Risk Factors Amenable to Primary Prevention of Type 2 Diabetes Among Disaggregated Racial and Ethnic Subgroups in the U.S.

Type 2 diabetes currently affects ∼11% of adults in the U.S., with 1.4 million newly diagnosed adults each year (1). Major risk factors, such as obesity and physical inactivity, continue to be highly prevalent, despite being amenable to lifestyle interventions with the potential to substantially decrease risk (2). Other amenable risk factors, such as unhealthy dietary patterns, smoking, and hypertension, may also increase risk (2). Moreover, for the 12% of U.S. adults who are uninsured or underinsured, preventive health interventions may be inaccessible or inadequate (3), potentially contributing to earlier onset or more severe type 2 diabetes (4). Even among those adequately insured, other barriers to health care access, such as provider availability and accessibility (e.g., lack of transportation), may limit access to preventive care (5).

Racial and ethnic disparities in type 2 diabetes risk factors, such as obesity, physical inactivity, and health insurance status, are well-documented—showing, for example, the highest prevalence of insufficient physical activity among non-Hispanic (NH) Black adults (6). However, less research has focused on evaluating these risk factors using disaggregated race and ethnicity data. Data disaggregation breaks down commonly used racial and ethnic categories into more detailed subgroups to highlight underlying patterns and heterogeneity that would otherwise not be seen if the data were aggregated. Therefore, data disaggregation can be vital to reveal potentially large heterogeneity in risk factors among subgroups attributed to differences in factors such as country of birth, language, cultural norms, socioeconomic status, and other social determinants and drivers of health inequities.

Given the substantial health and economic burden of type 2 diabetes, increasing the effectiveness of primary prevention through racial and ethnic subgroup data disaggregation is important from a population health perspective. As a first step in describing the population’s risk profile, we evaluated the prevalence of both lifestyle-related type 2 diabetes risk factors and barriers to preventive care—among aggregated racial and ethnic groups and disaggregated subgroups in a nationwide population of adults without diagnosed type 2 diabetes in the U.S.

The study sample derives from the Behavioral Risk Factor Surveillance System (BRFSS), an annual random-digit-dialed landline– and cellular telephone–based survey of a randomly selected representative sample of noninstitutionalized adults aged ≥18 years, from the 50 states, the District of Columbia, and participating U.S. territories. BRFSS collects information including but not limited to health risk behaviors, health care access, and chronic conditions. Further information is available online from https://www.cdc.gov/brfss/. The study period included data from 2013, when disaggregated data on select, self-identified racial and ethnic subgroups started to be collected, to 2021. The overall median response rate, defined as the percentage of adults who completed the interview among all eligible adults, ranged between 44.0 and 49.9% across the survey years. Among 4,030,567 respondents, 535,026 had self-reported diagnosed diabetes, 58,743 had missing data on age, sex, and/or race and ethnicity, and 146 did not have data on any of the variables measuring lifestyle risk factors or health care access, resulting in a sample of 3,437,640 respondents. The corresponding distribution of respondents by survey year and race and ethnicity is shown in Supplementary Table 1. Informed consent was attained from respondents. This study was reviewed by the Centers for Disease Control and Prevention (CDC) and was conducted consistent with applicable federal law and CDC policy [45 CFR §46.102(l)(2), 21 CFR §56; 42 U.S.C. §241(d); 5 U.S.C. §552(a); 44 U.S.C. §3501 et seq.].

Sociodemographic information included age, sex, educational attainment (less than high school, high school, more than high school), and employment status (employed, not employed, retired). Race and ethnicity were first categorized as Hispanic or NH. Thus, respondents who identified as both Hispanic and an NH race group were categorized as Hispanic. Race and ethnicity variables and corresponding disaggregated subgroups on the questionnaire included the following: Hispanic (Cuban, Mexican, Puerto Rican, Other Hispanic), NH American Indian/Alaska Native, NH Asian (Chinese, Filipino, Indian, Japanese, Korean, Vietnamese, Other Asian), NH Black, NH Pacific Islander (Guamanian/Chamorro, Native Hawaiian, Samoan, Other Pacific Islander), NH White, NH Multiracial, and NH Other.

Self-reported type 2 diabetes risk factors were categorized as dichotomous variables (yes/no) and included the following: 1) current smoking; 2) hypertension diagnosed by a health care professional; 3) overweight or obesity, defined as BMI >25 kg/m², or >23 kg/m² for NH Asian adults (7); and 4) physical inactivity, defined as no leisure-time physical activity in the past month. As information on hypertension was collected biennially, analyses including this variable are based on a subset of the total sample (n = 1,907,017).

Risk factors related to health care access and use, hereafter referred to as barriers to preventive care, were categorized as dichotomous variables and included 1) being uninsured; 2) not having a primary care doctor, on the basis of the respondent not reporting a doctor or group of doctors thought of as their personal health care provider(s); 3) cost concerns, based on the respondent reporting at least one instance in the past 12 months of avoiding seeing a doctor due to cost; and 4) not having a general physical exam within the past 12 months. As the questionnaire item on health insurance status changed in the 2021 survey year, analyses including this variable are based on 2013–2020 survey data (n = 3,053,052). Questionnaire items used for each risk factor are shown in Supplementary Table 2.

Prevalence of each risk factor with 95% CIs was estimated from multivariable logistic models. Covariates included age and sex to adjust for basic demographic factors that may affect risk factor prevalence, as well as survey year, to account for any changes in sampling over the study period. Other factors that may affect risk factor prevalence, such as socioeconomic factors, were not included as covariates for two reasons: 1) to highlight the descriptive epidemiology of potential heterogeneity in prevalence among disaggregated racial and ethnic groups, since further adjustment for socioeconomic factors might obscure actual differences in prevalence as socioeconomic factors are a likely driver of some disparities; 2) to keep the models parsimonious, as additional variables may have resulted in model instability due to small samples for some racial and ethnic subgroups. Sample weights and complex designs were used to produce nationally representative estimates. Analyses were conducted with SAS 9.4 (SAS Institute, Cary, NC) and SUDAAN 11.0.1 (RTI International, Research Triangle Park, NC).

Data that support the findings of this study are not publicly available due to terms of the data use agreement with the Population Health Surveillance Branch, CDC. Data that do not include information on disaggregated race and ethnicity are publicly available online from https://www.cdc.gov/brfss/annual_data/annual_data.htm.

Characteristics of the 3,437,640 adults without diagnosed diabetes are displayed in Supplementary Table 3. Overall, mean (SE) age was 45.9 (0.02) and 51.6% (SE 0.1) were women. Among all aggregated racial and ethnic groups and disaggregated subgroups, mean age ranged from 36.1 (0.56) years for Vietnamese adults to 53.6 (0.54) years among Japanese adults. Sex distribution also varied from the lowest proportion of female adults in the Samoan group (43.0% [SE 3.3]) to the highest among Filipino adults (60.3% [1.1]). Educational attainment varied considerably from 31.8% (0.2) of Mexican adults reporting more than a high school degree to 84.9% (0.6) among Indian adults. Employment status ranged from 53.1% (1.6) of Japanese adults reporting being employed to 70.0% (0.7) among Indian adults.

Figure 1 and Supplementary Table 4 show the prevalence of type 2 diabetes lifestyle risk factors (current smoking, hypertension, overweight/obesity, and physical inactivity) among adults without diagnosed diabetes, by aggregated racial and ethnic groups and disaggregated subgroups.

Overall prevalence of current smoking was 16.1% (95% CI 16.0–16.2), with the highest prevalence of 29.4% (28.5–30.2) observed among NH American Indian/Alaska Native adults. Prevalence varied within disaggregated subgroups. For example, among NH Asian adults, the lowest prevalence was 5.2% (4.6–5.8) among Indian adults and 12.2% (10.5–14.1%) among Korean adults (Fig. 1A).

Prevalence of hypertension showed similar heterogeneity, although individual estimates show greater variability due to the smaller sample size (Fig. 1B). Overall prevalence of hypertension was 27.7% (95% CI 27.6–27.9), with the highest prevalence, of 37.5% (37.1–37.9%), among NH Black adults. Prevalence also varied among subgroups, such as among Hispanic adults, where prevalence ranged from ∼25% in both Mexican and Other Hispanic adults to 31.9% (31.2–32.6) among Puerto Rican adults.

Overall, approximately two in three adults without diagnosed diabetes had overweight or obesity (Fig. 1C). Prevalence varied among all subgroups. For example, among NH Pacific Islander adults, prevalence ranged between 64.2% (95% CI 61.6–66.7) for Other Pacific Islander adults to 79.8% (73.5–84.9) among Samoan adults, with the latter finding representing the highest prevalence among all racial and ethnic groups and disaggregated subgroups.

Approximately one-quarter of adults without diagnosed diabetes were physically inactive (Fig. 1D). The highest prevalence was observed among Puerto Rican adults (37.7% [95% CI 37.0–38.3]). Compared with other risk factors, prevalence of physical inactivity varied less within most racial and ethnic subgroups.

Figure 2 and Supplementary Table 5 show the prevalence and corresponding 95% CI of barriers to preventive care. Approximately 14% of adults without diabetes reported being uninsured (Fig. 2A). Prevalence varied approximately threefold within Hispanic subgroups, from 11.4% (95% CI 10.9–11.9) among Puerto Rican adults to 33.0% (32.5–33.5) among Mexican adults, with the latter finding representing the highest prevalence among all aggregated racial and ethnic groups and disaggregated subgroups.

In Fig. 2B, the overall prevalence of adults who reported having no primary care doctor was 23.8% (95% CI 23.7–23.9). Among both aggregated racial and ethnic groups and disaggregated subgroups, the distribution of prevalence estimates for this metric was similar to that of being uninsured, with the largest subgroup differences among Hispanic subgroups (Mexican, 38.1% [95% CI 37.7–38.6]; Puerto Rican, 22.5% [22.0–23.1]).

The overall prevalence of adults reporting health care cost concerns was 12.8% (95% CI 12.7–12.9), with the highest prevalence among Mexican, Other Hispanic, and NH American Indian/Alaska Native adults (18–21%) (Fig. 2C). Prevalence of having cost concerns was heterogeneous among disaggregated subgroups. For example, among NH Asian subgroups, the lowest prevalence, of ∼6%, was observed among Chinese and Japanese adults, and the highest prevalence, 11.1% (95% CI 9.3–13.1), was observed among Korean adults.

In the total sample, 29.6% (95% CI 29.5–29.7) reported no physical exam in the past 12 months (Fig. 2D), with the highest prevalence, 36.8% (36.3–37.3), reported among Mexican adults. Among disaggregated subgroups, there was an ∼10 percentage point difference between the lowest and highest prevalence observed within both Hispanic and NH Asian subgroups.

In a large nationwide sample of adults without diagnosed diabetes, we observed substantial heterogeneity in both lifestyle risk factors for type 2 diabetes and barriers to preventive care among both aggregated racial and ethnic groups and disaggregated subgroups.

As overweight or obesity is a major risk factor of type 2 diabetes, it is important to assess disparities in overweight or obesity by disaggregated racial and ethnic subgroups. Investigators of prior studies among NH Asian subgroups, which have predominately included evaluation of only regional populations, reported similar findings with the highest prevalence of overweight and obesity among Filipino and South Asian subgroups (8,9). Because studies of obesity among NH Asian subgroups are scarce, reliable explanations for subgroup differences are lacking. Given the nature of race and ethnicity as social constructs, it is important to examine the potential role of social determinants of health in explaining subgroup disparities. For instance, among Filipino and Indian adults, the high prevalence of overweight and obesity may be attributed to increased acculturation (10). The highest prevalence of overweight or obesity observed among Guamanian/Chamorro and Samoan adults may be attributed to factors such as a poor food environment, transition from a traditional to Western diet, and urbanization (11). To our knowledge, no current nationwide studies have described the prevalence of overweight and obesity among disaggregated subgroups of NH Pacific Islander adults in the U.S. However, these findings are partly reflected in those of a regional study with data from a 2007 health plan survey, showing a higher prevalence of obesity among Samoan adults compared with other Pacific Islander adults (12). Previous evidence has suggested a higher BMI threshold for Polynesian populations to define overweight (26–32 kg/m²) and obesity (>32 kg/m²) (13), given a higher mean proportionate lean muscle mass. However, these modest increases in BMI thresholds would not meaningfully account for the high observed prevalence of overweight and obesity among NH Pacific Islander adults as a whole. A more important point is the contrast to lower BMI thresholds for NH Asian adults due to higher mean proportionate visceral adiposity (7). As NH Asian and NH Pacific Islander adults are frequently aggregated into a single category, this highlights the role of data disaggregation in accurately characterizing adiposity, given the large differences in prevalence of overweight and obesity between NH Asian and NH Pacific Islander adults.

The high prevalence of current smoking among NH American Indian/Alaska Native adults confirms previous evidence (14) and is attributed to factors such as increasing use of commercial tobacco products for ceremonial purposes, target marketing, and low costs of tobacco products on tribal land (14). Despite the high prevalence, heterogeneity in smoking prevalence likely exists among tribal subgroups, due to differences such as historical influences and cultural practices (15). Among disaggregated subgroups, Guamanian/Chamorro and Samoan adults also showed a high prevalence of current smoking. As the existing evidence base among disaggregated subgroups of NH Pacific Islander adults is very limited, it is unclear whether this finding reflects actual elevated prevalence, due to factors such as psychosocial influences and reduced access to smoking cessation programs (16,17), or high statistical variability in the estimates.

Among NH Asian subgroups, Filipino adults showed the highest prevalence of hypertension, a finding reflecting those of prior studies among disaggregated NH Asian subgroups (18). Due to a paucity of studies, there is uncertainty regarding potential explanations, but these may include a greater risk profile from behavioral factors such as smoking, physical inactivity, and high alcohol consumption (18). Furthermore, since hypertension measurement was self-reported, actual prevalence is likely underdiagnosed. The degree of underdiagnosis may differ by aggregated racial and ethnic groups (19), although it is unknown if and how hypertension underdiagnosis may also vary among disaggregated subgroups.

All Hispanic subgroups showed the greatest prevalence of physical inactivity, similar to findings of prior studies, although findings are difficult to compare due to differing measures of physical activity (6,20). Among NH Pacific Islander subgroups, Guamanian/Chamorro and Samoan adults showed the highest prevalence of inactivity. While there was wide variability in estimates, these findings were consistent with analogous findings for the prevalence of overweight or obesity and current smoking and therefore may reflect an overall profile of unhealthy lifestyle risk factors. While prior studies assessing lifestyle risk factors among NH Pacific Islander subgroups are scarce, we can hypothesize as to potential explanations for these findings. For example, Micronesian populations have faced a history of discriminatory policies that have impacted socioeconomic status and increased health care barriers (21). Moreover, the absence of culturally appropriate questionnaire items may have contributed to misclassification (22), although it is unclear how any potential misclassification may differ among subgroups. Additionally, although the definition for physical inactivity represented a low threshold (any leisure-time activity in the past 30 days) compared with recommended guidelines (23), nearly one-quarter of respondents were below the threshold and there was sufficient variability in the distribution of prevalence, highlighting the low frequency of leisure-time physical activity in the overall population.

Potential explanations for the large subgroup disparities in health insurance access among Hispanic adults are multifactorial and may include factors such as differences in qualification for public insurance, country of birth, occupational characteristics, and health literacy (24). For example, Central American adults, likely classified as Other Hispanic adults in the current study, may be more likely to have low-wage occupations that do not provide health insurance (24). Hispanic adults’ participation in the health care system may also be affected by unfair treatment due to language barriers, which may differ among subgroups (24). While findings of previous studies with data sets from two decades ago are consistent with our findings (25,26), prior studies have not included comparison of the scale of subgroup disparities in relation to other aggregated racial and ethnic groups and disaggregated subgroups. As the difference between the lowest and highest prevalence of being uninsured among Hispanic subgroups was larger than the greatest differences between any two aggregated racial or ethnic groups, these results are a pronounced example of how data aggregation can mask large disparities among subgroups.

The prevalence of cost concerns for Mexican adults was nearly one-half of the prevalence of being uninsured, a difference not observed for other Hispanic subgroups. As lack of insurance is typically a major contributor to cost concerns (27), these results suggest that barriers other than lack of insurance may disproportionately impact Mexican adults compared with other Hispanic subgroups. Because there is a lack of empirical research on potential causes of differences in cost concerns among Hispanic subgroups, we can only hypothesize that differences may be attributable to common socioeconomic barriers to health care access such as language barriers, health literacy, and rurality (27,28). In addition, among all NH Asian subgroups the prevalence of having had no physical exam in the past 12 months was comparable with that among all other groups and subgroups, despite NH Asian subgroups having the lowest prevalence of being uninsured and having cost concerns. With few and inconsistent prior similar studies, it is uncertain to what extent this finding may be attributable to other barriers to health care use, such as health literacy, language barriers, and discrimination in the health care setting (29,30). Heterogeneity in this metric among NH Asian subgroups may also be explained by variations in these health care barriers.

Although our findings highlight the utility of collecting disaggregated race and ethnicity data, institutions seldom deviate from minimum reporting standards, which are over two decades old (31). Efforts to facilitate collection of disaggregated race and ethnicity data may include culturally appropriate interventions and assurances of privacy protection to address hesitancy among individuals due to concerns such as mistrust of research, perception of potential discrimination, and lack of understanding of the purpose of data collection (32). Such efforts may facilitate the use of disaggregated data to advance health equity—particularly because racial and ethnic groups facing the greatest health disparities may also be the most hesitant to participate in health research (33). With the increasing use of administrative data (e.g., insurance claims, electronic health records) in health research, buy-in from health care organizations may also mitigate organizational challenges such as a lack of information, limited technology resources, legal concerns, and perceived negative patient reaction (32). Furthermore, more granular disaggregation may also be needed. For example, Middle Eastern/North African individuals are typically aggregated into the NH White category, despite facing potentially large health disparities (34).

Despite these barriers, disaggregated data may advance our understanding of disease etiology, improve research methodology, and increase the effectiveness of primary prevention by appropriately allocating health care resources with culturally appropriate interventions. For example, among Native Hawaiian adults, recognition of culturally relevant physical activity such as hula may more accurately capture physical activity when included in survey questionnaires (35) as well as reduce hypertension when implemented as a clinical intervention (36). Among Puerto Rican adults with diabetes, compared with a control group, modest improvements in glycemic control were observed among adults receiving an intervention adapted to the language dialect, cultural beliefs, dietary patterns, and other factors distinct from other Hispanic subgroups (37). The primary challenges in successfully applying disaggregated data to improvements in population health may be in both the collection of disaggregated data and sustainability of culturally tailored interventions. While both patients and health care systems express sufficient interest in culturally tailored interventions, their long-term sustainability may be limited by insufficient financial and other resources (36,37).

One strength of this study is the large sample size, which allowed reliable estimation of prevalence among disaggregated racial and ethnic subgroups. This study also includes limitations. First, self-reported information may result in misclassification, such as underreporting of overweight or obesity (38). Although laboratory data are available in other nationwide surveys such as the National Health and Nutrition Examination Survey, sample sizes were not sufficient to conduct a disaggregated analysis. Second, temporal changes over the sampling period may have affected prevalence estimates, although adjustment for survey year would likely account for any introduced bias. Third, the questionnaires did not provide disaggregated data for NH Black, NH American Indian/Alaska Native, and NH White adults, who may also show health disparities among disaggregated subgroups (34,39,40). For example, among NH Black adults, African immigrants may have a lower prevalence of diabetes and related risk factors compared with African Americans (40). Similarly, among NH American Indian/Alaska Native adults, there may be a higher prevalence of diabetes risk factors among groups in the Southwestern U.S. compared with Alaska Native adults (39). Fourth, adequate dietary measurements were not available; fruit and vegetable intake survey items were substantially changed during the study period, and the resulting data could not be aggregated. Fifth, the available physical activity measure represented a low threshold for classifying physical inactivity. Lastly, as data were pooled across 9 years, period effects could not be analyzed.

While there is an increasing focus on health equity in public health, assessment regarding race and ethnicity in health studies frequently includes aggregation into broad categories representing heterogeneous groups. Numerous barriers at both organizational and individual levels may need to be addressed before disaggregated data collection becomes standard practice. Given the small number of published studies of diabetes with use of disaggregated data, the field may benefit from additional studies to strengthen the collective evidence. In particular, in additional studies investigators may investigate the specific mechanisms underlying the observed subgroup disparities. With our current limited evidence, potential explanations for subgroup disparities may be speculative rather than evidence based. With further research, to build a stronger evidence base, investigators may then be able to target risk factors specific to each racial and ethnic subgroup—to better shape health care practice and policy and to advance health equity.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.K.K., K.M.B., S.O., F.X., R.S., Y.M., and M.E.P. were involved in the conception, design, and conduct of the study and the interpretation of the results. A.K.K. conducted statistical analysis and wrote the first draft of the manuscript, and all authors edited, reviewed, and approved the final version of the manuscript. A.K.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.