The separation of psychosocial and medication treatment by organization and financing mechanisms creates different incentives for their use. Organizations that manage psychosocial benefits under a capitated agreement have strong financial incentives to conserve on psychosocial treatment expenditures, whereas psychotropic medication generally represents a "free" good. In fact, no inherent benefits would accrue to such organizations if they reduced medication use. However, there are strong incentives to maintain, if not enhance, drug therapy. Psychotropic medications may reduce the need for expensive and intensive psychosocial treatment, notably inpatient care, giving managed care organizations every incentive to use medication treatment to protect their limited resources. In such a situation, the most likely explanation for decreases in medication treatment that are attributable to managed care is poor-quality psychosocial treatment that inhibits access to or maintenance of medication treatment.
Psychosocial treatment and medication treatment should act as complements for this to be true—a perspective that is consistent with guidelines for the treatment of serious and persistent mental illnesses. Best practice involves appropriate combinations of both, and failure to provide one type of treatment—or failure to provide it effectively—may influence the effectiveness of the other. Another view is that medication and psychosocial treatments can be substituted for one another. Decreases in medication that are related to psychosocial managed care could reflect changes in psychosocial treatment that reduce the need for or duration of medication treatment. This scenario is more likely to apply to treatment of persons with less severe and persistent mental illness.
Overall, and at least for persons with severe and persistent mental illness, decreases in drug therapy resulting from psychosocial carve-outs are almost surely an indicator of reduced quality of psychosocial treatment. Decrements in the quality of psychosocial treatment and medication treatment, in turn, are likely associated with reductions in measures of treatment outcome. Alternatively, decreases in the duration or intensity of psychosocial treatment are not inherently reductions in quality and therefore may or may not be associated with either reduced medication treatment or poorer outcomes.
The study used a quasi-experimental, pre-post design derived from the implementation of mental health managed care in some parts of Colorado and not others. This design allowed for estimation of the changes in treatment patterns and consumer outcomes that were attributable to implementation of the managed care program. A key element of this program was a shift from retrospective fee-for-service reimbursement to prospective, capitation payment. Detailed descriptions of the Colorado program have been published elsewhere (
+2,
+6).
Two distinct administrative models emerged, which were used in the study. In some areas, managed care contracts were awarded directly to free-standing, nonprofit community mental health centers (CMHCs). These areas are referred to in this article as the CMHC areas. In other areas, contracts were awarded to joint ventures between a for-profit managed behavioral health care organization (MBHO) and a CMHC. These are referred to as the MBHO areas. The remaining areas of the state that retained fee-for-service reimbursement are referred to as the FFS areas.
Changes under the two managed care models were compared with changes in the FFS areas. The two-year study period was determined by the extent of available medication data spanning January 1, 1995, through December 31, 1996. Selected variables were compared over three eight-month periods (January to August 1995, September 1995 to April 1996, and May to December 1996). The program was implemented on August 1, 1995, in the CMHC areas and on September 1, 1995, in the MBHO areas, which provided one preintervention study period and two postintervention study periods.
Study participants were drawn from a sample of 522 adult Medicaid-eligible consumers with severe and persistent mental illness that was developed to investigate the impact of the managed care program in Colorado (
+6). Only consumers who had a diagnosis of schizophrenia and who received their pharmacy benefit under fee-for-service reimbursement were selected for this study. Consumers who had a diagnosis of schizophrenia could be meaningfully compared in terms of antipsychotic medication use. Medication data for consumers enrolled in medical HMOs with a pharmacy benefit were not available for the precapitation periods.
The original sample of consumers was randomly selected from geographic areas that were matched on socioeconomic characteristics. The consumers were Medicaid-eligible adults aged 18 years and older who had a diagnosis of schizophrenia or bipolar disorder or who had at least one 24-hour inpatient stay with a primary DSM-IV diagnosis. The original sample consisted of 176 consumers from CMHC areas, 195 from MBHO areas, and 151 from FFS areas. Additional exclusion criteria used for the study reported here reduced the sample to 89 from CMHC areas, 70 from MBHO areas, and 76 from FFS areas, for a total of 235 participants. The reduction in sample size was predominately a result of excluding consumers who did not have a diagnosis of schizophrenia.
The Colorado Department of Health Care Policy and Finance supplied Medicaid claims data for specialty mental health services that were used during the pre-managed care period and in the post-managed care period for the FFS areas. Colorado Mental Health Services provided encounter data ("shadow claims") for the same services provided by the managed care sites as well as data on use of the state hospital. Information about consumers' outcomes was collected directly from consumers, who provided informed consent, by trained interviewers employed as part of the original study. Partial data from Medicaid pharmacy claims were recently identified from an unrelated research project. These partial claims contain only the consumer's Medicaid identifier, the National Drug Classification (NDC) drug code, and the date that the prescription was filled. All pharmacy claims with NDC codes for antipsychotics that fell within the time frame of the study were identified.
Encounter data, or shadow claims, were corroborated through matching of records in the shadow claims to those in the Colorado Client Assessment Record (CCAR), which document consumers' symptoms and functioning at admission and discharge from treatment. These data were consistently collected for all consumers in Colorado's publicly financed treatment system for several years before the managed care program was implemented. Total expenditures calculated from the managed care shadow data were also compared with annual expenditure reports submitted to the Colorado Mental Health Services by the managed care organizations. In both cases, the shadow data figures compared very closely despite some differences in reporting rules and requirements.
Use of psychosocial treatment was measured as a binary variable indicating whether or not any service was received in a given period and as the sum of treatment expenditures if services were used. Treatment expenditures were log-adjusted for the typical rightward skew in individual expenditure data. For this study, psychosocial treatment included the mental health specialty inpatient and outpatient services covered under managed care. State hospital services for consumers aged 22 to 64 years, which are not covered by Medicaid and not present in the FFS claims or managed care shadow data, were also included in the measure of psychosocial treatment, because the hospital services were a close substitute for covered services and allowed comparability across age groups.
Use of antipsychotic medication was measured as a binary variable indicating whether or not any prescription for an antipsychotic was filled in a given period. It was also measured as the number of months in a given period in which at least one prescription was filled and as a binary indicator of whether or not any prescription for a second-generation antipsychotic (clozapine, risperidone, or quetiapine) was filled in a given period. Change in the schizophrenia subscale of the Brief Psychiatric Rating Scale (BPRS) from the pre-managed care to the post-managed care period was chosen as a single outcome measure relevant to both medication and psychosocial therapy. This measure averages symptom severity scores (on a scale of 0 to 6, with higher scores indicating more severe symptoms) for the three positive symptom areas most closely associated with schizophrenia: conceptual disorganization, hallucinations, and unusual thought content.
BPRS scores and data on other consumer outcomes were collected across five six-month periods beginning three to six months before managed care was implemented (
+2). The first and fourth data collection points occurred during the second half of the first and third study periods, respectively. Outcome measurements from the second and third data collection points were combined and averaged to provide a comparable point of measurement for the second eight-month study period. Overall sample retention was high (more than 80 percent), but data were missing for some consumers. Baseline scores were included in the regression models to account for regression to the mean.
Measures of subject characteristics incorporated as control variables included age, gender, ethnicity, and whether psychosocial treatment expenditures two years before managed care were higher ("high cost") or lower ("low cost") than median expenditures for consumers treated in that two-year period.
All measures were summarized for the pre-managed care baseline period by area (FFS, CMHC, and MBHO). Chi square tests or F statistics were computed to assess differences in the sample characteristics across areas. Ordinary least-squares estimation was used in all analyses. Using linear regression for binary variables allows for direct interpretation of the magnitude and statistical significance of "difference in difference" results, unlike the nonlinear alternatives such as logistic or probit estimation (
+11).
All regression analyses used the same empirical "difference in difference" specification consistent with the general research design. Dummy variables CMHC and MBHO measured consumer differences between these areas and the FFS areas during the pre-managed care period. Post 1 and post 2 measured general trends over time. The model-by-time interactions (CMHC and MHBO post 1 and 2—that is, the first and second postcapitation periods in the study) measured the effect of managed care in that particular area and period relative to the initial differences and secular trends. The standard errors of the coefficients were adjusted to account for potential lack of independence among observations for each participant over time and for possible heteroskedasticity.
In the context of the sample and measurements used for this study, the findings supported the hypotheses initially proposed. The Medicaid mental health managed care program in Colorado did not decrease use of antipsychotic medications for consumers who had a diagnosis of schizophrenia and in some cases appeared to increase their use. The findings for antipsychotic use are mirrored by the maintenance or decrease of symptoms measured by the BPRS change score. At the same time, the reductions in psychosocial treatment that were evident under managed care were not associated either with changes in medication use or with treatment outcomes as measured. This picture is consistent with a positive definition of care management in which individual consumers are selected into treatment patterns on the basis of effectiveness and efficiency. This implies reductions in treatments that may have been previously overprescribed, such as psychosocial treatment, and maintenance or increases in treatments that may be adequately or underused, such as medication therapy.
Some limitations of this study should also be recognized. The small sample allowed only for detection of moderate to large effects. At the same time, the results for psychosocial treatment were strongly consistent with results based on the full sample (
+6) and as yet unpublished results for the full population of treated consumers with a diagnosis of schizophrenia. Similarly, the results for second-generation antipsychotic use (post-managed care) and outcomes are consistent with the results of previous studies of the full sample (
+2,
+10). Thus, to the extent that this small sample appears representative of its larger population, the greater efficiency of a larger sample would continue to support, if not enhance, the findings in regard to the hypotheses presented.
Unobserved differences among consumers across areas and between managed care and other service systems may have influenced the study results. The sample represents only adults with a diagnosis of schizophrenia, a class of consumers that is expected to be particularly sensitive to inappropriate changes in treatment patterns. The measures of medication use are quite general and do not account for potentially important treatment characteristics such as dosage. Finally, the single outcome measure may not adequately capture effects of both psychosocial and medication treatment patterns.
The combined findings of this study and those of the Utah, Tennessee, and Massachusetts studies discussed in this article suggest that changes in patterns of psychiatric medication use that are attributable to typical mental health managed care arrangements that carve out psychosocial treatment may be a strong proxy for changes in the quality of the psychosocial treatment provided. Specifically, decreases in medication treatment appear to be a consistent indicator of decreases in the quality of psychosocial treatment, whereas decreases in the amount of psychosocial treatment do not. As such, medication treatment patterns may be more closely associated with overall treatment outcomes.
Further investigation of the medication therapy patterns under managed care in the context of more current treatment conditions and for other important consumer groups that have experienced change in psychosocial treatment, such as children and adolescents (
+12,
+13,
+14), would help determine whether these results can be generalized. Psychosocial and medication treatment may act as either substitutes for or complements to different consumers, between specific types of medication and psychosocial treatment, and across different periods in an episode of treatment. The imposition of distinct (or even similar) organizational and financial arrangements on these two central modes of psychiatric treatment is likely to have important effects on combined treatment patterns and merits greater understanding and further research.
This study was supported by grant R01 54136, Capitating Medicaid Mental Health Services in Colorado, from the National Institute of Mental Health (Joan Bloom, Ph.D., principal investigator).