A 1-page survey and cover letter were mailed to 400 primary care clinicians in Minnesota and Wisconsin in October 2002. Within each state, a simple random sample of 200 clinicians was selected from among those with an active medical license in any of the following specialties: emergency medicine, family practice, general practice, general internal medicine, pediatrics, nurse practitioner, or physician assistant. The sampling frame was obtained from the licensing agencies in each state and included 5,800 clinicians in Minnesota and 6,510 clinicians in Wisconsin. Two reminder letters were sent to all survey recipients to encourage participation.

The anonymous survey contained 12 questions to obtain information on specialty, county of practice, outpatient practice activity during the last quarter of 2001, requests for anthrax prophylaxis from October through December of 2001, and distribution of antimicrobials for anthrax prophylaxis or stockpiling. Clinicians who indicated that they prescribed or distributed anthrax-related antimicrobial agents were asked to specify the number and types of people (i.e., patients, family members, acquaintances, self) who received the drugs, and the specific drugs that were used. The occurrence of suspected anthrax-related exposures was not assessed. Clinicians who indicated they were not in full- or part-time outpatient medical practice from October through December of 2001 were asked to return the survey without answering the other questions.

The survey procedures were reviewed and approved by the institutional review board at Marshfield Clinic.

The primary measures of fluoroquinolone use were based on prescribing data (Xponent, IMS Health, Inc., Plymouth Meeting, PA) and volume retail distribution data (DDD, IMS Health, Inc.) for the states of Wisconsin and Minnesota. The prescribing and distribution data represent independent measures of antimicrobial use that were obtained to evaluate a statewide program to promote appropriate antimicrobial use. Prescribing data were available for the years 2000 through 2002. Volume distribution data were available for the years 1999 through 2002. For this analysis, oral antimicrobial agents were grouped into the following three categories: fluoroquinolones (ciprofloxacin, levofloxacin, lomefloxacin, moxifloxacin, ofloxacin), tetracyclines (doxycycline, tetracycline, minocycline), and other oral antimicrobial agents (amoxicillin, amoxicillin-clavulanate, ampicillin, cephalosporins, extended-spectrum macrolides, macrolides, penicillin, and trimethoprim-sulfamethoxazole). The first two categories represented drug classes that were most likely to be used for anthrax prophylaxis or stockpiling, since they were recommended as first-line agents by CDC (14). The third category provided a measure of seasonal and annual trends in outpatient use of antimicrobials, since the drugs in this category (with the possible exception of penicillin and amoxicillin) would not be appropriate for managing anthrax exposure. Penicillin or amoxicillin use was recommended for anthrax use only when the first-line agents were contraindicated, and it is therefore unlikely that a substantial number of these prescriptions were generated for anthrax-related concerns. Prescribing and volume distribution data were not available for individual drugs within each class. Only oral agents in solid (capsule or tablet) formulation were included in the analyses to avoid including prescriptions for young children.

Xponent prescribing data included all dispensed prescriptions in Minnesota or Wisconsin from licensed prescribers. The total number of prescriptions written in each state was available by month for the years 2000, 2001, and 2002. The prescribing databases were compiled by IMS Health with proprietary methods. The Xponent prescribing database was derived from transactional data provided by 59% of all retail pharmacies in Wisconsin and Minnesota, including 65% of chain pharmacies and 51% of independent retail pharmacies. Prescriptions from unsampled stores were estimated on the basis of prescription totals from matched nearby stores with weighting to adjust for differences in total retail sales volume, which was available for nearly all stores. Estimates were also weighted to account for the distance between sampled stores and matched unsampled stores, with closer stores contributing more to the estimated prescription volume. The proportion of all prescriptions in each state based on estimated data from unsampled stores was 33%–37%.

Retail distribution was measured based on the volume of antimicrobial agents distributed to retail outlets on a monthly basis for the years 1999 through 2002 (DDD data, IMS Health, Inc.). Retail distribution data (measured in kilograms) were ultimately derived from wholesalers and distributors serving pharmacies in both states. Volume was based on distribution to retailers rather than actual sales to patients, and distributed antimicrobial agents could be returned to the wholesalers without being sold. In this situation, returned antimicrobial agents were subtracted from the total distributed in a given month to yield the net retail distribution for each drug class. As a result, the retail distribution data may overestimate or underestimate actual distribution to patients, particularly in short time periods. Inpatient pharmacies, prisons, veterinarians, nursing homes, dialysis clinics, and federal government sites were excluded from the volume distribution data. The DDD database captured 93% of actual volume of antimicrobial distribution in these states.

To account for the state-level sampling in the clinician survey, analysis weights were generated for each survey responder. These weights were computed by multiplying the inverse of the sampling probability for each responder by the inverse of the response rate in the appropriate state. Weighted analyses for complex sample designs were performed by using the SAS 8.2 (SAS Institute Inc., Cary, NC) and SUDAAN 8.0 (RTI International, Research Triangle Park, NC) software. The impact of the weighting was minimal in our study; estimated percents for the weighted and unweighted analyses differed by <l percentage point. We therefore present unweighted point estimates, except for the estimate of anthrax-related antimicrobial drug prescribing. All statistical tests for the clinician survey were based on weighted data to reflect the complex sample design. Categorical variables were compared by using the chi-square test.

For the Xponent prescribing data, we fit models for the absolute number and proportion of prescriptions for fluoroquinolones and tetracyclines. Absolute numbers of prescriptions were modeled by using linear regression, while proportions of prescriptions were modeled by using negative binomial regression. All of the models contained categoric effects for year and month and indicator variables for October, November, and December of 2001. We generated predictions for the fourth quarter of 2001 by computing linear combinations of the appropriate parameters from the above models. The modeling procedures for the DDD data were the same as those for the Xponent data, except that we modeled total retail distribution volume by month for each drug category.

Surveys were returned by 239 (60%) of 400 clinicians, including 123 in Wisconsin and 116 in Minnesota. Twenty-nine (12%) of the 239 respondents were excluded from subsequent analyses because they were not engaged in full-or part-time outpatient practice during the last 3 months of 2001. The respondent medical practices were located in 68 counties; 58 (28%) practices were located in the 11-county Minneapolis–St. Paul metropolitan area, and 24 (11%) were located in the 4-county Milwaukee-Waukesha metropolitan area. One hundred fifty-seven (75%) were physicians (MD or DO), and 52 (25%) were nurse practitioners or physician assistants. Physician specialties included family practice (42%), internal medicine (30%), pediatrics (18%), emergency medicine (4%), and other (6%).

Fifty-eight (28%) of the clinicians reported that someone had asked them to prescribe an antimicrobial drug to prevent anthrax or stockpile in case of future bioterrorist attacks. Physicians were significantly more likely than nonphysicians to receive requests for antimicrobial agents (Table). The occurrence of patient requests by state, practice specialty, or practice location (Minneapolis–St. Paul and Milwaukee–Waukesha metropolitan areas vs. other areas) did not differ significantly.

Nine (4%) of the clinicians provided antimicrobial agents to 11 persons for anthrax prevention or stockpiling. Seven clinicians provided such agents for a single person, and 2 clinicians provided them for 2 persons. All 9 clinicians were MDs or DO, and 8 (89%) practiced in Wisconsin. Among 58 clinicians who received such requests, 8 (14%) provided them; 1 additional clinician did not receive requests but provided these drugs for family members. Nine (82%) of the 11 courses of antimicrobial agents were given to patients seen in the clinician’s practice, and 2 were given to family members. One clinician reported that prophylactic antimicrobial agents were given to a patient who had exposure to a building in Washington, D.C., where anthrax release was confirmed; the risk for anthrax exposure was unknown for the other 10 persons. A fluoroquinolone (levofloxacin or ciprofloxacin) was used for 10 (91%) of the 11 courses; amoxicillin was given to 1 person. No other classes of antimicrobial agents were used.

Survey responses were used to estimate the total number of anthrax-related fluoroquinolone prescriptions written in Wisconsin and Minnesota during the last quarter of 2001. State medical licensing records indicate that ≈10,807 primary care clinicians were engaged in full- or part-time outpatient practice in Minnesota and Wisconsin at that time. If the survey responses are representative of this group, an estimated 420 clinicians (3.9%) in these states provided anthrax-related antimicrobial agents. The 95% confidence limits (CI) for this proportion are 1.2% and 6.5%, corresponding to 135 and 706 clinicians, respectively. The total number of anthrax-related courses of antimicrobial agents prescribed during the last quarter of 2001 was estimated to be 523 (95% CI, 394–653) in the two states. If these prescriptions were mostly for fluoroquinolones, they would have represented ≈0.3% of all fluoroquinolone prescriptions written in Minnesota or Wisconsin from October through December of 2001.

The annual number of fluoroquinolone prescriptions in Minnesota and Wisconsin increased 20% from 2000 to 2002, while the annual number of prescriptions declined during this period for tetracycline drugs and other antimicrobials. In October 2001, the number of fluoroquinolone prescriptions was 22% higher than in October 2000, and the number of fluoroquinolone prescriptions exceeded the 95% CI based on the predictive model (Figure 1). The proportion of all antimicrobial prescriptions (excluding tetracyclines) represented by fluoroquinolones gradually increased from 2000 to 2002, and this proportion was higher in October 2001 (17.5%) than in October 2000 (15.5%). However, the observed proportion of fluoroquinolone prescriptions was lower than predicted in October and November of 2001, after adjusting for the secular trend (Figure 1). Retail distribution of oral fluoroquinolones was significantly elevated in October and November of 2001 (Figure 2). The fluoroquinolone proportion of total retail distribution of antimicrobial agents (excluding tetracyclines) also increased significantly in October 2001, but it was lower than predicted in November (Figure 2).

The proportion of all antimicrobial prescriptions (excluding fluoroquinolones) that were for tetracyclines was near the upper limit of the 95% CI in October 2001, and it exceeded the upper limit in November 2001 (Figure 3). The proportion of antimicrobial drug retail volume represented by tetracyclines was not elevated during those months (data not shown).