Patients and Setting
The study was conducted from November 27, 2001 to February 28, 2002 in the Massachusetts General Hospital Medical Walk-In Unit (Boston, Mass.). The Medical Walk-In Unit is an adult acute care clinic, open 7 days a week, where most patients are seen on a first-come, first-served basis.
At check-in, the unit secretary directed patients to a flyer that stated: “Do you have runny nose, cough, congestion or stuffiness, sore throat, ear pain, sinus pain or other symptoms that makes you think you may have pneumonia, bronchitis, sinusitis, a cold, the flu or another respiratory infection?” If the patient had respiratory symptoms, the secretary offered the patient a “Respiratory Illness Survey” to complete in the waiting room. Patients could also receive a survey by responding to flyers posted in the waiting room or by being asked by the triage nurses. The secretaries and nurses were not required to verify the presence of respiratory symptoms. Patients were not informed of the content or purpose of the study prior to being given the survey.
Surveys were filled out and returned to a lockbox in the waiting room before the patient's visit with the physician. Although efforts were made to include as many patients as possible, participation in the study was voluntary. Patients gave their consent to participate by returning a survey. No data are available concerning the demographics or diagnoses of the patients who declined participation.
Information on the survey was not available to the treating physician. Information about recorded history, physical exam findings, diagnosis, and medications prescribed was later abstracted from the chart. The Institutional Review Board of Massachusetts General Hospital approved the study protocol.
Respiratory Illness Survey
Patients answered questions about characteristics of the present illness, reason for the visit, expectations for the visit, desire for antibiotics, demographics, and comorbid illnesses. Questions were derived from prior studies.12–14,19,20
Prior to study use, clinicians and patients reviewed the questionnaire to ensure understandability.
Information about the present illness included asking about the duration of illness, whether the patient missed work due to illness, and 24 specific symptoms using yes/no questions. Patients were asked how bothered they were by the present illness on a 5-point scale, from “extremely bothered” to “not bothered.” Patients indicated their primary reason for coming to the clinic by answering the question “For your visit today, of the reasons for coming to the clinic, which one is most important to you?” With this question, we sought to understand the main goal served by getting better. Possible reasons for the visit included getting better because the symptoms are annoying, getting better because of pain, getting better to care for family, getting better to return to work, getting better for leisure time or travel, or getting better for another reason.
Patients indicated their primary expectation for the visit by answering the question “For your visit today, of the expectations below, which one is most important to you?” With this question, we sought to know what the patient wanted to happen at the visit. Possible expectations included having tests or x-rays, getting a diagnosis, getting reassurance, getting a referral, getting a nonantibiotic treatment, getting an antibiotic prescription, or getting an estimate of how long the symptoms would last.
Desire for antibiotics was assessed using a 5-point Likert scale (strongly agree to strongly disagree) in response to the statement “I want antibiotics for my illness today.” Similar response frames were used to assess whether patients felt that antibiotics work for them when they have a cold, repeated use of antibiotics could be harmful to them personally, or if they planned on asking the doctor for antibiotics.
Patients answered questions about their age, sex, primary language, race/ethnicity, education, insurance status, household income, employment status, and whether or not they had a primary care doctor they have seen before. Patients were asked the number of medicines they took, if they smoked, and if they had heart disease, lung disease, diabetes, or cancer (excluding skin cancer), or were pregnant.
For the present analysis, we included patients who had a physician-assigned upper respiratory tract infection and who answered the question regarding desire for antibiotics. Because we wanted patients to respond to questions about treatment expectations prior to their visit with the doctor, we anticipated a need to exclude patients who were not given a primary diagnosis of a common upper respiratory tract infection. Included diagnoses were upper respiratory infection, viral syndrome, influenza, otitis media, sinusitis, nonstreptococcal pharyngitis, streptococcal pharyngitis, tonsillitis, infectious mononucleosis, acute bronchitis, and acute cough. Patients with influenza (3 patients) and infectious mononucleosis (1 patient) were considered to have viral syndrome; patients with cough (10 patients) were considered to have acute bronchitis. We excluded patients who reported symptoms for more than 30 days.
We dichotomized patients into those who strongly agreed with the statement “I want antibiotics for my illness today,” and those who had no opinion, disagreed, or strongly disagreed. To assess prior antibiotic use, we dichotomized patients into those who reported using, on average, zero antibiotic courses per year and those who reported using 1 or more antibiotic courses per year.
We did not evaluate diagnosis as a predictor of antibiotic prescribing. The signs and symptoms of common upper respiratory tract infections overlap and a diagnosis might be used to legitimize an antibiotic prescription. For example, if a patient has headache, myalgias, and sinus pain, and insists on an antibiotic, the physician may diagnose the patient with sinusitis—an antibiotic-appropriate diagnosis—instead of upper respiratory tract infection, for which antibiotics would be inappropriate.
To detect a 20% absolute increase—from 40% to 60%—in antibiotic prescribing for patients who wanted antibiotics, assuming 50% of patients would want antibiotics, with Type I error of .05 and 80% power, 180 total patients would be required. Our goal was to enroll 300 total patients to account for exclusions, missing data, and loss of power using a clustered analysis.
We used standard descriptive statistics. We used Fisher's exact test, the χ2 test, the χ2 test for trend, Student's t test, and the Wilcoxon rank-sum test where appropriate.
We developed two multivariable logistic regression models. In the first model, we sought to identify independent predictors of wanting antibiotics. We evaluated any variable associated with wanting antibiotics (P≤ .1 on univariate testing) and adjusted for confounding while minimizing collinearity between variables. We considered confounding or collinearity to be present when the addition of a new variable into the model changed the beta coefficient or standard error, respectively, of another variable by more than 10%.
In the second model, we sought to determine whether wanting antibiotics was independently associated with receiving antibiotics, adjusted for clustering by provider.21
To begin the model building process, we used a forward selection algorithm to identify the strongest predictors of antibiotic prescribing. We also evaluated any other covariate associated with either wanting antibiotics or receiving antibiotics on univariate testing (P
All analyses were performed using SAS 8.1 (SAS Institute, Cary, NC). P values were two-tailed where possible and P values ≤ .05 were considered significant.