Assessing the Impact of Parity in the Large Group Employer-Sponsored Insurance Market: Final Report. 4. RESULTS

02/27/2019

We divide our results into several sections:

  1. Utilization Outcomes.

  2. Average Insurer and Enrollee Spending Outcomes.

  3. Insurer Reimbursement and Enrollee Out-of-Pocket Outcomes.

  4. Out-of-Network Outcomes.

  5. Outcomes for Service Utilizers at the 95th Percentile and Subpopulations.

For most MH and SUD primary outcomes (Sections 1-4), we present a graph of the primary outcome trend, followed by the time series regression results. We also present the ITS regression results for the OUD versus non-OUD comparison for each outcome. In later sections for secondary outcomes (Section 5), we present graphs of the trends for spending only followed by related ITS regression results.

To ground the reader in the analyses and to provide a framework for interpretation for all findings in later sections of this results chapter, we introduce this chapter with a detailed description of the impact of MHPAEA on any MH and SUD outpatient services in Section 1. Supplemental information about the impact of MHPAEA on multiple service outcomes are included in Appendix B and Appendix E.

Primary Outcomes--Utilization

In this section, we focus on the following primary outcomes: percentage of enrollees with any use of services and average number of services used per service user. We present results for non-BH services, as a comparison to MH and SUD results. For each primary outcome, we also stratify SUD services into services for OUD and services for all other SUDs (non-OUD).

Percentage of Enrollees with Any Use of Outpatient Services

Summary of Findings: Any Use of Outpatient Services
MHPAEA did not have an impact on percentage of enrollees with any use of MH outpatient services (the percentage of enrollees who used one or more service). However, MHPAEA did have a small but meaningful effect on any use of SUD outpatient services. We observed a similar significant impact of MHPAEA on any use of outpatient SUD services for those receiving treatment for OUD compared with those receiving treatment for other SUD condition (non-OUD).

We examined MHPAEA's impact on percentage of any use of outpatient services to understand whether the percentage of the population accessing outpatient MH and SUD services has changed as a result of parity. If the parity law resulted in improved coverage of MH and SUD services overall, we would expect that parity would make access to treatment (e.g., seeking any treatment) easier for enrollees with BH conditions who previously had not sought treatment. Additionally, given that parity may have led to improved coverage of MH and SUD services, some enrollees who previously paid for certain MH/SUD treatments out-of-pocket now may be covered under their insurance plan. This second scenario is more likely to be the case with SUD services, because MHPAEA extended parity provisions to include SUD services. In both scenarios, if parity improved coverage, we would expect to see an increase in the percentage of enrollees accessing MH and SUD outpatient services.

Trend Analysis

Figure 1 presents our trend analysis for the outcome percentage of enrollees with any use of services. There are three separate trends that are plotted over time--for non-BH, MH, and SUD outpatient services. We plotted a data point for each month from January 2005 through September 2015. Each data point represents the percentage of enrollees with at least one outpatient service use of the relevant service type (i.e., non-BH, MH, or SUD). Thus, the percentage with any outpatient use is separated by MH, SUD, and other non-BH outpatient services. For example, the first data point in the SUD trend is approximately 0.4 percent, which represents the percentage of enrollees with at least one (i.e., one or more) outpatient SUD service use in January 2015.

In this trend analysis, we were interested in seeing how trends change over time, with particular focus on comparing the pre-parity years (2005-2010) with the post-parity years (2011-2015). The post-parity period in this analysis begins at the start of 2011, indicated in the graph by the red vertical line. We considered January 1, 2011, to be the beginning of the post-parity period because MHPAEA compliance was required for calendar year employer-sponsored insurance plans effective at the beginning of 2011, and most employer-sponsored insurance plans are on a calendar year.

We also were interested in comparing BH services (both MH and SUD services) with non-BH services. Because non-BH services are not subject to the parity law specifically, we do not expect the law to have major impacts on this category of services. In our analysis, we were generally concerned about being able to attribute impacts to the parity law, separate from general health care trends. In part, we were able to do this by examining the timing of the observed change with our understanding of the law's implementation. A second step was to examine non-BH trends where we do not expect parity to have an effect. If we find similar impacts for non-BH services, that would suggest that our results were driven by overall health care trends. However, if we do not observe similar effects for non-BH services, that finding would strengthen our ability to conclude that impacts were due to the parity law.

When examining changes in trends between the pre-parity and post-parity periods, we looked for changes in both the level and the slope of the trend. An observed change in the level, the slope, or both at the time of parity's effective compliance date is evidence that the change is attributable to the law. A change in the level suggests an immediate impact of the law, whereas a change in the slope suggests an impact that occurs over time.

  • An analysis of parity's impact on the level of the outcome (in this case, percentage of enrollees with any use) is graphically represented by a change in the level of the trend line in the pre-parity period (i.e., 2005-2010) versus the post-parity period (i.e., 2011-2015). This is would be seen in the graph as a vertical shift at the start of the post-parity period (2011); the red vertical line indicates the start of the post-parity period.

  • An analysis of parity's impact on the slope of the outcome (in this case, percentage of enrollees with any use) is graphically represented by a change in the slope of the trend line in the pre-parity period (i.e., 2005-2010) versus the post-parity period (i.e., 2011-2015). The slope of the trend line refers to the steepness of the curve.

We are interested in whether there was a change in the level of the trend line at the start of parity implementation or whether there was a change in the slope of each trend line between the pre-parity years (2005-2010) and the post-parity years (2011-2015). All three of the trend lines in Figure 2 have a relatively flat slope (meaning neither a large increase nor a large decrease in the level of the trend line over time), and there was no noticeable change for either MH or SUD outpatient services in the level at the beginning of 2011 or in the slope between the pre-parity and post-parity periods. This finding suggests that parity did not have a strong impact on the percentage of enrollees who used any MH or SUD outpatient services. However, the ITS regression results allowed us to empirically test this observation.

FIGURE 1. Percentage of Enrollees with Any Outpatient Service Use by Non-BH, MH, and SUD
FIGURE 1, Trend Graph: G005 through September 2015.

We then used ITS regression analysis to estimate the magnitude of any change in the level and slope on the outcome and to test the significance of the measured change for each (see Table 3).

We performed an ITS regression for each of three outcomes: average monthly percentage of enrollees with any use of non-BH outpatient services, MH outpatient services, and SUD outpatient services. In each ITS regression, we included three primary independent variables (in addition to the GDP variable to control for inflation).

  • The parity (pre-post) binary indicator was equal to zero in 2005-2010 year-months and one in 2011-2015 year-months. The parity (pre-post indicator) measured the change in the level of the outcome trend at the start of the post-parity period.

  • A Parity*Month interaction term measured the change in the slope of the outcome trend in the post-parity period.

  • The linear time trend variable (called month, row 3 in Table 3) controlled for changes in the monthly percentage with any use due to general trends in the marketplace during this time period.

The ITS regression results for any use of outpatient services, by diagnosis group, confirmed our observations from the trend analysis for MH in Figure 1 that there were no large changes in any use of MH services due to parity (see Table 3). Neither of the two estimated coefficients on our two MH variables of interest, parity (pre-post indicator) and Parity*Month, were significantly different from 0 (p-values=0.291 and 0.570, respectively), meaning that parity had no impact on any use of MH services.

However, the results for SUD show a nominally small but meaningful impact. Both variables of interest, the parity (pre-post indicator) and Parity*Month results had p-values that were <0.001, which means the coefficients were significantly different from 0. Both coefficients were positive, meaning that parity did lead to an increase in the number of enrollees with any use of SUD services. The coefficient of 0.011 on the Parity (pre-post indicator) variable means that the parity law was estimated to have increased the level of any use of SUD services by 0.011 percentage points. For the Parity*Month variable, the coefficient was 0.001, which means that the parity law was estimated to have increased the slope by 0.001 percentage points per month. These results together suggest that parity increased the percentage of enrollees with any use of outpatient SUD services by 0.023 percentage points in the first year following parity's implementation.

TABLE 3. Monthly ITS Regressions on Any Use of Outpatient Services by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) -0.483 0.245 0.093 0.291 0.011 <0.001
Parity*Month -0.025 0.028 -0.001 0.570 0.001 <0.001
Month (linear time variable) 0.012 0.073 0.016 <0.001 0.001 <0.001

Although the impact on any use of outpatient SUD services was nominally very small, it is helpful to put that finding into context. Among enrollees, 0.1927 percent had any SUD outpatient service use in the last month of 2010. If parity increased the percentage with any use by 0.023 percentage points in the next year, that represents an 11.9 percent increase. It is difficult to see this increase in the trend line for SUD in Figure 2 because the monthly SUD percentage with any use of services was so small to begin with. We can infer that this change was due to parity (and not just general health care trends unrelated to parity) because: (1) there was a measured change at the time that parity was implemented; (2) the linear time trend (which is a measure of general health care trends) controls for the general change over the time period; and (3) we did not see any similar impacts on non-BH outpatient services, for which we did not expect to see impacts due to parity.

OUD Compared with Non-OUD--Percentage With Any Use of Services

For each primary outcome, we also examine findings in the SUD group, separating service use for OUD and non-OUD SUDs. We observed a positive increase in the percentage of any outpatient service use for both groups (Table 4). The coefficients for both the level and slope for the OUD and non-OUD SUD results were very similar, suggesting a similar impact of parity on the use of both types of SUD services.

TABLE 4. Monthly ITS Regressions on Any Use of Outpatient Services by SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 0.007 <0.001 0.006 0.007
Parity*Month 0.000 <0.001 0.001 <0.001
Month (linear time variable) 0.001 <0.001 0.001 <0.001

Average Number of Services Used per Service User

Summary of Findings: Frequency of Outpatient Services
MHPAEA had a significant impact on the frequency of outpatient services for both MH and SUD (average number of outpatient services used per service user). The magnitude of the impact of MHPAEA on SUD outpatient services was roughly 10 times larger than the magnitude for MH outpatient services. The impact of parity on SUD outpatient services continued well into year 2015, and translates to an estimated increase of more than 3 additional SUD outpatient monthly services per service user, over the entire post-parity period. We observed a similar significant impact on frequency of outpatient services for both OUD and other SUD conditions, although the average number of outpatient services used per service user was slightly higher for the OUD diagnosis group than for the non-OUD diagnosis group.

A second potential driver of spending is the number of outpatient services used per service user. If we again assume that parity positively improved coverage of MH and SUD services in large group commercial plans, we would expect an increase in the frequency of outpatient service use (i.e., the number of services used per service user). Examining Figure 2, for MH and non-BH services, we see a very similar pattern to Figure 1 in which there is not much evident change in either trend line. For SUD services, however, there is a discernable change in the slope of the SUD trend line between the pre-parity and post-parity periods for the average number of monthly outpatient services used per service user.

FIGURE 2. Average Number of Outpatient Services Used per Service User by Non-BH, MH, and SUD
FIGURE 2, Trend Graph: Graph shows the trend analysis for the average number of outpatient services used per service user for non-behavioral health, mental health, and substance use disorder, from January 2005 through September 2015.

Looking at the Parity (pre-post indicator) coefficients, the ITS regression results show no change in the level as a result of parity for either the MH or SUD in the average number of outpatient services (see Table 5). However, the Parity*Month variable estimated coefficients were significantly different than 0 for both MH (coefficient=0.005) and SUD (coefficient=0.054), with p<0.001. Looking further at the magnitude of the differences in average number of outpatient services, we see differences between SUD services and MH services. The magnitude of the SUD Parity*Month coefficient, was roughly ten times the size of the same MH coefficient. This means that parity had a small impact on MH frequency of service use, whereas it had a large impact on the SUD frequency of service use. For example, in a 1-year period following parity's implementation, the average monthly number of outpatient services used per service user was expected to increase by 0.06 services for MH services due to parity (0.005 times 12), whereas the impact on SUD services was roughly 0.65 services (0.054×12). As we see from the SUD trend in Figure 2, the impact of parity continued well into year 2015, which translates to an estimated increase of more than three additional SUD outpatient monthly services per service user, over the entire post-parity period. A similar analysis for MH indicates that the changes would amount to less than a 0.3 increase in outpatient monthly service use over same the post-parity period.

TABLE 5. Monthly ITS Regressions on Average Monthly Number of Outpatient Services Used per Service User by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) -0.025 0.504 0.000 0.993 -0.059 0.247
Parity*Month 0.001 0.621 0.005 <0.001 0.054 <0.001
Month (linear time variable) 0.006 <0.001 0.002 <0.001 -0.005 <0.001

OUD Compared with Non-OUD--Average Number of Services Used per Service User

Looking next at the number of services used comparing OUD with non-OUD services, here we again see significant increases in the slope for both the OUD and non-OUD groups (Table 6). The change in the slope for average number of outpatient services used per service user was slightly higher for the OUD diagnosis group than for the non-OUD diagnosis group (0.062 vs. 0.050). In a 1-year period following parity's implementation, parity's effect on the average monthly number of outpatient services used per service user for OUD was an increase of 0.744 compared with 0.60 for non-OUD services.

TABLE 6. Monthly ITS Regressions on Number of Outpatient Services Used by SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 0.008 0.909 -0.052 0.353
Parity*Month 0.062 <0.001 0.050 <0.001
Month (linear time variable) -0.005 <0.001 -0.003 0.002

Primary Outcomes--Average Insurer and Enrollee Spending

In this section of the report, we present our findings for the spending outcomes, focusing on average monthly spending for outpatient services. We first present the average monthly outpatient spending by insurer and then by the enrollee. As above, we compare the MH and SUD spending to non-BH spending, and also stratify the SUD findings by those with OUD and those with other SUD disorders (non-OUD).

Average Monthly Outpatient Insurer Spending per Service User

Summary of Findings: Average Monthly Insurer Spending
MHPAEA had a significant positive impact on average monthly insurer spending on MH and SUD outpatient services. For insurer spending on MH outpatient services, the impact was moderate, but the impact on SUD outpatient services was greater. The patterns for OUD and non-OUD spending were essentially the same, although the magnitude of this change was slightly larger for the OUD category.

Our findings for Average Monthly Outpatient Insurer Spending per Service User demonstrate a significant impact of MHPAEA on both MH and SUD insurer spending. However, the impact on MH insurer spending was very moderate, whereas the impact on SUD insurer spending was larger. To draw this conclusion, we looked first at the trend analysis of average spending for non-BH, MH, and SUD outpatient services. Then, we examined the coefficient results of the ITS regression analysis, which test the significance of the impact of parity on outpatient insurer spending.

Trend Analysis

In Figure 3, the average monthly outpatient insurer spending per service user is plotted over time for non-BH, MH, and SUD services, separately. For both non-BH and MH services, average monthly outpatient insurer spending per service user increased gradually over time, but there was little to no noticeable change in either the level of the trend or the slope of the trend in the post-parity period. However, for SUD services, the pre-parity slope of the trend was similar to both the non-BH and MH trends, but there was a noticeable increase in the slope of the SUD trend at the start of the post-parity period.

FIGURE 3. Average Monthly Insurer Spending (in dollars) on Outpatient Services per Service User by Non-BH, MH, and SUD
FIGURE 3, Trend Graph: Graph shows the trend analysis for the average monthly insurer spending on outpatient service use by service user for non-behavioral health, mental health, and substance use disorder, from January 2005 through September 2015.

Interrupted Time Series Regression Analysis

Looking at the results of the ITS regression analyses, our findings reinforce what we observed in the trend graphs. For SUD outpatient services, Table 7 results indicate almost a $48 increase in the level of average monthly insurer spending per service user. The Parity*Month meaning that the post-parity slope increased by $6.88 per month in the post-parity period, above the linear trend captured by the month variable. This means that over a 1-year period, parity was expected to increase the average monthly insurer spending per service user for SUDs by $82.56.

Table 7 results for MH services were much smaller in magnitude than those for SUD services. The parity pre-post indicator indicated no statistically significant change in the level associated with MHPAEA, but the Parity*Month interaction term showed a small increase in the post-parity slope. The estimated increase in the slope was $0.35 per month, which over a 1-year period, amounts to a $4.20 increase.

Comparing both the MH and SUD results with non-BH results, we see in Table 7 that MHPAEA was associated with a slight decrease in the slope of non-BH average outpatient insurer spending per service user. These findings demonstrate that the implementation of parity did not have the same effect on average monthly outpatient insurer spending for non-BH services as it did for the same outcome for the MH and SUD service groups.

TABLE 7. ITS Regression Results on Average Monthly Outpatient Insurer Spending per Service User by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) -18.303 0.179 -3.674 0.413 47.674 <0.001
Parity*Month -0.320 0.394 0.348 0.005 6.879 <0.001
Month (linear time variable) 1.646 0.066 1.519 <0.001 0.372 0.643

Thus, taken together, these findings show that parity had a large and significant impact on average monthly outpatient insurer spending for SUD services and a significant but moderate impact on average monthly outpatient insurer spending for MH services. We can infer that this change was due to parity (and not just general health care trends unrelated to parity) because: (1) there was a distinct change at the time that parity was implemented; (2) the linear time trend (which is a measure of general health care trends) controls for the general change over the time period; and (3) we also controlled for inflation in these models by including a variable for the quarterly measure of the GDP deflator. In addition, we controlled for seasonality in spending by including monthly indicators.

OUD Compared with Non-OUD--Average Monthly Insurer Outpatient Spending per Service User

The trend graph presented in Figure 4 illustrates that the patterns for average monthly insurer outpatient spending per user were essentially the same for OUD and non-OUD diagnosis groups. Both OUD and non-OUD diagnosis groups experienced an increase in spending around the time of parity implementation and continued to trend upward during each post-period month. In these descriptive results, however, it appeared that the difference in the slope of the trend pre-parity versus post-parity was larger for the OUD group because of the downward sloping trend in the pre-parity period.

FIGURE 4. Average Monthly Insurer Outpatient Spending by Diagnosis Category
FIGURE 4, Trend Graph: Graph shows the trend analysis for the average monthly insurer outpatient spending by Opioid Use Disorder and non-Opioid Substance Use Disorder, from January 2005 through September 2015.

The ITS regression model for average monthly outpatient spending by insurer showed that, in fact, there were significant changes in both the level and the slope for both OUD and non-OUD SUD (Table 8). However, the magnitude of this change was larger for the OUD category. The impact on the level of average monthly spending was $65.32 for OUD versus $47.36 for non-OUD, with an impact on the slope for OUD of $9.58 versus $6.71.

TABLE 8. Monthly ITS Regressions on Average Monthly Insurer Outpatient Spending By SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 65.32 <0.001 47.36 <0.001
Parity*Month 9.58 <0.001 6.71 <0.001
Month (linear time variable) 3.32 <0.001 0.88 0.33

Average Monthly Out-of-Pocket Enrollee Spending

Summary of Findings: Enrollee Out-of-Pocket Spending
We observed no impact of MHPAEA on average monthly out-of-pocket enrollee spending for outpatient MH services. However, results for SUD outpatient services do indicate a small but non-trivial impact of MHPAEA on the average enrollee spending for SUD outpatient visits. The impact of MHPAEA on OUD and non-OUD outpatient services was virtually identical.

Results from the trend analyses for out-of-pocket spending reveal that out-of-pocket spending has been increasing over the full study period across all service categories (Figure 5). For MH outpatient services, parity does not appear to have substantially affected the level or slope of the out-of-pocket spending trend. For SUD outpatient services, we saw some indication that parity may have affected the level and slope of SUD outpatient services, particularly during some outlier months.

We also see in Figure 5 that there is substantial seasonal variation in average monthly out-of-pocket spending. Out-of-pocket spending includes the deductible, copayment, and coinsurance amounts paid over the course of the month. This seasonal variation is evident in the shape of the line within each year, which has a downward slope over the course of the year. This pattern corresponds to the restarting of deductibles at the beginning of each calendar year. Recall that when the deductible resets at the beginning of the calendar year, average monthly out-of-pocket spending is higher and subsequently falls over the course of the year as enrollees meet the deductible.

FIGURE 5. Average Monthly Out-of-Pocket Spending on Outpatient Services by Non-BH, MH and SUD
FIGURE 5, Trend Graph: Graph shows the trend analysis for average monthly out-of-pocket spending on outpatient services by non-behavioral health, mental health, and substance use disorder, from January 2005 through September 2015.

The trends in Figure 5 show that some average monthly out-of-pocket spending amounts at the beginning of the calendar year in the post-parity period were very high for SUD services. For example, average monthly out-of-pocket spending was $168 in the first month of 2012, $194 in the first month of 2013, and $250 in the first month of 2014. Spending in subsequent months in the calendar year (February-December) was, however, lower than in January. This result suggests that although the general health care trend is toward higher deductibles and as a result higher average out-of-pocket spending in our full study period, parity appears to have increased SUD service average out-of-pocket spending above the general health care trend level.

Regression results (see Table 9) for MH outpatient services indicate no impact of parity on average monthly out-of-pocket spending. However, results for SUD outpatient services do indicate an impact on the level of SUD services of $16.78 and a small but non-trivial impact on the slope of SUD services of $0.76 per month. In the first year following parity, this impact on SUD outpatient services amounted to an increase of $25.90, and in subsequent years, the increase was $9.12 per year. From Figure 5, we know that a substantial portion of that impact was occurring for SUD outpatient service users in the first month of service use in the calendar year when the plan deductible had not yet been met.

TABLE 9. Monthly ITS Regressions on Average Monthly Outpatient Out-of-Pocket Spending by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) 9.091 0.150 4.223 0.299 16.781 0.046
Parity*Month 0.158 0.363 0.070 0.534 0.757 0.001
Month (linear time variable) 0.437 0.289 0.164 0.537 -0.150 0.784

OUD Compared with Non-OUD--Average Monthly Out-of-Pocket Spending on Outpatient Services

Comparing OUD with non-OUD found that average monthly out-of-pocket outpatient spending was similar between these two diagnosis groups. The trend analyses presented in Figure 6 illustrate very similar patterns for the OUD and non-OUD SUD average out-of-pocket outpatient spending. We saw the same pattern for both OUD and non-OUD services of much higher average out-of-pocket spending in the beginning of the calendar year in the post-parity period and in general continued increasing average monthly out-of-pocket spending over the full post-parity period.

FIGURE 6. OUD Versus Non-OUD Average Monthly Out-of-Pocket Spending on Outpatient Services by Diagnosis Category
FIGURE 6, Trend Graph: Graph shows the trend analysis for average monthly out-of-pocket spending on outpatient services by Opioid Use Disorder and non-Opioid Substance Use Disorder, from January 2005 through September 2015.

The regression analyses confirm the initial descriptive findings (see Table 10). Both OUD and non-OUD diagnosis groups experienced a significant increase in the level and the slope of the average monthly out-of-pocket spending on outpatient services by enrollees and the magnitude of the effects were nearly identical.

TABLE 10. Monthly ITS Regressions on Average Monthly Outpatient Out-of-Pocket Spending by SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 18.211 0.030 17.126 0.050
Parity*Month 0.741 0.002 0.769 0.002
Month (linear time variable) 0.190 0.726 -0.157 0.783

Primary Outcomes--Insurer Reimbursement and Enrollee Spending per Outpatient Visit

Average Insurer Reimbursement Amount Paid per Outpatient Visit

Summary of Findings: Insurer Reimbursement Paid per Outpatient Visit
There was no statistically significant effect of parity on the average reimbursement amount paid per outpatient visit for SUD services. There was a statistically significant positive impact of MHPAEA on reimbursement rate paid per MH outpatient visit. But, there was a similar statistically significant coefficient for non-BH services suggesting that this impact was not in fact due to parity but was a result of general health care trends not otherwise captured in our linear time variable.

 

FIGURE 7. Average Insurer Reimbursement Among Paid per Outpatient Visit by Non-BH, MH, and SUD
FIGURE 7, Trend Graph: Graph shows the trend analysis for average insurer reimbursement amount paid per outpatient visit by non-behavioral health, mental health, and substance use disorder, from January 2005 through September 2015.

In addition to any use of outpatient services, and frequency of service use, the third potential driver of changes in average insurer spending is changes in the average amount paid to the provider per outpatient service over time. Figure 7 displays the three trend lines for non-BH, MH, and SUD services. There was no discernable impact of parity on the average reimbursement amount paid per outpatient visit from the trend lines. SUD services on average had more variation in the average reimbursement paid per outpatient visit over time than Non-BH and MH trends, as indicated by the dispersion of the green dots in Figure 7. This outcome was higher in almost all months for SUD than Non-BH and MH services. Note that the outpatient visit category is broad and includes intermediate care such as intensive outpatient and partial hospitalization (see Appendix B for outpatient service category coding). These modalities often are components of outpatient SUD treatment and help explain the higher mean costs for the SUD group.

ITS regression results confirm that there was no statistically significant effect of parity on either the level or the slope of the average insurer reimbursement amount paid per outpatient visit for SUD services (see Table 11). There was a statistically significant estimated coefficient for the parity (pre-post indicator) variable for MH services (coefficient=$2.55, p=0.019). Yet, there also was a similar statistically significant coefficient for non-BH services (coefficient=$4.98, p<0.001). This result suggests that the statistically significant impact for MH services was not in fact due to parity but was a result of general health care trends not otherwise captured in our linear time variable.

TABLE 11. Monthly ITS Regressions on Average Insurer Reimbursement Amount per Outpatient Visit by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) 4.982 <0.001 2.548 0.019 5.954 0.180
Parity*Month -0.297 <0.001 -0.072 0.015 -0.074 0.548
Month (linear time variable) 0.102 0.080 0.273 <0.001 0.851 0.004

OUD Compared with Non-OUD--Average Insurer Reimbursement Amount Paid per Outpatient Visit

For both SUD diagnosis groups, parity had no impact on average insurer reimbursement amount paid per outpatient visit (Table 12).

TABLE 12. Monthly ITS Regressions on Average Insurer Reimbursement Amount Paid per Outpatient Visit by SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 2.655 0.828 6.381 0.217
Parity*Month 0.554 0.102 -0.211 0.140
Month (linear time variable) 0.267 0.738 0.947 0.006

Average Outpatient Out-of-Pocket Paid by Enrollee per Visit

Summary of Findings: Enrollee Outpatient Out-of-Pocket Paid per Visit
There was no statistically significant effect of parity on the average out-of-pocket amount paid per service by the enrollee for MH or SUD outpatient services. These results indicate that increases in spending were not due to increased cost sharing by the enrollee. We did not find an impact of parity on enrollee out-of-pocket amount paid per visit for either OUD or other non-OUD SUD services.

Our analysis further considers the drivers of increased SUD out-of-pocket spending as a result of parity. In Table 9, we examine whether parity also had an effect on the average outpatient out-of-pocket amount paid per visit. We see that there were no statistically significant impacts on MH or SUD services for this outcome. We already found that parity had an effect on the number of enrollees with any SUD service use (see Table 3) and the average number of services used per service user (see Table 5). The null results in Table 13 indicate that parity's impact on average out-of-pocket spending (see Table 9) was not driven by an effect on the amount paid per outpatient visit (e.g., actual value of the cost sharing). Rather, this impact was driven by effects on the number of enrollees with any outpatient service use and the average number of outpatient services used per service user.

TABLE 13. Monthly ITS Regressions on Average Outpatient Out-of-Pocket Amount Paid per Visit by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) 1.812 0.205 1.781 0.178 1.156 0.423
Parity*Month 0.018 0.656 0.017 0.650 -0.030 0.449
Month (linear time variable) -0.036 0.702 -0.064 0.458 -0.060 0.525

OUD Compared with Non-OUD--Average Outpatient Out-of-Pocket Amount Paid per Service

As illustrated in our regression analyses (Table 14), we did not find effects from parity on the average out-of-pocket spending by enrollees comparing OUD with non-OUD. As in the primary analysis above, across both types of SUD diagnosis groups, we concluded that increases in average out-of-pocket SUD outpatient spending were being driven by effects on the percentage of enrollees with any use of services and the average number of outpatient services used per service user.

TABLE 14. Monthly ITS Regressions on Average Outpatient Amount Paid per Visit by SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 1.781 0.503 1.064 0.507
Parity*Month 0.092 0.211 -0.069 0.120
Month (linear time variable) 0.130 0.456 -0.073 0.486

Primary Outcomes--In-Network Versus Out-of-Network Outpatient Spending

Summary of Findings: Ratio of Out-of-Network Spending to Total Spending
There has been a general trend shifting spending to in-network for MH outpatient services as well as non-BH services. These findings suggest that this shift is due to general health trends, and not the impact of MHPAEA. However, for SUD services, we observe a strong impact of MHPAEA on out-of-network spending for outpatient services, as depicted by the negative coefficient of the month time variable. We observed a large and significant positive impact on out-of-network spending for both OUD and non-OUD outpatient services as well, indicating that this impact was not driven exclusively by the opioid crisis.

In order to better understand the impact of parity on MH/SUD service delivery, we also examined patterns of spending by insurer on in-network and out-of-network services. Here, we present our findings, first for the overall population and then comparing the OUD and non-OUD groups.

Ratio of Total Out-of-Network Outpatient Spending to Total Spending

The trend analyses presented in Figure 8 demonstrated that for non-BH services and MH services there has been a decrease in the ratio of out-of-network over in-network spending over time. The decrease is evident over all pre-parity and post-parity years. In contrast, we saw a very different pattern for SUD services. For these services, we observed a similar decrease in the years 2005-2009, but starting at the beginning of 2010, we saw that pattern start to reverse. By 2012, there was a discernable and quite dramatic reversal, which continues throughout the remainder of the post-period. These findings suggest a strong impact of parity on spending for out-of-network SUD services due to parity.

FIGURE 8. Ratio of Out-of-Network Outpatient Spending to Total Outpatient Spending, by Non-BH, MH, and SUD
FIGURE 8, Trend Graph: Graph shows the trend analysis for the ratio of out-of-network outpatient spending to total outpatient spending for non-behavioral health, mental health, and substance use disorder, from January 2005 through September 2015.

In Table 15, we present the ITS results for these analyses. Those results show that there has been a general trend shifting spending to in-network, as depicted by the negative coefficient of the month time variable. The positive and significant coefficients for MH services initially suggested that parity had a small impact on the level and slope of the MH out-of-network trend. It is worth noting that the size of the positive coefficient on the Parity*Month variable was much smaller than the size of the negative coefficient on the month linear time variable. Therefore, the suggested impact of parity on the slope of the MH services percentage of out-of-network spending was only a small lessening of the downward trend in the outcome. However, because we saw very similar results for non-BH services, meaning that there were similar small positive coefficients on the parity and Parity*Month variables, we suspect that this is evidence of a general health care trend not otherwise captured by the linear time variable. Therefore, we are unable to assert that the impact on MH services is in fact due to parity.

However, the size of the coefficient on the Parity*Month variable was much larger for SUD services. In effect, it canceled out the downward trajectory in the pre-parity period of shifting SUD service spending to in-network services to a strong upward trajectory in the post-parity period that is shifting SUD services to out-of-network. Although results for MH and SUD coefficients in the regression results were similar in size and statistical significance, the magnitude of the Parity*Month variable for SUD services was close to ten times the estimated effect for MH services. This difference in magnitude sets the SUD results apart from both the MH and Non-BH results and gives us confidence that the result was due to parity's implementation.

The regression result, in particular the coefficient on the Parity*Month variable, is evidence of parity's large and positive effect on the ratio of total out-of-network spending to total overall spending. We found similar trends when assessing the ratio of total monthly out-of-network services used of total overall services used.

TABLE 15. ITS Regression Results on the Impact of Parity on the Ratio of Out-of-Network Outpatient Spending to Total Outpatient Spending by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) 1.602 0.000 1.418 0.000 2.826 0.000
Parity*Month 0.068 0.000 0.054 0.000 0.443 0.000
Month (linear time variable) -0.144 0.000 -0.185 0.000 -0.078 0.000

Stratifying the OUD and non-OUD groups, we saw identical patterns for both groups in the trend analyses of the impact of parity on the ratio of total out-of-network spending of total overall spending (see Figure 9). Having similar trends in the OUD and Non-OUD diagnosis groups gives us confidence that the effect was in fact due to parity and not due to outside trends associated with the rise of the opioid crisis.

FIGURE 9. Ratio of Out-of-Network Outpatient Spending to Total Outpatient Spending, by OUD and Non-OUD
FIGURE 9, Trend Graph: Graph shows the trend analysis for the ratio of out-of-network outpatient spending to total outpatient spending for Opioid Use Disorder and non-Opioid Substance Use Disorder, from January 2005 through September 2015.

Our regression analyses support the patterns evident from the trend depicted in Figure 9 (see Table 16). One difference in the results between OUD and non-OUD services was a significant increase in the level for the ratio of total out-of-network spending to total outpatient spending for non-OUD SUD services but not for OUD services. However, the more important effect was on the slope of the trend, for which there was a significant change in the slope for both OUD and non-OUD services post-parity. These findings demonstrate that the dramatic shift in the slope of the ratio of out-of-network spending for SUD services in Figure 8 was not driven only by the opioid crisis. It is important also to note that the timeline for the increase in the shift for both OUD and non-OUD services began at the start of 2010. This observation may reflect changes in employer-sponsored insurance plans prior to the interim effective date of compliance for most plans.

TABLE 16. ITS Regression Results on the Impact of Parity on the Ratio of Out-of-Network Outpatient Spending to Total Outpatient Spending by SUD Diagnosis Category
Variable OUD SUD Coefficient OUD SUD p-value Non-OUD SUD Coefficient Non-OUD SUD p-value
Parity (pre-post indicator) 0.021 0.976 3.292 0.000
Parity*Month 0.491 0.000 0.476 0.000
Month (linear time variable) -0.140 0.000 -0.080 0.000

Sensitivity Analyses on Primary Outcomes

Two supplemental sensitivity analyses were conducted as part of these analyses: (1) an analysis of the interim period 2009-2010 to assess our confidence in using 2011 as the time for measuring the effect of MHPAEA; and (2) a standard statistical test of first order serial correlation.

Interim Results

A full set of regression analyses were performed that included an additional indicator for the interim period, years 2009-2010, and an additional month*interim period variable. This alternative specification allowed us to test whether parity impacts on use and spending were evident in the 2-year interim period prior to the 2011 effective compliance date. Our results indicated some small effects in the interim period, however, overall there was little evidence of changes in outcomes as a result of parity in the interim period. While we selected not to present results from interim regressions, they are available on request.

The one outcome in which we did find sizeable and significant effects of parity was on the ratio of out-of-network spending of total outpatient spending. Coefficients in the interim period for this outcome were of similar magnitude to the results presented in Table 13 and Table 14. These results confirm what is evident in the Figure 8 SUD trend and Figure 9 trends, that the shift toward out-of-network SUD services began in 2010.

First Order Serial Correlation

The other sensitivity analysis that was performed for all primary outcomes was controlling for first order serial correlation. Tests of serial correlation using a Durbin-Watson test statistic were marginally significant, signifying the presence of some serial correlation. However, upon comparison of coefficients for results with and without controls for first order serial correlation, the decision was made to present unadjusted results because results were very similar across the full set of outcomes.

Secondary Outcomes--95th Percentile and Subpopulations

In this last section, we consider the impact of parity on high utilizers, examining outcomes at the 95th percentile, and also consider outcomes for individuals with SMI, and those with an OUD.

95th Percentile Outcomes

Summary of Findings: Outcomes at the 95th Percentile
Outcomes were similar to the analyses at the mean presented above, with much higher magnitude of effect. MHPAEA had a positive and significant impact on frequency of outpatient visits and total spending at the 95th percentile by the insurer for both MH and SUD treatment increased. However, there was no significant impact of parity on out-of-pocket costs to the enrollee for MH outpatient visits. For high utilizers of SUD services, there was a modest increase in out-of-pocket costs following parity. Similar to results at the mean, we expect that this increase is primarily driven by increased frequency of use though the shift to out-of-network spending may also be a factor.

Regression analyses examining outcomes at the 95thpercentile demonstrate that parity had an impact that was similar to that observed in the population as a whole, but the impact was greater, as we would expect (see Table 17). Interestingly, there was an initial decrease in the level of the number of outpatient visits at the 95th percentile for both MH and SUD. However, over time, the positive effect on the slope canceled out this one-time decrease in the first post-parity year. There was a much larger effect on SUD service use at the 95th percentile than for MH services.

As before, we saw an increase in the slope for monthly spending by insurer due to parity for both MH and SUD with no similar increase for non-BH. The magnitude of increase was higher for both MH and SUD compared with the outcomes at the average. Here, as well, the effect on the slope was much larger for SUD than that for MH ($26.56 compared to $1.2).

For out-of-pocket spending, it is reassuring that we did not see significant impacts of parity overall on MH out-of-pocket spending at the 95th percentile. For SUD services, however, the picture was different. There was not a statistically significant change in the level of SUD out-of-pocket spending at the 95th percentile; however, the slope for SUD out-of-pocket spending at the 95th percentile did increase by $2.76 per month. These findings are consistent with the analyses presented above on average SUD out-of-pocket spending.

TABLE 17. ITS Regression Results on Average Monthly Outpatient Visits, Average Monthly Outpatient Insurer Spending, and Average Out-of-Pocket Spending per Enrollee for the 95th Percentile Group by Non-BH, MH, and SUD Groups
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Outpatient visits at 95th percentile
Parity (pre-post indicator) -0.142 0.353 -0.313 0.004 -0.849 0.000
Parity*Month 0.015 0.000 0.016 0.000 0.174 0.000
Month (linear time variable) 0.012 0.000 0.015 0.000 0.003 0.332
Outpatient spending at 95th percentile by insurer
Parity (pre-post indicator) -111.345 0.074 -6.157 0.560 65.262 0.286
Parity*Month -1.893 0.269 1.199 0.000 26.561 0.000
Month (linear time variable) 2.722 0.502 3.768 0.000 0.313 0.938
Outpatient out-of-pocket spending at 95th percentile by enrollee
Parity (pre-post indicator) 43.259 0.138 16.559 0.355 44.752 0.264
Parity*Month 0.787 0.327 0.588 0.235 2.763 0.013
Month (linear time variable) 1.471 0.439 -0.022 0.985 -0.920 0.725

SMI Subpopulation

Summary of Findings: SMI Subpopulation
For those with SMI, we observed an increase in average spending, primarily for SUD outpatient services. However, importantly, for this group, there was no impact on average out-of-pocket spending per outpatient visits.

Trend analyses for the SMI subpopulation indicated that parity had a somewhat positive impact on average spending by insurer on SUD outpatient services but no evident impact on average spending for MH outpatient services (see Figure 10). The effect on average spending for SUD outpatient services for this subpopulation appears to be much less dramatic than was found in analyses on average SUD outpatient spending for the full population.

FIGURE 10. Average Spending by Insurer for Outpatient Services for the SMI Subpopulation
FIGURE 10, Trend Graph: Graph shows the trend analysis for the average spending by insurer for outpatient services for the serious mental illness subpopulation, from January 2005 through September 2015.

The ITS regression results confirm the effect of parity on the slope of average spending for SUD outpatient services by insurer (see Table 18). They also show an effect on average MH outpatient spending for the SMI subpopulation. The findings demonstrate that the magnitude of the impact on the slope for SUD outpatient spending for this subpopulation was larger than the impact on the slope for MH outpatient spending but that parity had a positive effect on average outpatient spending for BH services in general for individuals with SMI.

TABLE 18. ITS Regression Results Estimating the Impact of Parity on Average Spending by Insurer for Outpatient Services for the Subpopulation of Individuals With SMI by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) -44.857 0.010 -2.613 0.680 18.373 0.608
Parity*Month 0.098 0.836 0.884 0.000 4.093 0.000
Month (linear time variable) 2.653 0.019 2.001 0.000 -0.417 0.859

In Table 19, we present findings for average monthly out-of-pocket spending by enrollee for the SMI population. There were no significant impacts on out-of-pocket spending by enrollee for this high utilizer group, suggesting that despite increases in utilization and associated increases in insurer spending due to parity, cost sharing did not increase among this subpopulation. This supports our hypotheses that parity may have a protective effect on financial burden among those with the greatest need. However, it also suggests that the 2008 parity law may not have made a substantial difference among this group, who were probably most likely to be protected already by the original 1996 law.

TABLE 19. ITS Regression Results Estimating the Impact of Parity for the Subpopulation of Individuals with SMI on Average Out-of-Pocket Spending by Enrollee by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) 12.232 0.154 5.449 0.229 13.084 0.208
Parity*Month 0.031 0.629 0.122 0.330 0.198 0.489
Month (linear time variable) 0.270 0.897 0.336 0.257 -0.028 0.967

OUD Subpopulation

Summary of Findings: OUD Subpopulation
For those with an OUD, we observed a similar increase in outpatient spending following parity as we observed for those with SMI, with a much larger magnitude of increased spending for SUD. However, here we saw a substantial increase in out-of-pocket spending for those receiving SUD outpatient services. We also saw a significant increase in the level of out-of-pocket spending on MH services for this population at the point of MHPAEA implementation.

Looking at the OUD subpopulation in Figure 11, we see an expected large increase in overall spending for SUD outpatient services by insurer, but we also see a less dramatic increase in the slope for spending on MH outpatient services. These results are consistent with our overall analysis of SUD outpatient spending for the full population.

FIGURE 11. Average Outpatient Insurer Spending for the OUD Subpopulation
FIGURE 11, Trend Graph: Graph shows the trend analysis for the average spending by insurer for outpatient services for the opioid use disorder subpopulation, from January 2005 through September 2015.

In Table 20 we present the ITS findings for outpatient spending by insurer for the OUD subpopulation. As evident in the trend analysis, for this population, parity has had a very large impact on SUD average outpatient insurer spending. Interestingly, compared with the SMI subpopulation, parity has had a larger effect on average outpatient spending for MH services for the OUD subpopulation. We can only speculate on why this is the case; however, one possible explanation is that the increase in SUD service use has also resulted in increased referrals to receive care for MH conditions, which are often a comorbidity for individuals with OUD.

TABLE 20. ITS Regression Results Estimating the Impact of Parity of Average Outpatient Spending by Insurer, for the Subpopulation of Individuals with OUD, by Non-BH, MH, and SUD
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) -57.968 0.011 14.679 0.202 88.042 0.000
Parity*Month 2.915 0.000 2.310 0.000 7.780 0.000
Month (linear time variable) 1.198 0.419 0.648 0.388 1.327 0.184

In Table 21, we present the findings for average out-of-pocket outpatient spending for the OUD population. There was a significant impact on the level of out-of-pocket spending for MH outpatient services and a quite dramatic impact on the level and slope of SUD out-of-pocket outpatient spending. It is important to note two things. First, these findings are consistent with our findings that parity had an effect on average out-of-pocket outpatient spending for SUD services for the full population. Second, in our overall analyses, average out-of-pocket spending for MH outpatient spending did not increase. The fact that out-of-pocket spending for MH outpatient services increased for the OUD group is noteworthy and may be linked to the higher effect of parity on access to MH outpatient services for this subpopulation.

TABLE 21. ITS Regression Results Estimating the Impact of Parity, for the Subpopulation of Individuals With OUD, on Average Out-of-Pocket Spending by Enrollee by Diagnosis Category
Variable Non-BH Coefficient Non-BH p-value MH Coefficient MH p-value SUD Coefficient SUD p-value
Parity (pre-post indicator) 11.786 0.122 10.681 0.019 19.620 0.003
Parity*Month 0.310 0.141 0.201 0.109 0.902 0.000
Month (linear time variable) -0.059 0.906 -0.306 0.302 0.104 0.810