The best magazine
Methods
The subjects included in this cross-sectional study with chart review were internal medicine residents and attending physicians at three academic primary care clinics affiliated with Baystate Medical Center (Springfield, MA). For each physician, we included all of the adult patients he or she saw during the study session. Patients who were younger than 21 years old were excluded. We also excluded physicians who had no obese patients during any of the sessions they were surveyed because it was not possible to calculate either obesity recognition or documentation. The institutional review board at Baystate Medical Center approved the study.
Participation in the survey was voluntary. Physicians provided personal demographics (physician's height, weight, year in training, and number of years he or she has lived in the United States) and answered seven questions that measured attitudes about obesity and its treatment. At the end of their clinic session, physicians were provided with a list of the patients seen during that session. Survey sessions were selected at random and the participating physicians were not aware of the study until the time of survey, when the patient visits were complete. Only a small portion of each physician's sessions were surveyed. Physicians were asked to indicate whether each patient was of a normal weight (body mass index [BMI] <25 kg/m), overweight (BMI 25–29.9), or obese (BMI ≥30) based on recall. Information regarding the patient's BMI was available to the physician before and during the patient's visit via the electronic health record (EHR), but not at the time the questionnaire was completed. For each patient seen and categorized, a research nurse reviewed the corresponding chart and recorded the following information: demographics, anthropometrics, and obesity-related comorbidities (diabetes, osteoarthritis, hyperlipidemia, coronary disease, and sleep apnea). The nurse also noted whether the physician had documented obesity as a problem or had addressed it within three visits before the index visit.
Statistical Analysis
Descriptive statistics (mean/standard deviation; n[%]) were used to characterize physicians and patients included in the study. For each outcome (obesity recognition, obesity documentation/counseling), fixed effects logistic regression models were fitted to derive adjusted odds ratios (AORs) and proportions as a function of patient- and physician-level predictors and covariates. For outcomes related to one putative exposure (eg, obesity recognition and patient's ethnicity), covariates were considered for inclusion in the model if their univariable (Fisher exact, unpaired t test) associations were significant at P ≤ 0.20 (two-sided) with both exposure and outcome. Candidate covariates remained in the model if their removal changed the principal exposure coefficient by >10%. For research questions requiring predictive models, characteristics were considered for inclusion if their likelihood ratio test (intercept-only model vs single-predictor model) was significant at P ≤ 0.2. Two-sided P values of ≤0.05 were considered significant for all hypothesis tests. Stata/MP 12.1 for Windows (StataCorp, College Station, TX) was used for all of the analyses.
Source: ...