Assessing the Impact of Parity in the Large Group Employer-Sponsored Insurance Market: Final Report. APPENDIX E: Supplemental Regression Analyses Results

02/27/2019

Annual Descriptive Trends and Simple Regression Analysis

We produced annual outcome trends for the average and 95thpercentile for all service categories, comparing non-BH, MH, and SUD diagnosis groups as well as OUD and non-OUD SUD trends. We then produced plotted graphs to visually assess whether there were distinct patterns in results or discontinuities in trends. We quickly could see that the largest impacts of parity were for SUD services. However, it was important that we narrowed our focus as we moved toward monthly regression analyses.

To narrow our focus, we performed a multitude of simple ITS regression analyses on both the average and 95th percentile annual trends. For each regression analysis, we had ten data points for years 2005-2014, six of which were pre-parity trends and four of which were post-parity trends. We reasoned that we were less likely to find statistical significance with so few data points, but that using the sign of coefficients and statistical significance where present would allow us to better identify where our monthly regression resources would be best spent. In each simple regression, we included a parity indicator (0=2005-2010; 1=2011-2014), an annual linear time variable, and a parity*annual linear time variable interaction term.

We then compiled results of the full set of simple regressions into two sets of tables--one for the average results and a second for the 95th percentile results. We present the simple regression average results below.

Results for Simple Regressions

We present the results from the full set of regression analyses in Table E1. The signs of the parity and parity*year interaction coefficients are represented by "+" and "–" symbols, and statistical significance at the 0.05 level also is indicated by blue shading and two "+" symbols in the case of a positive coefficient and yellow shading and two "–" symbols in the case of a negative coefficient. We present the primary outcomes from these analyses across the six service categories, including inpatient admissions with and without a preceding emergency department visit, treat-and-release emergency department visits, outpatient, pharmacy, and an overall service category. As stated in the Methods, we do not present laboratory and radiology services separately because diagnosis codes often are not included on service claims.

The strongest evidence of an impact of parity across all service categories was for all services and outpatient services. For almost every outcome in these two service categories, non-BH services decreased or did not change significantly in the post-period, whereas at least one or more BH categories increased significantly in the post-period. Because the results for all services and outpatient services were very similar (and results for other services categories were not), we concluded that outpatient services, which constitute the bulk of utilization and spending, were driving the results for all services.

TABLE E1. Summary Table of Simple Regression Results for Specific Service Types, by Diagnosis Group
  Non-BH Parity coeff. Non-BH Parity*yr coeff. MH Parity coeff. MH Parity*yr coeff. SUD Parity coeff. SUD Parity*yr coeff. Non-OUD Parity coeff. Non-OUD Parity*yr coeff. OUD Parity coeff. OUD Parity*yr coeff.
Access (percentage of enrollees with any service use)
All Services – – + + + + + + ++
Inpatient with ED + – – + ++ + ++
Inpatient with No ED + + + ++ ++ ++ ++ ++
Outpatient – – ++ + + + ++
Treat-and-Release ED + + + ++ ++ ++ +
Pharmacy + + + – – + ++ – –
Utilization (number of services)
All Services – – ++ ++ – – ++ ++
Inpatient Preceded by ED + + + +
Inpatient with No ED + + + +
Outpatient + + ++ ++ ++
Treat-and-Release ED + + + + + + +
Pharmacy ++ + – – – – ++ – –
Inpatient Days Utilization
Number of Days + ++ + + + + + +
Insurer Spending
All Services + + + ++ + ++ ++ ++
Inpatient Preceded by ED + – – + +
Inpatient with No ED + ++ – – + + + + ++ ++
Outpatient – – + + + ++ + ++ + ++
Treat-and-Release ED + + + + +
Pharmacy ++ + – – + – – + – –
Enrollee Spending
All Services + + + + ++ + ++ + +
Inpatient Preceded by ED + ++ + + ++ ++ ++
Inpatient with No ED + + ++ ++ ++ ++
Outpatient + + + + ++ + ++ + ++
Treat-and-Release ED + + + + + + + +
Pharmacy + – – + – – + – – + – – + – –
Insurer Reimbursement
All Services + – – ++ – – ++ – – + – – ++ – –
Inpatient Preceded by ED + – – + – – +
Inpatient with No ED + ++ – – + ++ + + + ++
Outpatient + – – + ++ – – ++ ++
Treat-and-Release ED + + + + +
Pharmacy ++ + – – + – – – – + – –
Enrollee Reimbursement
All Services + + + – – ++ – – + – – + – –
Inpatient Preceded by ED + + + + ++ ++ ++
Inpatient with No ED + + ++ ++ ++ ++
Outpatient + + + ++ – – + – – + – –
Treat-and-Release ED + + + + + + +
Pharmacy + – – + – – + – – + – – + – –
NOTES: The Parity coeff. (i.e., the parity pre-post indicator coefficient) measures the effect on the level of the outcomes, and the Parity*yr coeff. (i.e., the parity indicator * the yearly linear time variable) measures the effect of parity on the slope of the outcome over time. The signs ++ and – – indicate the sign of the coefficient from the ITS regression. A significant effect for a negative coefficient is indicated by orange shading and 2 minus signs; a significant effect for a positive coefficient is indicated by blue shading and 2 plus signs.