All data used for this study were openly, publically available . Raw data were extracted from datasets from the National Health and Nutrition Examination Survey, 2007-2008 (NHANES 2007-2008) . This survey contained data for 10,149 individuals of all ages. Data were collected between January 2007 and December 2008 using a complex, multistage, probability sample design. The data from NHANES consist of 2-year representative samples of the non-institutionalized, civilian U.S. population. All participants read, understand and sign informed consent forms based on the approved study protocol from the National Center for Health Statistics Research Ethics Review Board . The study was limited to male and female adults ages ≥ 21 years who provided self-report diagnosis of diabetes or who had hemoglobin A1c (A1C) ≥ 6.5, underwent an examination at the Mobile Examination Center (MEC), and belonged to one of the following race/ethnicity classifications: Black non-Hispanics (BNH), Mexican American (MA), and White non-Hispanics (WNH). Of the total sample size for the 2007-2008 participants, there were 2,064 MA, 1,147 other Hispanics, 2,141 BNH, 3969 WNH and 441 persons classified as "other". From the combined sample, there were 777 persons (7.7%) of the 9372 valid cases who responded to the screening question for diabetes diagnosis. The categories for "other Hispanics" and "other races" were 10.8% and 2.7%, respectively and did not have a sufficient sample sizes for comparative analyses. Approximately 2% (n = 14) of the subjects were minors (< 18 years). The final sample size of participants who met the inclusion criteria for the main study question was [N = 654 (130 MA, 224 BNH and 300 WNH)]. There were n = 161 classified as 'undiagnosed diabetes'. The final sample size, including individuals with 'undiagnosed diabetes', was [N = 815(171 MA, 281 BNH and 364 WNH)].
Several questions were considered for the construction of the variable 'diagnosed diabetes': 1) "The next questions are about specific medical conditions. Other than during preganncy, have you ever been told by a doctor or health professional that you had diabetes or sugar diabetes?"; 2) "How old were you when a doctor or other health professional first told you that you had diabetes or sugar diabetes?"; and 3) "When was your diabetes diagnosed?"
Each question was by self-report and subject to recall-bias. Since there were missing values for questions 2 and 3, construction of the variable, 'diagnosed diabetes' was based on question 1.
Both fasting blood glucose (FBG) and hemoglobin A1C (A1C) were initially considered for construction of the 'undiagnosed diabetes' variable. The latter was chosen, since there were approximately twice as many missing values for FBG as compared to A1C. The inclusion criteria were the same as for 'diagnosed diabetes' with the exception that both conditions were met: 1) 'no self-report of being diagnosed with diabetes'; and, 2) percent A1C ≥ 6.5.
The ecological model suggests that changes in knowledge, skills and attitudes will change behaviors and consequently health outcomes. Behaviors that are hypothesized to impact DSM health outcomes were measured in the NHANES  and in the applied DSM model using the 'current behaviors' (dietary plans, weight management, and physical activity changes). Patient provider communication is an assumption of the model and it was not measured in this study.
Measurements using the ecological model
Medical advice (on diet, weight management and physical activity and 'told by a doctor or health provider in the past 12 months'), fits into the ecological model as an environmental influence which acts on the individual and helps to promote health behavior change. Influences were measured in NHANES  and in the applied model (Figure 1) using 1) 'medical advice' (reduction of fat or calories, control or loss of weight, increased physical activity or exercise); 2) 'diabetes education' (enhancing skills related to self-care); and, 3) 'race/ethnicity' (culture/health disparities). The NHANES database  measures reported 'medical advice' without making a distinction among the healthcare provider (the response choice for these questions were 'told by a doctor or health provider'). The response for diabetes education was 'When was the last time you saw a diabetes nurse educator or dietitian or nutritionist. Do not include doctors or other health professionals'.
'Health Outcomes' measured in this study were the changes in behavior related to diabetes self-management (reporting reducing fat or calories in the diet, increasing physical activity or exercise and controlling or losing weight) (Figure 1). Questions on medical advice were used for the first time by NHANES  and correspond to recommendations from the Global Guideline for Type 2 Diabetes, lifestyle management standard care . Medical advice questions examined in this study all were phrased with the standard language 'to lower your risk for certain diseases, during the past 12 months have you ever been told by a doctor or health professional to' and the advice added for each question were 1) reduce the amount of fat or calories in your diet; 2) increase physical activity or exercise; and, 3) control weight or lose weight. The health behavior questions were phrased 'To lower your risk for certain diseases, are you now doing any of the following' and behavior added for each question were: 1) reducing the amount of fat or calories in your diet; 2) increasing your physical activity or exercise; and, 3) controlling weight or losing weight. A variable for recent diabetes education was created with two categories (two years or less and more than 2 years or never) based on responses from the following 'When was the last time you saw a diabetes nurse educator or dietitian or nutritionist for your diabetes? Do not include doctors or other health professionals'.
Logistic regression models were performed with reporting receipt of medical advice within the past 12 months by race/ethnicity for each outcome of diabetes self-management behavior. Specifically, the independent variable for model 1 was binary responses to 'told to reduce fat or calories' and the dependent variable was 'currently reducing fat or calories'. Similarly, model 2 assessed the likelihood of 'told to increase physical activity' with 'currently increasing physical activity' and model 3 contained 'told to control or reduce weight' as the independent variable and 'currently controlling or reducing weight' as the dependent variable. Full models contained age, gender, education, health insurance, overweight/obese, and diabetes education.
Additional analysis dependent variable: obesity
A binary variable was constructed using a proxy measure of obesity, body mass index (BMI). Classification was ≥ 30 for 'obese' and < 30 'not obese' with measurements of weight in kg divided by height in m2 (kg/m2). Values used for BMI for this study were calculated by direct measurements taken in the MEC by NHANES. Next, quartiles of BMI were compared with each medical advice and behavior by the Chi Square Test.
In this study, we assessed the following.
1. Whether the effect of medical advice on the likelihood of performing the corresponding recommendations differs by race/ethnicity.
2. Whether individuals having 'undiagnosed diabetes' (no self-report of being diagnosed with diabetes and a percent hemoglobin A1C ≥ 6.5) versus 'diagnosed diabetes' (self-report of being diagnosed by a doctor or health professional) would differ in the relationship between given medical advice (yes/no) and its effect on the corresponding behaviors.
3. Will the effect of reporting being told to control or loss weight by reporting performing the behavior (controlling or losing weight) be associated with obesity.
Sample weights were constructed and included in the data sets to account for complex sample design and achieve unbiased national estimates. The choice of sample weight was based on the data file with the smallest sample size as recommended by the National Center for Health Statistics (NCHS) guidelines . Data analysis was conducted with IBM-SPSS version 18 with a complex sampling add-on. Prior to analysis, continuous variables were assessed for normality by Q-Q plots and when needed, transformed. Post-analysis, continuous variables were tested by residual graphs for skew. Hierarchical logistic regression models were conducted for medical advice by race/ethnicity predicting adequate/inadequate DSM adding variables associated by the literature as covariates. We examined 3 different types of medical advice, therefore a Bonferroni correction was used to ensure an overall error rate of 0.05 and p < 0.017 was considered significant. Models were estimated with and without covariates. Covariates considered included age, gender, health insurance, diabetes education and education. In addition, obesity was added for 'told to reduce fat or calories' and 'told to increase physical activity; whereas overweight and obesity was included for 'told to control or loss weight'. The final models retain covariates with p-values of ≤ 0.2 for diagnosed diabetes. Undiagnosed diabetes models were adjusted for health insurance, age, gender, and body mass index categories. The Wald F statistic was used to determine model significance for complex logistic regression analysis  where the degrees of freedom are constrained to a constant value (the primary sampling units minus the strata which equal 17 for this data). Results are only presented for models which significantly predicted the outcomes (values available upon request). Odds Ratio (OR) and 95% confidence limits are presented for the model without covariates and adjusted odds ratio (AOR) and 95% confidence limits are presented for the models with covariates. Where there was information missing, list-wise deletion was used and the number for each analysis was provided.