Interventions and Main Measures
Each patient (or parent) participating in the study was asked to complete a questionnaire detailing the patient’s symptoms, duration of symptoms, expected treatment for the current illness, history of previous respiratory illness and prior treatment, and smoking history. The nurse caring for the patient noted the patient’s temperature, location of visit (regular office or urgent care), and the type of clinician (physician, physician assistant, or nurse practitioner) who saw the patient. Finally, the clinician completed a questionnaire identifying the patient’s chief complaint, history of lung disease, physical findings, treatment, the clinician’s belief about the patient’s expectations, diagnosis, and secondary factors affecting the decision to give antibiotics, if given. Secondary factors were reasons for prescribing an antibiotic that were not directly related to the diagnosis. The list of options for secondary factors included: patient expected an antibiotic, patient requested an antibiotic, patient leaving town, patient not improving, patient getting worse, patient sick too long, patient smokes, patient has chronic lung disease, patient has comorbidity, and patient is extremely ill. Finally, we asked the clinicians, “What is the likelihood of an adverse outcome if this patient does not receive an antibiotic today?” Response options were: very unlikely, unlikely, moderately likely, and very likely.
Our study was approved by the Michigan State University Committee on Research Involving Human Subjects, and we obtained informed consent from all subjects.
All analyses were done using SPSS 7.0 (SPSS Inc, Chicago, Ill). We calculated frequency distributions of individual variables and used chi-square analysis to assess univariate associations between each variable and both antibiotic prescribing and patient expectation for an antibiotic prescription. Continuous variables were converted to categorical data for the univariate analyses but treated as continuous data for the logistic regression. Because we measured secondary factors only in encounters in which an antibiotic was prescribed, that analysis was descriptive only and was not included in the logistic regression models.
We developed 2 logistic regression models, 1 to identify the factors independently associated with antibiotic prescribing (antibiotic prescribing model) and 1 to identify the factors independently associated with patients’ expectations of receiving an antibiotic prescription (patient expectation model).
For the antibiotic prescribing model, dichotomous variables that had a statistically significant univariate association with prescriptions for antibiotics and all continuous variables were included in the first step of the logistic regression modeling. We used the following method in an attempt to avoid type 1 errors due to the many variables we tested: All variables with a P value of .05 or less on univariate analysis were included in the initial logistic regression model, then removed sequentially, in order of ascending statistical significance, if they did not affect the odds ratio of the remaining model by at least 20%. Only main effects were assessed. We calculated odds ratios and their 95% confidence intervals from the final antibiotic prescribing model.
The patient expectation model was developed in a similar fashion. For the logistic regression, continuous variables with clinically implausible outlying values (0%) or missing values (7%) were replaced with the mean value if normally distributed and with the median value if skewed. We excluded from analysis records missing dichotomous variables necessary for logistic regression analysis (1%).
Results
Nine hundred twenty-eight patients completed questionnaires; 11 declined participation. We eliminated from analysis 222 patients with primary diagnoses of otitis media, pneumonia, pharyngitis, allergy, or other. An additional 113 were eliminated because their questionnaires were lacking essential information, they were out of the age range, or they did not give informed consent. The remaining 593 had a main diagnosis of bronchitis, sinusitis, or nonspecific URI. There were 482 records entered in the analysis: 80 were eliminated because of a secondary diagnosis of otitis media (38), pharyngitis (33), or pneumonia (9), and 31 were eliminated because the patient reported being sick longer than 30 days. Mean temperature and mean days sick were assigned respectively to 17 records (3%) missing temperature and 21 records (4%) missing number of days sick.
The subjects were primarily adult women (67%) from a predominantly white (95%) population. Physicians saw most of the patients (67%), and most visits were in a physician’s office (81%) rather than an urgent care facility. Most patients were being seen for the current illness for the first time (91%) and had been sick for no more than 14 days (86%).
The principal diagnoses assigned by the clinicians were: sinusitis for 176 patients (37%), nonspecific URI in 167 patients (34%), and bronchitis in 139 patients (29%). Antibiotics were prescribed for 319 (66%) of all diagnoses, 111 (80%) of bronchitis diagnoses, 173 (98%) of sinusitis diagnoses, and 35 (21%) of nonspecific URI diagnoses.