This is the first large study of the relationship between educational attainment and multiple health behaviors among fully insured patients enrolled in a managed care setting. We found that even in managed care settings with uniform access to and quality of diabetes care across levels of SES , less education was associated with poorer health care behaviors among particularly high risk patients for whom these behaviors would be unquestionably indicated and recommended by their health care providers.
Alternative SES indicators yielded similar, although not always significant, findings. In general, education and income more strongly predicted health behaviors than social class or parental education. Each indicator has characteristics which influence its association with each outcome. Income varies over time and is likely to be influenced by adult health status, since progressing disease severity may reduce income. Educational attainment, on the other hand, is generally more stable over the life course, and less likely to be influenced by diabetes health status, since diabetes related complications typically occur later in the life course. Moreover, since education is a strong predictor of subsequent income, income may be considered as on the causal pathway between education and behaviors. Parental education may be less predictive of adult behavior given that childhood social exposures may be overshadowed by later life experiences and opportunities for social advancement. The Olin Wright social classification was only weakly associated with health behaviors in this study. This measure of social stratification is questionable among homemakers and the unemployed; furthermore, class may be a less pertinent social determinant of health in the United States than in Europe.
The observed relationships further our understanding of the poorer health outcomes persisting in lower SES patients with diabetes in managed care settings. The strong SES gradients in smoking and physical activity are consistent with other population-based studies including diabetic patients, although the SES indicators varied across studies [28, 29]. A population-based US study (Americans' Changing Lives Survey) reported that poor behavioral patterns failed to fully explained the greater all-cause mortality rates in lower SES individuals . Although differences in model specification make the findings difficult to compare, a study of individuals from Finland, a country with socialized medicine, showed that the association between SES and mortality was practically eliminated by adjustment for health behaviors, biologic factors, and psychosocial factors . Subjects with widely varying degrees of health coverage, access and quality are included in population-based studies in the U.S., while, like countries with socialized medicine, subjects in managed care settings such as in TRIAD, have relatively uniform access to care.
Mirowsky and Ross's theory of human capital  suggests that education improves health by allowing people to develop healthy lifestyles and prosperity (i.e., economic resources to buffer against illness and want). We found evidence that education has an independent effect on health behaviors in models adjusted for income. Moreover, education was predictive of a count of healthy behaviors, giving an indication of the degree of congruence of these health behaviors. Mirowsky and Ross's second hypothesis, the theory of personal control, suggests that education facilitates a sense of control over one's life, encouraging one to seek information that may improve health . We found that education was strongly predictive of health-seeking activities in diabetes, and this was independent of income. The third theory suggests that educated parents practice and transmit a healthy lifestyle to their offspring. We found that parental education was modestly predictive of health-seeking behavior and regular exercise, although less strongly than SES indicators from mid- and later adult life.
Some study limitations and strengths should be considered. Our study analyses are based on self-reported education and health behaviors, and may be subject to some misclassification due to social desirability of some response options. The cross-sectional design precludes causal inferences. Although we found no significant interactions, the nonlinearity of logistic regression models limits our ability to detect significant interactions . The Olin Wright classification applies only to currently employed subjects and led to exclusion of approximately 200 unemployed subjects who likely had lower SES. By creating indicators and proxies for childhood (parental education), mid-life (educational attainment) and late-life (income) socioeconomic status on health outcomes, we were able to consider a "life course approach ." While current health behaviors may reflect early SES influences over the life course and may have an accumulated impact , we find that health behaviors are more strongly associated with proximate social indicators than those from early life. Unfortunately, disentangling the separate and propagated effects of early-, mid- and late-life SES was not possible . Additionally, we were unable to address potential mediators (particularly health literacy [36–38]) that may explain observed social differences in behavior. Our sample came from the managed care populations participating in the 6-state TRIAD study. Findings may not apply to uninsured subjects given they would not be included in this sample. Thus our sample of subjects was not randomly selected from all people with diabetes in the United States, and our health plans may not represent the larger population of managed care settings. Our sample only included small numbers of subjects whose primary language was Spanish. It is important to note that our findings were observed in samples selected so that each examined health care behavior was unquestionably indicated as a beneficial, preventive measure (e.g., daily foot examination in diabetic patients with peripheral neuropathy, or a history of ulcers or amputations).