We designed a population-based, prospective, observational study and recruited participants who were enrolled with the Henry Ford Health System (HFHS) in Detroit, Mich., USA. The HFHS comprises the Henry Ford Medical Group (HFMG), which is a large medical group practice, as well as the Health Alliance Plan (HAP), which is a health maintenance organization. Our study population was drawn from a pool of approximately 350,000 patients at the intersection of HFMG-HAP. When patients first visit the health system, they are provided a copy of the systems ‘Notice of Privacy Practices’. The notice includes explanations of the use of personal information for approved research. A master patient index is maintained with enrollment and basic demographic information. From this list we were able to identify study subjects who had been enrolled for at least 1 year in HAP, were aligned with a HFMG primary care provider and were between 25–40 years of age. Individuals whose electronic medical record included an ICD-9-coded clinical appointment for diabetes mellitus (type 1 or 2), atherosclerotic cardiovascular disease or osteoporosis were excluded. HFHS is part of the Detroit Surveillance, Epidemiology and End Results (SEER) tumor registry and has been participating since the registry's inception in 1973. This registry was matched to the pool of potential participants to exclude individuals with a prior history of cancer (except non-melanoma skin cancer).
As men and African Americans have been underrepresented in prior genetic research studies, we oversampled individuals from these groups. We used the master patient index data to identify patients’ self-reported gender and race. The initial sample was 59% African American versus 41% white. We also oversampled those predicted to have less education. To approximate education status, we mapped the patient's address to information from the 2000 US Census. Individuals from census block groups where 10% or more of the residents whose educational level was high school or below were considered to be from ‘low education neighborhoods’ and likely to be of a lower education status themselves. By this method, 55% of people in our sample were from low education neighborhoods.
Using this oversampling scheme, we created a participant file containing 6,600 potential participants. We accepted only 1 individual from each household. In a staggered fashion, we mailed advance letters advising the potential participants that they would be asked to participate in a survey. A USD 2 bill was included with this mailing and a toll-free number to call if the potential participant did not want to participate. Approximately 2 weeks after the advance letter was mailed and the participant had not called to decline participation, the Center for Survey Research at Group Health Cooperative (Seattle, Wash., USA) telephoned each participant to conduct a 35-min baseline interview. The survey included confirmation of the participant's medical history as well as psychological and behavioral assessments. Those who self-reported a personal medical history of conditions that excluded them from the study (e.g. diabetes, heart disease and osteoporosis) were given an abbreviated survey and excluded from further participation in the study. At the conclusion of the survey, eligible interviewees were sent a brochure describing the Multiplex study.
After observing initial participation rates for 6 weeks, we increased study incentives to achieve 2 objectives, namely to increase overall recruitment and to encourage those who were not interested in testing to log on to review and evaluate the information provided about the Multiplex test. We added text to the survey's exit script specifically mentioning that we would be mailing a brochure about the next phase of the study with a USD 20 bill. In addition, we added exit script text that also mentioned the incentives available through the study's Web site. The cover letter text also mentioned the incentives available for completing the Web-based assessments. Finally, we asked for participants’ e-mail addresses and sent them e-mail reminders that included a link to the study Web site. The brochure mailing included the study identifier and directions for logging into the Multiplex Web site. In the brochure, participants were offered a paper copy of the Web site, if they so desired. The Web site included information modules on genetic testing, specifics about the genes being tested and 4 questionnaires. Incentives (in the form of gift certificates to a major retailer) were offered for completing each assessment on the Web site, up to a total of USD 50 for completing all Web assessments. The focus of the first 3 assessments was to assess understanding of the information presented. The fourth assessment was simply one question, i.e. ‘Are you interested in genetic testing?’ Three responses were offered: ‘Yes’, ‘No’ and ‘Maybe’. Participants who answered ‘Yes’ or ‘Maybe’ were contacted by a research educator (RE) to answer any additional questions and/or to schedule a clinic visit for blood collection.
To maximize participant convenience, multiple HFHS clinics throughout metropolitan Detroit served as sites for participants to provide written consent and blood samples. During the clinic visit, the RE used a PowerPoint presentation to explain the pros and cons of testing prior to obtaining the participant's written consent. After the presentation, the participant completed a questionnaire to evaluate their understanding of the content presented. The participant then signed the consent and provided a blood sample for the genetic test.
Both the National Institutes of Health and HFHS Institutional Review Boards approved all aspects of this study.
Comparisons were made between participation in each of the 3 major research contacts (i.e. completing the baseline survey, visiting the study Web site and having the genetic test performed) by each of our oversampled groups (race, gender, and education neighborhood). Comparisons were made using χ2 tests and quantified with logistic regression modeling. All p values were assigned using two-tailed tests. Logistic regression models were used for multivariate analyses.