Lisa Hutchinson, PharmD, MPH, BCPS, FCCP

The Medicare Part D program was implemented to provide the elderly with access to prescription medications and to protect them from high future out-of-pocket (OOP) prescription costs.

Part D coverage was delivered through 2 types of plans: Medicare Advantage plans with prescription drug coverage (MA-PD) or Medicare prescription drug plans (PDP). The program was uniquely designed so that participation was optional and enrollees could switch Part D plans to adjust to changes in their prescription needs.

 The result was that elderly Part D enrollees had to choose, on average, from at least 40 PDPs (1429 plans from 34 regions) or from multiple MA-PDs (1314 plans over 26 regions), and thus felt overwhelmed in selecting their initial plan.

An estimated 10% of Part D beneficiaries in 2006 enrolled in the lowest-cost plan, while only 6% to 12% switched coverage for 2007.

 Two single center, pharmacist-led intervention studies indicated that switching could result in annual OOP savings ranging from $100 to $522.

 To date, there has not been a formal examination to identify which Part D enrollees switched plans and why switching occurred.

This study used survey data to estimate the proportion of Part D enrollees who reported switching plans and to examine the predictors of self-reported switching behavior. Based on health policy literature from Europe, switching was expected to be demographically associated with age, education, and health status, and to be closely related with the initial enrollment decision.

 This study also tested whether self-reported switching behavior was associated with future changes in annual OOP prescription costs, under the hypothesis that switching Part D plans would, in practice, result in decreased OOP costs.

This study used the 2006, 2008, and 2010 RAND-HRS version L file to measure self-reported switching behavior. The RAND-HRS version L file is a cleaned-up version of the Health and Retirement Study (HRS),

which has been a longitudinal biennial survey of near elderly and elderly Americans since 1992.

Survey response rates approached 88%. Previous research has validated the use of surveys over administrative data to estimate actual plan switching behavior and the application of HRS data to estimate Part D enrollment.

The rates and predictors of self-reported switching were each analyzed using a retrospective cross-sectional study design, while the association of self-reported switching with changes in annual OOP prescription costs was analyzed using a retrospective cohort difference-indifference study design. Since the HRS fields its survey every 2 years, self-reported switching behavior was evaluated in 2 different time periods: between 2006 and 2008, and between 2008 and 2010. Baseline information for the 2006-to-2008 time period was drawn from the 2006 HRS survey, and drawn for the 2008-to-2010 time period from the 2008 HRS survey.

 for a visual description of the sample selection process for each study analysis. HRS participants were included in the analysis of switching rates for a given time period if they completed both HRS surveys, had Part D coverage during that time period, and were at least 65 years old at baseline. To measure the incidence of self-reported switching of nondisabled, elderly Part D beneficiaries, this analysis excluded those who reported switching because their existing plan closed,

In order to ensure that study subjects had responded to each HRS question used to populate each potential predictor of self-reported switching behavior, subjects were excluded from the sample population used in the analysis of switching rates if they were missing any baseline characteristic information from the sample population used in the analysis of the predictors of switching. Subjects were excluded from the analysis of OOP costs if they did not regularly use prescriptions, 

were enrolled in the Low-Income Subsidy (LIS) program (and therefore did not have to pay for their prescriptions),

or reported excessively high OOP prescription costs (above the 99th percentile).

Since government subsidies largely covered the OOP prescription costs of LIS enrollees, these subjects were excluded from the analysis of OOP costs. By contrast, LIS enrollees were not excluded from the analysis of switching rates or the analysis of the predictors of switching because they were permitted to switch plans on a monthly basis, while non-LIS enrollees could only switch during the annual open enrollment period.

Since OOP prescription costs were measured differently in the 2006 and 2008 HRS surveys than in the 2008 and 2010 surveys, the analysis of OOP costs was only performed during the 2008-to-2010 time period.

 For the analyses of switching rates and of the predictors of switching, the outcome variable dichotomized Part D enrollees as self-reported switchers or nonswitchers based on well-established HRS questions. In 2008 and 2010, HRS participants with Part D coverage in the current and previous surveys were asked, “Do you still get your Medicare drug coverage through this plan?” Those who directly responded “yes, same company, different plan” or “no” were designated as switchers. Those who directly responded “yes” were designated as nonswitchers. However, this question was only asked to HRS participants who specifically named their Part D plan during the previous HRS survey.

For HRS participants who did not directly report whether they switched plans, this study indirectly determined switching behavior from reported changes in their Part D plan type across HRS surveys during a given time period (ie, PDP coverage in 2006 and MA-PD coverage in 2008, or MA-PD coverage in 2006 and PDP coverage in 2008). Between 2006 and 2008, 53% directly reported and 47% indirectly reported whether they switched Part D plans. Between 2008 and 2010, 21% directly reported and 79% indirectly reported whether they switched.

 The outcomes for the analysis of OOP costs were continuous values of annual OOP prescription costs for 2008 and 2010. HRS participants were asked, “On average, about how much have you paid OOP per month for your prescriptions since [previous wave/in the last two years]?” HRS respondents who did not report an exact dollar amount were directed into a series of follow-up questions meant to elicit a minimum and a maximum dollar amount from nonresponses. From the range of annual OOP prescription costs reported by HRS respondents, RAND imputes an exact dollar amount while accounting for inflation and demographic factors such as employment status, education, health status, age, race, marital status, occupation class, cognition, and bequest motive.

RAND-HRS medical expenditure data includes measured OOP prescription costs reported by the more than three-fourths of HRS respondents who provided exact dollar amounts and imputed OOP prescription costs for those who reported a range, not an exact dollar amount.

The RAND imputation procedure was shown to reduce item nonresponse rates from approximately 21% to less than 5% and to result in valid and reliable estimates of annual OOP prescription costs of the elderly population when benchmarked against cost data from the Medicare Current Beneficiary Survey (MCBS) and Medical Expenditure Panel Survey (MEPS).

 Lastly, total OOP costs (plan premiums plus OOP prescription costs) were not measured as a potential outcome of self-reported switching because premiums were measured categorically; thus, they could not be combined with continuous values of OOP prescription costs.

For the analysis of the predictors of switching, each baseline characteristic was evaluated as a potential predictor of self-reported switching. For the analysis of OOP costs, each baseline characteristic was evaluated as a covariate, attempting to control for potential differences between reported switchers and nonswitchers. Please refer to 

 for a description of each baseline characteristic.

All statistical analyses were weighted to adjust for the complex sampling design of the HRS.

For the analysis of switching rates, rates were calculated as the proportion of Part D enrollees who reported switching plans among the total nondisabled, elderly Part D population.

For the analysis of the predictors of switching, this study first tested for differences in the distribution of each baseline characteristic between self-reported switchers and nonswitchers using the Pearson χ

 statistic with Rao and Scott correction for complex survey data.

We then predicted the probability of reported switching using a multivariable logistic regression model, which controlled for baseline Part D plan characteristics, baseline individual characteristics, baseline demographic characteristics, and 2 interaction terms: monthly premiums by plan type and monthly OOP prescription costs by plan type. Baseline characteristics and interactions were assessed as potential predictors of switching using average marginal effects (AMEs) rather than odds ratios, a practice recommended by Norton et al (2004) and Karaca-Mandic et al (2012).

 This approach is recommended for 2 reasons: first, the common interpretation of logistic regression coefficient estimates as roughly equivalent to log-transformed risk ratios relies on a low frequency outcome. Second, the interpretation of coefficients of interaction terms is notoriously difficult in logistic regression. For these 2 reasons, we reported AMEs rather than odds ratios or risk ratios. The AME is an appropriate and interpretable measure of effect for this study, and its validity does not rely on a low-frequency outcome.

For the analysis of OOP costs, changes in annual OOP prescription costs for 2008 and 2010 were compared between self-reported switchers and nonswitchers using a multivariable generalized estimating equation (GEE) regression model with gamma distribution and log link function that accounted for skewed results in the distribution of OOP prescription costs and for clustering around study subjects, who each had 2 observations: OOP costs for 2008 and for 2010.28 A modified Park test failed to reject the model specification of gamma distribution with log link function (

The differencein- difference analytic framework was structured around an interaction term, which allowed for changes in annual OOP prescription costs between 2008 and 2010 to vary between switchers and nonswitchers. Lastly, the association between switching and OOP costs was stratified by baseline plan type.

All statistical analyses were performed using StataMP 12.0 (StataCorp LP, College Station, Texas).

Statistical significance was set a priori to 

Of nondisabled elderly Part D enrollees, 20.6% (95% CI, 18.6% - 22.6%) reported switching plans between 2006 and 2008, and 23.4% (95% CI, 21.7% - 25.1%) reported switching between 2008 and 2010. Across both study periods, a greater percentage of switchers compared to nonswitchers never graduated high school and qualified in the bottom categories of net worth and annual income. Fewer switchers were of white race/ethnicity and lived in the west (

Controlling for all baseline characteristics, the probability of switching Part D plans between 2006 and 2008 was 3 times greater for PDP enrollees (28.9%) than for MAPD enrollees (9%). However, between 2008 and 2010, the probability of switching from an MA-PD plan increased to 17.1% while it dropped slightly for PDP enrollees (

Paying $60 or more in monthly premiums was also associated with self-reported switching. Between 2006 and 2008, paying $80 or more in OOP prescription costs was associated with a lesser likelihood of self-reported switching (

= .042). Over time, LIS enrollment was associated with a higher probability of switching between 2008 and 2010 (P = .036) but not between 2006 and 2008 (

= .073). The probability of switching for beneficiaries of black race/ethnicity compared with white enrollees was consistently greater between 2006 and 2008 (26.8% vs 19.3%;

= .052) and between 2008 and 2010 (31.3% vs 21.6%;

= .014). With respect to net worth, beneficiaries in the lowest category ($75,000 and lower) consistently had the highest probability of switching.

The multivariable regression model (Table 4) also tested for interactions between monthly premiums andplan type and between monthly OOP drug spending and plan type. Only during the second time period was there a significant (

= .009) interaction between monthly OOP prescription spending and plan type (Table 4). Within the interaction of OOP prescription spending stratified by plan type, higher OOP drug spending was correlated with higher rates of plan switching between 2008 and 2010 (

illustrates the results from the differencein-difference, multi-variable GEE model to predict mean changes in annual OOP prescription costs for self-reported switchers and nonswitchers. While adjusting for baseline covariates, the mean annual OOP prescription costs for nonswitchers increased by $116 (17.5%; 

 <.001) between 2008 and 2010, but only by $95 (12.5%; 

= .121) for switchers; the difference-in-differences of annual OOP prescription costs from 2008 to 2010 between self-reported switchers and nonswitchers was not statistically significant (

When stratified by plan type, a somewhat meaningful association was evident only among beneficiaries with previous PDP coverage. Among PDP enrollees, the mean annual OOP prescription costs for nonswitchers increased by $152 (21.2%; 

<.001) between 2008 and 2010, but only by $70 (9.2%; 

= .300) for switchers; ; the difference-indifference of annual OOP prescription costs from 2008 to 2010 between self-reported switchers and nonswitchers just missed statistical significance (

Though difficult to compare against other 1-year estimates of switching, this study confirmed that most elderly, nondisabled Part D enrollees did not report switching plans from 2006 to 2010. Since the Part D program was the first of its kind to allow beneficiaries to switch their prescription coverage, the literature from the United States on continued plan choices is limited. The literature from Europe about the predictors of health plan switching is more thoroughly developed because of previous government-funded health insurance programs that allowed consumers to choose their coverage each year.

Inconsistent with the internationally developed health insurance literature,

 switchers from the Part D program were more likely non-white in ethnicity, less formally educated, and of lower socioeconomic status. This group of Part D beneficiaries was largely dual-eligible for Medicare and Medicaid and likely switched for 1 of 2 reasons: first, an estimated 2.2 million LIS enrollees who switched plans after their initial random assignment to a benchmark plan, also known as “chooser dual eligibles,” were responsible in 2009 for finding qualified benchmark plans to avoid paying plan costs.

Choosers faced the challenge of high turnover in the market for low-cost benchmark plans—a market in which less than 20% of the approximately 400 qualifying benchmark plans in 2006 still qualified in 2010.

This phenomenon was potentially reflected by the finding that LIS enrollment was at first not associated with switching between 2006 and 2008, but over time, came to be associated with switching between 2008 and 2010. This study may underscore how some LIS enrollees without the financial means to pay for their prescriptions were compelled to switch plans because of volatility in the market. Second, LIS enrollees who never switched plans, also known as “nonchooser dual eligibles,” were randomly assigned to a low-cost benchmark plan each year. Study findings reflect how the reassignments of nonchoosers likely contributed to the rate of switching, though deciphering the economic behaviors of nonchoosers from chooser dual eligibles was not possible with the HRS survey.

The strongest predictors of switching were the baseline plan characteristics, a finding that was consistent with the expectation that the factors influencing the initial enrollment decision would also influence ongoing enrollment decisions.

While multiple baseline demographic characteristics were significantly associated with plan switching in a simple bivariate relationship, these associations were no longer statistically significant when controlling for plan

characteristics in the multivariable model.

There were several reasons why MA-PD enrollees were more likely than PDP enrollees to experience plan stickiness and that the likelihood of switching among MA-PD enrollees doubled over time. MA-PD enrollees may have held on to their existing coverage because many MA-PD plans in 2006 had been renamed Medicare Choice plans that beneficiaries had previously been enrolled in for a long time; by contrast, PDP plans did not exist before 2006. MA-PD enrollees also faced substantial “transaction costs” because switching Medicare managed care plans often involved changing other aspects of one’s healthcare.

For example, MA-PD enrollees were required to drop their Medicare Advantage coverage and enroll in original Medicare before switching to PDP coverage. As enrollment in Medicare Advantage grew from 14% in 2006 to 24% of the Medicare population in 2010, so did the availability of MA-PD alternatives, which potentially led to the greater likelihood of switching among MA-PD enrollees between 2008 and 2010.

Consistent with economic theory that pricing influences the decision to change health insurance, higher monthly premiums were consistently associated with switching.

 This is an indication of how premiums are likely the most apparent plan cost to Part D beneficiaries. Other plan costs, such as OOP prescription costs, could be more ambiguous and difficult to compare across plans without properly using the Medicare Plan Finder or consulting with a trained professional.

PDP enrollees were conversely shown to be more sensitive to premium differences than MA-PD enrollees.

 The same reasons which potentially contributed to plan stickiness among MA-PD enrollees also likely reduced the responsiveness of these enrollees to higher MA-PD premiums. Higher OOP monthly prescription costs were associated with reduced likelihood of switching during the initial program years (between 2006 and 2008). Assuming that the elderly were averse to potential changes to the convenience and company name of their existing prescription drug coverage,

his result may provide evidence that nonswitchers who initially paid high OOP prescription costs exhibited plan stickiness.

When stratified by baseline plan type, only among MA-PD enrollees between 2008 and 2010 was self-reported switching behavior sensitive to differences in OOP prescription costs. Since around 50% of MA-PD plans did not require a monthly premium, it was conceivable that MA-PD enrollees were more sensitive to OOP costs than premiums.

Consistent with plan switching literature, switching Part D plans was variously associated with those of African-American ethnicity and lesser net worth.

Since Part D enrollees were less familiar with their Medicare prescription drug coverage between 2006 and 2008 than they were between 2008 and 2010, the predictors of switching were expected to be generally inconsistent across time, which supports why this study examined switching over 2 separate time periods.

Annual OOP prescription costs were not significantly different for self-reported switchers and nonswitchers, which is inconsistent with previous Part D literature that suggested switching could limit future increases in OOP costs.

 This lack of correlation may reflect how switching is not entirely motivated by plan costs, but instead, could be influenced by noncost factors. Possible scenarios include an MA-PD beneficiary who switched because their doctor no longer accepts that particular insurance, an LIS enrollee who received poor customer service from their PDP, or a patient who is prescribed a new medication not covered by their current plan. Study findings indicated that patient satisfaction, which was used as a proxy for such intangible factors not available in the HRS survey, was marginally associated with switching Part D plans, at most; however, the variable labeled “satisfaction with care” was more general in scope and not specific to the Part D benefit.

This study differed from previous studies by attempting to measure the potential relationship of switching plans with OOP prescription costs on a national level. The analysis of OOP costs was also performed without knowledge of whether self-reported switchers enrolled in the plan with the lowest cost available, which may highlight an underlying association between switching and reduced OOP prescription costs.

For example, it is possible that the complexity of the Part D benefit led some beneficiaries to switch to a plan that did not reduce their OOP costs or keep their same coverage. It is also possible that beneficiaries did understand the Part D benefit but incorrectly predicted their prescription utilization in the following year because of changes in their health.

With advancements in, knowledge of, and education on the Medicare website plan finder, Part D beneficiaries can more readily search for plans that best meet their needs. Efforts to promote annual patient evaluations of Part D coverage may increase switching rates and possibly improve the continued plan choices of beneficiaries.

First, this study examined self-reported, rather than measured, switching of Part D plans. Since the specific plans for each study subject were not known, the results were presented without knowledge of potential inconsistencies in the actual and reported continued plan choices of Part D beneficiaries. This study could not account for whether beneficiaries switched plans more than once during a given study period or whether LIS enrollees were compelled to switch plans because of volatility in the market for low-cost benchmark plans. Though the limitations of using HRS data to model plan choice conditional on existing enrollment are well documented,

this study used questions from a well-established survey of the aging US population to present results that best approximated the switching behavior of nondisabled, elderly Part D beneficiaries. In the absence of comprehensive, nationally representative data that measure actual switching behavior, this study attempts to address an important gap in the Medicare literature by providing useful insight on the continued plan choices of Part D beneficiaries.

Second, the study definition of “plan switching” lacked sensitivity, which was expected to bias results from the analyses of the predictors of switching and of OOP costs toward the null hypothesis. Findings from HRS data were therefore interpreted as conservative estimates.

Third, because LIS enrollees were excluded from the OOP costs analysis, these results may not be generalizable to all income groups. Specifically, the results from the OOP costs analysis largely reflect the behavior of the middle- to upper-income strata of Part D enrollees; the results from the analysis of the predictors of switching reflect the behavior of more widely represented income groups.

Fourth, study subjects were not randomized to switcher or nonswitcher groups. Since Part D-enrolled HRS participants chose whether or not to switch plans, this study performed adjusted analyses to control for systematic differences between switchers and nonswitchers.

Lastly, the medical expenditure files from RAND-HRS data did not represent actual expenditures, but rather a self-reported estimate of annual total OOP prescription costs that excluded premiums. It was also possible that the multiple imputation method used by RAND to assign continuous estimates of OOP prescription costs to each study subject broadened the confidence intervals of results from the analysis of OOP costs. As a result, our ability to imply causality in the temporal association between switching Part D plans and total OOP prescription costs was limited.

Less than a quarter of nondisabled, elderly Part D beneficiaries reported switching plans. Self-reported switching of Part D plans was consistently associated with both PDP coverage (compared with MA-PD) and higher monthly premiums, but marginally associated with LIS enrollment over time (though not initially). Self-reported plan switching was not associated with significant reductions in annual OOP prescription costs, although study findings do not discourage switching Part D plans.

Articles

Less than one-fourth of Part D beneficiaries reported switching plans. Elderly beneficiaries who reported switching did not realize significant savings in prescription costs over nonswitchers.