Dr. Donohue is affiliated with the Graduate School of Public Health, University of Pittsburgh, 130 De Soto St., A613, Crabtree Hall, Pittsburgh, PA 15261 (e-mail: jdonohue@pitt.edu). Dr. Frank is with the Department of Health Care Policy, Harvard Medical School, and the National Bureau of Economic Research, Cambridge, Massachusetts. This article was presented at the National Institute of Mental Health's 13th Biennial Mental Health Economics conference, Bethesda, Maryland, September 25—26, 2006.

Psychotropic drugs are playing an increasingly central role in the treatment of mental disorders (1). Since 1997 spending on psychotropic medications has grown at a higher rate than spending on health care or prescription drugs overall (2). The Medicare Prescription Drug Improvement and Modernization Act of 2003, which established a new Medicare drug benefit (Part D), required the six million beneficiaries who were dually eligible for Medicare and Medicaid to transition from Medicaid prescription drug coverage to Medicare drug plans in January 2006. Mental disorders are highly prevalent among dually eligible beneficiaries, particularly those under the age of 65 who are eligible for Medicare because of a disability (3). Because of Medicaid's key role in paying for psychotropic drugs, particularly antipsychotics (2,4), the new way of paying for drugs may have important implications for the treatment of mental disorders.

Most state Medicaid programs have been cautious about applying cost control measures to psychotropic drugs even though they have been drivers of increased Medicaid drug spending (2,5,6). However, Medicare drug plans may structure their formularies or apply utilization controls (such as prior authorization and step therapy) in order to simply save money or to discourage beneficiaries with high expected drug costs from enrolling (7). In fact, a study by the Medicare Payment Advisory Commission found that Medicare drug plans were more likely to apply utilization controls to second-generation antipsychotics and newer antidepressants than to other medication classes (8). Medication disruptions stemming from formulary exclusions or utilization controls may be particularly difficult for dually eligible individuals with mental disorders because they are likely to have comorbid medical conditions, cognitive or physical impairments, and little ability to pay for needed medicines on their own because of their very low incomes (9).

In this study we assessed the impact of the transition of dually eligible beneficiaries with mental disorders to Medicare Part D by estimating the number of beneficiaries who will encounter disruptions in psychotropic medication use. Specifically, we estimated rates of medication switching among dually eligible beneficiaries who use antidepressants, antispychotics, and mood stabilizers. We hypothesized that some beneficiaries face access restrictions because of exclusion of their drugs from a Medicare drug plan's formulary or because of prior-authorization or step-therapy requirements. Although a full assessment of the actual impact of the transition to Part D would be preferable to an estimate, we have reason to believe that it will be quite some time before such an assessment is possible. Data from Part D plans are not yet available, and it is unclear whether researchers will be able to link pre-2006 data from state Medicaid programs to post-Part D data from the plans via publicly available data. Once data are available, it will take some time to generate estimates of the impact of this policy change for this highly vulnerable population. In the meantime, our analyses provide an early sense of the potential impact of the policy as well as an analytical framework to guide future research in this area. We begin with a brief overview of the policies under Medicare Part D that aim to ensure continuity of medication treatment in the dually eligible population.

Medicare beneficiaries participating in Part D may either enroll in a Medicare Advantage prescription drug plan to obtain coverage for all Medicare-covered benefits, including prescription drugs, or stay in traditional fee-for-service Medicare and enroll in a stand-alone prescription drug-only plan (PDP). In addition, beneficiaries with private drug coverage through an employer may keep that coverage. To prevent gaps in prescription drug coverage, Medicare automatically assigned dually eligible beneficiaries to certain PDPs (those with premiums at or below a national benchmark set by the Centers for Medicare and Medicaid Services [CMS]) across the nation in September 2005. In 2006 the number of plans eligible for automatic assignment of dually eligible beneficiaries in each region ranged from six to 16, depending on the region (the United States was divided into 34 regions that form markets for drug coverage). In some areas, dually eligible beneficiaries were enrolled instead in Medicare Advantage Special Needs Plans (described in the Discussion section); however, in most areas dually eligible beneficiaries were assigned only to stand-alone PDPs. Therefore, the focus of our analysis is on PDPs. Medicare randomly assigned dually eligible beneficiaries to plans and did not take the current or future medication needs of dually eligible beneficiaries or plan features into account (9). As of June 2006 dually eligible individuals made up approximately 30% of total Part D enrollment (10).

The Medicare Prescription Drug Improvement and Modernization Act (MMA) contains two provisions related to use of formularies and pharmacy management tools: plans must offer at least two drugs from each therapeutic category or class, and plans are not permitted to use pharmacy management tools in such a way as to discourage enrollment of certain groups of beneficiaries (the so-called nondiscrimination provision) (11). Whereas the first rule is relatively easy to monitor and enforce, the second is far less so. CMS formulary guidelines imposed additional requirements for plans' coverage of some psychotropic medications. Plans are required to list "all or substantially all" antidepressants, antipsychotics, anticonvulsants, anticancer drugs, immunosuppressant drugs, and HIV-AIDS drugs (12). Annalysis of the 35 plan options offered by 15 nationwide or nearly nationwide plans found that they covered nearly all antidepressants (13). However, other psychotropic drug classes were not studied.

The MMA outlines a standard benefit design with a deductible, cost-sharing levels, and a coverage gap. For those who do not qualify for dual coverage, plans can vary all aspects of the benefit design as long as they offer coverage that is actuarially equivalent or enhanced. However, dually eligible beneficiaries pay a fixed copayment for all generic drugs ($1 or $2, depending on their income) and a higher fixed copayment for all brand-name drugs ($3 or $5, depending on income) regardless of the plan in which they are enrolled.

Because of these requirements for formulary structure for psychotropic drugs and cost sharing for dually eligible beneficiaries, plans will have to rely on pharmacy management tools such as prior authorization (requiring preapproval before coverage) and step therapy or "fail first" policies (that is, enrollees must try specified drugs before moving on to other drugs) to manage utilization and costs for this population. For antidepressants, antipsychotics, and anticonvulsants, plans can apply utilization tools in only these categories to new-start enrollees according to CMS guidelines. However, most plans will lack pre-2006 utilization data; therefore, it may be difficult for plans to comply with this requirement. Thus we would expect prior authorization and other utilization management tools to be the focus of cost control efforts.

Overview

We estimated the amount of medication switching that would occur among dually eligible beneficiaries by using data from several sources. First, we obtained baseline information on the medication use patterns of dually eligible beneficiaries with mental disorders in 2002. Second, we extracted information on formulary coverage and use of pharmacy management tools for psychotropic medications from a sample of Medicare Part D drug plans. Third, we obtained estimates of utilization response to Medicaid prior-authorization programs by conducting a literature review. Estimates from the literature review were used as proxies for patient and provider response to utilization management tools under Part D. Using the results from these inquiries, we estimated the likelihood of switching medications conditional on use of drugs and assignment to plans.

Medicare plan data

We obtained data on formulary coverage and use of pharmacy management tools for psychotropic medications from Michigan, one of the 34 Medicare Part D regions. We chose this region because it had an average number of Medicare beneficiaries and offered an average number of PDP plans. In that region in 2006, 14 PDP plans were eligible for automatic assignment of dually eligible beneficiaries. Of the 14 plans, seven were offered in all 34 regions, and an additional three were nearly nationwide plans (they offered PDPs in at least 30 of the 34 regions) (8). Therefore, we considered this region to offer a reasonable reflection of plans available nationally to dually eligible beneficiaries. We obtained data on the psychotropic drugs listed on each plan's formulary and whether each plan required prior authorization or step therapy for specific drugs.

Estimates of utilization response

To obtain estimates of the utilization response associated with prior-authorization programs, we conducted a literature review. We searched the Ovid database using the search terms "prior authorization," "Medicaid," "step therapy," and "pharmacy." We limited our search to studies published after 1990. Studies that included a comparison group or used time-series methods to control for secular trends in medication use patterns were included. We obtained estimates of the effect of these policies on rates of medication switching or changes in the market shares of restricted drugs. We found six studies that examined the impact of prior-authorization programs on prescription drug utilization among Medicaid beneficiaries and no studies on step therapy in Medicaid. We excluded three studies because they examined the revocation, not the imposition, of the policy on rates of medication switching (14) or used aggregate prescription data and did not report results at the individual level (15,16). Three studies reported 58.9% to 70.5% reductions in the use of nonsteroidal anti-inflammatory drugs (NSAIDs) or proton pump inhibitors (PPIs) after implementation of a prior-authorization policy (17,18,19).

Baseline data on psychotropic utilization patterns

Data on pharmacy utilization and spending among dually eligible Medicare beneficiaries were obtained from the 2002 Medicare Current Beneficiary Survey (MCBS). MCBS is a continuous, multipurpose survey of a representative sample of approximately 12,000 Medicare beneficiaries. CMS collects prescription drug data from face-to-face interviews for community-dwelling beneficiaries only.

We identified as dually eligible any beneficiary enrolled in Medicaid for at least one month in 2002. Individuals were defined as having a mental disorder if they had been told by a physician that they had a mental or psychiatric disorder, including depression, and used at least one psychotropic drug in 2002. We investigated use of a claims-based case finding but found that approximately one-third of MCBS respondents were missing claims data because they were in managed care plans, were in long-term care, had not used any health services, or because they were in the first wave of MCBS data collection so claims were not retrieved . We calculated descriptive statistics on the full sample of dually eligible beneficiaries with mental disorders.

Switching analysis

For the switching analysis, we narrowed the focus to individuals with any use of antidepressants, antipsychotics, or mood stabilizers in 2002. We applied a sample weight to the use and spending estimates to obtain population-based estimates for the entire Medicare population enrolled at any time in 2002.

We conducted separate analyses to estimate the effect of formulary exclusions, prior-authorization, and step-therapy requirements on medication switching for each drug category as well as for specific drugs within the category. We assumed that baseline rates of medication switching held constant over the study period and focused only on switching attributable to Medicare Part D. We determined the number of beneficiaries using each drug in the three classes. Our unit of analysis was the person-drug.

Table 1 shows the key assumptions of our analysis and a description of the sensitivity analyses we conducted. First, we assigned a probability that each person taking a specific drug would switch medications if he or she enrolled in a particular plan. We assumed that the probability (P) of switching equaled 0 if the drug was on the plan's formulary and no restrictions were applied and 1 if the drug was not on the plan's formulary. With the studies reviewed, we assumed that dually eligible beneficiaries' rates of switching for psychotropics as a result of prior authorization or step therapy equaled 70% in our analysis. The probability that someone taking a particular drug would switch medications was equal to the product of the probability of switching conditional on enrollment in a plan and the probability of enrollment in the plan. We randomly assigned each person-drug combination to one of the 14 plans in our sample; therefore, the probability of being assigned to a particular plan was equal to 1 of 14, or .07.

Initially, we assumed that each dually eligible beneficiary was taking only one drug per category and simply summed the probability of switching for all drugs. Finally, we multiplied the number (N) of dually eligible beneficiaries taking each drug by the probability of switching, and we divided this number by the total number of beneficiaries using the medication class to attain a population-based estimate of switching.

Thus the calculations to estimate rates of switching are as follows:

• P(switch) for each person-drug combination=P(switching given enrollment in a plan) × P(enrollment in that plan)

• N people taking each drug who switch=P(switch) × N taking each drug

• Total proportion in class who switch=sum of N people taking each drug who switch across all drugs in class / total N using that class

We examined the sensitivity of our estimates to alternative assumptions (see Table 1). First, because our estimates of the impact of prior authorization came from studies of NSAIDs and PPIs, which are viewed as more substitutable than psychotropic drugs, we applied a more conservative measure of utilization response (30% switching rate) to test the sensitivity of our estimates. Second, we relaxed the assumption that individuals used one drug per class, and we accounted for polypharmacy by adjusting our estimates on the basis of rates obtained from the MCBS data. Third, we tested alternative assumptions about automatic assignment to PDPs. We calculated class-specific rates of switching for each plan, assuming 100% enrollment of dually eligible beneficiaries with mental disorders in each of the 14 plans to determine a range of rates of switching across plans.

Study population

In 2002 a total of 2,617, or 20.6% of the total MCBS sample of 12,697, were dually eligible for Medicaid. Of those, 467, or 18%, had a self-reported diagnosis of a mental disorder and used a psychotropic medication within the year. Drug utilization data from these individuals provided our baseline rate of use for the analysis. Table 2 shows sociodemographic statistics for the study population as well as the population's frequency of use of psychotropic drug classes. Among the dually eligible, disabled beneficiaries under age 65 with mental disorders, 59.9% had incomes ≤100% of the federal poverty level ($8,860 for an individual in 2002), and 56.7% of dually eligible beneficiaries aged 65 and over had incomes below this level. The most commonly used psychotropic drug classes were antidepressants and antipsychotics among beneficiaries with disabilities and antidepressants and benzodiazepines among beneficiaries 65 and older.

Drug spending among dually eligible beneficiaries

Figure 1 shows the distribution of dually eligible individuals and the subset of the dually eligible beneficiaries with mental disorders by quintile of drug spending among all Medicare beneficiaries. Forty-nine percent of all dually eligible individuals and 71% of dually eligible individuals with mental disorders were in the top two quintiles of total drug spending among all Medicare beneficiaries in 2002, demonstrating the high level of drug spending in this population.

Formulary exclusions and utilization controls

Table 3 summarizes the limits that PDPs placed on psychotropic medications in 2006. Plans were much more likely to exclude at least some antidepressants or antipsychotics from their formularies than they were to exclude mood stabilizers. The number of drugs excluded varied greatly among plans. For instance, among the 11 plans that excluded some antidepressants, the median number of drugs excluded from the formulary was two but ranged from one to 12. The most common antidepressants excluded from the formulary were escitalopram (Lexapro) and nefazodone (Serzone). Similarly, 13 of the 14 plans excluded some antipsychotics from their formulary; the number of drugs excluded varied from one to 11, with a median of three. The most commonly excluded antipsychotics were conventional antipsychotics, such as pimozide (Orap) and prochlorperazine (Compazine). Only one plan excluded a second-generation antipsychotic (clozapine) from its formulary.

Fewer plans used prior authorization for psychotropic drugs, but the median number of drugs for which prior authorization or step therapy was required was slightly higher per plan than the number of formulary exclusions. Depending on the drug class, between three and six of the 14 plans had prior-authorization or step-therapy requirements (Table 3). Utilization controls most commonly applied to venlafaxine, nefazodone, escitalopram, sertraline, and amitriptyline among the antidepressants; clozapine, olanzapine, and ziprasidone among the antipsychotics; and lamotrigine and topiramate among the mood stabilizers.

Switching results

Table 4 shows the results of the switching analysis. Assuming that dually eligible individuals were randomly assigned to one of the 14 PDPs and did not switch plans, we estimated that 4.6% to 7.3% of dually eligible beneficiaries using antidepressants would switch medications because their plan did not cover the drug or the plan required prior authorization or step therapy. We estimated that the bulk of medication switching (88%) would occur among individuals taking four drugs: sertraline, nefazodone, amitriptyline, and venlafaxine. We estimated that a slightly higher percentage of dually eligible beneficiaries taking antipsychotic medications would switch (5.8% to 10.2%). In this class we estimated that most (73.6%) of the medication switching would take place among individuals taking second-generation antipsychotics that were exposed to utilization controls (risperidone, olanzapine, quetiapine, and clozapine). Among individuals taking mood stabilizers, we estimated that between 2.7% and 4.0% would switch. Fully 84.3% of switching would occur among individuals taking gabapentin, topiramate, or lamotrigine.

The last two columns in Table 4 indicate that the estimated percentage of dually eligible beneficiaries who switched medications varied dramatically from plan to plan, depending on how tightly the plan managed psychotropic medications. If all dually eligible beneficiaries were assigned to a single plan, the lower-bound utilization response (30% switching rate) indicates that the percentage of persons switching medications would vary from .5% to 26.3% for dually eligible beneficiaries taking antipsychotics, from 0% to 20.6% for dually eligible beneficiaries taking antidepressants, and from 0% to 11.4% for those using mood stabilizers, depending on the plan.

We found that 30.2% of antidepressant users, 23.7% of antipsychotic users, and 12.3% of mood stabilizer users had used more than one drug in the class in 2002. We were unable to determine from the MCBS data whether use of multiple drugs in the class represented switching or augmentation (polypharmacy). Nevertheless, assuming that these percentages approximate rates of polypharmacy and applying the lower-bound utilization response estimates, we estimated that 3.2%, 4.4%, and 2.4% of dually eligible beneficiaries using antidepressants, antipsychotics, and mood stabilizers, respectively, would switch medications as a result of enrollment in Part D plans (results not shown).

Dually eligible beneficiaries with mental disorders have high drug costs relative to other Medicare beneficiaries; 71% of these beneficiaries have drug spending that puts them in the top 40% of total drug spending among all Medicare beneficiaries. The results of our simulation analysis indicate that a minority (less than 10%) of dually eligible beneficiaries with mental disorders will experience medication treatment disruptions after being directed to a different drug by their plan in the transition to the Medicare drug benefit. However, we did not estimate the impact of other aspects of the transition, such as the increase in out-of-pocket costs faced by dually eligible beneficiaries or problems with the enrollment process that could disrupt medication use among this population (20). In accordance with the CMS formulary guidelines, the plans we studied list "all or substantially all" antidepressants, antipsychotics, and mood stabilizers on their formularies. With a few exceptions (one plan excluded sertraline, or Zoloft) plans limit formulary exclusions to drugs that are seldom used. A subset of PDPs do, however, apply management tools like prior authorization or step therapy to commonly used psychotropic medications, and these tools may lead to treatment disruptions for dually eligible beneficiaries.

Our results also indicate that estimated rates of medication switching vary substantially from plan to plan because of differences with respect to plans' use of these management tools. This finding calls into question the appropriateness of random assignment of the dually eligible population to PDP plans. CMS may want to consider adjusting the assignment algorithm to take into account plan management of drug classes commonly used by dually eligible beneficiaries. Another alternative to assignment to stand-alone PDPs is to enroll the dually eligible in special needs plans (SNPs). SNPs are plans that specialize in care for the dually eligible beneficiaries residing in long-term care facilities or for individuals with chronic illnesses (21). In 2006 a total of 276 SNPs served Medicare beneficiaries and most plans served the dually eligible. Although SNPs offer the promise of better care coordination for these vulnerable populations, they must deal with numerous differences between Medicare and Medicaid in rules for bidding, contracting, enrollment, marketing, and rate setting, among other issues (22). Moreover, enrollment of dually eligible beneficiaries in Medicare Advantage remains low (18% nationally), and as long as enrollment in managed care is voluntary, it is not clear that these plans offer much protection for dually eligible beneficiaries in the short run.

In the absence of special plans to meet the needs of this vulnerable population, it will be important for clinicians to counsel their patients on finding the plan that best meets their needs. Unlike other Medicare beneficiaries who can change Part D plans only once per year during the open enrollment period, dually eligible beneficiaries may change plans at any time through the special enrollment period provision. Clinicians should refer patients with complex treatment needs to community-based organizations that can point them to plans that do not tightly manage their medications to allow for flexibility in treatment. Medication disruptions among individuals with mental disorders lead to significant increases in symptoms and use of other health services (23).

Our analysis of Medicare Part D's impact on medication switching has some limitations. First, we estimated the impact of formularies and management tools on utilization by using historical data on medication use patterns. These patterns are likely to have changed since 2002 because of the introduction of new medications. For instance, two antipsychotic drugs, aripiprazole (Abilify) and a long-acting version of risperidone (Risperdal Consta), which are widely used and tightly managed by some PDPs, were not on the market in 2002. Therefore, we may have underestimated the rates of switching depending on how the market shares of drugs have changed in the past few years. Second, we could not identify the indication for which a drug is used. Presumably, by narrowing the sample to individuals with a mental disorder, we increased the likelihood that use reflects treatment for mental disorders and not other uses.

Third, our estimates of utilization response to prior-authorization programs were obtained from studies of Medicaid programs. Medicaid prior-authorization programs typically grant nearly all (95%) requests for prior authorization (18,24) so we may underestimate the influence of these tools on switching and discontinuation in Part D. A recent survey of psychiatrists suggests that Part D utilization controls may lead to more treatment disruptions among dually eligible beneficiaries with mental disorders than what we found (20). Moreover, the utilization response to prior-authorization policies for NSAIDs and PPIs may not be generalizable to psychotropic drug classes because psychotropic drugs are viewed as less therapeutically interchangeable (25,26). Specifically, the response may be weaker for psychotropics than for other drug classes. Our sensitivity analyses account for these differences by applying a much more conservative estimate of utilization response. In addition, we assume that the effect of prior-authorization programs on rates of switching will not vary across plans, yet the impact of these management tools on treatment patterns will depend on how they are implemented in practice. For instance, some prior-authorization programs require physicians to document the dates and outcomes associated with all previous medication trials before they will pay for the restricted medication, whereas others require less documentation. It will be important to study actual experience with prior authorization in Medicare Part D.

Fourth, our study does not capture data from all Medicare drug plans. Although formulary information is available from all Medicare drug plans from CMS, data on use of pharmacy management tools, particularly step therapy, are incomplete. We are confident that the region we selected for study provides a representative picture of PDPs' management of psychotropic medications nationally, for two reasons. First, parent organizations of half of the Michigan plans offered PDPs in most regions and made up 34% of national Medicare Part D enrollment in 2006 (27). Second, the drugs subject to formulary exclusions and use of prior authorization in our sample plans were very similar to those reported in an analysis of all Part D plans (28).

A final challenge to our analysis is that we assumed that individuals assigned to plans would stay in those plans. Data are not yet available on the number of dually eligible beneficiaries who remained in the plans to which they were assigned. However, one survey of Medicare beneficiaries suggests that relatively few dually eligible beneficiaries changed plans in 2006 (8).

Our analyses provide an important first look at the challenges of transitioning dually eligible individuals with mental disorders from Medicaid to Medicare drug plans. It will be important to conduct analyses of the impact of Part D formulary exclusions and utilization controls once data are available from the PDP plans. Given that access restrictions vary substantially from plan to plan and that dually eligible beneficiaries are entitled to switch plans at any time, clinicians should take a more active role in referring dually eligible beneficiaries with serious health needs to social service agencies and other community-based organizations charged with helping Medicare beneficiaries navigate the complex Medicare Part D market.

The authors gratefully acknowledge support from a professional services contract (Medication Use and Cost for the Dually Eligible Under MMA) from the National Institute of Mental Health and from grant KL2-RR024154-01 from the National Center for Research Resources, a component of the National Institutes of Health (NIH) and NIH Roadmap for Medical Research (to Dr. Donohue). The authors thank Arnold Epstein, M.D., M.A., Judith Lave, Ph.D., Tami Mark, Ph.D., M.B.A., and Harold Pincus, M.D., for helpful comments.

The authors report no competing interests.