Combined intervention. We hypothesized that the combination of the conference, consultation, and prevention coordinator interventions would produce the greatest improvement in cardiovascular disease preventive services, because it provided professional education through the conference; facilitated individual practice assessment, planning, and implementation through the consultation visits; and provided prevention coordinator time to coordinate and support system development. Prevention coordinator hiring occurred after the conference and the randomization of practices to intervention groups. Prevention coordinators in combined intervention practices attended all consultation and reinforcement meetings and often served as a prevention leader in the practice.
Sample Size and Power Analysis
Determination of sample size for this hierarchical analysis involved an estimation of the number of primary care practices, physicians, and patient medical records needed to approximate the screening and management of individual physicians, then a practice, and subsequently, the practices within an intervention group. The number of medical records needed to represent physician performance was based on an assessment of the changes in variance in the expected physician performance from pilot data for 2 outcome variables (documented screening of blood cholesterol and smoking history on the medical record), as a function of the number of records sampled. It was determined that 20 medical records would provide a stable estimate of physician performance. Power was based on practice group mean differences among physicians, and calculation of an appropriate sample size was conservatively based on the smallest effect size detected for a single outcome variable.
Data Collection and Management
Data was collected using physician and staff questionnaires, patient questionnaires, medical record reviews, and physician and staff phone interviews. The measures assessed the intervention effects at several levels and provided cross-validation.
Patient questionnaires included a consent form allowing a review of the medical record, patient demographics, family and personal cardiovascular disease history, attitudes toward and experience with practice preventive services, and personal cardiovascular disease risk factors including hypertension, smoking, diabetes, and lack of exercise. Physician and patient care staff working full time and at least 50% part time completed questionnaires that assessed attitudes, beliefs, and estimates of cardiovascular preventive services. Periodic telephone interviews with a sample (22%, n = 239/1075) of physician and patient care staff during the interventions assessed goal setting and validated practice cardiovascular disease screening and management activities.
The medical record review included the number of practice visits, cardiovascular disease status, family cardiovascular disease history, hypertension diagnosis and management, smoking status and management, diabetes diagnosis, height, weight, lipoprotein levels within 5 years, diet advice, cholesterol medication, and exercise information. We noted the record location of screening and management data. Information was entered directly into a customized Filemaker Pro computer database by HEART reviewers blinded to intervention group at baseline but not at 12 and 18 months. At baseline, 100% of the medical records were reviewed by 2 reviewers for data entry reliability. At 12 and 18 month there were random second reviews on 10% of the charts to assess consistency and ensure accuracy. Half of the second reviews showed no data differences, and the error rate for the other records was 1.3%.
Data Analysis Model
The study hypotheses were: An increase in documentation of patient heart disease risk factors on patient records would occur as a result of the interventions, and there would be an additional increase as a result of the combined intervention. Our hypotheses also stated that these changes would demonstrate durability at 18 months, 6 months after the end of the intervention. General descriptive statistics and testing of univariate associations between selected variables were done using the chi-square statistic on categorical variables and 1-way analysis of variance where appropriate. Since each physician worked within a practice and there were 10 to 11 practices per intervention group, we used a hierarchical analytic modeling strategy to test the original study hypotheses. The hierarchical statistical tests first assess the differences between each of the 3 intervention groups and the conference-only group at baseline. A separate hierarchical model was built for the 12-month and 18-month data to perform baseline adjusted contrasts of each intervention group with the conference-only group.
Results
All 45 practices completed the study. Practice, physician, and staff characteristics are described in Table 2. None of the characteristics show a significant difference between intervention groups, primarily because of wide standard deviations. Of the enrolled practices, 87% (n = 39) consisted of only or mainly family physicians, and 11% (n = 6) were only or mostly internists. Physician specialties were: family physicians 82% (n = 131), internists 14% (n = 23), general practitioners 2% (n = 5), and 1 geriatrician. Eleven (7%) participating physicians were from ethnic minority groups, and 19% were women. Seventy-one percent of the practices (n = 32) were autonomous entities at baseline, and 29% (n = 13) were administered by a larger organization (health maintenance organization [HMO], hospital, and so forth). Only 2 practices had a participating physician leave the practice during the interventions, and 4 practices had a physician depart between the 12- and 18-month data collections. All other physicians completed the study. The response rate for the 3 questionnaires sent to the 160 study physicians was 96%.