We used data from the 2004 wave of the Health and Retirement Study (HRS) as well as data from the 2005 HRS Prescription Drug Study (PDS). The HRS is a longitudinal study of a nationally representative cohort of older Americans that was designed to assess the predictors and consequences of transitions out of the workforce in later life, and it includes detailed questions about participants' health insurance including prescription drug coverage . The PDS is a subsample of the HRS drawn from respondents who participated in the HRS in 2004. It was designed to help track potential changes in prescription medication utilization among beneficiaries as Medicare Part D was phased in (a second wave of PDS data was collected in 2007 and is not yet available). To be eligible for inclusion in the present study, respondents needed to be aged 65 or older in 2005 (e.g., age eligible for Medicare when the PDS data were collected), and be self-respondents (i.e., they were able to provide responses without a proxy respondent). Approximately 40% of the 2004 HRS sample was approached for possible inclusion in the PDS study, which had a response rate of 88%. Out of the 4684 people who completed the PDS, 3997 were aged 65+, 3394 had data on CRN, and 3071 responded without a proxy. Therefore, our final study sample size was 3071.
The HRS and PDS provide excellent data for testing our conceptual model because they include indicators of several of the posited domains, including indicators of financial pressures for CRN (income, insurance coverage, and number of prescriptions), as well as nonfinancial factors that could mitigate or exacerbate patients' risk of CRN: patient sociodemographic characteristics, health status indicators, and medication characteristics. Less information is included in these datasets regarding other domains described in the conceptual model such as clinician counseling and health system characteristics [6,18]. Variables from the PDS included: CRN, OOP medication costs, drug coverage, adverse medication effects, number of monthly prescriptions, and age. All other variables were taken from the HRS. These variables are included in a modification of Piette et al.'s  conceptual model in Figure 1.
Figure 1. Conceptual framework for the factors influencing patients' risk of cost-related nonadherence (adapted from Piette et al. ). ADL, activities of daily living; IADL, instrumental activities of daily living.
Download figure to PowerPoint
All analyses were weighted and adjusted for HRS' complex sampling design (stratification and clustering) and used SAS 9.2 (SAS Institute, Cary, NC). The Institutional Review Board of the University of Michigan Medical School approved this research project and it received exempt status.
We chose independent variables based on data from the HRS and PDS that were in accordance with the domains of our conceptual model:
Measures of financial pressures included out-of-pocket expenditures (measured in quartiles: $0 to $20, $20.01 to $50, $50.01 to $110, >$110) for a month supply of “regular” drugs, other OOP medical expenses from the previous 2-year period (measured in quartiles: $0 to $580, $580.01 to $1792.50, $1792.51 to $4570, >$4570), net worth (measured in quartiles: $0 to $38,000, $38,000.01 to $154,500, $154,500.01 to $425,000, >$425,000), annual household income (measured in quartiles: $0 to $14,042.11, $14,042.12 to $25,660, $25,660.01 to $48,384, >$48,384), and any drug coverage (yes or no and including employer, private purchase, Medicaid, VA, Medicare HMO or Medicare + Choice plan, or state pharmacy assistance program).
Demographic characteristics included age (65–74, 74–85, 85+), education (high school graduate or less, at least some college), sex, employment status (working, not working, retired), current marital status (married, separated/divorced, widowed, never married), and race/ethnicity (white, black, Hispanic, other).
Patients' burden of chronic illness was measured using indicators for each of eight chronic medical conditions: 1) high blood pressure or hypertension; 2) diabetes or high blood sugar; 3) cancer or a malignant tumor of any kind except skin cancer; 4) chronic lung disease except asthma such as chronic bronchitis or emphysema; 5) heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems; 6) stroke or transient ischemic attack; 7) emotional, nervous, or psychiatric problems; and 8) arthritis or rheumatism. Other health-related characteristics included patients' perception of their overall health (excellent vs. very good, good, fair, or poor), limitations in each of five activities of daily living (eating, getting in and out of bed, dressing, bathing, and walking across a room, measured as 0 vs. ≥1) and five instrumental activities of daily living (preparing meals, grocery shopping, making phone calls, taking medications, and managing money measured as 0 vs. ≥1) .
To determine depressive symptoms, each respondent was asked the following eight depressive symptoms questions taken from the Center for Epidemiologic Studies Depression (CES-D) scale [27,28] (with response options of “yes” or “no”): 1) Much of the time during the past week, I felt depressed; 2) I felt everything I did was an effort; 3) My sleep was restless; 4) I was happy; 5) I felt lonely; 6) I enjoyed life; 7) I felt sad; and 8) I could not “get going.” The total number of “yes” responses to questions 1, 2, 3, 5, 7, 8, and the “no” responses to questions 4 and 6 were summed to arrive at a total depressive symptom score that ranged from 0 to 8, which has been shown to have a Cronbach's alpha of >0.8 in the HRS data . In our multivariable models, we used a three-level aggregated measure of CES-D scores of 0 (no symptoms), 1–3 (depressive symptoms), and ≥4 (depressed). This cut point of ≥4 has been found to produce comparable results to the 16-symptom cutoff for the well-validated 20-item CES-D scale .
Prescription regimen characteristics included the number of different prescriptions a respondent reported using in the last month (0–2, 3–4, 5–6, ≥7). Patients also reported whether they experienced any adverse effects associated with medication use (yes or no).
An additional variable was used from the clinician factors domain from the theoretical model. This question asked respondents who they trust to make decisions about health insurance and response choices included family members (spouse, child, other), friends, financial advisors, as well as doctors, nurses or other health-care providers (measured as no one, family/friend, or professional).