and 2002 was low, and the samples were considered independent in the analyses.

The preintervention questionnaire was mailed to Wisconsin clinicians in April 1999 and to Minnesota clinicians in November 1999. During March–May 2002, the postintervention questionnaire was mailed to clinicians in both states. The questionnaires contained a cover letter explaining the purpose of the survey; 2 follow-up reminders were sent to maximize compliance. The preintervention and postintervention questionnaires were identical in their measures of knowledge, beliefs, and decision making and differed only in the addition of questions to the preintervention questionnaire regarding clinician opinion for effective campaign materials (for planning purposes) and the addition of questions to Wisconsin's postintervention questionnaire about the television advertisements.

After determining practice setting and basic demographics, clinicians caring for adults were asked to estimate the proportion of adult patients who requested antimicrobial agents for cough, cold, or flulike symptoms. Likewise, those caring for children were asked to estimate the proportion of parents who requested antimicrobial agents for their child. The survey questionnaire (Figure 1) then asked a series of questions to assess 1) the influence of 2 nonpredictive clinical factors (i.e., clinical symptoms or signs characteristic of both viral and bacterial infections, which therefore did not necessarily warrant antimicrobial drug therapy) and 1 social factor on the decision to prescribe antimicrobial drugs; 2) the likelihood of antimicrobial agent prescribing in adult and pediatric clinical case scenarios for URIs and bronchitis; and 3) perceptions and beliefs regarding patient expectations and peer-established norms. In addition, questions regarding exposure to and perceived impact of the WARN campaign were asked on the postintervention questionnaire (Wisconsin clinicians only). For most questions, the responses were based on a 5-point Likert-scale (e.g., "strongly disagree to strongly agree"). The Likert responses were dichotomized into desired and undesired responses (Figure 1); responses were classified as "desired" if they were consistent with national pediatric and adult clinical practice guidelines or the educational goals of the WARN campaign. The influence of social factors, patient expectations, and peer-established norms on clinical decision-making was considered "undesired" since each was an inappropriate reason for antimicrobial drug prescribing.

For the clinician-reported estimates of the percentage of patients or parents who requested an antimicrobial drug, we compared the distribution of responses for each year and state. We used the 1-sided Jonckheere-Terpstra test, a generalization of the nonparametric Mann-Whitney U test, to compare within-state distributions for 1999 and 2002. The null hypothesis was that the distributions did not differ between these 2 periods.

For each Likert-scale question, we calculated a ratio based on the proportion of clinicians with a desired response in 2002 (numerator) divided by the proportion with a desired response in 1999 (denominator). A ratio >1.0 indicated improvement in 2002 versus 1999. After calculating this ratio for each state, we compared the ratios between Wisconsin and Minnesota. All ratios were adjusted for clinician sex, years in practice, practice setting, and clinician type. All adjusted ratios and corresponding statistical test results were obtained directly from multivariable models similar to logistic regression models but with a log (rather than logit) link function (24). Such models permit comparison of proportions rather than odds. The models included terms for state, year, their interaction, and the control variables (clinician sex, years in practice, practice setting, and clinician type). Examining appropriate combinations of the estimated parameters from these models permitted within-year comparisons of Wisconsin versus Minnesota (e.g., baseline comparisons), within-state comparisons of 2002 versus 1999, and between-state comparisons of within-state ratios (i.e., comparisons of the 2002/1999 ratios for Wisconsin to those in Minnesota for estimating the effect in Wisconsin beyond that observed in Minnesota).

Ten percent of questionnaire responses were entered in duplicate for quality assurance. Statistical analyses were performed by using SAS version 8.2 (SAS Institute Inc., Cary, NC, USA) and EpiInfo 6 (CDC, Atlanta, GA, USA).

Baseline responses were compared for Wisconsin and Minnesota in 1999, before initiation of the WARN campaign in Wisconsin. Clinicians in Wisconsin and Minnesota perceived similar proportions of their adult patients requesting antimicrobial agents (p = 0.217) (Figure 2). The median percentage of patients perceived to request antimicrobial agents was 50% in Wisconsin and 40% in Minnesota; this difference was not significant. The perceived demand by parents for antimicrobial agents to treat their child's respiratory illness was also similar between the 2 states (p = 0.473) (Figure 3); the median reported percentage of parents requesting antimicrobial agents was 25% in both states.

Clinicians in Wisconsin and Minnesota gave similar baseline responses regarding the influence of a social factor (e.g., patient states antimicrobial agents given for similar symptoms in the past) and a nonpredictive clinical factor (e.g., purulent nasal discharge) (Table 2, baseline p values not presented). However, Minnesota clinicians were significantly more likely to report that productive cough with purulent sputum would not influence their decision to prescribe an antimicrobial agent (Wisconsin 14%, Minnesota 20%, p = 0.027). For the bronchitis and viral URI case scenarios, the overall proportion of clinicians who would withhold antimicrobial agents was similar in each state. The proportion who would withhold antimicrobial agents was greater for the pediatric case scenarios than for the adult case scenarios (Table 3). Responses to the belief questions regarding patient expectations and clinician peer norms were similar between the states during the baseline period (Table 4).

We also compared baseline responses between states for clinicians in practice ≤10 years and those in practice >10 years. For those in practice ≤10 years, a higher proportion of Minnesota clinicians indicated that their decision to use antimicrobial agents was not influenced by purulent nasal discharge (Wisconsin 36%, Minnesota 51%, p = 0.024) or cough with productive sputum (Wisconsin 12%, Minnesota 25%, p = 0.005). Responses to the influence of the social factor and responses to the adult and pediatric case scenarios were similar between the states. For clinicians in practice >10 years, responses were similar between the 2 states regarding the influence of nonpredictive clinical factors and the social factor. Compared to Wisconsin clinicians practicing >10 years, a higher proportion of Minnesota clinicians in long-term practice indicated they would withhold antimicrobial agents in the pediatric case scenarios (p = 0.048) and the adult case scenarios (p = 0.118).

Within Wisconsin, some baseline responses differed according to length of time in practice. A significantly higher proportion of clinicians practicing ≤10 years gave the desired responses for the pediatric (p = 0.027) and adult (p = 0.002) case scenarios. They were also more likely to give the desired response regarding the influence of a social factor (i.e., patient states antimicrobial agents were given for similar symptoms in the past) (p = 0.043). Wisconsin clinicians in practice ≤10 years and those in practice >10 years gave similar responses regarding the influence of the nonpredictive clinical factors (i.e., purulent nasal discharge and productive cough).

When specialties were compared, a higher proportion of pediatric clinicians gave desired responses than clinicians in other specialties on most outcome measure in both Wisconsin and Minnesota. Baseline comparisons between physicians and nonphysicians did not show a consistent tendency for 1 group to perform better than the other.

In 2002, Wisconsin clinicians perceived less demand for antimicrobial agents among adult patients compared with 1999 (p<0.001) (Figure 2). Based on clinician estimates, the median percentage of patients who requested an antimicrobial agent for cough, cold, or flu symptoms decreased from 50% in 1999 to 30% in 2002. Minnesota clinicians also perceived a decrease in the percentage of patients who requested antimicrobial agents, but the difference was not significant (p = 0.152) (Figure 2); the median percentage of Minnesota patients requesting antimicrobial agents decreased from 40% in 1999 to 30% in 2002.

In both states, a decline was noted in the perceived parental demand for antimicrobial agents to treat pediatric respiratory illness (Figure 3). The temporal change was significant in Wisconsin (p = 0.004) and approaching significance in Minnesota (p = 0.064). The median reported percentage of parents who requested an antimicrobial agent decreased from 25% in 1999 to 20% in 2002 in both states, but the distribution around the medians differed significantly between the states.

In Wisconsin, significant improvement occurred in the responses to the 2 questions about nonpredictive clinical factors and the social factor that may increase the likelihood of prescribing antimicrobial agents (i.e., purulent nasal discharge, productive cough, and patient or parent statement that antimicrobial agents were prescribed for similar symptoms in the past) (Table 2). Wisconsin clinicians were significantly more likely to report that each factor did not influence antimicrobial agent prescribing practices in 2002 compared with 1999. In Minnesota, a significant improvement occurred in responses regarding the influence of purulent nasal discharge, but no significant change occurred for the other 2 factors. Overall, Wisconsin clinicians demonstrated significant improvement regarding the influence of purulent nasal discharge (p = 0.044) and productive cough (p = 0.010) after accounting for temporal changes in Minnesota.

Both Minnesota and Wisconsin clinicians improved in their responses to the adult case scenarios for URI and bronchitis (Table 3). The magnitude of improvement was greater for Wisconsin clinicians, but the improvement in Wisconsin was not significant after accounting for the secular trend in Minnesota. In the pediatric case scenarios, Wisconsin clinicians improved from 1999 to 2002 (p = 0.058), while the responses of Minnesota clinicians were essentially unchanged (p = 0.807).

Wisconsin clinicians demonstrated a modest improvement from 1999 to 2002 in response to questions concerning perceived clinician peer norms and patient expectations, but the changes in Wisconsin were not significant after accounting for temporal changes in Minnesota (p = 0.103 and 0.519, respectively, Table 4).

Responses were analyzed separately for clinicians who had practiced >10 years (1999, n = 198; 2002, n = 243) and those in practice ≤10 years (1999, n = 187; 2002, n = 239). In Wisconsin, clinicians who were in practice for the longer period demonstrated significant improvements regarding the likelihood of prescribing antimicrobial agents for purulent nasal discharge (2002 to 1999 ratio = 1.61, p = 0.005) and productive cough (2002 to 1999 ratio = 2.35, p = 0.001). They also improved in their responses to the influence of patient/parent statement that antimicrobial agents were prescribed for similar symptoms in the past (2002 to 1999 ratio = 1.36, p = 0.012). Wisconsin clinicians practicing >10 years also demonstrated significant improvements in the adult case scenarios (2002 to 1999 ratio = 2.00, p<0.001) and the pediatric case scenarios (2002 to 1999 ratio = 1.43, p = 0.002)] and in the questions concerning patient expectations and peer norms. The 2002 to 1999 ratio was 1.40 (p = 0.031) for patient expectations and 1.28 (p = 0.021) for peer norms. However, only the responses to the pediatric case scenarios improved significantly among physicians in practice >10 years (p = 0.027) after accounting for the secular trend in Minnesota.

Wisconsin clinicians in practice ≤10 years improved in fewer areas. They improved significantly in responses regarding the influence of purulent nasal discharge (p<0.001) and productive cough (p<0.001) on antimicrobial agent prescribing practices, and both factors remained significant after accounting for the secular trend. No significant change occurred in the other responses. A direct comparison between Wisconsin clinicians practicing ≤10 years to those practicing >10 years demonstrated no significant difference with regard to improvement in knowledge or the response to the clinical scenarios.

Ninety percent of primary care clinicians in Wisconsin had heard of WARN. Of those, 70% had used WARN patient education materials. Of those using the materials, 41% reported that they were "very useful," and 59% said that they were "somewhat useful."