The ethics committee of the Canton Zurich approved the study protocol, and we obtained informed consent from all participants.
Recruitment of physicians
Physicians from two group practices in Zurich (mediX group practice and “Praxisgemeinschaft Altstetten,” both located in the urban Zurich area) were invited to participate in this study. They were provided with detailed information about the study, including the questionnaire for patient histories and the patient information leaflet.
In- and exclusion criteria
Patients aged 16 years or older with the main symptom of new or worsened cough lasting at least 24 hours and who experienced the feeling of increased body temperature were potentially eligible for this study. Only patients in which our decision aid was negative, meaning patients with CRP levels less than 50 μg/ml (normal < 10 μg/ml), no dyspnea reported by the patient, no daily subjective feeling of increased body temperature since the onset of cough, and no treatment with antibiotics after the first consultation. We excluded patients with known chronic lung diseases (except chronic bronchitis), HIV-positive patients, patients who had taken oral corticosteroids within the last month, patients on chemotherapy, patients after organ transplantation, pregnant women, and patients incapable of reading the information leaflet and/or providing informed consent. Patients who were prescribed antibiotics at the initial consultation were excluded from the validation sample [7].
Data gathering
Physicians assessed medical history after obtaining informed consent from the eligible patients. In particular, they asked patients about the presence of dyspnea and the duration and regularity (daily or not) subjective feeling of increased body temperature. Body temperature was not measured during consultation and patients were not instructed to measure body temperature at home. Venous blood samples for measuring CRP levels were drawn from all patients, and blood was analyzed using standard procedures. Other diagnostic procedures, treatment type, and the decision to initiate antimicrobial treatment were left to the discretion of the treating physician.
At a minimum of one week after the first consultation (inclusion of patients), physicians or physician assistants contacted the patients for follow-up only once by phone or during a planned office consultation. All patients were asked four questions: Have you received a prescription for antibiotics since the time of the consultation leading to inclusion in the study? Have you been hospitalized during this time? Has the severity of your cough decreased, increased, or remained stable since the first consultation? Has the subjective feeling of increased body temperature disappeared, remained stable, or worsened? When a patient had been hospitalized since inclusion in the study, we contacted hospitals to obtain the information whether pneumonia or another illness was the reason for the hospitalization.
When patients at the follow-up consultation (by phone or in the office) reported that symptoms improved and no antibiotics had been prescribed to them since the first consultation we assumed that a clinically relevant pneumonia could be excluded with a high degree of certainty.
Statistical analysis
The sample size of this study was planned to estimate the proportion of falsely classified non-pneumonia cases with a specified precision. For this we assumed p to be 2% (p = 0.02), and the standard error (p) = 0.015 such that the upper bound of the 95% confidence interval for p does not exceed 5%. These specifications led to a sample size of n = 88; we aimed to include at least 90 patients.
All patients fulfilling the inclusion criteria, and with complete answers to the four interview questions, were included in the statistical analysis. Descriptive statistics were calculated for patient demographic information, clinical signs of disease, and answers to the four interview questions. Categorical variables were displayed as percentages, and continuous variables as median and range. For the estimation of the proportion of falsely classified non-pneumonia cases , we determined the number of patients who received an antibiotics prescription for the diagnosis of pneumonia (question 1), and divided that number by n. We estimated the corresponding 95% confidence interval with the Wilson score interval [8]. All analyses were performed with R 2.14 [9].