Increasingly, there have been calls for reducing health disparities based on socioeconomic position and race and ethnicity (1) and for implementing community interventions that address segments of the population in which risk for chronic disease is concentrated (2,3). Few papers in the literature, however, address the methodological challenges in recruiting participants for studies of such interventions. This paper describes and presents the results of the sample selection and survey methods for two group-randomized trials of cancer prevention interventions designed for multiracial and multiethnic working-class populations.

The Harvard Cancer Prevention Program Project, Healthy Directions, was designed to develop and evaluate cancer prevention interventions for multiracial and multiethnic working-class populations (4). The project comprised two intervention studies, Healthy Directions–Small Business (HD–SB) and Healthy Directions–Health Centers (HD–HC), and a cancer prevention policy model-analysis project. The intervention projects were group-randomized controlled studies that tested the shared primary hypotheses that mean levels of dietary and physical activity outcomes would improve more significantly in the intervention group than in the control group. The interventions developed for the two projects were based on a common conceptual framework (4) drawing on social ecological theory (3,5). Using this framework, the social context in which people live was incorporated into the design and delivery of the interventions. This framework encompasses several factors, including individual factors (e.g., material circumstances), interpersonal factors (e.g., family roles and responsibilities), organizational factors (e.g., access to health care), and community factors (e.g., neighborhood safety). In contrast to interventions designed for a specific racial or ethnic group, we used this framework to design interventions that were suitable for a multiracial and multiethnic population.

HD–SB was a worksite-based intervention study designed to test the effectiveness of an integrated health promotion and occupational health protection intervention in 26 small manufacturing businesses in Massachusetts (6). HD–HC was a health-center–based intervention in 10 community health centers in metropolitan Boston (7). The two intervention studies were aimed at four primary outcomes: increasing fruit and vegetable consumption, decreasing red meat consumption, increasing daily multivitamin use, and increasing physical activity. In both studies, the organization was the unit of randomization and intervention, and the individual worker or health center member was the unit of observation.

Group-randomized trials are those in which groups of individuals are randomized to study conditions, but observations are made on the individuals within the groups (8). An advantage of this design is the ability to enhance the intervention effectiveness through the social interactions among members of the groups randomized. The main disadvantage is the loss of statistical efficiency due to the correlation in behavior among members of the same group (9). This study design has been increasing in popularity over the last 25 years, especially for the evaluation of community-based interventions (10). Planning for such studies requires estimates of the within-group correlation of the proposed outcome measures, yet published estimates for specific behaviors and populations are hard to find because there are few publications that include these values in reports of results.

This report focuses on our success in recruiting multiracial and multiethnic working-class participants. We compare the characteristics of the participants with selected characteristics of the larger population within which they reside, and we provide point estimates of the outcome measures and estimates of the intraclass correlation coefficients (ICCs). The ICC is the fraction of the total variation in a measure that is attributable to the clustering of the behavior by members of the same group in comparison with members of different groups (i.e., the health center or worksite). This information is important for researchers planning group-randomized trials in diverse working-class populations.

The methods of both studies were approved by Dana-Farber Cancer Institute's Office for the Protection of Research Subjects and the Harvard School of Public Health's Human Subjects Committee. Additionally, the methods of the small business study were approved by Beth Israel Deaconess Medical Center's Committee on Clinical Investigations, and the methods of the health centers study were approved by Harvard Vanguard Medical Associates–Department of Ambulatory Care and Prevention.

These two studies were successful in sampling a multiracial and multiethnic subpopulation of eastern Massachusetts residents. The sampling strategies of both studies reached a subpopulation that is more heterogeneous in racial and ethnic make-up than the greater Boston area. Furthermore, the HD–SB sample has a larger percentage of members with working-class occupations and those with a high school education or less than the general population of adults.

Health disparities in the United States are often described in terms of racial or ethnic inequalities; yet, within racial and ethnic groups, there is variability in both socioeconomic position and morbidity and mortality risk. Nevertheless, populations of color bear a disproportionate burden of poverty (25-27). It is well known that socioeconomic deprivation adversely affects health and increases mortality (12,28). The concepts of social class and socioeconomic position are complex and encompass occupational class, income, poverty, wealth, education, and prestige or status at the individual, household, and area levels (12). We have measured two dimensions of race and ethnicity and three of socioeconomic position. Maintaining these separate characteristics, rather than attempting to define a single measure of socioeconomic position, will allow us to explore the interactions among them in understanding the determinants of successful interventions.

Small percentages of the study sample respondents lived in households that were below the poverty threshold, as expected among a population of working-class participants and those with health insurance. Nevertheless, a substantial proportion of our samples would be eligible for food aid — 15% in the HD–SB sample and 18% in the HD–HC sample. These categorizations do not take into consideration regional differences in cost of living. The greater Boston area is one of the most expensive areas in the country; the self-sufficiency standard for a family of four in that area was $42,564 in 1998 and $54,612 in 2003 (29).

Although both study samples represent multiracial and multiethnic working-class populations, the two samples differ from one another. The HD–SB sample was somewhat younger and included more men than the HD–HC sample. The HD–SB sample had a higher percentage of Asian and Hispanic respondents and a higher percentage of recent immigrants than the HD–HC sample. The HD–HC sample had a higher percentage of participants with household incomes below 185% of poverty than the HD–SB sample. The HD–SB sample has a higher percentage of respondents with less than a high school education than the HD–HC sample.

The differences between the two study samples in the levels of the health behaviors reported by the participants reflect these sociodemographic differences. The workers in the HD–SB sample were more physically active than those in the HD–HC sample were, and a higher percentage of the HD–HC sample members reported taking a multivitamin daily. These differences may be attributable to the differences in age, sex, education, or other factors. Despite these differences in health behavior practices, the percentages of respondents in both samples who were at lower levels of the target behaviors are high, indicating a need for the behavior change interventions in both subpopulations.

The mean hours of physical activity reported by members of both samples is surprisingly high. Although we asked about leisure time activity, participants may have conflated their reports to include activities related to occupational activities, domestic chores, childcare, and walking for transportation, for example. A small validity study done in conjunction with this project indicated, however, that total hours of activity were reported accurately (data not shown).

The response rate to the baseline survey in HD–HC was low, due in part to the fact that potential participants were agreeing to participate in a randomized trial, not just a health survey. This is similar to the low recruitment rate for the businesses in the HD–SB study. The internal validity of the intervention trials is assured by randomization, and the survey is a valid baseline assessment of the levels of behaviors in the two intervention groups prior to intervention implementation. For other researchers who might use the baseline measures and ICCs to plan studies, the generalizability to health center members and workers in worksites who would consent to participate in such a study is also appropriate.  

Planning for group-randomized trials of the effectiveness of interventions targeting modifiable health behaviors requires estimates of the means, variances, and ICCs of the behaviors within the study population of interest (8,10). The estimates reported here apply to four specific health behaviors and two types of randomization groups. The estimated ICCs in the HD–SB sample are similar to those found in other worksite-based intervention studies (30). The estimated ICCs in the HD–HC sample are lower than those in the HD–SB sample are but are similar to those at the district health authority level in England (31). Nevertheless, the ICCs in both studies are sufficient to influence the error variance of the test statistic for evaluating the effectiveness of the intervention and must be included in power calculations for group-randomized studies.

In summary, the procedures developed by these two intervention studies to sample multiracial and multiethnic working-class populations in eastern Massachusetts were successful in identifying such groups. These samples are more diverse in their racial and ethnic make-up and other sociodemographic characteristics than the greater Boston  population. Although the subpopulations resided in the same geographic area and may overlap in other ways, the HD–HC sample was restricted explicitly to exclude anyone in the HD–SB sample. Despite the close proximity of these two subpopulations, they differ in ways that would be expected by their provenance. Furthermore, both samples represent populations with high percentages of members who have cancer-related risk behaviors.

There has been a call for research on the effectiveness of interventions targeting modifiable health behaviors (28), yet intervention approaches have not been designed for or sufficiently tested in working class, ethnically diverse populations (32). Our explicit aim was to recruit from the large multiracial and multiethnic group of working-class men and women at elevated risk for adverse health outcomes. This group is confronted with constraints and limited resources that may influence patterns of health behaviors. We have developed behavioral interventions that respond to the social contextual realities of this group as an approach to reducing the excess burden of cancer borne by communities of color and lower socioeconomic position (4). To fully evaluate the effectiveness of these interventions, it is important to study a diverse population. Future manuscripts will report on the effectiveness of the interventions in promoting change in the behaviors and the influence of the social context on behavior and behavior change.