Behavioral Health Insurance Parity for Federal Employees

Equivalency Examination in Mental Health and Substance Abuse Care in the FEHB Program

FEHB enrollees are 8.5 million; Approximately 25 percent of current federal employees, 25 percent of retirees, and 50 percent of spouses or dependents are current or retired employees. Registrars choose from over 350 health insurance products.7 Beginning January 1, 2001, the Office of Personnel Management required parity in coverage for mental health care and substance abuse, defined as coverage “identical with respect to the conventional Medicare deduction, co-insurance, co-pay, and day and visitation limitations.”8 Parity applies only to insurance benefits within the network.9 Providers and beneficiaries were informed of the policy change by direct mail from the plans.

The Office of Personnel Management encouraged plans to use managed care techniques to control projected increases associated with expanding mental health and substance use coverage. Prior to 2001, some plans had already contracted with managed behavioral health care organizations to control costs (in a process known as “cutting”).9

We analyzed the results of this naturalistic trial using a quasi-experimental design to account for secular trends in use and spending on mental health and substance abuse care not associated with coverage equivalency implementation. (Previous evaluations of parity examined one health plan before and after implementing parity and were not able to account for secular trends.) We compared spending in seven large FEHB plans from 1999 to 2002 with spending in a matched group of health plans without parity in coverage or changes in Mental health and substance abuse coverage from Medstat’s MarketScan database. Most comparison plans have been run by large, self-insured employers. We matched plans by location and plan type.

Table 1. Table 1. Characteristics of nine FEHB plans and their market comparison plans before and after the implementation of parity in coverage for mental health and substance abuse services in 2001.

Initially, nine FEHB plans were selected for study on the basis of location and type of plan (Health Maintenance Organization [HMO] or point of service plan versus preferred provider organization [PPO]population size and interest in participation. Enrollees between the ages of 18 and 64 years were included in the study. Table 1 Distinguishes the nine FEHB plans and comparison plans. The parity policy is shown to have improved the mental health and substance use benefits of seven of the FEHB plans. Two HMOs, which were close to parity in 2000, showed no significant change in benefits. The analysis focuses on the seven PPO plans on which the impact of equivalence implementation can be expected. (Data on impacts on HMO plans are available at

We examined responses of subjects who were continuously enrolled in a plan before and after implementation of coverage parity. Use of data from all enrollees may confuse equivalence effects with those of changes in plan composition. We examined the benefits of the plan to evaluate equivalency implementation and then evaluated the results. The main outcomes examined were the rate of use of mental health and substance abuse services, the total expenditure of these services among users, personal expenditure on these services, a single measure of quality of care, and the duration of follow-up—even for depression.


Of the seven plans, we obtained four years of data on benefit design, registration, and medical and pharmaceutical claims, including two years before equivalence in coverage was applied for FEHB plans and two years after. We analyzed data from a random sample of 20,000 participants for each plan. We also obtained data on benefits, enrollment, and claims for the matched comparison group over the same period from the MarketScan database.

Identify mental health and substance abuse services

We classified inpatient and outpatient services associated with a specific mental health, substance abuse, and psychiatric diagnosis as mental health and substance abuse services. (Detailed description is available at Mental health and substance use diagnoses were defined as those with diagnostic codes 291, 292, 295 through 309 (excluding 305.1 and 305.8) and 311 through 314 in International Classification of Diseases, 9th revision, clinical modification (ICD-9-CM). An inpatient is considered a user of mental health and substance abuse services if the initial diagnosis is the last and most of the initial diagnoses in the inpatient registry are mental health and substance abuse diagnoses. An outpatient is considered a user of mental health and substance abuse services if any of the following are indicated: an initial diagnosis of mental health and substance abuse, a procedure for mental health and substance abuse care, or a face-to-face meeting with a provider of such care or treatment in a specialized facility in mental health care and substance abuse. To define the use of psychotropic substances, we developed two lists: a restricted list of drugs that are used only for mental health and substance use disorders and an expanded list of drugs used for both mental health and substance abuse disorders and other conditions. Expenditures on any medications on the restricted list count as spending on mental health and substance abuse care. If the patient makes any other use of mental health services and substance abuse or incurs any related expenses during the year, the expenses for any medications on the Extended List are counted as spending on mental health care and substance abuse.

To assess the quality of care for depression, we examined data from patients diagnosed with major depressive disorder (codes 296.2 and 296.3). Outpatient patients were included only if the diagnosis appeared at at least two service appointments; Inpatients were included if an initial diagnosis of major depressive disorder was the reason for hospitalization.

statistical analysis

We estimated the economic impact of equivalence by the variance difference method. The difference in variances is the mean difference (before and after equivalence is implemented) in the outcomes of interest in the comparison plans minus the mean difference before and after equivalence is implemented in the FEHB plans. This approach allowed us to explain any secular trend in the results. Any remaining significant differences in the score were attributed to equivalence.

To estimate the difference in variances, we had to factor in two important features of the data on health care spending. Most people do not receive mental health and substance abuse care in any given year (ie have zero spending), and among those who receive such care, a disproportionate number have high levels of spending. To account for these features, we examined a number of competing approaches discussed in the literature.10 After testing the competing models, we agreed on the two-part model because it fits the data better. We used a generalized linear model to estimate the relationship between mental health spending and substance abuse care and equity. After checking several correlation functions and distribution assumptions, we used a normal model to describe the expenditure. The correlation between repeated annual observations was calculated by using a generalized estimation equation approach.

The first part of the two-part model used logistic regression to estimate the effect of implementing coverage parity on the likelihood that a person would use mental health and substance abuse services. The monitoring unit was the public person. In these regressions, we adjusted for a person’s demographic characteristics (age and gender) and the person’s relationship to the policyholder (child or spouse). The age variable was used to adjust the direction for any time. The main variables of significance were an indicator variable assigned a value of one for the post-equivalence period and zero for the setting period, an indicator variable assigned a value of one for members of FEHB plans and zero for members of comparison groups, and the interaction of the two indicator variables. Because the logistic model is nonlinear, the net effect of the equivalence policy on the outcome cannot be calculated directly from the interaction term coefficient.11 Instead, we calculated the average effect on the likelihood of mental health service use and substance abuse by using simulation methods based on an estimated regression model. Using bootstrap samples, we generated 95 percent confidence intervals for our final estimates.12

The second part of the two-part model used a least squares regression approach to analyze individual spending on mental health and substance use services for those who used any of these services. In this model, we used the same independent variables as in the first part, as well as the indicator variables for the diagnosis for which the service user received treatment. The interaction term coefficient allowed us to estimate any change in spending on mental health and substance abuse care due to the parity policy, taking into account the secular trend in such spending among users of mental health services and substance abuse. A generalized estimation equation was used to estimate the standard errors of the model parameters.

We performed a before-and-after analysis of administrative data to assess any changes in quality of care, as measured by duration of treatment follow-up for acute-phase depression. Receipt of services (ie, mental health care visits and substance abuse or antidepressant prescriptions) for four months or more is considered an indication of the quality of treatment for acute depression.13-15 Depression care episodes from six of the seven FEHB plans were studied before and after equivalence was implemented to assess the proportion of patients with four or more months of treatment follow-up. National PPO (Table 1) because the data was different from that of the other plans, and the differences limited comparability. Logistic regression was used to estimate the association between the post-equivalence period and the quality scale. We constructed a 95 percent confidence interval (CI) for adjusted odds ratios and used a generalized estimation equation approach to calculate repeated observations.

Leave a Comment