1. INTRODUCTION
Several studies have demonstrated the importance of healthy lifestyles, and primary and secondary preventive measures in reducing or delaying the burden of chronic health conditions, even among the elderly [1-6]. For example, several studies have examined the value of physical activity in preventing heart disease and colon cancer, limiting the effects of chronic diseases such as arthritis, and improving coordination and flexibility to help avoid falls and alleviate depression among older adults [1-4]. There is also evidence concerning smoking cessation, and for utilization of some primary prevention and disease screening services [5,6]. Historically, Medicare did not cover routine preventive services, but beginning with coverage of pneumococcal vaccination in 1981, the program has extended coverage to an increasing range of screening and vaccinations. Most recently, the Medicare Modernization Act of 2003 added a one-time “Welcome to Medicare” physical examination and health risk appraisal, and most vaccinations are covered under Medicare Part D plans. With passage of the Patient Protection and Affordable Care Act in 2010, all preventive services with a grade of A or B by the US Preventive Services Task Force (USPSTF) are available free to Medicare beneficiaries. While Medicare coverage of services is an important step, it is not the only factor to affect behaviors. Furthermore, the challenge for clinicians, program planners, and policy makers is how to encourage Medicare beneficiaries to initiate and maintain desired lifestyle and service use behaviors.
There is considerable published research on the characteristics of Medicare beneficiaries who use covered preventive services as well as other health practices. Influenza vaccination is perhaps the most thoroughly analyzed preventive health behavior among the elderly [7-12], but the research covers a broad spectrum of other health practices as well [13-20]. The general consensus from this body of research is that preventive services are underused by Medicare beneficiaries, particularly among disadvantaged segments of the population [11,16-18].
With rare exception, these studies provide point-in-time estimates and do not consider whether elderly individuals are persistent in their health behaviors over time [12,19, 21,22]. Knowledge of persistence in health behaviors is important for two major reasons. First, the health returns from some behaviors—exercise and smoking avoidance for example—improve with persistence. Persistence in annual vaccinations also appears to increase the immune response to influenza vaccine [23,24]. A second reason is that some screening interventions such as mammography and cervical cancer screening are not recommended on an annual basis for this population (and in the very elderly may not be appropriate at all) [25,26]. Annual surveys thus tend to underestimate the true rate of adherence to guideline receipt of these tests.
This study had two aims: 1) to assess the persistence with which Medicare beneficiaries practice common health promotion and disease screening behaviors; and 2) to determine whether there are common factors that explain why some beneficiaries are more persistent in practicing healthy behaviors than others. We selected 7 common health behaviors for study, 2 measures of healthy lifestyle (routine exercise, smoking avoidance), 1 measure for disease prevention (influenza vaccination), and 4 measures of disease screening (cholesterol testing, mammography, cervical and prostate cancer screening). Persistence in each behavior was tracked over 4 years for a nationally representative sample of community-dwelling Medicare beneficiaries.
2. METHODS
2.1. Data Source and Study Sample
Data for the study were drawn from the Medicare Current Beneficiary Survey (MCBS) Access to Care (ATC) surveys from 1997 through 2005. The MCBS is a nationally representative rotating panel survey of the Medicare population conducted by the Centers for Medicare and Medicaid Services. Approximately 4500 beneficiaries are inducted into the MCBS survey each fall and are followed for up to three additional years. The sample is refreshed with a new induction cohort each year and an equal number are retired from the survey. The ATC survey captures a rich set of data on demographic and socioeconomic characteristics, health status (self-reported health, diseases, functioning), health behaviors, and access to care. Medicare claims with diagnostic codes are supplied with the MCBS files.
To assess behavioral persistence, we selected community-dwelling MCBS respondents who had complete annual ATC surveys and survived their 4-year selection period. This process resulted in 6 cohorts (1997-2000, 1998-2001, etc.) representing 13,913 unique individuals. Because we required complete reporting for each behavior under study, persons who had missing values for behavioral questions of interest were excluded from the analyses of those behaviors. The MCBS initiated questions for two of our selected behaviors (exercise and cholesterol testing) in 2001, which limited the available samples for analyses of these behaviors. The study was approved by the University of Maryland Baltimore Institutional Review Board (IRB).
2.2. Measures
All 7 behaviors studied were measured based on selfreports during the fall ATC survey rounds. Question wording and rules for coding responses are provided in Table 1. Persistence was measured as the proportion of years a respondent indicated they practiced the behavior, with values ranging from 0 (no affirmative response in any year), 0.25 (behavior reported in 1 of the 4 years), through 0.50, 0.75, and 1.0 (behavior reported in all 4 years).
The exercise question in the MCBS is only asked on alternate years, which limited each respondent to 2 responses over the 4 years observation period. We coded persistence in this case as 0 (no activity in the 2 years in which the question was asked), 0.5 (1 year), and 1.0 (both years).
We included a wide array of explanatory variables in each behavioral persistence model to test for common factors expected to predict behavioral persistence. These variables included basic demographics (age, sex, race, living situation based on information about marital status and household composition, and acculturation as indicated by taking the MCBS in a language other than English), basis for program entitlement (under age 65 recipient of Social Security Disability Income (SSDI)), attitudes about seeking medical care, socioeconomic factors (educational attainment, income in relation to the Federal poverty level, presence of supplemental medical coverage), and heath status measures (self-reported health, body mass index, count of limitations in activities of daily living (ADL) [27], and count of comorbidities). The count of comorbidities is based on 189 hierarchical co-existing conditions (HCCs) derived from diagnostic codes in each beneficiary’s annual Medicare claims. The HCC is used by the Centers for Medicare and Medicaid Services (CMS) to risk adjust capitation payments to Medicare Advantage plans, and is widely used as a comorbidity index in studies of the Medicare population [28-30]. Because several of the behaviors evaluated are associated with specific clinical indications either for or against their use, we included 17 self-reported disease and function indicators relevant to specific health behaviors. These behavior specific measures are listed in Table 1.
Table 1. Measurement of persistence in healthy behaviors using MCBS panel data.
We estimated behavior-specific models using 4-year proportional persistence measures as dependent variables and the explanatory variables measured at baseline (year 1). Each model was estimated using both a partial proportional odds model (PPOM) [31] and an ordinary least squares (OLS) linear probability model. The PPOM models used a generalized estimating equation algorithm to determine if the impact of a particular covariate was equivalent across the range of possible outcomes (e.g., whether the impact of being female was the same for individuals with 0 to 0.25 persistence in influenza vaccination compared to those with 0.75 to 1.0 persistence in this behavior). In all but a few instances, the PPOM models output consistent effects for each covariate across the spectrum of behavioral persistence, thus indicating linearity of response. The alternative OLS estimator presumes linearity of response and is preferred on efficiency grounds. We report only the OLS results here.
3. RESULTS
Table 2 reports the percentage of beneficiaries who reported each behavior from 0 to 4 years. For the 3 health promotion and disease prevention behaviors, persistence was highest in smoking avoidance (81% reported not smoking at all, 11% always smoked, and 8% smoked at some point), followed by influenza vaccination (52% reported it each year), and lowest for exercise (only 24% reported participating in weekly exercise in both years in which they were asked the question). There also was wide variation in persistence in disease screening and monitoring. Two-thirds of the sample reported cholesterol testing each year. Slightly more than 42% of women reported having 3 (19%) or 4 (23%) mammograms over the 4-year period. Fewer women (26%) reported having at least 3 Pap smears over the 4 years, but 42% reported having pap smears in 2 or more years. More than half of all men (53%) reported a prostate cancer screen in 3 or 4 of the