Our period of study was from January 2005 through September 2015. We selected these years to have an adequate number of pre-parity and post-parity months for analysis. Within the full study period, we defined three separate time periods to assess the impact of MHPAEA based on the timeline of MHPAEA's implementation:
Pre-Period: January 2005-December 2008. This period approximately corresponds to the time frame before MHPAEA was signed into law. MHPAEA was signed into law on October 3, 2008.
Interim Period: 2009-2010. This period corresponds to the time frame during which health plans knew that the law had been passed and were provided with the implementation standards and requirements for parity, including treatment classification categories, criteria for applying the predominant and substantially all requirements, and clarification of the requirements for NQTLs.
Post-Period: January 1, 2011-2015. To examine the impact of MHPAEA, we consider January 1, 2011, as the start of MHPAEA's large group commercial implementation. This is because MHPAEA's interim final rule, that was passed on February 2, 2010, required most large group employer plans to comply with the law at the start of 2011.
Our primary analyses identify the interim period (2009-2010) as part of the pre-parity period. Therefore, our primary analyses use a pre-parity period of January 2005-December 2010 and a post-parity period of January 2011-September 2015. In a separate sensitivity analysis, we include the years 2009 and 2010 as a separate interim period to test whether large group employer plans responded to the law prior to the effective compliance of the start of 2011.
Our primary data source was the Truven Health MarketScan® Commercial Claims and Encounters Database (CCAE). The MarketScan CCAE Database contains private insurance claims from approximately 150 large employers for employees, their dependents, and early retirees. The database includes claims from roughly 50 million lives per year. Although MarketScan is a convenience sample that has fluctuated in size and contributors over time, the database has maintained the same age and sex distribution as reported by the U.S. Census Bureau for individuals with employer-sponsored insurance. To further confirm that the trends were not being influenced by changes in the sample of employers, we ran analyses on a subset of 65 employers that continuously contributed to MarketScan in the study time period.
We used four MarketScan files in our assessment of MHPAEA's impact on the employer-sponsored insurance market: (1) the inpatient file; (2) the outpatient file; (3) the outpatient drug claims file; and (4) the enrollment file. All service claims, including prescription drugs, had an enrollee identification, allowing us to link services between files and across years. We also linked monthly enrollment information to identify enrollees with continuous enrollment.
In this study, we examined enrollees younger than 65 years with continuous enrollment who were covered under large group employer-sponsored insurance plans. We chose to include children and adults under age 65 in order to capture the full population covered by private insurance, where we could be reasonably sure that the employer-sponsored insurance was the primary source of coverage. For adults 65 and older, individuals are more likely to have Medicare as their primary insurance coverage, and hence we excluded this group. In initial analyses, we explored whether there may be differences in parity's impact on outcomes for children and adolescents, compared to adults. We produced separate annual spending trends for children and adolescents aged 0-17 years and adults aged 18-64 years. These results did not show any major differences in the trends between the two age groups. We therefore chose to focus our monthly trend and regression analysis on the full population under age 65. Table 2 is an attrition table that presents the total number of enrollees and the total number of contributing employers after each of several exclusion criteria were applied.
First, we excluded enrollees covered under any plans that were not fully insured by the employer. Second, because we were interested in having data on the complete set of health care services used by enrollees, we excluded enrollees covered under plans that did not provide prescription drug data. There were very few employer-sponsored insurance plans that did not provide prescription drug data. Third, we required continuous enrollment, meaning that enrollment data on enrollees must indicate that the enrollee was enrolled for all 12 months in each calendar year.
|TABLE 2. Attrition Table for Sample Used in Study|
|All Individuals in Truven Health MarketScan CCAE Database (enrollees in millions)||25.0||31.9||35.0||49.3||53.1||51.7||55.6||56.5||45.1||47.4||28.3|
|I. Restrict to self-insured employers (enrollees in millions)||13.6||14.1||15.0||18.0||18.3||19.5||21.2||22.1||22.5||20.8||19.6|
|No. of employers submitting data that meet restriction||123||133||140||146||151||157||163||162||160||151||147|
|II. Restrict to enrollees with prescription drug data (enrollees in millions)||13.3||14.1||15.0||18.0||18.3||19.5||21.2||22.0||22.5||20.8||19.6|
|No. of employers submitting data that meet restrictions I and II||122||132||139||146||151||157||163||162||160||151||147|
|III. Restrict to individuals enrolled for at least 12 out of 12 months (enrollees in millions)||10.2||10.6||11.3||13.7||14.2||15.1||16.6||17.3||17.5||15.9||14.8|
|No. of employers continuously contributing across all years||65||65||65||65||65||65||65||65||65||65||65|
In sensitivity analyses, we explored whether results were sensitive to the inclusion of plans with capitated payments. The plan types with capitated payments include health maintenance organization plans and point of service with capitation plans. We performed this sensitivity analysis because plan types with capitated payments may have missing payment information for some claims. However, results were very stable across these sensitivity analyses, leading us to present results including plans with capitated payments. We also performed sensitivity analyses for the set of 65 employers that continuously contributed to MarketScan CCAE data during the study period. We conducted these analyses to test whether our findings varied because of plans cycling in and out of the MarketScan CCAE Database. We discuss sensitivity analyses in more detail at the end of the results section on primary outcomes and in Appendix B.
We took a population-level analytic approach in this study. First, we graphically present population-level outcomes over the study period, during which time parity was implemented. We then use a regression model to estimate the size of the parity impact and the statistical significance of the estimated impacts for each outcome.
Focus on Outpatient Services
The outpatient service category used in this report includes all services in the MarketScan CCAE Outpatient file, with the exceptions of treat-and-release emergency department visits and laboratory and radiology tests. The MarketScan CCAE Outpatient file does not include inpatient admissions or prescription drug fills, both of which are included in separate analytic files. Thus, our outpatient service category is broad in scope and by definition incorporates an array of provider types and service settings (see Appendix B for service setting details). For example, this broad service category includes both office-based physician visits and outpatient surgery in a hospital. Important to MH and SUD services, the outpatient service category also includes intensive outpatient, partial hospitalization, and outpatient residential services.
We made the decision to present outpatient service results in this report after performing an extensive preliminary assessment of impact of MHPAEA on a full set of service categories using regression analysis. Given expectations from prior literature that MHPAEA would not have a strong impact on average, and that the greatest impacts would be for high utilizers, we first examined impacts at the mean for all service categories. In addition to outpatient services, the preliminary assessment included the following service categories: inpatient admissions, emergency department visits, prescription drug fills, and laboratory and radiology services. Summary results for the full set of service categories is included in Appendix E. Our preliminary assessment showed that parity impacts at the mean were, in fact, evident in the outpatient service category. Therefore, we selected the outpatient service category as the primary focus for our subsequent regression analyses presented here.
We selected a monthly time interval for outcomes in our analysis in this report because it gave us a sufficient number of data points to model the pre-parity and post-parity periods.
Specific Outcomes of Interest
Outcomes of interest were utilization outcomes and financial (i.e., spending-related) outcomes. The financial outcomes were further broken down by those related to spending by the insurer (i.e., the employer-sponsored insurance health plan) and spending by the enrollee. The following are specific outcomes of interest considered in this report:
Percentage of enrollees with any service use.
Number of services used per service user.
Financial outcomes (insurer):
Average insurer spending per service user (over 1-month period).
Average insurer reimbursement amount paid per service use (visit).
Financial outcomes (enrollee):
Average enrollee out-of-pocket spending per service user (over 1-month period).
Average enrollee out-of-pocket amount paid per service use (visit).
Other spending outcome (including insurer AND enrollee spending):
Ratio of total out-of-network spending to total overall spending.
Spending Decomposition Framework
A spending decomposition framework provides a useful theoretical structure for interpreting results. This framework is based on the understanding that health care spending for an individual in a specified period (e.g., a month) is composed of several parts. The first part is whether an enrollee uses any services in the specified period. The second part is the number of services used in the specified period. The third part is the amount paid per service.
Similarly, looking at this framework at the population level, we are able to decompose population-level spending (e.g., average insurer spending per enrollee) into its relevant parts. This type of analysis allowed us to examine what is driving any changes in spending at the population level. It separates the components of population-level spending per service user into three parts:
Percentage of enrollees with any service use.
Average number of services used per service user.
Average insurer reimbursement paid per service use.
For example, if we find that parity affects average insurer spending per enrollee, we can use this framework to analyze whether that impact is due to changes in the number of enrollees using any services, changes in the frequency of services used, or changes in the reimbursement amount paid by the insurer per service use.
Alternatively, the same spending decomposition framework may be used to understand changes in average enrollee out-of-pocket spending per enrollee. The only difference in the framework for out-of-pocket spending is that the out-of-pocket amount paid is considered in the analysis rather than the insurer reimbursement amount paid. We examined both insurer spending and out-of-pocket spending in this study.
Primary Results Versus Secondary Results
We separated our results into primary and secondary results. Our primary results include population-level outcomes at the average for the full population that meets our inclusion criteria. Secondary results include population outcomes at the 95th percentile, as well as average outcomes for two subpopulations, individuals with an SMI and individuals with an OUD.
In our primary results focused on average outcomes for the full population, we examined changes in outpatient services for the full set of outcomes by diagnosis group. First, we assessed parity's impact on MH and SUD utilization and spending outcomes, including non-BH services as a comparison. For each of these analyses, we also separated SUD services into those that have OUD diagnoses and those that have other SUD diagnoses (i.e., non-OUD diagnoses). For our final analysis in the primary results, we examined parity's impact on the percentage of total outpatient spending that is out-of-network.
Our first set of secondary results focused on the 95thpercentile of service utilizers. These results demonstrate parity's impact on outcomes for those who use a higher frequency of services and those who incur higher levels of spending, both for the insurer and for the enrollee. Because one of parity's requirements is to eliminate quantitative limits on services, we expected that service users with more frequent service use would be more likely to be affected following implementation of the law. We expected that elimination of quantitative service limits (e.g., the number of allowable outpatient SUD services in a calendar year) would be more likely to occur for SUD services than for MH services because the prior 1996 Mental Health Parity Act required comparable annual and lifetime limit on MH services and not SUD services.
Our second set of secondary results focus on the subpopulations of SMI and OUD. These results help us understand how parity has affected health care use and spending among two of the most vulnerable groups with MH and SUD diagnoses. For this part of the analysis, we first identified individuals in the MarketScan data who had either one inpatient admission with a primary diagnosis of the associated disorder or two outpatient visits for any listed diagnosis with the associated disorder. The set of diagnoses used to identify each subpopulation is detailed in Appendix B. After identifying the two subpopulations, we examined outpatient average service use and spending outcomes by diagnosis group, for MH, SUD, and the comparison non-BH services.
We first present graphs that show how each population-level outcome measure changed over time between 2005 and 2015. To assess descriptively whether parity had an impact from the trend graphs, we conducted visual examinations to see whether there were changes in the trend line that corresponded with our expected timing of parity implementation. For most of our results presented in the next chapter, we used the start of 2011 as the beginning of the post-parity period and all months prior in the years 2005-2010 as the pre-parity period. In all trend graphs presented, a vertical red line demarcates where the pre-parity period ends and the post-parity period begins. We do however consider whether large employer-sponsored plans responded earlier to the parity law, prior to the 2011 effective date. In the additional analysis to assess early response to the parity law, we considered calendar years 2009 and 2010 as an interim period and calendar years 2005-2008 as the pre-parity period.
In the trend graphs, we were interested in whether there was a change in the level (up or down) of the trend line for each outcome at the start (or close to the start) of the pre-parity period. An impact on the level of the trend line indicates a one-time impact. We also were interested in whether there was a change in the slope of the trend line between the pre-parity and post-parity periods. The slope of the trend line is a measure of the change in the outcome level over time. A more horizontal trend line indicates a smaller rate of change over time, whereas a more vertical line indicates a higher rate of change over time. Parity has the potential to not only have a one-time effect on the trend level, as described above, but also affect the outcome over time. This second impact will show up in the trend graph as a difference in the slope (i.e., the change in the outcome level over time) between the pre-parity and post-parity periods. The first section of the results looks in detail at the outcome of any use of outpatient services and presents a description of how to interpret the descriptive trend graphs presented here (see Chapter 4).
Our approach to our regression models is to use a population-level interrupted time series (ITS) regression to estimate the impact of parity on each outcome, similar to methods used in other recent parity analyses. This ITS regression approach uses as the dependent variable population-level summarized measures at regular intervals (i.e., months), similar to the trend analyses. We include three predictor variables: (1) a linear time variable month; (2) a binary parity pre-post indicator that distinguishes between the pre-parity and post-parity periods (0=2005-2010; 1=2011-2015); and (3) a Parity*Month interaction variable. The month linear time variable measures the overall slope of the trend line, whereas the parity pre-post indictor measures the one-time parity impact on the level of the trend line and the Parity*Month variable measures the impact of parity on the trend line over time.
ITS is most applicable to impacts that occur relatively quickly following the measured change. We believe that the evaluation of the impact of MHPAEA, where the interim rules were effective for most large employer-sponsored plans by January 1, 2011, is a good candidate for ITS.
In addition to the three predictor variables described above, we also controlled for seasonality by including indicators for each month. In all spending-related outcomes, we controlled for inflation by including a quarterly measure of inflation, the Gross Domestic Product (GDP) deflator. We also tested all regression analyses for serial correlation. More details on ITS regression specification and sensitivity analyses are described in Appendix B and Appendix E.
Because MHPAEA is federal legislation that affected all large group employer-sponsored insurance plans nationally, it is difficult to find a suitable comparison population (e.g., a population enrolled in health plans that were similar but not subject to parity). Instead of using a comparison group, we chose to compare trends in BH services with trends in non-BH services. The logic for this comparison is that parity was expected to influence MH/SUD outcomes, but not necessarily medical/surgical outcomes. This is an approach that we took in prior MarketScan analyses that proved useful in distinguishing BH trends from other broader health care trends in similar analyses.[32, 33, 34]