The CDC Division for Heart Disease and Stroke Prevention provided initial funding for ARCHES ($760,000 over 2 years), through the ADH Heart Disease and Stroke Prevention program. To maximize the scope of ARCHES, we used this funding to encourage participation from several partners, which resulted in funding, donated materials, volunteer assistance, and collaboration from many programs within ADH and several external collaborators (Figure 1). Total cash available became $1.08 million. With this additional funding, we expanded ARCHES beyond its initial CDC mandate and included several questionnaire domains, covering many risk factors and health conditions, and the collection of examination data, including anthropometric measures and biological samples. The Science Review Council of the ADH and its institutional review board approved all protocols, instruments, procedures, consent forms, and other documents used for recruitment and data collection.

ARCHES was a population-based, cross-sectional survey of noninstitutionalized adult (aged ≥18 y) residents in Arkansas, using a 3-stage clustered sample design (Table 1). In the first stage, we divided the 623 inhabited census tracts into 1) 188 tracts with a black population greater than 22.7% and 2) the remaining 435 tracts. We designated the first group, and the 5 largest tracts from the second group, as certainty clusters (ie, they were included in the sampling frame). We selected an additional 182 clusters from the remaining 430 tracts by using probability proportional to size, resulting in 375 sampled clusters. In the second stage, letters were sent to a random sample of households within the selected clusters, and then we randomly called households in each cluster until we had 4 households in that cluster that agreed to participate. In the third stage, in each selected household we determined the number of adults and sampled 1 person by using a computerized algorithm based on Kish sampling methods (2,3). If this person refused to participate, the next household on the list was called, until 4 respondents had agreed in each sampled cluster.

Our goal was to recruit approximately 1,500 participants, based on available funding and sample size calculations indicating that a sample of 1,344 would yield statistical power adequate to compare blacks and whites for several key cardiovascular variables. We verified the adequacy of the sample size by simulating results using BRFSS household data, Census 2000 racial data, and NHANES prevalence data.

We weighted the data in 2 steps to represent the Arkansas population. First, to account for the complex sampling plan, we computed a structural weight as the product of the inverse of the cluster sample probability (1 for certainty sample tracts, otherwise as computed by PROC SURVEYSELECT of SAS [SAS Institute, Inc, Cary, North Carolina]), households per sampled tract represented by each sample household, and adults per sampled household represented by each sample adult. Second, we accounted for nonresponse by using a postsampling weight that adjusted age, race, and sex categories to the 2007 Arkansas population estimates (4).

Exclusion criteria were not speaking English and having psychiatric, cognitive, or developmental disorders, which were identified during the recruitment call. We excluded non-English speakers because, at the time of sampling, only 5% of the state's population were estimated to be Hispanic — not enough to form a significant part of the sample — and funding limitations did not allow oversampling of this subgroup.

We recruited participants through a letter followed by a telephone call. We first mailed letters and brochures (written at an average 7th-grade level) to selected households, explaining the survey and informing them that a telephone call would be made on behalf of ADH asking for participation of a person from the household. Within 2 weeks, we made calls to households to explain the survey, answer questions about it, and ask for enough household information to allow random selection of 1 person.

We briefly screened the selected person by telephone to ascertain eligibility, to answer the participant's questions about the survey, and to obtain initial verbal consent for participation. We then made an appointment with the selected person for an in-person interview and examination, at a location of the participant's choosing (in all cases, the home of the participant). We informed participants that blood and urine samples would be taken and asked them to fast overnight unless there were medical reasons not to. We also informed participants about benefits of participation, including the service that they would provide, provision of all laboratory test results (valued at approximately $260) to each participant, and gift cards of up to $50 ($40 for home visits and $10 for a returned Food Frequency Questionnaire [FFQ]).

To maximize participation and data completeness, we completed the interview, examination, and collection of blood and urine samples in 1 home visit. The only exception was the FFQ, which we left with the participant with instructions for completion and return to ADH, using self-addressed and stamped envelopes. Home visits ranged from 60 to 90 minutes, depending on skip patterns for questions and ease of anthropometric measurements and sample collection. Examination Management Services, Inc (EMSI) (Irving, Texas), a provider of specimen collection services for clinical trials and epidemiologic studies, collected all data.

Interviewers were nurses or other health professionals employed by EMSI and trained in phlebotomy and interviewing and examination techniques. In addition, we required that interviewers take training courses related to human subjects research and Health Insurance Portability and Accountability Act (HIPAA) privacy rules and participate in a 3-day training and certification session conducted by ARCHES investigators. Training included administering the questionnaire and standard protocols for drawing blood, measuring blood pressure, and performing anthropometric measurements. In accordance with Arkansas law, we also trained interviewers to follow required procedures if they observed child or adult abuse while in homes.

The ARCHES questionnaire consisted of up to 285 questions (depending on skip patterns) covering behavioral, psychosocial, socioeconomic, and demographic variables, personal and family medical history, cardiovascular and other chronic disease risk factors, health care access, and other subjects (Box). Questions were mainly from BRFSS (5) and NHANES (6). The complete questionnaire was pretested through cognitive interviews with a convenience sample of low-income community volunteers, and necessary changes were made. Nutrient intake data were collected by using the FFQ developed by the Nutrition Assessment Shared Resource (7) of the Fred Hutchinson Cancer Research Center (FHCRC). Questionnaires are available on the ARCHES page of the ADH website (http://www.healthy.arkansas.gov/programsServices/chronicDisease/Initiatives/ Pages/Arches.aspx).

After the participant provided written informed consent, interviewers administered the questionnaire and recorded responses on paper forms. Interviewers also examined medicine bottles for all medications (prescribed and over-the-counter, including dietary supplements), and recorded medication names. We measured participants' height, weight, and abdominal circumference by using standard NHANES protocols, while they wore light clothing and no shoes. We used a Tanita digital, self-calibrating scale (model HD-351) (Tanita Corporation of America, Inc, Arlington Heights, Illinois) to measure weight. After measurement of arm circumference and use of an appropriate-sized cuff, interviewers recorded blood pressure 3 times at intervals during the interview process by using standard protocols (6), with an Omron HEM-907XL monitor (Omron Healthcare, Inc, Bannockburn, Illinois). For each recording, the monitor recorded the average of 3 separate readings, for up to 9 readings and 3 recordings.

After administering the questionnaire and taking anthropometric measurements, interviewers collected blood and urine samples. Interviewers placed the samples in containers with frozen-gel bags, processed them in the field according to protocols provided by Clinical Reference Laboratories, Inc (CRL) (Lenexa, Kansas), and shipped them to CRL, a laboratory certified according to the CDC Lipid Standardization Program (http://www.cdc.gov/labstandards/lsp.html), for analysis and reporting. The laboratory also froze aliquots of blood and urine and shipped them to ADH for storage and future analyses.

We managed all activities, including tracking of interviews, movement of forms and biological samples, data entry, and quality control, centrally at ADH, through contracts with several external entities (Figure 1). After completing the interview, EMSI personnel shipped questionnaires, forms, and biological samples to CRL, where blood and urine samples were further processed and analyzed. CRL sent the results of analyses, along with forms and questionnaires, to ADH, where they were logged and entered in the main database. The University of Arkansas at Fayetteville Survey Research Center scanned the main questionnaires, digitized the data, and transmitted them electronically to ADH. ADH personnel logged and shipped the FFQs that had been mailed back by participants to FHCRC for analysis. FFQ results were electronically transmitted back to ADH. After all data were received and entered into the main database, we created an analysis dataset, stripped of personal identifiers, for use in further analyses.

We took several steps to ensure data integrity and quality. One of the senior investigators telephoned a 5% random subsample of respondents within a few days of the interviews to ascertain interviewer accuracy and find out about participants' experiences with the interviewers. We reported inaccuracies or problems to EMSI for corrective action. Before scanning and data entry, we hand-checked all questionnaires for readability, correct skip patterns, and missing data, and made corrections to the extent possible (calling respondents when needed). We pulled a 5% random subsample of paper copies of the main questionnaires and checked them against the electronic data to ensure accuracy in digitization. We hand-checked all FFQs before scanning and analysis. Laboratory analyses were subject to internal laboratory standards and checks by CRL and to range and consistency checks of all data by ARCHES staff.

Interviewers informed respondents with high blood pressure of their measurements and advised them to seek appropriate medical care. We monitored the results of blood and urine tests as they were received by ARCHES; respondents with results above predetermined critical values were contacted immediately and advised to seek medical attention, with an offer to fax results directly to their health care provider. We sent all blood and urine results, within 2 weeks of receipt, to each respondent along with a cover letter, with abnormal values flagged for attention and discussion with their health care provider.