Economic incentives and contracts: the use of psychotropic medications.
"The past decade has seen an outpouring of new drugs introduced for the treatment of mental disorders. New medications for treatment of depression and schizophrenia are among the achievements stoked by research advances in both neuroscience and molecular biology" (U.S. Department of Health and Human Services, 1999, p. 68).
Innovation in prescription pharmaceuticals has been dramatic during the past two decades. The impact of new psychotropic medications for the treatment of mental disorders has been particularly profound. In 1977, 5.2% of the U.S. population received treatment for a mental disorder, and among those people, about 63% of the treatments received involved the use of prescription drugs. By 1996, 7.7% of the U.S. population received treatment for a mental disorder and 77% of those people were treated with prescription drugs. In fact, nearly the entire growth in the rate of treatment of mental disorders is attributable to expansion in cases treated with prescription drugs (Frank and Glied, 2006a).
At the same time these psychotropic medication innovations were being introduced, dramatic changes were occurring in the institutions that ration mental health services. Managed care arrangements became widespread, and specialized managed care organizations emerged, known as managed behavioral health care organizations (MBHOs), focusing on treatment of mental disorders. Mental health was increasingly "carved out" of general health plans. That is, the insurance risk for mental health care was separated from general health care and managed under specialized contracts by MBHOs.
While the emergence of MBHO carve-outs is a development specific to the behavioral health care sector, it has much greater implications in the broader overall study of contracts and market incentives. (1) Most importantly, studies have shown that this new form of contracting appears to be successful in circumventing two long-standing issues in health care markets--moral hazard and adverse selection (Frank et al., 1996; Ma and McGuire, 1998). Furthermore, two features of carve-outs are particularly significant for understanding the utilization of new technologies. First, MBHO carve-outs apply specialized expertise and economies of scale in contracting to managing care of mental disorders. Second, the carve-out contracts all separate prescription drug financial risks from other mental health care financial risks. The effect of this latter phenomenon is to take the costs of one important set of inputs--psychotropic drugs--"off budget" for the organization that has been delegated responsibility for managing care for those suffering from mental illness. Thus, MBHOs have strong economic incentives to shift treatment strategies to those that favor use of prescription drugs over other nondrug inputs (e.g., psychotherapy, inpatient care, and other psychosocial interventions). (2) An implication is that the study of carve-outs may shed light on the role of demand-side factors, specifically the role of prices in the substitution between old and new technologies.
Our analysis of the substitution between old and new medications within Medicaid population builds on Griliches' (1957/1997) seminal study on hybrid corn. Griliches identified three main determinants that affect the adoption and utilization of new technology: supply-side factors (such as the determinants of potential profitability of entry), demand-side factors (such as the long-run equilibrium profitability differential from users adopting the innovation), and the rate of reaching market saturation (such as the spread of information concerning new products). Given the stochastic and long drug approval process and our relatively short study period of 10 yr from 1991 through 2000, we do not examine the research and development (R & D) origins of this innovation but instead treat U.S. Food and Drug Administration (FDA) approvals as exogenous or predetermined. (3) Thus, we focus on demand-side factors and, more specifically, on the extent to which utilization of new technology may be encouraged or facilitated by recent institutional changes in the organization and financing of care, particularly for the Medicaid population. (4)
The Medicaid programs provide a unique laboratory for studying issues of institutional change and the utilization of new technologies in mental health care. Medicaid accounts for a disproportionate share of spending on psychotropic drugs. While a general trend of substituting older generations of psychotropic medications with newer medications may have occurred among the Medicaid population, we hypothesize that the implementation and the incentive structure of Medicaid carve-outs may have accentuated this trend. In particular, among the Medicaid population, in 1991, traditional antidepressants (tricyclics [TCAs] and heterocyclics -[HCAs]) accounted for 54% of sales, but by 2000, this share had fallen to 4%. Similarly, the atypical antipsychotic drug share rose from less than 6% of Medicaid use in 1991 to about 94% in 2000. (5) During the 1990s, 16 states implemented mental health carve-out arrangements within their state Medicaid programs. The dates of MBHO implementation spanned the decade, although a great deal of activity occurred early in the 1990s (see Table 1). (6)
The rate of substitution between old and new medications holds significant welfare implications, since it determines the speed at which patients may realize and reap the benefits from these technological advances. In this paper, we apply long-standing concepts of "general-purpose technologies" (7) and "technology-skill complementarity" (8) to examine the role of contracts and incentive structures in the diffusion processes of new technologies. The concept of general-purpose technologies is based on the observation that significant economic growth and technological progress have often been precipitated by a few fundamental innovations, such as the steam engine, electricity, microelectronics, and other technologies (Helpman and Trajtenberg, 1998). Besides promoting investments in complementary inputs and affecting the utilization and prices of inputs such as skilled and unskilled labor, these general-purpose technologies have also spurred and become embodied in later inventions.
In the current context, advances in antidepressant and antipsychotic medications represent distinct types of innovation. In the class of antidepressant medications, newer drugs known as selective serotonin reuptake inhibitors (SSRIs), serotonin norepinephrine reuptake inhibitors (SNRIs), and other new antidepressants offer therapeutic advantages over the older generations of antidepressants (TCAs and HCAs), primarily in terms of safety, tolerability, and ease of administration. Results from clinical trials indicate that the efficacy of the drugs is about equal across these different classes of agents (U.S. Department of Health and Human Services, 1999). The innovations can therefore be viewed as improvements in "user-friendliness" for both patients and their physicians. Patients no longer risk overdose and now face fewer side effects from these medications. One result has been lower rates of medication discontinuation. The newer drugs are simpler to administer in that appropriate dosages fall within a relatively narrow range, and thus, drug titration is considerably simpler (Depression Guideline Panel, 1993). A consequence is that primary care physicians can more effectively administer antidepressant medications, and patients need to rely less on specialty physicians such as psychiatrists for appropriate treatment. The new antidepressants also often represent a more efficacious treatment for anxiety disorders, some of which are highly comorbid with depression and are also frequently treated by primary care physicians. Both because of their user-friendliness and their multiplicity of uses, one can view the new antidepressants as being an example of general-purpose technologies (Bresnahan and Trajtenberg, 1995).
Atypical antipsychotic drugs offer greater efficacy for people with treatment-resistant schizophrenia and comparable efficacy for others (U.S. Department of Health and Human Services, 1999). Unlike depression that can often be treated in primary care settings, schizophrenia typically requires specialty care. But like the new antidepressants, the atypical antipsychotic drugs offer fewer side effects than their predecessors, though they do increase treatment complexity. Unlike the case of antidepressants, use of atypical antipsychotics tends to require greater specialized medical expertise (U.S. Department of Health and Human Services, 1999, p. 280). They are also typically delivered in the context of a complicated array of other treatment inputs. Because they augment the skill set of specialized psychiatrists relative to that of primary care physicians, one can view the new atypical antipsychotics as embodying technology-skill complementarity (Acemoglu, 2002). In this paper, we examine implications of these differing types of innovation.
The remainder of this paper is organized as follows. In Section II, we provide a background on psychotropic drug innovations in the context of general-purpose technologies and technology-skill complementarity, as well as the institutional features of behavioral carve-outs in the utilization of new technologies. In Sections III and IV, we discuss the data sources and analytical methods used, as well as descriptive evidence on the substitution between old and new psychotropic drugs. In Section V, we outline econometric issues including formulating a set of panel data econometric models to assess empirically the effects of carve-out implementation on new technology utilization. In Section VI, we present empirical evidence, and then finally in Section VII, we conclude and outline policy implications.
A. Antidepressants as General-Purpose Technologies
By definition, general-purpose technologies have three main characteristics (Bresnahan and Trajtenberg, 1995; Helpman and Trajtenberg, 1998). First, their designs are based on generic concepts that may be used for other applications (such as continuous rotary motion or binary logic for steam engines and microelectronics, respectively) and are often used as inputs in a number of economic sectors. Second, general-purpose technologies have the potential to be the catalysts for a series of sustained technological advances. Third, positive spillovers may exist for other sectors of the economy through providing incentives for R & D of inputs and complements to general-purpose technologies, thus leading to further economic growth. In this section, we argue that new antidepressants embody several of these vital characteristics of general-purpose technologies.
As documented in Table 2, the past decade has been marked by the rapid development of pharmacological innovations and significant expansion of options available for the treatment of mental illnesses. New antidepressants such as the SSRIs introduced in the late 1980s and 1990s quickly replaced conventional antidepressant treatments such as TCAs and monoamine oxidase inhibitors (MAOIs). Evidence from clinical trials suggests that in terms of their efficacy, full courses of psychotherapy, TCAs, and SSRIs are typically similar. TCAs and SSRIs have been found to be more efficacious than psychotherapy for treating severe forms of major depression (see Berndt et al., 2002). In addition, SSRIs are associated with lower risk of overdose, improved side-effect profile, and convenient and relatively simple dosing compared to the older medications. The introduction of SSRIs has changed the treatment of several mental illnesses including depression, obsessive-compulsive disorder (OCD), and other anxiety disorders. By 1996, nearly half of depressed patients were treated with an SSRI (Berndt, Busch, and Frank, 2001; Berndt et al., 2002; Frank, Berndt, and Busch, 1999). SSRIs also have multiple indication approvals and are widely used in treating a variety of mental illnesses (e.g., anxiety disorders, depression, and OCD). For these reasons, the newer antidepressants manifest characteristics of general-purpose technologies. (9)
B. Atypical Antipsychotics and Technology-Skill Complementarity
In contrast, the nature of treatment for schizophrenia together with the side effects and more complex dosing of antipsychotics make the prescribing of atypical antipsychotics dependent on more specialized skill and medical knowledge and is therefore more likely to be bundled with other specialty services (e.g., psychosocial treatments, family counseling). Unlike the general-purpose technology aspect of antidepressants, biased technological change embodied in use of the atypical antipsychotics instead can be envisaged as reflecting technology-skill complementarity, in that the skill set of specialist providers such as psychiatrists is augmented more than that of primary care physicians. (10)
It is therefore reasonable to expect very different impacts of carve-outs on the utilization of new antipsychotics relative to new antidepressants. In particular, we hypothesize that while MBHO carve-outs are likely to utilize more intensively the general-purpose technology antidepressants, the induced additional utilization of the technology-skill complementary atypical antipsychotics is likely to be much smaller.
The concept of technology-skill complementarity stems from the view that technological progress and skill are relative complements. In a different context, the rise of computerization and the concurrent increase in relative demand for the educated and the widening wage structure are viewed as resulting in part from technology-skill complementarity (see Berman, Bound, and Griliches, 1994; Goldin and Katz, 1998).
The introduction of new medications for the treatment of schizophrenia (e.g., atypical antipsychotics) has yielded numerous therapeutic benefits. Clozapine, the first atypical antipsychotic, was introduced in the United States in 1986. Conventional antipsychotic medications had been used to treat psychotic illnesses since the 1950s. While quite effective, these older medications have been associated with a number of side effects such as muscle stiffness, tremor, and tardive dyskinesia (involuntary muscle movements of face and limbs). Atypical antipsychotics have fewer and milder side effects than conventional or typical antipsychotics and are frequently better tolerated than the previous generation of conventional antipsychotic drugs, although they are increasingly associated with weight gain and risk of diabetes. Following clozapine, three new second-generation atypical antipsychotics were introduced in the 1990s: risperidone (1993), olanzapine (1996), and quetiapine (1997) (see Table 2).
Nevertheless, atypical antipsychotics still carry serious side-effects profiles, thus requiring continual professional mentoring. For example, clozapine (a first-generation atypical antipsychotic) carries a risk of agranulocytosis (a condition that diminishes white blood cell counts and can be life-threatening) and thus requires regular laboratory testing and monitoring. Second-generation atypical antipsychotics are not associated with agranulocytosis but in some cases have other types of side effects such as weight gain and risk of diabetes (Leslie and Rosenheck, 2002). Not only are patients assessed in terms of improvements in treating the negative and positive symptoms of schizophrenia, but changes in their physical health must also be monitored. In these ways, the innovation of atypical antipsychotics represents an example of technology-skill complementarity.
C. The Adoption of Carve-Out Contracts and Their Hypothesized Impacts on the Utilization of New Medical Technology
Even though carve-out contracts in health care delivery have existed for some time in the United States, (11) it was during the same period as the introduction of these psychotropic drugs that significant transformations occurred in the delivery of mental health services and in the broader adoption of carve-out contracts. These institutional changes included the rise of managed care organizations and the grounds gained by managed behavioral health carve-outs in occupying a central place in the public delivery of mental health care, replacing the "integrated" fee-for-service systems (see Table 1).
The effects of MBHO carve-outs on quality and costs of mental health care remain controversial. Previous studies have found that MBHO carve-outs are adept in reducing mental health expenditures through lower intensity of care, but maintaining similar or higher levels of access to mental health care, when compared to traditional arrangements (Berndt et al., 1997b; Goldman, McCulloch, and Sturm, 1998; Grazier et al., 1999). Specifically, anecdotal evidence suggests that drug utilization is encouraged under MBHOs, while use of psychotherapies is reduced based in part on cost incentives arising from these contractual arrangements. Such treatment substitutions are indeed controversial in the mental health field, where widespread agreement on the optimal levels of psychotherapy and drug treatments is still elusive.
From the vantage point of transaction costs and contract theory, the rationale and benefits of carve-outs for behavioral health coverage include economies of specialization, better monitoring of utilization and control of moral hazard, as well as the mitigation of adverse selection. (12)
By the nature of mental illness, specialized knowledge is needed for diagnosis and choice of mental health treatments. In addition, asymmetric and sometimes imperfect information regarding efficacy of treatments as well as a broad choice of effective treatment with very different costs make quality and supply of mental health care difficult and challenging to monitor and to be evaluated by patients and generalists, hence creating an advantage for specialized management. The contractual arrangements of carve-outs can circumvent the important problem of adverse selection and allow purchasers to promote efficiency through single sourcing. Specifically, in a common MBHO arrangement, a purchaser will conduct competition among vendors of the carve-out services for a single fixed-length contract (usually for 3 yr). Competition for a contract is focused on price and quality of services.
By 1998, 36 states had received waivers of certain Medicaid rules to establish new managed care programs for mental health services (Office of Inspector General, 2000). In 1999, 30 states offered their low-income disabled individuals some Medicaid managed care program on a mandatory basis (Donohue, Hanson, and Huskamp, 1999). States also increasingly carved out their mental health services, separating mental health coverage from physical health coverage and placing the former under separate financial and administrative arrangements. Nevertheless, the implementation of carve-outs among Medicaid populations is a relatively recent phenomenon. In particular, 10 out of 16 states implemented carve-out programs in 1995 or later. Furthermore, the characteristics of the Medicaid MBHO procurement process and contracts support and illustrate the theories of adverse selection and cost control discussed above. Thirteen of the 16 behavioral health carve-out programs enroll all eligibility categories on a mandatory basis. In 14 of the 16 programs, the state or local government contracts with only one MBHO per service area; thus, these programs do not offer a choice of behavioral health plans, thereby avoiding the problem of adverse selection. Fifteen of the 16 states reimburse MBHOs on a capitated basis for nondrug medical services to better control costs (Donohue, Hanson, and Huskamp, 1999).
These MBHO carve-out implementations may potentially affect the use of new technology by setting limits on utilization or expanding reimbursements for the new technologies and by influencing the practice of medicine through utilization review and application of practice guidelines. Previous studies have found that MBHO carve-outs set prices for services and affect medical practices through the setting of relative prices for different services. Network providers must agree to comply with the carve-outs' utilization review and protocols. Carve-outs thus could affect rates of utilization and medical practice directly through exercising the authority to approve service use and setting up various checkpoints for the dispensing of inpatient and outpatient care (Ma and McGuire, 1998).
Carve-outs also influence medical practices through changing the organization of medical care and hence may affect utilization by determining the ability and commitment of providers to provide innovative care. Carve-outs create networks of specialists, including both specialty inpatient and outpatient providers, thus facilitating economies of specialization and use of specialized expertise to manage care. If new pharmacological treatments are efficacious and have improved side-effect profiles, then we would expect the implementation of carve-outs to increase utilization of new medications and also to constrain nonmedication services. This is especially true if drugs are "off-budget" inputs to the carve-out.
In the case of the Medicaid populations, public policy and political economy also play important roles in the utilization of new technologies. (13) For instance, after a drug has been approved by the FDA, formulary provisions and other restrictions (e.g., caps on the number of prescriptions at one time) may impede the utilization of new medications. In general, however, there are fewer formulary exclusions under Medicaid than under most private insurance, (14) and these will likely not be important factors in the utilization of new medications. For Medicaid, in the case of antidepressants and antipsychotics, the only two drugs affected by formularies and other restrictions have been clozapine and sertraline. (15)
Other market variables, such as price, that are typically envisaged as affecting the utilization of new technology may play a smaller role among the Medicaid populations than in the general population due to institutional structures. Specifically, reimbursements for drugs in the Medicaid program are results of bargaining and regulations. In order for manufacturers' drugs to be included on the Medicaid formulary, under the 1990 Omnibus Budget Reconciliation Act (OBRA), manufacturers were required to enter into a drug rebate program with the federal Department of Health and Human Services. That law was later amended by the Veterans Health Care Act of 1992, whereby a drug manufacturer is required in addition to enter into a national rebate program with the Department of Veteran Affairs in order for drugs to be reimbursed for Medicaid. Currently, Medicaid purchasers pay the lowest prices for branded drugs sold in the United States (Frank, 2001).
III. DATA AND METHODS
A. Data Sources
The data used in this study are compiled from quarterly reports submitted by manufacturers to the Health Care Finance Administration (HCFA), later renamed Centers for Medicare and Medicaid Services (CMS), under reporting requirements from the 1990 OBRA and the Veterans Health Care Act of 1992. Approximately 500 pharmaceutical companies participate in these programs.
The Medicaid Drug Utilization files from HCFA/CMS supply information at the level of an 11-digit National Drug Code (NDC) number on a quarterly basis. The data include product name, units paid, total number of prescriptions reimbursed, and total dollar amount paid. Based on this data source, we create analytical variables for overall antidepressant and antipsychotic prescriptions and dollar expenditures, as well as at the therapeutic class level, by state and by quarter, between 1991 and 2000. NDC numbers can be matched with presentational form, for example, bottles of 100 tablets each having a 20 mg strength. The number of prescriptions reimbursed is then summed up to the therapeutic class level for each state during each time period.
B. Construction of Dependent Variables
There are two sets of dependent variables in the analyses reported below: (a) total number of prescriptions reimbursed by Medicaid for antidepressants, and for antipsychotics and (b) logarithm of (s/1 - s) where s is the unit share of prescriptions reimbursed for a particular drug grouping within the therapeutic class level. Before proceeding with a discussion of econometric models and results, we first provide a description of data trends for these two sets of dependent variables.
IV. DESCRIPTIVE ANALYSIS OF MARKET EXPANSION AND SUBSTITUTION AMONG PSYCHOTROPIC DRUGS
Much growth has taken place in recent years in the use of new antipsychotics and new antidepressants, reflecting effects from both substitution (e.g., new drugs replacing older drugs) and market expansion (e.g., increases in the total number of patients treated by psychotropic drugs). (16) According to Intercontinental Marketing Services (IMS) data (17) for the overall U.S. market (i.e., Medicaid plus non-Medicaid), the average annual growth rates of expenditures on SSRIs, SNRIs, and atypical antipsychotics between 1997 and 2001 were 19.7%, 47.6%, and 64.7%, respectively (see Table 3). During the same time period, decreases occurred for gross expenditures and market shares on older medications. For example, gross expenditures on TCAs and other older antidepressants fell by an average of 2.8% per year between 1997 and 2001, while gross expenditures on the conventional or typical antipsychotics fell by an average 11.1% per year, with the expenditure share of typical antipsychotics falling from 8.5% in 1997 to 1.7% in 2001. Similarly, expenditures shares of the older antidepressants fell by 2.9 percentage points, from 4.7% in 1997 to 1.8% in 2001. In aggregate, gross expenditures for antipsychotics and antidepressants increased by 33.2% and 23.9% per year on average, respectively, between 1997 and 2001.
The IMS data include both Medicaid and non-Medicaid (e.g., private insurance) populations. Among the Medicaid-only population, we find similar patterns of change but different magnitudes of substitution and market expansion effects. As seen in the bottom panel of Table 3, growth rates in gross Medicaid reimbursements for SSRIs, SNRIs, and atypical antipsychotics between 1997 and 2000 were 25.6%, 94.2%, and 54.8%, respectively. Hence, gross Medicaid reimbursements for SSRIs and SNRIs increased by a greater percentage than for the overall population, while reimbursements on atypical antipsychotics increased by a smaller percentage than that for the overall population. Gross reimbursements on older drugs also fell at a lower percentage rate among the Medicaid population than in the overall population. In particular, gross reimbursements for typical antipsychotics fell by 3.4% per year on average rather than the 11.1% reported for the overall population in the IMS data, while Medicaid reimbursements for TCAs and other older antidepressants rose by 9% per year on average rather than the 2.8% decrease reported for the overall population. The average annual percentage growth rates of gross reimbursements for all antidepressants and all antipsychotics were also higher among the Medicaid population than in the overall population (34.1% vs. 23.9% and 45.6% vs. 33.2%, respectively). While the market expansion effects were therefore greater among the Medicaid population than in the overall population, the substitution effect, as measured by changes in expenditures share over the period, was of similar magnitude in the Medicaid and the overall populations. Keeping in mind this finding on the similarity in the patterns of substitution among the general and Medicaid populations, in the regression analyses reported below, we will assess the magnitude of market expansion effects relative to the substitution effect between old and new technologies, separately for carve-out and non-carve-out states.
It is possible that, given changes in prices for the general population and the set price formula between the federal government and the pharmaceutical companies in determining drug prices for Medicaid population, these observed trends in increased expenditures may merely reflect rapid increases in prices for newer drugs. To remove price considerations, we now examine the changing quantity shares of prescriptions for new antidepressants and new antipsychotics among Medicaid populations, thereby focusing on changes in quantities reimbursed between 1991 and 2000.
Figure 1 documents a significant growth in the number of Medicaid prescriptions for antidepressant and antipsychotic medications between 1991 and 2000, from approximately 1 and 1.1 million prescriptions to 5.4 and 3.9 million prescriptions, respectively. Figures 2 and 3 demonstrate the remarkable market expansion in the growth in the number of prescriptions for atypical antipsychotics, SSRIs, and SNRIs, from 0.007 million, 0.18 million, and 0.01 million to 2.9 million, 1.4 million, and 2.9 million between 1991 and 2000, respectively. The combined share of SSRI and other new antidepressant prescriptions grew by approximately 60 percentage points during that period, from 18.4% at the beginning of 1991 to 78.9% at the end of 2000 (see Figure 4). The share of atypical antipsychotics prescriptions increased by about 74 percentage points during the same period, from 0.6% to 74.5% (see Figure 5).
[FIGURE 1 OMITTED]
For modeling purposes, it is useful to note that, as seen in Figure 4, the shares of SSRI and other new antidepressants markedly followed S-shaped diffusion curves, similar to those observed in other studies on the adoption process of new technologies, notably Griliches' seminal hybrid corn study.
V. ECONOMETRIC MODELS FOR CHARACTERIZING THE UTILIZATION OF NEW TECHNOLOGY
The logistic growth curve is commonly used to analyze the process through which new technology comes to replace old technology (Griliches, 1988). The idea underlying the logistic growth model is an epidemic-based formulation describing the spread of an infectious disease. Specifically, if personal contact is important in the adoption of an innovation by a limited population, the process of substituting new technology for old ones may be viewed as akin to the spread of an infectious disease:
(1) [d[n.sub.t]]/dt = [beta][[n.sub.t]/[N.sub.t]]([N.sub.t] - [n.sub.t]),
where [n.sub.t] is the number of individuals who have adopted the innovation at time t, [N.sub.t] is the number of potential adopters, and in most formulations of this equation, [beta] is a parameter reflecting the likelihood of adoption. The number of new adoptions at period t is then equal to the number of remaining potential adopters ([N.sub.t] - [n.sub.t]) multiplied by the probability of adoption. The probability of adoption can in turn be expressed as the product of the proportion of the existing adopters at time t ([n.sub.t]/[N.sub.t]) and the likelihood of adoption ([beta]). In order for the adoption process to be based on imitative behavior (or to display a bandwagon effect), the number of adopters at any time t needs to be a function of the number that have already adopted the innovation.
[FIGURE 2 OMITTED]
The solution to Equation (1) is:
(2) [n.sub.t] = [N.sub.t][1 + exp(-[alpha] - [beta]t)][.sup.-1].
This is the cumulative density function of the logistic frequency distribution, [alpha], a constant that defines the initial starting point of the process, is affected by supply-side factors. Transformation of Equation (2) to logarithms yields:
(3) log([n.sub.t]/[[N.sub.t] - [n.sub.t]]) = [alpha] + [beta]t,
where t is the period variable and is equal to 1, 2, 3,...
[N.sub.t] is determined by demand-side factors. We estimate the following equation:
(4) [N.sub.t] = [[delta].sub.0] + [delta][X.sub.t]
where [X.sub.t] are demand-side factors including the period variable (t).
Dividing Equation (3) through by [N.sub.t] to express the left-hand side in terms of shares ([s.sub.t] = [n.sub.t]/[N.sub.t]) and adding the determinants of [N.sub.t] from Equation (4), we can express Equation (3) as
(5) log([s.sub.t]/[1 - [s.sub.t]]) = [[eta].sub.0] + [eta][X.sub.t],
where [s.sub.t] is the quantity share of users of a set of drugs within a larger therapeutic class. We estimate parameters in Equation (5).
A. Explanatory Variables
The focus of our research is on the relationship between organizational aspects of insurance and the utilization of new mental health medications. An indicator variable that equals 1 after carve-out implementation and 0 otherwise captures the effects of carve-out implementation. Since the characteristics of behavioral carve-outs (e.g., mandatory enrollment, no choice of plan, and competitive procurement process) are relatively similar across states, no other variables are used to describe alternative carve-out characteristics.
[FIGURE 3 OMITTED]
We also control for differences in deflated per capita income by state and number of Medicaid recipients by state. Consistent with traditional technological adoption models, a time counter is included in these regressions to capture the use of new technology over time. To account for time-invariant state characteristics, we allow for state fixed effects within a panel data framework.
This framework yields two sets of equations. The first equation is the "ceiling" or potential market at time t, corresponding to Equation (4):
[FIGURE 4 OMITTED]
(6) [N.sub.t] = [[delta].sub.0] + [[delta].sub.1]Time + [[delta].sub.2]C[O.sub.t] + [[delta].sub.3](Medicaid recipients) + [[delta].sub.4](deflated stpcin) + [[xi].sub.t],
where [N.sub.t] is the total number of prescriptions for antidepressants (or antipsychotics) in the Medicaid population for a specific state, Time is a linear time trend, C[O.sub.t] denotes an indicator that takes on value 1 after a behavioral carveout implementation and is equal to 0 otherwise, Medicaid recipients is the number of Medicaid recipients in that state, and deflated stpcin is the deflated state per capita income. [[xi].sub.t] is a random error term.
The second equation that corresponds with Equation (5) is the rate of substitution between old and new medications of various therapeutic class groupings of antidepressants (or antipsychotics):
(7) log([[s.sub.t]/[1 - [s.sub.t]]]) = [[eta].sub.0] + [[eta].sub.1]Time + [[eta].sub.1]C[O.sub.t] + [[eta].sub.3]log(Medicaid recipients) + [[eta].sub.4]log(deflated stpcin) + [[xi].sub.t],
where [s.sub.t] is the share of prescriptions for antidepressants (or antipsychotics) in the Medicaid population for a specific state and for a specific therapeutic class grouping of antidepressants (or antipsychotics), C[O.sub.t] is the carve-out indicator variable, log(Medicaid recipients) is the log of the number of Medicaid recipients in that state, and log(deflated stpcin) is the log of deflated state per capita income. Parameters in both Equations (5) and (6) are estimated with state fixed effects included. In the case of the antipsychotics equation, we also add a variable NURHOME measuring the number of nursing homes as there has been clinical evidence of high rates of dispensing of antipsychotic medications in treating cognitive impairments and behavioral symptoms in nursing homes despite federal regulations limiting their use (Lindesay, Matthews, and Jagger, 2003; Snowden, Sato, and Roy-Byrne, 2003).
[FIGURE 5 OMITTED]
In a previous study (Ling, Frank, and Berndt, 2006), we reported finding significant incentives for cost shifting to pharmacotherapies following carve-out implementations. Here, our goal is to differentiate between the increase in utilization of new antidepressants and new antipsychotics from cost-shifting incentives (a shift in the ceiling, [N.sub.t], analogous to the idea of a market expansion effect in that the effects are uniform across drug classes) and incentives to adopt new technologies (a diffusion effect, [s.sub.t], analogous to the idea of substitution effects in that one might expect greater rates of growth in shares among newer drugs). In order to differentiate between the magnitudes of these two effects, we compare findings from separate analyses for antipsychotics and antidepressants. We exclude sertraline and clozapine from the analyses due to restrictive policies by a number of state Medicaid agencies on these two drugs but report the sensitivity of regression results to excluding these drugs in Appendix C.
B. Analytic Approach and Identification Strategies
The empirical analysis of substitution between old and new medications over time consists of a series of fixed-effect regressions on total prescriptions and log-shares of prescriptions. The carve-out effect is identified through the different observed timing of carve-out implementations among states. Through the carve-out indicator variable, we identify the average before-and-after difference in the utilization of new medications for states that have implemented behavioral health carve-out relative to those that have not implemented carve-outs.
This identification strategy is clearly based on the implicit assumption of an exogenous carve-out effect on pharmaceutical spending and utilization. To assess the validity of this identifying assumption, we also examine and compare characteristics of states that had implemented carve-outs relative to states that had not in the late 1980s or early 1990s (see Table 1). While other studies (e.g., Danzon and Pauly, 2001) have suggested a two-way causal relationship between insurance coverage and design and changes in health care and pharmaceutical spending, our analyses revealed no endogenous relationship between pharmaceutical spending and state-specific implementations of the behavioral carve-out. Specifically, on average, states that have implemented behavioral carve-outs were found to have higher percentage of persons enrolled in managed care in 1990 (14.25% vs. 6.4%), greater number of Medicaid recipients in 1991 (833,313 vs. 402,853), larger growth in total Medicaid expenditures between 1985 and 1990 (110.02% vs. 86.5%), and higher growth in state mental health agency per capita expenditures between 1981 and 1990 (88.47% vs. 77.75%) than states that had not implemented behavioral carve-outs. However, we do not find material large differences in the growth rates of total prescription drug spending (225.53% vs. 214.96%) and growth rates in Medicaid spending on drugs and other medical nondurables between 1980 and 1990 for states that had implemented carve-outs (290.98%) relative to states that had not implemented them (270.87%). We also undertook Cox proportional hazard and logistic multivariate analyses of the timing and probability of carve-out implementation but found no statistically significant relationships with Medicaid drug spending (see Appendix B for further details). As a result, we take the decision and timing to implement behavioral carve-outs among the Medicaid populations as being largely exogenous to Medicaid prescription drug spending and utilization.
VI. REGRESSION RESULTS
Table 4 presents results of the linear regression of the total number of prescriptions for antidepressants and antipsychotics (the "ceiling" or expansion effect equation), allowing for state fixed effects and excluding sertraline and clozapine. (18) The time trend has a positive and significant impact on the number of prescriptions, with a coefficient of 647 prescriptions per quarter for antidepressants and 400 prescriptions per quarter for antipsychotics (each p < 0.001). The coefficient on the carve-out indicator is positive and significant for antidepressants, while the coefficient on carve-out indicator for the antipsychotics equation is negative and smaller in absolute magnitude. On average, states with carve-out implementation have an addition of 7,784 antidepressant prescriptions each quarter relative to states without carve-out implementation (p < 0.001). States with carve-out implementation have 4,074 fewer antipsychotics prescriptions per quarter relative to states without carve-out implementation (p < 0.02). Hence, the carve-out ceiling effect for antidepressants is about twice the magnitude of the ceiling effect for antipsychotics and different in sign. As expected, the total number of Medicaid recipients has a positive and significant impact on the total number of antidepressant and antipsychotic prescriptions reimbursed (p < 0.0001). We also find positive and significant income effects (p < 0.001). Specifically, higher deflated per capita state income is significantly related to a greater number of prescriptions. A greater number of nursing homes is also associated with an increase in the number of antipsychotic prescriptions (p < 0.001).
Table 5 presents the regression results from the logistic share models for varying classes of antidepressants and antipsychotics (the "substitution" equation). In general, we find positive and significant effects of time trend on the utilization for newer drugs (SSRIs, other new antidepressants, and atypical antipsychotics) but negative and significant effects for older drugs (tricyclics, tetracyclics, MAOIs, and typical antipsychotics). For example, the coefficients on time trend are 0.056, 0.099, and 0.280 for SSRIs, other new antidepressants, and atypical antipsychotics, respectively (p < 0.0001) compared to negative coefficients for tricyclics (-0.065), tetracyclics (-0.076), and MAOIs (-0.043), respectively (all p < 0.0001). Surprisingly, we find negative and occasionally significant coefficients on deflated per capita income for all older drugs and even for newer drugs such as SSRIs and atypical antipsychotics although not for other new antidepressants.
The coefficients on the carve-out indicator variable are positive for newer drugs (SSRIs, other new antidepressants, and atypical antipsychotics), as well as for the tetracyclics but are negative for the older MAOIs and tricyclics; only the positive SSRI and negative tricyclic estimates reach statistical significance. Hence, the adoption or utilization rate of SSRIs is significantly greater in states with carve-out implementation than in those without carve-outs.
The coefficient on the logarithm of number of Medicaid recipients is positive and significant for tricyclics and tetracyclics, negative and significant for MAOIs, other new antidepressants, and atypical antipsychotics, and positive but insignificant for the SSRIs. The number of nursing homes has a positive but insignificant effect on the adoption rate of atypical antipsychotics.
Several reasons may account for the positive but insignificant effect of carve-out implementation on the adoption process of atypical antipsychotics. As hypothesized above, the need for specialty care monitoring and the need for high-skilled labor and specialty knowledge in the prescriptions of the atypical antipsychotics may have placed limitations on the use of antipsychotics due to their complementarity with high-skilled specialty care. Given the complexity of the treatments and the illness, the intensity of use of the atypical antipsychotics was essentially unaffected by the implementation of MBHO carve-outs, consistent with the technology-skill complementarity hypothesis and in contrast to the general-purpose technologies feature of SSRIs.
VII. CONCLUSION AND POLICY IMPLICATIONS
New medications, advances in diagnostic and surgical procedures, and other remarkable technological innovations have transformed health care delivery, improved the quality of life, and prolonged the longevity of millions of Americans. Particularly in the area of mental health, the introduction of new and more effective psychotropic medications, along with developments in psychotherapy and the growth of behavioral health carve-outs, has significantly altered the organization of and treatments for mental illness. These developments not only hold significance for the study of health care policy and delivery but also have important implications for the broader study of the role of contractual innovation and cost incentives in the substitution between old and new technologies.
In addition, our model and empirical findings hold particular significance for Medicaid program design and spending. Medicaid has been a major purchaser and source of payment for prescription drugs. Prescribed drugs have been one of the most frequently used and fastest growing segment of Medicaid services, with over 26 million Americans receiving at least one drug paid for by Medicaid in 2000 and Medicaid drug expenditures amounting to approximately $29.7 billion in 2002 (Bruen and Ghosh, 2004). Moreover, central nervous system medications, the broad therapeutic category that psychotropic drugs fall under, was found to be the most expensive drug category for 29 State Medicaid Programs in 1998 (Baugh et al., 2004). The data we analyze reveal significant growth in spending for antidepressants and antipsychotics among the Medicaid populations, growth that even outpaced spending growth in the overall population. We also find faster growth under Medicaid for new drug categories such as SSRIs and SNRIs than for the overall population. Given the differences in the timing of state-specific implementation of carve-outs in the Medicaid population, we have examined how the significant increase in utilization of new medications might be attributed to carve-out implementation. Following Griliches' hybrid corn model, we employ the logistic equation to model the utilization of new medications. Notably, we find that behavioral health carve-outs raise the number of potential users for the newest antidepressants but do not do the same for the newest antipsychotics. We also find that the effects of carve-out implementations on the utilization of new antidepressants and new antipsychotics differ in magnitude and statistical significance.
Specifically, after accounting for possible restrictions from formularies and other policies, we find positive and significant effects of carve-out implementation on the adoption of SSRIs. We find positive but insignificant effects of carve-out implementation on the adoption of second-generation atypical antipsychotics. (19) We attribute the differences in adoption patterns among antidepressants and antipsychotics to the degree of user-friendliness and the degree of technology-skill bundling with other specialty services. Similar to the hypothesis of general-purpose technologies, we expect technology that is relatively easy to use and treats more related conditions to therefore be accessible to more users and to have more rapid penetration. Perhaps even more importantly, financial incentives from carve-outs may encourage further use and adoption of technology that has a lower specialty skill requirement and that is used to treat illnesses in a general care setting. The lack of a significant carve-out effect on the utilization of the second-generation atypical antipsychotics is consistent with the constraining impact of carve-outs associated with requisite technology-skill complementarity. Unlike general-purpose technologies, those involving technology-skill complementarity are constrained by the required specialist treatments as carve-outs ration specialist care. Both these hypotheses require further empirical analyses. The adoption paths of new technologies that can be characterized as varying in their different skill requirements and user-friendliness in other medical fields merit additional scrutiny.
Our study documents the importance of financial incentives in the utilization of new technology. Clearly, the adoption of new medical technology has significant health and quality-of-life benefits as well as pecuniary and nonpecuniary costs. These benefit issues are outside the scope of the current study but are nonetheless important in the overall assessment of technological adoption and utilization.
This study is particularly relevant to the current debate on universal coverage, Medicaid reforms, as well as the recent implementation of the Medicare Prescription Drug, Improvement, and Modernization Act (MMA) of 2003.
The findings in our study suggest that the implementation of Medicare Part D may slow the utilization of new medications overall by raising the cost faced by some patients (Donohue, 2006). The MMA established a new prescription drug benefit (Medicare Part D) for Medicare beneficiaries in 2006. Preceding the implementation of Medicare Drug Benefit, Medicaid had been responsible for dually eligible beneficiaries' drug coverage. With the implementation of Medicare Part D in 2006, these dually eligible beneficiaries were moved from Medicaid drug coverage to Medicare prescription drug plans that typically would have greater cost-sharing provisions and hence would increase financial burdens to patients. For example, with greater cost sharing, these beneficiaries would face a co-payment of $1-$2 for generic drugs and $3-$5 for brand-name drugs under the new plan rather than a co-payment of $3 or less under Medicaid drug plan (Elliott et al., 2005).
Due to changes in financial incentives, we expect to see greater rationing of antipsychotic medications in nursing homes, which previously had been a driver in Medicaid drug spending. Nevertheless, the actual effects of these policies are difficult to quantify since the cost-sharing provisions are small in magnitude. In addition, the cost-sharing provisions are unevenly applied to different populations.
Another insight from our study is that fragmentation in the overall benefit design and coverage may create incentives for cost shifting and changes in medical care utilization with unintended consequences. While the debate regarding universal coverage tends to focus on the working poor who do not qualify for Medicaid or Medicare, our study suggests that there is no magical solution to contain costs and provide access. Health plans respond strongly to financial incentives in rationing health care. For instance, while the intent of the MMA is to increase consumer choice and to expand drug coverage, studies have found that through restrictive formularies or requiring prior authorization for expensive drugs, health plans have attempted to select potential enrollees on the basis of risk (Donohue, 2005). Hence, while MBHO carve-outs address the problem of adverse selection by design and limit consumer choice, the fragmented nature of mental health and drug coverage creates opportunities to shift the burden of payment among payers (Frank and Glied, 2006b). In this way, piecemeal responses to growth in mental health spending and medical care coverage overall are not likely to create a sustainable long-term solution.
APPENDIX A Policy Restrictions on Utilization and Prescriptions of Prescription Drugs among Medicaid Populations
Mental Health Drugs Requiring Prior Carve-Out/ Authorization Formulary Limits on All Drugs States by State Exclusions Prescriptions Carve-Out (a) Arizona [check] California [check] [check] Colorado Florida [check] [check] [check] Hawaii [check] [check] [check] [check] Iowa [check] [check] [check] (a) Kentucky [check] [check] [check] Massachusetts [check] [check] [check] Michigan [check] [check] [check] [check] Nebraska [check] [check] [check] [check] (a) Pennsylvania [check] [check] [check] Tennessee Texas Utah [check] [check] [check] [check] (a) Washington [check] [check] Wisconsin (a) All prescription drugs are carved out and covered by a specialty firm.
APPENDIX B. IDENTIFICATION ANALYSES
To examine the possibility of an endogenous timing of the MBHO carve-out implementation, we modeled the probability and timing of carve-out implementation in a political economy framework, whereby policy makers and politicians do not necessarily maximize social welfare but instead are also motivated by self-interests and expected gains. We hypothesize that the decision for policy makers to implement carve-outs may be affected by both budgetary pressures and fiscal constraints to reduce state spending on mental health, as well as the political climate and political persuasion of constituents. We employ both Cox proportional hazard and logistic regression models to examine the timing and probability of carve-out implementation and their relationships with Medicaid drug spending.
Our explanatory variables for carve-out implementation include both institutional variables and characteristics of state Medicaid spending. Specifically, we include the percentage of individuals enrolled in managed care in the state to capture managed care penetration and to proxy for the pressure from private sector institutional changes in health insurance in reforming the public mental health care system. (In addition, managed care employs strong mechanisms to ration care and shorten inpatient care, thereby potentially shifting care from private to public institutions.) Mental health spending variables such as state mental health agency spending per capita is used to proxy for budgetary pressures to cap public mental health spending. Medicaid prescription drug spending per recipient is included to test for the possible endogenous relationship between carve-out implementation and Medicaid drug spending, while the number of supplemental security income recipients is used to proxy for the size and burden of the Medicaid system. Other political economy variables include total state revenue per capita to proxy for the amount of fiscal budget constraint, Republican representation in the state Senate and the House to proxy for political affiliation, and policy stance of constituents in the state.
The regression results are presented below. HMO penetration is positive and significant in increasing the probability (1.243, p < 0.05) and the hazard of carve-out implementation (1.156, p < 0.05). An increase in HMO penetration by 1% increases the probability of carve-out implementation by 24.3%. Hence, institutional changes in private health insurance place strong pressures for reforms in the public sector. State mental health agency spending per capita is significant and negative in affecting the probability of carve-out implementation (0.925, p < 0.05). An increase in state mental health spending per capita actually lowers the probability of carve-out implementation by 7.5%. This finding counters our hypothesis that public sector mental health delivery reforms may be a response to budgetary pressures. Notably, we find that Medicaid prescription drug spending per recipient and total state revenue per capita are also not significant in influencing the probability and timing of carve-out implementation. Neither of the political economy variables are found to be significant in the regressions.
These findings support our identification strategy in that prescription drug spending does not seem to motivate carve-out implementation. Specifically, the evidence presented here provides empirical support for our assumption that the timing and probability of carve-out implementation are predetermined and exogenous to our modeling of Medicaid prescription drug spending and utilization.
Models of Medicaid Mental Health Carve-Out Implementation A. Logit Model of Carve-Out Implementation
Standard Odds Ratio Errors (a) Z Ratio HMO penetration in 1990 1.243* 0.115 2.350 State mental health agency per capita 0.925* 0.036 -2.030 expenditures in 1987 Total Medicaid drug spending per 0.870 0.312 -0.390 Medicaid recipient in 1990 Number of SSI recipients in 1991 1.000 0.000 1.570 Total state revenue per capita in 1991 0.484 0.450 -0.780 Percentage of Republican state 0.510 0.665 -0.520 representatives in the Senate in 1990 Percentage of Republican state 2.993 5.531 0.590 representatives in the House in 1990 Note: Pseudo [R.sup.2] = 0.291; [chi square] = 8.25. (a) Adjusted for heteroskedasticity. *p < 0.05.
B. Cox Proportional Model of Time to Carve-Out Implementation
Hazard Ratio Standard Errors (a) Z Ratio HMO penetration in 1990 1.156* 0.063 2.670 State mental health agency 0.963 0.032 -1.130 per capita expenditures in 1987 Total Medicaid drug spending 0.876 0.247 -0.470 per Medicaid recipient in 1990 Number of SSI recipients in 1.000 0.000 -0.390 1991 Total state revenue per 0.459 0.406 -0.880 capita in 1991 Percentage of Republican 0.393 0.371 -0.990 state representatives in the Senate in 1990 Percentage of Republican 1.384 1.616 0.280 state representatives in the House in 1990 Note: Pseudo [R.sup.2] = 0.114; [chi square] = 36.23. (a) Adjusted for heteroskedasticity. *p < 0.05.
APPENDIX C. REGRESSION ANALYSES INCLUDING SERTRALINE AND CLOZAPINE
Regression on Prescriptions for Antidepressants and Antipsychotics (Including All Antidepressants and All Antipsychotics)
Antidepressants Prescriptions Coefficient (Standard Error) pValues Time 877.81** (117.84) <0.0001 Carve-out indicator 4451.73* (2496.43) 0.075 Medicaid recipients 0.05** (<0.005) <0.0001 Deflated State 4.63** (0.90) <0.0001 Per Capita Income) Number of nursing homes Constant -100,287** (19,148.67) <0.0001 N = 1,492, [R.sup.2] = 0.675, F(4,1438) = 322.53 Antipsychotics Prescriptions Coefficient (Standard Error) p Values Time 749.71** (100.90) <0.0001 Carve-out indicator 391.38 (2128.53) 0.854 Medicaid recipients 0.03** (<0.005) <0.0001 Deflated State 1.36** (0.77) 0.079 Per Capita Income) Number of nursing homes 25.94** (11.70) 0.027 Constant -14,454.23 (16,295.84) 0.375 N = 1,492, [R.sup.2] = 0.836, F(5,1437) = 136.27 **p < 0.05; *p < 0.10.
Acemoglu, D. "Technical Change, Inequality, and the Labor Market," Journal of Economic Literature, 40(1), 2002, 7-72.
Baugh, D. K., P. L. Pine, S. Blackwell, and G. Ciborowski. "Medicaid Prescription Drug Spending in the 1990s: A Decade of Change," Health Care Financing Review, 25(3), 2004, 5-23.
Berman, E., J. Bound, and Z. Griliches. "Changes in the Demand for Skilled Labor within U.S. Manufacturing Industries: Evidence from the Annual Survey of Manufacturing," Quarterly Journal of Economics, 109(2) (1994), 367-398.
Berndt, E. R. "Trends and Drivers of Expenditures on Psychotropic Drugs in the US," presentation at National Institute of Mental Health Pharmacoeconomics Research workshop: Public Health Perspectives, Bethesda, MD, 23 May 2002.
Berndt, E. R., S. H. Busch, and R. G. Frank. "Treatment Price Indexes for Acute Phase Major Depression," in David M. Cutler and Ernst R. Berndt, eds., Medical Care Output and Productivity, NBER Studies in Income and Wealth, Volume 62, Chicago and London: University of Chicago Press, 2001.
Berndt, E. R., R. G. Frank, and T. G. McGuire. "Alternate Insurance Arrangements and the Treatment of Depression: What are the Facts?" American Journal of Managed Care, 3(2), 1997b, 243-252.
Berndt, E. R., R. S. Pindyck, and P. Azoulay. "Network Effects and Diffusion in Pharmaceutical Markets: Antiulcer Drugs," Journal of Industrial Economics, Vol. LI, June 2003, 243-270.
Berndt, E. R., L. T. Bui, D. H. Lucking-Reiley, and G. L. Urban. 'The Roles of Marketing, Product Quality and Price Competition in the Growth and Composition of the U.S. Anti-Ulcer Drug Industry,' chapter 7 in Timothy F. Bresnahan and Robert J. Gordon, eds., The Economics of New Goods, Studies in Income and Wealth, Volume 58, Chicago: University of Chicago Press for the National Bureau of Economic Research, 277-322, 1997a, 277-322.
Berndt, E. R., A. Bir, S. H. Busch, R. G. Frank, and S. -L. T. Normand. "The Medical Treatment of Depression, 1991-1996: Productivity Inefficiency, Expected Outcome Variations, and Price Indexes," Journal of Health Economics, 21(3), 2002, 373-396.
Bresnahan, T.F. and M. Trajtenberg, "General Purpose Technologies: 'Engines of Growth'?" Journal of Econometrics, 65(1), 1995, 83-108.
Bruen, B., and A. Ghosh. "Medicaid Prescription Drug Spending and Use," Kaiser Commission on Medicaid and the Uninsured, Washington, D.C.: Kaiser Commission on Medicaid and the Uninsured, June 2004.
Cutler, D. M. and M. McClellan. "Technological Change in Medicine Worth it?" Health Affairs, 20(5), 2001, 11-29.
Danzon, P. M. and M. V. Pauly. "Insurance and New Technology: From Hospital to Drugstore," Health Affairs 20(5), 2001, 86-100.
Depression Guideline Panel. Depression in Primary Care, Vol. 2, Rockville, MD: U.S. Department of Health and Human Services, AHCPR publication, 1993, 93-0551.
Donohue, J. M. "The Economics of the New Medicare Drug Benefit: Implications for People with Mental Illnesses," Psychiatric Services, 56(6), 2005, 645-647.
Donohue, J. M. "Mental Health In the Medicare Part D Drug Benefit: A New Regulatory Model?" Health Affairs, 25(3), 2006, 707-719.
Donohue, J. M., K. W. Hanson, and H. A. Huskamp. "Persons with Disabilities and Medicaid Managed Care: Issues in Programmatic Design, Enrollment, Procurement, and Contracting," Unpublished Report prepared for the Center for Health Care Strategies, Inc., and the Robert Wood Johnson Foundation, November 1999.
Elliott, R. A., S. R. Majumdar, M. R. Gillick, and S. B. Soumerai. "Benefits and Consequences for the Poor and the Disabled," New England Journal of Medicine, 353(26), 2005, 2739-2741.
Ellison, S. F., I. Cockburn, Z. Griliches, and J. Hausman. "Characteristics of Demand for Pharmaceutical Products: An Examination of Four Cephalosporins," Rand Journal of Economics, 28(3), 1997, 426-446.
Frank, R. G. "Prescription Drug Prices: Interpreting Differences." Health Affairs, 20(2), 2001, 115-128.
Frank, R. G, E. R. Berndt, and S. H. Busch. "Price Indexes for the Treatment of Depression," in Triplett, J.E., ed., Measuring the Prices of Medical Treatments, Washington, D.C.: Brookings Institute Press, 1999, 72-102.
Frank, R. G., and S. A. Glied. Better but Not Well: Mental Health Policy in the United States since 1950, Baltimore: John Hopkins University Press, 2006a.
Frank, R. G., and S. Glied. "Changes in Mental Health Financing Since 1971: Implications for Policymakers and Patients," Health Affairs, 25(3), 2006b, 601-613.
Frank, R. G., and T. G. McGuire. "The Economic Functions of Carve-outs in Managed Care," The American Journal of Managed Care, 4(SP), 1998, SP31-SP39.
Frank, R. G., H. A. Huskamp, T. G. McGuire, and J. P. Newhouse. "Some Economics of Mental Health 'Carve-Outs," Archives of General Psychiatry, Vol. 53 (1996), 933-937.
Goldin, C., and L. F. Katz. "The Origins of Technology-Skill Complementarity," Quarterly Journal of Economics, Vol. 113, June 1998, 683-732.
Goldman, W., J. McCulloch, and R. Sturm. "Costs and Use of Mental Health Services Before and After Managed Care," Health Affairs, Vol. 17, March/April 1998, 40-52.
Grazier, K. L., L. L. Eselius, T. Hu, K. K. Shore, and W. A. G'Sell. "Effects of a Mental Health Carve-Out on Use, Costs, and Payers: A Four-Year Study," The Journal of Behavioral Health Services and Research, Vol. 26, 1999, 381-389.
Griliches, Z. Technology, Education, and Productivity: Early Papers with Notes to Subsequent Literature, Oxford and New York: Blackwell, 1988.
Griliches, Z. "Hybrid Corn: An Exploration in the Economics of Technological Change", in Wolff, E.N., ed., The Economics of Productivity, Volume 1. Elgar Reference Collection. International Library of Critical Writings in Economics, Vol. 77, 1997. Cheltenham, U.K. and Lyme, N.H.: Elgar, 12-33, originally published in 1957.
Helpman, E., and M. Trajtenberg. "Diffusion of General Purpose Technologies," in Helpman, E., ed., General Purpose Technologies and Economic Growth, Cambridge and London: MIT Press, 1998, 85-119.
The Judge David L. Bazelon Center for Mental Health Law. Medicaid Formulary Policies: Access to High-Cost Mental Health Medications, November 1999. http:www.webcom.com/bazelon/formulary.html.
Leslie, D. L. and R. A. Rosenheck. "From Conventional to Atypical Antipsychotics and Back: Dynamic Processes in the Diffusion of New Medications," American Journal of Psychiatry, Vol. 159, 2002, 1534-1540.
Lindesay, J., R. Matthews, and C. Jagger. "Factors Associated with Antipsychotic Drug Use in Residential Care: Changes Between 1990 and 1997," International Journal of Geriatric Psychiatry, Vol. 18, 2003, 511-519.
Ling, D. C., E. R. Berndt, M. Kyle. "Deregulating Direct to Consumer Marketing of Prescription Drugs: Effects on Prescription and Over-the-Counter Product Sales," 2002, Journal of Law and Economics: 691-723.
Ling, D. C., R. G. Frank, and E. R. Berndt. "Cost-Shifting and Implementation of Behavioral Health Carveouts among Medicaid Populations," 2006, unpublished manuscript.
Ma, C. A. and T. G. McGuire. "Costs and Incentives in a Behavioral Health Carve-Out," Health Affairs, Vol. 17, March/April 1998, 53-69.
Office of Inspector General. Mandatory Managed Care: Early Lessons Learned by Medicaid Mental Health Programs, Department of Health and Human Services, OEI-04-97-00343, January 2000.
Schwalberg, R., H. Bellamy, M. Giffin, C. Miller, S. S. Williams, and L. Elam. Medicaid Outpatient Prescription Drug Benefits: Findings from a National Survey and Selected Case Study Highlights, Kaiser Commission on Medicaid and the Uninsured, www.kff.org, October 2001.
Snowden, M, K. Sato, and P. Roy-Byrne, "Assessment and Treatment of Nursing Home Residents with Depression or Behavioral Symptoms Associated with Dementia: A Review of the Literature," Journal of American Geriatric Society, Vol. 51, Sept 2003, 1305-1317.
U.S. Department of Health and Human Services. Mental Health: A Report of the Surgeon General--Executive Summary. Rockville, MD: U.S. Department of Health and Human Services, Substance Abuse and Mental Health Services Administration, Center for Mental Health Services, National Institutes of Health, National Institute of Mental Health, 1999.
Vogelsang, I., "Economic Aspects of Mental Health Carve-Outs," The Journal of Mental Health Policy and Economics, Vol. 2, 1999, 19-41.
DAVINA C. LING, ERNST R. BERNDT and RICHARD G. FRANK*
*Research support from the National Institute of Mental Health grant R01MH43703 (to Frank and Ling), is gratefully acknowledged. We have benefited from discussions with Iain Cockburn and participants in the NBER Productivity Lunch. The views expressed in this paper are those of the authors only, and do not necessarily reflect those of any institutions with which they are related, or of any research sponsor.
Ling: California State University, Fullerton, Department of Economics, 800 N. State College Blvd, Fullerton, CA 92834. Phone (714) 278-8216, Fax (714) 278-3097, Email firstname.lastname@example.org
Berndt: National Bureau of Economic Research, Sloan School of Management, Massachusetts Institute of Technology, 50 Memorial Drive, Cambridge, MA 02142. Phone 617-253-2665, Fax 617-258-6855, Email email@example.com
Frank: National Bureau of Economic Research, Harvard Medical School, Harvard University, Department of Health Care Policy, 180 Longwood Avenue, Boston, MA 02115. Phone 617-432-0178, Fax 617-432-1219, Email firstname.lastname@example.org
CMS: Centers for Medicare and Medicaid Services
FDA: U.S. Food and Drug Administration
HCA: Heterocyclic Antidepressants
HCFA: Health Care Finance Administration
HMO: Health Maintenance Organization
IMS: International Marketing Services
MAOI: Monoamine Oxidase Inhibitor
MBHO: Managed Behavioral Health Care Organization
MMA: Medicare Prescription Drug, Improvement, and Modernization Act
NBER: National Bureau of Economic Research
NDC: National Drug Code
OBRA: Omnibus Budget Reconciliation Act
OCD: Obsessive-Compulsive Disorder
R & D: Research and Development
SNRI: Serotonin Norepinephrine Reuptake Inhibitor
SSRI: Selective Serotonin Reuptake Inhibitor
TCA: Tricyclic Antidepressants
1. For more theoretical and detailed discussions of carve-outs using transaction costs and industrial organization frameworks, see Ma and McGuire (1998) and Vogelsang (1999).
2. For empirical evidence of cost shifting to prescription drugs, see Ling, Frank, and Berndt (2006).
3. For a longer study period, one would also want to account for the influence of insurance reimbursements on incentives for innovation and drug development. In particular, producers often depend upon insurance reimbursement to create demand for the adoption and use of their innovations and indeed base their R & D investment decisions in part on the expected market for their future new products. See Danzon and Pauly (2001) for a discussion of the endogenous relationship between insurance and technological change and Berndt, Pindyck, and Azoulay (2003) who used a similar rationale in treating the regulatory process as predetermined.
4. See Griliches (1988) for a review of the literature of technology diffusion.
5. These findings are based on the authors' analyses using the Medicaid Drug Utilization files between 1991 and 2000 (http://cms.hhs.gov/medicaid/drugs).
6. The carve-outs considered in this paper are direct managed behavioral health carve-out contracts between states or regions and MBHOs and not states with managed care and managed behavioral health carve-out contracts between managed care and MBHOs. Since it is rare for individuals with mental illnesses to be covered under Medicaid Health Maintenance Organizations (HMOs), our analyses capture most of the individuals with mental illnesses.
7. See Bresnahan and Trajtenberg (1995) and Helpman and Trajtenberg (1998).
8. Based on the concept of capital-skill complementarity discussed in Goldin and Katz (1998) and Acemoglu (2002).
9. See Bresnahan and Trajtenberg (1995) and Helpman and Trajtenberg (1998) for more detailed discussions of the concept of general-purpose technologies.
10. For a discussion of evidence of biased technological progress and technology-skill complementarity, see Goldin and Katz (1998) and Acemoglu (2002).
11. See Vogelsang (1999).
12. See Frank et al. (1996), Ma and McGuire (1998), Frank and McGuire (1998), and Vogelsang (1999) for discussion regarding the functions and rationales of mental health "carve-outs."
13. Professional associations and interest groups are important in influencing formulary decisions as well. In particular, the Massachusetts Psychiatric Society was instrumental in changing the guidelines of one of the largest health plans in the state to promote SSRIs as the first-line treatment for depression. Groups such as the National Alliance for the Mentally Ill have been outspoken and have strongly opposed any restrictions on access to new medications such as atypical antipsychotics and SSRIs (The Judge David L. Bazelon Center for Mental Health Law, 1999).
14. Only eight states maintained closed formularies in 1999, while 33% of commercial HMO formularies in the private sector were considered closed and 52% of commercial HMO formularies were considered "selective" or "partially closed" in 1997. But considerations regarding Medicaid formularies have become important in the late 1990s and early 2000s due to recent implementations of more stringent restrictions on utilization of prescription drugs among the Medicaid population in a number of states (see Appendix A for policy restrictions on utilization and prescriptions for prescription drugs among the Medicaid populations).
15. According to Schwalberg et al. (2001), the use of clozapine requires prior authorization in states such as Alaska, Hawaii, Mississippi, and South Dakota. The use of sertraline requires prior authorization in states such as Montana.
16. See Cutler and McClellan (2001) for a detailed description of these two concepts.
17. Calculations presented by Berndt (2002). IMS's Retail Perspective[TM] tracks monthly shipments from manufacturers and wholesalers to retail warehouses and outlets. The revenue data are those to manufacturers and wholesalers and not to the retail outlets (who add retail margins). Although revenues are net of chargebacks (discounts given purchasers and channeled through wholesalers), rebates (payments made to providers who often do not take title to the pharmaceuticals, e.g., managed care organizations) are not included in the IMS revenue data, nor are prompt payment discounts. Many branded and generic pharmaceutical companies purchase and utilize the IMS data for their internal research. Examples of academic studies that use IMS data to examine prescription drug demand include Berndt et al. (1997a), Ling, Berndt, and Kyle (2002), and Ellison et al. (1997).
18. Regressions including sertraline and clozapine yield results that are qualitatively similar but quantitatively dissimilar (see Appendix C for results including sertraline and clozapine).
19. Second-generation atypical antipsychotics include risperidone, olanzapine, quetiapine, and ziprasidone, while first-generation atypical antipsychotics usually refer only to clozapine.
TABLE 1 States with Medicaid Behavioral Health Carve-Out Programs SSI Medicaid Beneficiaries are State Implementation Date Mandated to Participate Arizona 1990 Yes Utah 1991 Yes Massachusetts 1992 Yes Washington 1993 Yes Wisconsin 1993 No Hawaii 1994 No California 1995 No Colorado 1995 Yes Nebraska 1995 Yes Florida 1996 Yes Pennsylvania 1996 Yes Tennessee 1996 Yes Kentucky 1997 Yes Michigan 1998 Yes Iowa 1999 Yes Texas 1999 Yes TANF Medicaid Beneficiaries State are Mandated to Participate Reimbursement Form Arizona Yes Capitated Utah Yes Capitated (only CMHCs) Massachusetts Yes Risk sharing Washington Yes Capitated Wisconsin No Capitated Hawaii No Capitated California No State funds/cost-settled federal funds Colorado Yes Capitated (with a cap on profits) Nebraska Yes Capitated Florida Yes Capitated Pennsylvania Yes Capitated Tennessee Yes Capitated Kentucky Yes Capitated Michigan Yes Capitated Iowa Yes Capitated (with a cap on profits) Texas Yes Capitated Notes: CMHC = community mental health centers; SSI = Supplemental Security Income; TANF = Temporary Assistance for Needy Families. Source: Donohue, Hanson, and Huskamp (1999). TABLE 2 FDA Approvals of New Psychotropic Drugs SSRIs Prozac (fluoxetine) December 1987 Zoloft (sertraline) December 1991 Paxil (paroxetine) December 1992 Luvox (fluvoxamine) December 1994 Celexa (citalopram) July 1998 Other antidepressants Effexor (venlafaxine) December 1993 Serzone (nefazodone) December 1994 Wellbutrin (bupropion) December 1985 Remeron (mirtazapine) June 1996 Atypical antipsychotics Clozaril (clozapine) September 1986 Risperdal (risperidone) December 1993 Zyprexa (olanzapine) September 1996 Seroquel (quetiapine) September 1997 Geodon (ziprasidone) February 2001 TABLE 3 Overall and Medicaid-Only Expenditures on Antidepressants and Antipsychotics ($ Millions) 1997 2001 (A) Overall 1997 Share (%) 2001 Share (%) AAGR 97-01 (%) Antipsychotics Atypicals 1,434 84.2 5,147 96.4 64.7 Typicals, other 145 8.5 91 1.7 -11.1 Not orals, solids 123 7.3 99 1.8 -5.4 Totals 1,704 100.0 5,365 100.0 33.2 Antidepressants SSRIs 4,204 81.8 8,621 71.2 19.7 SNRIs 690 13.4 3,273 27.0 47.6 TCAs, others 242 4.7 216 1.8 -2.8 Totals 5,137 99.9 12,110 100.0 23.9 1997 2000 (B) Medicaid only 1997 Share (%) 2000 Share (%) AAGR 97-00 (%) Antipsychotics Atypicals 756 84.1 2,000 94.0 54.8 Typicals, other 143 15.9 128 6.0 -3.4 Totals 899 100.0 2,128 100.0 45.6 Antidepressants SSRIs 508 78.0 898 68.2 25.6 SNRIs 93 14.3 355 27.0 94.2 TCAs, others 51 7.8 64 4.9 9.0 Totals 651 1,317 34.1 Notes: SNRIs include bupropion, venlafaxine, mirtazapine, and nefazodone. Source: (A) IMS Health Retail and Provider Perspective Audit and (B) Medicaid Drug Utilization Data from CMS. TABLE 4 Regression on Prescriptions for Antidepressants and Antipsychotics with State Fixed Effects Antidepressants Prescriptions (a) Coefficient (Standard Error) p Values Time 647.34** (93.44) <0.0001 Carve-out indicator 7784.43** (1979.54) <0.0001 Medicaid recipients 0.042** (<0.005) <0.0001 Deflated state per 4.047** (0.71) <0.0001 capita income Number of nursing homes Constant -86,335.38** (15,183.95) <0.0001 N = 1,492, [R.sup.2] = 0.709, F(4,1438) = 363.51 Antipsychotics Prescriptions (b) Coefficient (Standard Error) p Values Time 399.72** (79.98) <0.0001 Carve-out indicator -4073.76** (1687.24) 0.016 Medicaid recipients 0.018** (<0.005) <0.0001 Deflated state per 2.025** (0.61) 0.001 capita income Number of nursing 33.04** (9.28) <0.0001 homes Constant -24,163.27* (12,917.4) 0.062 N = 1,492, [R.sup.2] = 0.826, F(5,1437) = 97.73 (a) Excluding sertraline. (b) Exclude clozapine. **p < 0.05; *p < 0.10. TABLE 5 Least Square Estimation of Logistic Diffusion Equations on the Substitution of Old and New Drugs within Therapeutic Classes (with State Fixed Effects) Tricyclics Coefficient (Standard Error) p Values Time -0.065** (0.002) <0.0001 Carve-out Indicator -0.138** (0.047) 0.003 Log (Number of Medicaid 0.169** (0.071) 0.017 recipients) Log (Deflated State Per -0.978** (0.422) 0.021 Capita Income) Log (Number of nursing homes) Constant 9.175** (4.298) 0.033 N = 1,490, [R.sup.2] = 0.684, F(4,1436) = 1,186.41 Tetracyclics Coefficient (Standard Error) p Values Time -0.076** (0.002) <0.0001 Carve-out Indicator 0.051 (0.034) 0.130 Log (Number of Medicaid 0.156** (0.050) 0.002 recipients) Log (Deflated State Per -0.350 (0.300) 0.243 Capita Income) Log (Number of nursing homes) Constant -2.111 (3.055) 0.490 N = 1,481, [R.sup.2] = 0.603, F(4,1427) = 2,781.24 MAOIs Coefficient (Standard Error) p Values Time -0.043** (0.002) <0.0001 Carve-out Indicator -0.026 (0.043) 0.551 Log (Number of Medicaid -0.164** (0.064) 0.011 recipients) Log (Deflated State Per -0.019 (0.385) 0.960 Capita Income) Log (Number of nursing homes) Constant -3.077 (3.919) 0.432 N = 1,466, [R.sup.2] = 0.270, F(4,1412) = 576.86 SSRIs (a) Coefficient (Standard Error) p Values Time 0.056** (0.003) <0.0001 Carve-out Indicator 0.219** (0.062) <0.0001 Log (Number of Medicaid 0.039 (0.092) 0.670 recipients) Log (Deflated State Per -1.030* (0.549) 0.061 Capita Income) Log (Number of nursing homes) Constant 7.942 (5.600) 0.156 N = 1,490, [R.sup.2] = 0.339, F(4,1436) = 411.83 Other New Antidepressants Coefficient (Standard Error) p Values Time 0.099** (0.002) <0.0001 Carve-out Indicator 0.035 (0.039) 0.370 Log (Number of Medicaid -0.229** (0.058) <0.0001 recipients) Log (Deflated State Per 0.448 (0.349) 0.199 Capita Income) Log (Number of nursing homes) Constant -5.999* (3.547) 0.091 N = 1,470, [R.sup.2] = 0.846, F(4,1416) = 3,484.83 Atypical Antipsychotics (b) Coefficient (Standard Error) p Values Time 0.280** (0.009) <0.0001 Carve-out Indicator 0.017 (0.146) 0.907 Log (Number of Medicaid -0.359** (0.172) 0.037 recipients) Log (Deflated State Per -11.141** (1.234) <0.0001 Capita Income) Log (Number of nursing 0.360 (0.229) 0.117 homes) Constant 107.107** (12.189) <0.0001 N = 898, [R.sup.2] = 0.200, F(5,844) = 623.33 Notes: The estimates for the equation for typical antipsychotics have the same magnitude as the estimates for the atypical antipsychotics but are of opposite signs. (a) Excluding sertraline. (b) Excluding clozapine. **p < 0.05; *p < 0.10. Logistic Regressions on Diffusion of Drug Classes (with State Fixed Effects and Including All Antidepressants and All Antipsychotics) Tricyclics Tetracyclics Coefficient p Coefficient p (Standard Error) Values (Standard Error) Values Time -0.071** (0.002) <0.0001 -0.080** (0.002) <0.0001 Carve-out -0.042 (0.047) 0.372 0.091** (0.033) 0.006 indicator Log(number of 0.121 (0.071) 0.086 0.155** (0.049) 0.002 Medicaid recipients) Log(deflated -0.649 (0.422) 0.124 -0.355 (0.296) 0.231 state per capita income) Log(number of nursing homes) Constant 6.387 (4.300) 0.138 -2.085 (3.020) 0.490 N = 1,490, [R.sup.2] = N = 1,481, [R.sup.2] = 0.685, F(4,1436) = 0.621, F(4,1427) = 1,303.47 3,072.37 MAOIs SSRIs Coefficient p Coefficient p (Standard Error) Values (Standard Error) Values Time -0.046** (0.002) <0.0001 0.063** (0.003) <0.0001 Carve-out 0.016 (0.043) 0.719 0.099 (0.062) 0.113 indicator Log(number of -0.163** (0.65) 0.012 0.081 (0.093) 0.383 Medicaid recipients) Log(deflated -0.026 (0.390) 0.947 -1.292** (0.557) 0.021 state per capita income) Log(number of nursing homes) Constant -3.055 (3.972) 0.442 10.137** (5.678) 0.074 N = 1,466, [R.sup.2] = N = 1,490, [R.sup.2] = 0.294, F(4,1412) = 0.342, F(4,1436) = 471.78 637.33 Other New Antidepressants Atypical Antipsychotics Coefficient p Coefficient p (Standard Error) Values (Standard Error) Values Time 0.095** (0.002) <0.0001 0.161** (0.006) <0.0001 Carve-out 0.079* (0.040) 0.050 -0.399** (0.111) <0.0001 indicator Log(number of -0.222** (0.060) <0.0001 0.243 (0.170) 0.152 Medicaid recipients) Log(deflated 0.368 (0.359) 0.306 -2.223** (1.014) 0.029 state per capita income) Log(number of -0.969** (0.161) <0.0001 nursing homes) Constant -5.330 (3.656) 0.145 20.492 (10.195) 0.045 N = 1,470, [R.sup.2] = N = 1,421, [R.sup.2] = 0.836, F(4,1416) = 0.484, F(5,1366) = 727.76 3,053.14 Notes: The estimates for the equation for typical antipsychotics have the same magnitude as the estimates for the atypical antipsychotics but have opposite signs. **p < 0.05; *p < 0.10.
|Printer friendly Cite/link Email Feedback|
|Author:||Ling, Davina C.; Berndt, Ernst R.; Frank, Richard G.|
|Publication:||Contemporary Economic Policy|
|Date:||Jan 1, 2008|
|Previous Article:||Why are Americans addicted to baseball? An empirical analysis of fandom in Korea and the United States.|
|Next Article:||Labor markets and health benefits: the offer and restrictions on it.|