The Translating Research into Action for Diabetes (TRIAD) study is a multi-center study of diabetes care in managed care settings . Within a subset of health plans participating in TRIAD, a cross-sectional survey was conducted from April-October 2007 to examine the implementation of the Medicare Part D drug benefit among beneficiaries with diabetes. Potential survey respondents were enrolled in one of three types of health insurance products in 2006: 1) for-profit Medicare Advantage Prescription Drug (MAPD) plans within eight states, 2) stand-alone, for-profit Medicare Prescription Drug Plans (PDPs) in the same eight states, and 3) an MAPD product offered by a large, integrated delivery system model HMO in California (IDS MAPD). The multi-state plan is a network-model system offering two different Part D benefit designs. One design had a standard coverage gap between $2,250 in total drug costs and $3,600 in out-of-pocket drug costs, and the other provided generic-only medication coverage during this gap. Beneficiaries could have either of the cost-sharing designs through an MAPD plan or through a stand-alone PDP plan. During the coverage gap, beneficiaries that were enrolled in a plan with generic-only coverage continued to have an $8.50 generic copayment but no coverage for brand name drugs, while those with a complete gap in coverage paid full cost for all drugs including generics. Patients in the IDS-MAPD may have had supplemental gap coverage.
To be eligible for the survey, patients were required to have been continuously enrolled in one of the two MAPD plans from 1/01/05 to 12/31/06, or newly enrolled in the PDP plan between 11/15/05 and 3/01/06, and must have hit the Part D coverage gap in expenditures (i.e. had total 2006 drug expenditures that reached $2,250) by October 1, 2006. Previous work suggests that more than 25% of diabetes patients enter the coverage gap annually . Among eligible patients who were beneficiaries in the participating plans, the survey randomly sampled members who were at least 65 years old, spoke English or Spanish. Drug expenditures were obtained using claims data. Patients who could not provide informed consent or were too ill to participate were excluded. Beneficiaries who were low-income subsidy (LICS) qualifiers were also excluded because their Part D benefit does not include a coverage gap. Potential participants were sent a $10 gift card as an incentive, and offered the option of completing a computer-assisted telephone interview or a written survey.
The survey response rate was 58%. Respondents were not significantly different from non-respondents in terms of gender, number of medications, or geocoded census-track income, and differed less than one year in mean age (data not shown); data are unweighted and did not adjust for non-response.
Survey participants were asked if, during 2006, they 1) thought the issue of prescription drug cost was important enough to raise with their doctor; 2) wanted their doctor to consider the cost to them when choosing medication; and 3) talked with any doctor about the amount they had to pay for prescription drugs. Responses to the first question was given on a 4-point Likert scale, and dichotomized for analysis into 'strongly agree/agree' vs. 'disagree/strongly disagree/don't know,' while responses for the last two questions were given as 'yes/no'. Patients were also asked if their doctor switched any prescriptions to a less expensive medication because of cost, or if they had used any medication less often than the doctor prescribed due to the amount they had to pay in 2006 (responses given as 'yes/no').
Multiple logistic regression models were used to create adjusted percents of patients across demographic and health plan characteristics responding 'strongly agree/agree' or 'yes' to the above questions (dependent variables), adjusting for patient demographics, self-rated health status, and a comorbidity score based on a simple sum of 14 self-reported health conditions obtained through the survey (independent variables). Adjusted percents are calculated by setting all characteristics except the variable of interest to the mean, using the coefficients from the model as multipliers. Since the average number of patients per prescribing physician was very low (mean of 1.5 patients per prescribing physician), models did not adjust for patient clustering within physician. Models also adjusted for month entering the gap, number of unique medications taken during 2006, percentage of medications in the first quarter that were generic, and difference between total and out-of-pocket medication costs in the first quarter (obtained through claims data). Analyses were performed using SAS v9.2.
This study was developed and approved by the Steering Committee of the Translating Research in Action for Diabetes (TRIAD) Study and conducted by researchers in two of TRIAD's Translational Research Centers, and approved by the appropriate Institutional Review Boards.