TRIAD is a study of diabetes care in managed care, involving 10 health plans in 6 metropolitan areas across the United States.22
These health plans include for-profit, not-for-profit, Medicare, and Medicaid providers. The current analysis used data from a TRIAD questionnaire administered between March and September 2006 to “case” and “control” diabetic patients identified from diagnostic claims and/or the medical record within 8 of the plans. Cases were sampled from the population with recent poor control of at least 2 of the 3 intermediate outcomes (HbA1c ≥8.0%, SBP ≥140 mmHg, LDL cholesterol ≥130 mg/dl), and controls were sampled from the population with 1) established hypertension and hyperlipidemia from diagnostic claims or the medical record and 2) good control of all 3 outcomes (HbA1c <8.0%, SBP <140 mmHg, and LDL cholesterol <130 mg/dl). We used the most recent intermediate outcomes recorded in the 12 months prior to the survey to define case or control status.
Patients were excluded from the sampling frame if they had a gap in plan enrollment of >45 days during either the 12-month study window or the preceding 12 months. Enrollees were also excluded if they did not have all 3 intermediate outcomes (HbA1c, SBP, LDL cholesterol) measured within the study window, were unable to speak English or Spanish, or were ≤18 years of age.
TRIAD investigators attempted to
contact 2609 persons to participate in the study. Of the 1615 persons who were successfully contacted, 1305 (80%) were eligible. Of these contacted, eligible persons, 1139 (87%) completed the survey. If persons who could not be contacted had the same rate of eligibility (80%)
as those who were contacted, and if they were counted in the denominator, the survey response rate would be 54% (Council of American Survey Research Organizations [CASRO] response rate).23
All study variables were drawn from the participant surveys, with the exception of intermediate outcomes and body mass index obtained from medical record review. We included participants in the analytic sample if they identified their race as either “white” or “African American,” even if they indicated other racial backgrounds as well. Participants who indicated their race as African American were classified as black, even if they also reported a white racial background. We excluded participants (n=375) with 1) Latino ethnicity, 2) only nonwhite/nonblack race, or 3) missing data for race. We identified mutable risk factors as potential “exposures” that may be associated with poor control of intermediate outcomes, including incomplete medication adherence, perceived poor quality of provider communication, depression, low self-efficacy for reducing cardiovascular risk, and low health literacy.
We evaluated 2 measures of adherence to medication over the previous 6 months: running out of any medications or missing any medication doses (Appendix
). Patients who indicated that they missed medication doses were asked follow-up questions to examine the underlying reasons. We modified 4 published scales to measure patients’ perceptions of the quality of provider communication (Appendix
Values for Cronbach’s α ranged from 0.59 to 0.81 for these modified scales. We defined depression as a score of ≥10 on the PHQ-8.25
We defined low self-efficacy for reducing cardiovascular risk as the acknowledgement that one was at high risk for heart disease, together with the belief that this risk could not be significantly lowered (Appendix
). Finally, we created a summary score of 4 individual health literacy items.26
The responses for each original item ranged from 0 to 4, with higher values corresponding to lower health literacy. In constructing our score, we assigned a point for each item with a response greater than 0. We also examined demographic characteristics, including age, sex, education, income, and body mass index (BMI).
We investigated the unadjusted distributions of demographic characteristics by race and case or control status. Using SAS’s PROC GLIMMIX Version 9.1.3 (SAS Institute, Cary, NC), we conducted separate unadjusted analyses and multivariate random effect models to examine associations between the presence of each risk factor and case or control status. In each unadjusted and adjusted model, we used interaction terms between risk factors and patient race to predict case and control status separately for black cases, black controls, white cases, and white controls. Each of the nine multivariate analyses controlled for fixed demographic characteristics, specifically age, sex, education, income, and study site. We used the 25th and 75th percentiles of the quality of provider communication scales to represent poor and good communication, respectively. Low health literacy was identified infrequently within our sample, and we were unable to fit a multivariate model for this variable.
Finally, we examined the unadjusted distributions of incomplete medication adherence among the subgroups of blacks and whites that missed medication doses. We considered results significant if P < 0.05. The Centers for Disease Control and Prevention, Division of Diabetes Translation, and the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, funded this study. Institutional review boards at all participating sites approved the study.