The U.S. obesity epidemic is accelerating (1, 2). Populations of color have higher levels of overweight and obesity and have experienced greater increases in overweight during the past decade compared with white populations (3, 4). Statistics on prevalence of overweight are implicated in substantive ethnic disparities in chronic disease morbidity and mortality (3, 4) and are rooted in less healthful physical activity and eating patterns (5, 6).

Cross-sectional and prospective cohort epidemiologic studies provide estimates of the population impact of small changes in body mass index, dietary intake, and energy expenditure. For example, population decreases in dietary fat of 1% to 3% could lower first-time heart attack rates by 25% (7). In a study of working-class African Americans, Type 2 diabetes risk was 50% lower among individuals physically active at any level, and two thirds lower among those who were at least moderately active (8). Recently, results from a 6-year observation of the Nurses' Health Study cohort revealed that 30% of new cases of obesity and 43% of new cases of diabetes could be averted by adopting a relatively active lifestyle (9). The potential diabetes prevention value associated with eating a lower-fat diet and increasing physical activity was realized in the Diabetes Prevention Program randomized controlled trial. In this study, intervention participants enjoyed a 58% reduced risk of diabetes after 3 years of follow-up (10).

Few intervention studies, however, have demonstrated sustained effectiveness in preventing or controlling overweight and obesity (11, 12, 13). Studies have mainly involved either 1) highly selected, relatively affluent whites engaged in costly, individually targeted educational or behavioral interventions; or 2) somewhat more heterogeneous, predominantly white populations exposed to low-intensity mass media efforts. These studies severely limit the ability to generalize to population-based public health approaches targeting lower socioeconomic status groups or communities of color. Despite the relatively optimal clinical circumstances of the individually targeted studies, they have generally lacked sustainable success (14). This lack of long-term success in improving most risk factors has also characterized most large population-based cardiovascular disease prevention projects (with large defined as annual budgets of $1 to $1.5 million for 10 or more years) (15). These failures have been increasingly attributed to a modern obesogenic environment that promotes physical inactivity and excessive food consumption (16, 17). Environmental obesogenicity is especially concentrated in communities of color (18). The disappointing collective experience of these studies led Winkleby to suggest that smaller, more focused studies within high-risk sub-groups such as minority and low-literacy populations are needed (19). In fact, a number of public health agencies and their academic, managed care, community health center, and other community partners have begun to implement smaller-scale cardiovascular disease prevention projects. A good example is the 15 WISEWOMAN projects, funded by the Centers for Disease Control and Prevention (CDC), which target the low-income, predominately ethnic minority women screened by the Breast and Cervical Cancer Early Detection Program.

Thus, the purpose of this paper is three-fold: 1) to review available studies of community-level interventions targeting substantial proportions of people of color in geographically defined populations; 2) to qualitatively aggregate their findings; and 3) to explain the implications of these findings for applied research and public health practice in weight-control-related lifestyle change to prevent chronic disease. In theory, there are many ways of defining populations (20). Operationally, populations are generally comprised of individuals who self-report ethnic/cultural status and who can be communicated with through defined channels (e.g., churches, magazine subscription lists, television audiences).

Investigators attempting to achieve ethnically diverse samples have faced major obstacles not only at the point of intervention and retention of subjects, but even earlier in the research process — at the point of outreach and recruitment (21, 22, 23). This paper will characterize the process by which inclusive studies have engaged communities and identify ways to facilitate effective outreach and recruitment. Additionally, the paper examines the extent to which ethnically inclusive interventions have focused on structural change beyond the individual level.

This review included the following study criteria:

We conducted a search for studies that met the criteria above on the following electronic databases: PubMed, AgriCOLA, Current Contents, and PsychInfo. We limited searches to English-language articles and to articles published between January 1970 and May 2003. The search strategy consisted of 2 steps. First, we identified population-based or community-level intervention research on diet, nutrition, physical activity, physical exercise, and/or exercise. Second, we examined each result to determine the extent of participation by communities of color. Two specific keyword phrases were used in PubMed to produce broad-based results: "population-based intervention adults United States AND (exercise OR diet)," which yielded 12 articles; and "community intervention adults United States AND (exercise OR diet)," which resulted in 111 publications. Five of the studies overlapped in these two PubMed searches, yielding 118 studies in total. We modified search phrases to exclude the limit of "United States" for the other electronic databases because that specification was too restrictive. In the AgriCOLA database, similar keyword phrases identified 17 additional studies. Using those keyword phrases, the PsychInfo and Current Contents searches did not yield additional studies. The combined, non-overlapping electronic database searches resulted in 135 studies, 3 of which met the selection criteria. For each of these 3 studies, the PubMed option of retrieving "related articles" was also explored, resulting in 614 additional articles, only 5 of which met the inclusion criteria. Thus, a total of 8 articles was identified through the electronic database search.

In addition, we retrieved non-electronically catalogued peer-reviewed, non-peer-reviewed, and unpublished studies from reference lists and materials received from expert colleagues. The decision to include such "grey literature" studies with limited distribution reflects our desire to fully represent the available evidence. The recruitment, retention, and resource generation challenges of inclusive intervention studies militate against publication in mainstream scientific journals (36, 37). We contacted CDC and National Institutes of Health (NIH) staff, local and state public health professionals, and authors of published articles by telephone and electronic mail to identify "in process" and other unpublished or uncatalogued intervention efforts. We evaluated these studies using the inclusion criteria.

The process of abstracting study data was performed in 3 phases independently by 3 study co-authors: first, to produce a descriptive project narrative (Results section); second, to generate a spreadsheet of individual study data which was then aggregated in constructing Table 2; and third, to verify the information in Table 2 using a systematic abstraction process. All 12 of the characteristics that were systematically assessed in the second step across all studies are listed in Table 2. The third step was performed by the co-author who was most familiar with the articles and another co-author who had not previously seen the articles or been a part of the review process, after agreeing on the appropriate elements for the abstraction form. Discrepancies were then highlighted for discussion among study collaborators to arrive at a consensus.

The lead author developed the criteria for assessing the studies. The criteria reflect salient elements not previously presented in past reviews focusing on communities of color — specifically, the prevalence of information on the following: 1) nutrition and obesity-related lifestyle change to prevent chronic disease; 2) facilitators of effective outreach and recruitment; and 3) outcome measures that included efforts to affect both individual, organizational and legislative/policy change. The 12 characteristics assessed systematically in each study are described below.

Ethnicity of Study Population: Each study targeted at least one racial/ethnic minority community. Categories were restricted to the Office of Management and Budget's (OMB) directive on racial and ethnicity reporting, which lists 5 races (American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, and white) and 2 ethnicities (Hispanic or Latino, or Not Hispanic or Latino). Although some studies targeted specific ethnic subgroups such as Cambodians and Mexicans, the paucity of data on communities of color in general warranted adherence to OMB standards. For studies reviewed here, ethnicity was usually determined through individual self-report (ethnic self-identification).

Setting: The type of geographical setting was evaluated by census and defined as urban, suburban, semirural, or rural. A category for interventions implemented in American Indian reservations was designated as reservation-based.

Theory: With one exception, all studies were characterized as invoking well-defined behavioral theory that fit one of the following categories: Social Learning (38); Organizational Development (39); Social Ecological (40); Stages of Change (41); Diffusion of Innovation (42); or Social Marketing (43).

Design: We evaluated studies by design type. Studies employed one of the following 5 variants of evaluation research design: 1) randomized controlled trial; 2) uncontrolled trial with pre- and post-test; 3) uncontrolled trial with pre-test only; 4) uncontrolled trial with post-test only; and 5) demonstration project. Randomized controlled trial and uncontrolled trial with pre- and post-test facilitated evaluation of intervention effect sizes. Uncontrolled trials were distinguished from demonstration projects by study instigation: if the investigators who implemented the intervention also conceptualized and evaluated the project, the project was considered an intervention trial.

Recruitment Strategy: Effective recruitment strategies engaging communities of color may differ from strategies that aim to impact a mainstream population. We characterized recruitment strategies as one of the following: 1) in-person (provider, community-based organizations, or CBOs, and social networks); 2) mass media (television, radio, mainstream newspaper or magazine, billboard); or 3) targeted media (direct mail, flyer/brochure, local/ethnically targeted newspaper or magazine, distribution posters, video showings).

Sample Type: This additional study dimension was included to collect information that represented a geographically defined population, even if the study design did not fit the "gold standard" of a randomized control trial.

Attrition Rate: High attrition rates have the potential to seriously hamper study results. Studies reviewed in this paper were grouped into 3 thresholds of attrition: less than 10%, 10% to 30%, or more than 30%. A fourth category includes studies for which no attrition data was provided.

Behavior Target: Interventions generally fell into one of the following categories: diet, physical activity, and diet and physical activity combined. Where possible, a behavior target was defined as one of the following: fat; fruits and vegetables, fiber, sugar; physical activity, nutrition and physical activity, or weight monitoring. Frequent weight monitoring appeared to be a salient characteristic of long-term weight control success in the National Weight Control Registry study (44).

Outcome Measures: Central to this review is the consideration of community-level transformations, as well as individually targeted behavioral and clinical changes. We identified the following outcome measures: 1) self-reported behavior; 2) observed behavior; 3) clinical measures; 4) morbidity/mortality rates; 5) organizational practice; and 6) legislative policy.

Study Duration: We defined the duration of a study as encompassing the following 3 phases: 1) the planning period preceding the intervention; 2) the intervention itself; and 3) post-intervention assessment. Long-term follow-up is defined here as follow-up lasting at least 12 months (45). Studies were grouped into 6 categories: less than one year, one to 2 years, 3 to 5 years, greater than 5 years, or undetermined.

Significant Findings (P < .05): Intervention studies that reported significant effects (P < .05) of diet, physical activity, or weight control were categorized by target outcome. The "Other" category included findings that were related to indirect target behavior, such as organizational policy changes supporting physical activity or healthier food choices.

Primary Sources of Funding: Primary sources of funding may govern the adequacy and representativeness of the sample and the scope and duration of the intervention. Three distinct categories depict the studies analyzed: federal, state and/or local, and private.

We aggregated results qualitatively for several reasons. One, we anticipated and observed the absence of outcome data for many interventions. Two, less-developed evaluation design, measures, and analytic approaches were available for capturing the range of more upstream intervention effects (46). Three, we recognized that intervention effects at the individual level may be small (not statistically significant, but meaningful in terms of population benefit) and temporally distant from intervention implementation (46), decreasing the likelihood of publication or dissemination.

The search yielded 23 interventions that met the selection criteria: the interventions were implemented between 1972 and 2000. The following narrative summarizes, in chronological order, the intervention methods and results for projects implemented during 2 periods: the early 1970s to early 1990s (n=7), and the mid-1990s to the present (n=16). Nine of the latter 16 were projects of a CDC-funded California Department of Health Services physical activity promotion initiative in underserved and understudied ethnic communities. Table 2 presents project data by study characteristic for early and later interventions.

Consistent with review findings (34, 35) for community-level interventions targeting general audiences, most of the inclusive community-level studies reviewed here used non-interpersonal channels for information dissemination directed at broad spheres of influence (e.g., mass media), promoted a wide spectrum of physical activities, and incorporated social marketing principles. Distributions of theories referenced or implied and behaviors targeted are similar to earlier review findings, with social learning theory, community organization, and ecological models predominating. However, a greater emphasis on the processes of intervening is evident in this review, paralleling processes observed in individual-level interventions targeting underserved and understudied groups. These processes include the following: involving communities and coalition building from inception; targeting captive audiences; mobilizing social networks, particularly using lay health advisors, community health workers or promotores; cultural tailoring of messages and messengers (ethnically relevant role models in positions of power) (53, 64, 66) or charismatic leadership of key staff (49); and implementing strategies consistent with social marketing principles and social learning theory (21-23, 63, 74, 75). In fact, the reluctance of people of color to participate in research, stemming from their history of exploitation, blurs the boundary between individual-level and community-level intervention more than in mainstream culture because of the considerable community engagement and support necessary to successfully mount even individual-level interventions (23, 76-78).

Given the presentation of outcome data in fewer than half of the studies, and the few significant effects and modest effect sizes, the best data available speak only to what it takes to engage and retain people of color, not what it takes to create and sustain weight loss, engagement in regular physical activity, or improved dietary quality. However, in 2 studies, outcomes for populations of color were the only significant positive outcomes demonstrated (47, 48). The contribution of cultural adaptations to outcomes is unclear, although an effect of these adaptations on recruitment and retention may be inferred from the availability of these data on ethnic groups largely absent from other studies.

One salient observation is that population-based approaches must not automatically be construed as upstream. Compared with the findings of Alcalay and Bell (34), the studies reviewed here focus less on the individual level (65% [studies reviewed here] versus 92% [Alcalay and Bell]) and more on community norms and activities (100% [studies reviewed here] versus 56% [Alcalay and Bell]). Also compared to Alcalay and Bell, the studies reviewed here focus less on influencing policy and legislation (13% [studies reviewed here] versus 92% [Alcalay and Bell]). The 13%, however, represents an increase from 0% of the earlier studies to 19% of the later ones. Overall, a clear progression toward using upstream approaches is apparent in later studies compared with earlier studies; the increasing use of ecological models also reflects the greater use of upstream approaches. The uniform program requirements (community coalition formation and governance, for example) of the 9 ON THE MOVE! projects created some skewing of results.

Only 2 out of 23 projects were funded by state and/or local health departments. This demonstrates the importance of leadership within local government and within communities of color to set priorities and direct local resources toward chronic disease risk reduction. It also has implications for project sustainability: federal and foundation funding are generally limited to specific grant or contract periods of up to 5 years, while local funds may continue substantially longer, subject to political support and regional economic stability (i.e., tax base preservation). Fourteen projects were funded primarily through federal sources (CDC, NIH, Indian Health Services and the Food and Drug Administration). Most federal support was — not surprisingly — from the CDC, given its applied and community improvement focus.

It is sobering to note that, as of 2001, fewer than 5000 participants in individual-level interventions had been studied (and reviewed elsewhere) (21) and fewer than 14,000 participants in population-based interventions had been studied (and reviewed here) to control obesity and reduce chronic disease risk among 100 million persons of color — more than one third of Americans. This is especially sobering when one considers that ethnic minority groups are heterogeneous culturally, both within and between racial/ethnic categories, and that many of these groups are known to have significantly elevated obesity and chronic disease risk and burden. As an added challenge, data derived from ethnically inclusive studies are not widely disseminated, with only about one third of the studies reviewed here identified through electronic database searches.

Insufficient evidence exists for drawing conclusions about the effectiveness of individual-level versus community-level approaches targeting underserved racial/ethnic groups. We view these approaches as complementary and possibly synergistic. Further investigation is needed on many fronts. The environmental context must be addressed for obesity epidemic control at the population level, but the environmental context may be too limiting for the more intensive, behavioral (downstream) approaches necessary for weight management in individuals at highest risk — those already obese, hypertensive, and/or hyperlipidemic, and living or working in socioeconomically challenged circumstances. None of the studies reviewed here offered a significant beneficial solution to weight management. The best approaches in each category deserve rigorous trials (including study design and level of resources) in multi-ethnic and ethnic-specific settings. The studies reviewed here also point to the critical need for government investment in greater surveillance at local (neighborhood and census tract) levels. Federal support would allow under-resourced and overextended community providers and organizations to focus on the service delivery that best reflects their competencies and missions, relieving them of some of the burden of evaluation. Also, the relative lack of outcome data and significant findings underscores the need for evaluation methods that are more effective at capturing upstream effects and small or delayed individual effects (46, 79).