Since 2000 the number of medications available to treat bipolar disorder has grown significantly (1). New drug approvals for bipolar disorder by the U.S. Food and Drug Administration included nearly all the second-generation antipsychotics and several second-generation mood stabilizers (2,3). Practice guidelines available in 2000 or in later years, including those published by the American Psychiatric Association (4), Expert Consensus Guideline Series (5), and others (6), have been substantially revised to include second-generation agents as first-line therapies. Baseline data from the Systematic Treatment Enhancement Program-Bipolar Disorder (STEP-BD), a multisite clinical trial funded by the National Institute of Mental Health, indicated that only 11% of patients received mood stabilizer monotherapy; the remainder received combinations of antipsychotics, antidepressants, and other medications (7).

Earlier studies of medication use indicate frequent departure from current established guidelines for the treatment of bipolar disorder. For instance, in Medi-Cal, California's Medicaid program, less than half of patients diagnosed as having bipolar disorder received lithium, carbamazepine, or valproate between 1994 and 1998 (8). Three studies of privately insured outpatients receiving treatment for bipolar disorder in the 1990s documented low rates of pharmacotherapy with mood stabilizers: only 60%—70% were receiving lithium, carbamazepine, or valproate and 30% were receiving an antidepressant without a mood-stabilizing medication (9,10,11). These studies either reported data from years before the approval of second-generation agents for bipolar disorder (8,9,10) or were not longitudinal and did not allow for analyses of changes in prescribing patterns (12,13). These studies typically excluded individuals whose treatment was covered by public insurance programs.

Consequently, there are few published data about recent use of medications and related costs for the treatment of bipolar disorder in the U.S. public mental health system (8,9,10,13). Given the wide range of treatment options currently available for bipolar disorder, it is important to understand longitudinal medication prescribing patterns for this disorder among public mental health care providers and how demographic variables interact with these patterns.

Uncertainty also surrounds the relationship between demographic characteristics, such as age, gender, and race-ethnicity, and pharmacotherapy for bipolar disorder. In a Department of Veterans Affairs (VA) sample, African Americans with bipolar disorder were less likely than Caucasians to receive mood stabilizers (14). In broader non-VA samples, African Americans were less likely to receive second-generation antipsychotics than first-generation antipsychotics (15) and were more likely to be treated with antipsychotics for nonpsychotic conditions (16). Few data are available to assess differences in pharmacotherapy for bipolar disorder among Latinos, a large and growing ethnic group in the United States.

We analyzed trends in the use of medications for fiscal years 2001—2004 among patients receiving treatment for bipolar disorder in a large public mental health system. This approach permitted us to assess medication utilization within a stable client cohort. We hypothesized that the proportion of patients treated with antipsychotics would increase over time. We also hypothesized that the rate of pharmacotherapy with mood stabilizers or antipsychotics would be statistically and significantly lower for persons from ethnic minority groups across the study period after we controlled for other covariates.

Data sources

Data from the encounter-based management information system (MIS) of San Diego County's Adult and Older Adult Mental Health Services (AOAMHS) were merged with data from California's Department of Health Care Services (DHCS) to identify Medi-Cal beneficiaries age 18 and older who received a diagnosis of bipolar disorder during fiscal years 2001—2004. As in earlier research (13,14), we included clients who received two or more outpatient diagnoses or one or more inpatient diagnoses of bipolar disorder I or II (ICD-9 diagnosis codes of 296.4x—296.7x, 296.80, and 296.89) during the study period. We excluded clients who also received a diagnosis of schizophrenia (295.xx). We limited the sample to clients with continuous enrollment during the four-year study period, permitting us to examine patterns and changes in medication utilization within the same set of patients.

We used pharmaceutical claims for mood stabilizers and antipsychotics to classify each client's pharmacotherapy by year into one of four categories: neither a mood stabilizer nor an antipsychotic, mood stabilizers alone, antipsychotics alone, or mood stabilizers plus antipsychotics. Patients were placed in the fourth category if they filled one or more prescriptions for each type of medication (mood stabilizer and antipsychotic) during the fiscal year. This category included clients filling medications concurrently, clients transitioning between medications, and clients alternately filling prescriptions for mood stabilizers and antipsychotics—for example, when a patient was receiving continuous pharmacotherapy with mood stabilizers and occasional therapy with antipsychotics during manic phases. Our claims data did not permit us to identify clients who were transitioning between medications from those who were switching back and forth between classes throughout the year.

Next, clients receiving antipsychotic medication were categorized according to the type of antipsychotic medication—that is, first or second generation—for which they had the greatest days' supply during that year. Clients with one or more claims for mood stabilizers were categorized according to whether the received pharmacotherapy with lithium, with older anticonvulsants (valproate or carbamazepine), or with newer anticonvulsants (lamotrigine, topiramate, or gabapentin). (Although gabapentin is now subject to prior authorization and is rarely used among AOAMHS clients, it was widely used during the study period.) We also identified clients who were receiving antidepressant treatment. Finally, we identified clients who were receiving antidepressant treatment as their only form of pharmacotherapy. Additional information obtained from the Medi-Cal and MIS data included age, gender, race-ethnicity, and diagnosis of a substance use disorder.

Previous studies using this database have examined the rate of adherence to antipsychotic medication, second-generation polypharmacy among persons with schizophrenia, and clinical features of bipolar disorder (17,18,19). However, this is the first study to examine medication use among patients with bipolar disorder and racial-ethnic variations in pharmacological treatment. The University of California, San Diego, Institutional Review Board and the San Diego County Mental Health Services Research Committee approved the use of these data for this study in accordance with the Privacy Rule of the Health Insurance Portability and Accountability Act of 1996.

Statistical analyses

For demographic statistics, we calculated the mean age, the percentage of women, and the distribution of racial-ethnic groups across the four study years. For each fiscal year, we then calculated the percentages of patients receiving neither mood stabilizers nor antipsychotics, mood stabilizers alone, antipsychotics alone, or both antipsychotics and mood stabilizers. For patients receiving antipsychotics, we calculated the percentages receiving first-generation and second-generation antipsychotics. For patients receiving mood stabilizers, we calculated the percentages receiving pharmacotherapy with lithium, with older anticonvulsants, and with newer anticonvulsants. We also calculated the percentage receiving pharmacotherapy with antidepressants. We assessed the significance of trends in pharmacotherapy using logistic regressions with a robust covariance matrix that allowed for clustering of observations within individuals.

Next we assessed variations in pharmacotherapy by age, gender, and race-ethnicity across the four fiscal years. We used logistic regression to estimate the probability of receiving mood stabilizers or antipsychotics as a function of age group (18—29, 30—49, and 50 and older), gender, and race-ethnicity (non-Latino white, Latino, and African American), while we controlled for comorbid substance use disorders. We calculated standardized estimates (or propensity scores) of the percentage of patients receiving either mood stabilizers or antipsychotics for each of these groups. Standardized estimates of pharmacotherapy (propensity scores) for a particular group were calculated as the mean predicted percentage receiving mood stabilizers or antipsychotics across all persons as if they belonged to that group. For example, we calculated the percentage receiving mood stabilizers or antipsychotics among women as the mean probability across all persons after all were classified as being female. Among those receiving some pharmacotherapy, we employed multinomial logistic regression to provide standardized estimates of the type of pharmacotherapy (mood stabilizers alone, antipsychotics alone, and mood stabilizers plus antipsychotics). Standard errors were computed by using the nonparametric bootstrap; p values were obtained from the regressions.

Finally, we assessed pharmacotherapy use by age, gender, and race-ethnicity over time. We employed logistic regression models to estimate the probability of pharmacotherapy with either mood stabilizers or antipsychotics and multinomial logistic regression to estimate the type of pharmacotherapy conditional on receiving at least a mood stabilizer or antipsychotic. These regression models included additional covariates for analyzing a time trend and interactions between time and the demographic variables of interest (for example, by time and gender). We calculated standardized estimates for probability and type of pharmacotherapy by demographic group and year as described above. We calculated p values from significance tests for differences between groups and for differences between groups over time (group-by-time interactions).

We identified 2,427 individuals with bipolar disorder, 1,473 (61%) of whom were continuously enrolled in Medi-Cal during the study period. Among those continuously enrolled, the mean±SD age was 48.0±13.1, 957 (65%) were women, 692 (47%) had a comorbid substance use disorder, and 692 (47%) had dual coverage under Medicare. A total of 795 (54%) were non-Latino white, 133 (9%) were Latino, 88 (6%) were African American, and 501 (34%) were of other or unknown race-ethnicity. Compared with clients who were not continuously enrolled in Medi-Cal for the four-year period, those who were enrolled were 6.2 years older on average; more likely to be women, to be non-Latino white, and to have Medicare coverage; and less likely to be diagnosed as having a substance use disorder.

Pharmacotherapy trends

Medication use patterns are shown in Table 1. The proportion of clients receiving either a mood stabilizer or an antipsychotic increased significantly over the four years, whereas the proportion receiving a mood stabilizer alone declined. The proportion receiving an antipsychotic alone increased. In 2004, 77% received either a mood stabilizer or an antipsychotic, 36% received antipsychotics alone, 20% received mood stabilizers alone, and 44% received both antipsychotics and mood stabilizers. Sixty-two percent of clients received an antidepressant, and 10% received an antidepressant without a concomitant antipsychotic or mood stabilizer. The use of second-generation antipsychotics increased over the four-year span, whereas use of first-generation antipsychotics declined. Within the category of mood stabilizers, the use of older anticonvulsants declined and the use of newer mood stabilizers increased. Use of lithium did not change significantly.

Pharmacotherapy by age, gender, and race-ethnicity

Significant variations in the proportion of clients receiving any mood stabilizers or antipsychotics were identified by age, race-ethnicity, and gender (Table 2). African Americans and Latinos were less likely to receive any mood stabilizer or antipsychotic, as were younger adults (ages 18 to 29). Among those receiving either a mood stabilizer or an antipsychotic, younger adults, African Americans, and Latinos were more likely to be prescribed antipsychotics alone.

Table 3 shows time trends in pharmacotherapy by demographic characteristics. The proportion of clients receiving any mood stabilizer or antipsychotic increased in the 30- to 49-year age group compared with the other two age groups (18—29 years and 50 years and older) and among women. Time trends in pharmacotherapy type were compared. Among those receiving some type of pharmacotherapy, no statistically significant differences in time trends by age, gender, or race-ethnicity were detected. [Tables presenting data on trends in pharmacotherapy type by age, gender, and race-ethnicity are available as an online supplement to this article at ps.psychiatryonline.org.]

This study examined longitudinal prescription patterns in a sample of patients treated for bipolar disorder in a large U.S. public mental health system from 2001 to 2004. The percentage of patients receiving pharmacotherapy increased over time—the result of a sharp increase among women. This finding can be seen as an indication of improvement in care. We also found increases in the use of antipsychotic medications and decreases in the use of mood stabilizers, and within these classes of medications, we found substantial trends toward use of newer medications. However, we also found that African Americans and Latinos were less likely than non-Latino whites to receive pharmacotherapy with antipsychotics or mood stabilizers.

Kilbourne and colleagues (20) proposed a conceptual framework for health disparities that is focused on three phases where research may identify, explain, and intervene to reduce health disparities, which the authors defined as "observed clinically and statistically significant differences in health outcomes or health care use between socially distinct vulnerable and less vulnerable populations that are not explained by the effects of selection bias." Following this definition, we propose that administrative data can be used to identify differences in pharmacotherapy between racial-ethnic subgroups, and to the extent that these differences in pharmacotherapy reflect differences in quality of care, they can be considered health care disparities. For example, in our study we observed that persons from minority groups were less likely to receive recommended pharmacotherapies (that is, a mood stabilizer or an antipsychotic), which constitutes a disparity in care. Our use of longitudinal data allowed us to demonstrate that this disparity persisted over time.

Among patients with bipolar disorder who were receiving pharmacotherapy, African-American patients were more likely than non-Latino white patients to receive pharmacotherapy with antipsychotics alone. This pattern was also identified in a sample of veterans (14). In our study Latinos were also more likely than non-Latino whites to receive pharmacotherapy with antipsychotic monotherapy, which, to our knowledge, has not been previously reported. We were unable to identify any time-by-demographic-category interactions, suggesting longitudinal stability in the treatment choice for racial-ethnic subgroups with bipolar disorder.

It is not clear that these differences in type of pharmacotherapy are related to quality of care, and thus it is not clear that they constitute health care disparities. Antipsychotics may be preferred over mood stabilizers because of the side effect profiles of the latter (including a more narrow therapeutic window with lithium and side effects such as hepatotoxicity of anticonvulsants) or monitoring demands associated with these concerns. These properties of mood stabilizers may be particularly important considerations for a public mental health system with few resources for monitoring. However, second-generation antipsychotics have been associated with significant health risks, such as metabolic syndrome (21). These risks need to be balanced against potential efficacy in a population of patients with severe mental illness that has been shown to have higher rates of morbidity and mortality (22,23) partly attributable to medication side effects. In addition, roughly 11% of clients were treated with antidepressants with no concomitant mood stabilizers or antipsychotics, which indicates a need for more provider education about evidence-based treatment for bipolar disorder.

In our 2001—2004 data, the shift toward greater use of second-generation antipsychotics predates the publication of randomized clinical trials comparing second-generation antipsychotics to divalproex and lithium (1). A notable difference between our findings and other published data on pharmacy patterns for bipolar disorder in privately insured populations is the greater use of antipsychotics compared with mood stabilizers in our sample. Furthermore, we found that use of mood stabilizer monotherapy declined and use of antipsychotic monotherapy increased over the four-year period. The increasing use of newer anticonvulsants, including gabapentin, was concurrent with a decline in the use of older anticonvulsants; the increase began before the accumulation of data from randomized controlled trials of the newer medications.

Within the antipsychotic class, significant trends were observed toward use of second-generation variants. We reviewed several pharmacy utilization studies and in none was the percentage of patients receiving antipsychotics higher than the percentage receiving mood stabilizers, as in our study. Only 30% of the STEP-BD participants were receiving an antipsychotic (24) as was only 11% of the privately insured population in the study by Baldessarini and colleagues (12). The single largest group of patients in our sample was taking both a second-generation antipsychotic and a mood stabilizer.

There are several potential reasons why antipsychotic use was more common in this sample than in previous studies of samples with private insurance (9,10,12). San Diego County lacks formulary restrictions governing the use of medications, which contrasts with other systems of care, such as the VA. Our study data are more recent than data in many previously published studies, which may have better captured the emergence of antipsychotics as a first-line treatment for bipolar disorder. The preponderance of antipsychotic use in this sample compared with privately insured populations may reflect greater severity of illness among persons covered by public insurance.

Several study limitations deserve mention. The diagnoses of bipolar disorder are not based on standardized instruments but on visit diagnoses from administrative data. To mitigate concerns about diagnostic nonspecificity, we selected only participants with two or more outpatient or one inpatient visit in which a bipolar diagnosis was made and excluded all clients who also had a diagnosis of schizophrenia or schizoaffective disorder. We lacked specific information on clinical characteristics, such as whether patients were in depressed or manic phases, which would aid in determining patient-level factors that may have influenced the choice of medications. Given the nature of administrative data, we were unable to determine whether medication utilization patterns differed by bipolar disorder subtype (that is, bipolar type I and type II). All study participants were eligible for public insurance, which implies a significant level of disability, lower income, and higher disease severity than observed in the general population. The trends we noted were limited to a four-year window; a longer window may have identified more distinctive trends. Our data reflect rates of prescription fills for the medications; thus we cannot determine the role of adherence and the extent to which these medications were actually taken by patients. Finally, we assessed change in medication utilization among continuously enrolled patients, which may have biased the sample toward patients with chronic psychiatric symptoms.

Our results suggest that there has been a significant shift in the types of medications used in the treatment of bipolar disorder in this public mental health system. We found that the rate of use of antipsychotics (either alone or together with mood stabilizers) increased over the four-year period, whereas use of mood stabilizers decreased. Our study raises questions about why treatment patterns for bipolar disorder differ by demographic characteristics. Given the range of treatment options currently available, practice-based research is needed to better understand clinicians' prescribing patterns for clients with bipolar disorder.

This work was supported in part by grants MH-077225 and P30-MH-066248 from the National Institute of Mental Health and grant 3-R01-DA-019829-S1 from the National Institute on Drug Abuse. The authors gratefully acknowledge Adult and Older Adult Mental Health Services and the County of San Diego Health and Human Services Agency for access to the management information systems.

The authors report no competing interests.