We performed a descriptive cross-sectional secondary analysis of data from the 2002 STEPwise approach to surveillance (STEPS) in Pohnpei [21]. Detailed STEPS design and methodology are available at: http://www.who.int/chp/steps/en/. The STEPS dataset, developed and administered by the World Health Organization (WHO), includes a sequential three-step process to collect information on risk factors for chronic diseases: 1) a personal interview, 2) anthropometric measurements, and 3) a biochemical assessment.

Prior to accessing the dataset, we completed an assessment of it for completeness and quality (Additional file 1) using established secondary analysis assessment criteria [22–24] and publicly available STEPS metadata archives. Generally, the assessment showed that the dataset was appropriate for secondary analysis with adequate power, a representative sample, validated instrumentation, standardized data collection, and limited issues regarding missing data. Thus, we were able to use the dataset to provide reliable data necessary for the primary objective of this study.

The 2002 STEPS sampling process used a multistage, probabilistic, cluster design (based on 2000 Pohnpei census enumeration areas) to randomly select households for participation. Data were obtained from 1638 adults (78% response rate) aged 25–64 years. Sample size calculations and details related to sampling procedures used in collecting STEPS data for the primary study are reported elsewhere [21]. We included the entire 2002 STEPS Pohnpei dataset to ensure that the study sample was representative of the overall population.

Before we downloaded data, the Uniformed Services University Institutional Review Board and the FSM Department of Health and Social Affairs approved this study.

Table 1 provides a summary of conceptual definitions and measures for the variables used in our study analyses. A brief description of outcome and predictor variables is provided below.Table 1

Summary of variable names, conceptual definition, and measures using the 2002 STEPS Pohnpei, FSM data set

Educational attainment is a preferred socioeconomic position indicator, as education can be determined for all individuals and is fairly stable after early adulthood. Additionally, education is a significant determinant of a person’s economic potential and acquisition of life-skills relative to adopting health-promoting behaviors [15]. Using self-report of years of education completed, we categorized participants into one of three educational-level groups: primary (<9 years), secondary (9–12 years), and postsecondary (≥13 years). Persons with a primary education were used as the reference group in the analysis.

We included estimated annual household income in the analysis, categorizing participants into one of three groups: low (<$5,000), middle ($5,000–$10,000), or high income (>$10,000). For the analysis, persons in a household with low income < $5,000 were used as the reference group. An “unknown income” category was included to account for missing values for the income field in the primary dataset [27].

For employment status, we categorized participants into one of three groups: paid, unpaid (i.e., retiree, volunteer, student, homemaker), and unemployed. For the analysis, persons who were unemployed were used as the reference group.

We included sex and age in the analysis as covariates. The literature supports sex and age possibly being mediating or confounding factors in the association between socioeconomic position and CVD risk factors. Age and sex may affect risk factor prevalence through variations in both biological characteristics and societal influences that vary across societies [15, 28].

Data were analyzed using SPSS version 20.0 complex samples module that accounts for the complex sampling design used in the STEPS survey, correctly calculating standard errors with weighted data. We applied sex-age structure survey weights (standardized to the FSM 2000 census for Pohnpei) to provide results representative of the adult Pohnpeian population aged 25–64 years. After data cleaning and recoding, we completed descriptive statistics for all study variables. We also completed exploratory analyses, using chi-square and one-way analysis of variance (with tests of normality and other assumptions and post-hoc comparisons) reported elsewhere [29]. Through this process, we identified the associations of socioeconomic position and demographic characteristics with outcome variables (using a priori probability value of <0.25, or had clinical relevance) for inclusion in the multivariate logistic regression models that this paper addresses.

We used multivariate logistic regression models to determine the extent to which socioeconomic position predicted CVD risk factors. First, we examined the crude associations between primary socioeconomic predictors (i.e., education, income, and employment) and each outcome measure. We also examined all three predictors and sex and age included as covariates. A priori, we also developed strata models based on evidence that sex and age may be confounding factors in the association between socioeconomic position and CVD risk factors [15, 28]. Magnitude of effect was calculated using the simple difference between crude and adjusted odds ratio estimates with values greater than 10% considered meaningful. Results were considered significant if p was less than 0.05 (p < 0.05).

Worldwide, as of 2006, smoking prevalence was higher for men (40%) than for women (nearly 9%), and men accounted for 80% of all smokers [30]. Our findings were consistent with a recent review of data from 48 LMICs participating in the World Health Survey, which revealed higher smoking rates among the less educated, decreasing proportions of smoking among persons in older age groups, and inverse associations between wealth and smoking within LMICs [31]. In contrast, our analysis found that, when controlling for other factors, women reporting paid employment and men reporting unpaid employment had twice the odds of daily tobacco use than those reporting unemployment.

Although other studies have assessed a variety of physical activity domains, in general our findings of higher likelihood of low physical activity for women, increasing age group, and those reporting primary education are consistent with other studies in LMICs. For example, an assessment of cross-national data from Estonia, Latvia, and Lithuania revealed that lower educational level is a strong and consistent predictor of leisure time inactivity across gender [32]. In most countries, men are more active than women [33, 34], and increasing age is associated with lower physical activity prevalence, especially in men [34].

In our study’s general population a higher income and paid employment status were associated with higher odds of high BP and central obesity, respectively. We also observed differences across gender. For example, men had twice the odds of high BP than women and women with incomes > $10,000 had more than twice the odds of high BP than those reporting incomes < $5,000 though not for men. We also found that women had twice the odds of central obesity than men. For men, paid employment was associated with three times the odds of central obesity. Some other studies in LMICs have reported different results for these associations. For example, in rural Vietnamese adults, aged 25–64 years, hypertension was inversely associated with education level [35]. Another study among urban Latin American adults aged ≥18 years found inverse associations for income and education with hypertension [36]. Additionally, in a landmark review (1989) on the relationship between socioeconomic position and obesity, researchers reported a strong inverse association for women in developed countries, with a higher likelihood of obesity among women in lower socioeconomic groups [37]. In 2007, McLaren [38] updated this review, reporting a gradual reversal of this social gradient (for both men and women), with an increasing proportion of positive associations and a decreasing proportion of negative associations when shifting from high-income countries to medium- and low- income countries.

While evidence is limited, the varied patterning among socioeconomic position and CVD risk factors in our study may suggest a gradual shift in the epidemiologic transition within Pohnpei. For many high-income countries, thru the progression of socioeconomic development, researchers have documented an epidemiological transition; from a direct to an inverse association between socioeconomic position and CVD risk factors [15]. Rising rates of mean BP has been identified as an early indicator of this transition [6]. Comparisons with previous research, although limited, suggest a rise in mean BP among adults living in Pohnpei. For example, seminal cross-sectional studies in 1947 and 1953 observed low BP among adults in Pohnpei and, in 1983, researchers reported significantly increased diastolic BP among urban males [39]. While our findings suggest a continued increase in mean BP among adults living in Pohnpei, particularly for those reporting higher income.

Researchers suggest that the epidemiologic transition occurs because wealthier and more educated persons tend to be early adopters of high-risk behaviors that contribute to CVD [40, 41]. As health-related consequences are realized, those in higher socioeconomic position embrace lifestyle change and other prevention measures, subsequently lowering their CVD risk. In contrast, the uneducated poor may experience later peaks in higher risk lifestyle behaviors—such as less physical activity, high-fat diets, or increased psychosocial stress—that lead to increased CVD after declines are seen among the wealthier [40]. Unlike high-income countries, the challenge for LMIC countries is that this transition may occur more rapidly with the changes in risk factors and CVD prevalence outpacing the development of policy, environmental and system changes needed to effectively target CVD reduction [6].

In 2011, a political declaration from the United Nations high-level meeting on chronic noncommunicable diseases recommended stronger country-level surveillance and monitoring of chronic diseases and associated risk factors, including socioeconomic determinants, to appropriately target public health policy and programmatic needs [42]. Although repeated surveys will provide trend data for CVD risk factors, our study results demonstrate how population health surveillance data can provide context-specific strategies to address CVD risk factors in that population, in this case that of Pohnpei. Over the last several years, the FSM Department of Health and Social Affairs has been building capacity to prevent and control chronic disease [43] and developing policy and environmental strategies to promote physical activity, consumption of local fruit and vegetables, and tobacco-free living, in partnership with governmental and community networks [43–45]. Possible interventions to address CVD risk factors within Pohnpei include:

 Work-site wellness programs that provide access to and support for tobacco cessation and obesity prevention among adults.

While the 2002 Pohnpei STEPS population health survey establishes baseline data, monitoring time trends can provide a consistent, up-to-date, and standardized database to support population health research, policy, and program development. While our analysis revealed several significant predictors for CVD risk factors, the strengths of the associations were small, suggesting that other determinants, not examined in our study, may play important roles in the health of the population. Further research to assess the behavioral, biologic, and environmental impact of socioeconomic position on CVD risk-factor clustering may also help increase understanding and inform population health efforts to reduce the burden of CVD disease [15, 46–48] in the people of Pohnpei.

The strengths of this study include engagement of the FSM Department of Health and Social Affairs leadership, a representative sample, and objective anthropometric measures. We also acknowledge the study limitations. Inherent methodological biases in our study model limit the generalizability of findings beyond Pohnpei and other subpopulations in FSM. For example, because the STEPS dataset did not provide probability variables for sample selection, we assigned standardized age-sex rates, which may not have been representative of the 2002 adult Pohnpeian population.

Additionally, we used a cross-sectional sample for this study, which does not allow examination of causal relationships. The study also includes use of self-reported behavioral risk-factor measures that are subject to participant recall, social desirability, and response bias. The limited data collection period, within the primary study, may have also introduced seasonal variation in responses. These potential biases could contribute to under- or over-reporting of risk [49].