Effects of the Vermont Mental Health and Substance Abuse Parity Law

Appendix C: Methods Used to Conduct the Claims/Encounter Data Analysis

This appendix presents an overview of the approach used to measure the effect of parity on the mental health and substance abuse (MH/SA) cost and utilization experience of two health plans: Kaiser/Community Health Plan (Kaiser/CHP) and Blue Cross Blue Shield of Vermont (BCBSVT). The first two sections describe the data sources and study sample, while the third section discusses the definition of MH/SA claims for analytic purposes. This appendix concludes with an overview of the approach used to conduct the descriptive and multivariate analyses.

A. Data Sources

Claims/encounter data were acquired for the two health plans for dates of service during the 4-year study period, 1996 through 1999 (2 years prior to parity and 2 years after initial implementation). In addition, each health plan provided a membership file, employer group file, and other supporting documentation to facilitate the claims analysis. The claims followed standard UB-92 and HCFA-1500 formats for inpatient and outpatient claims, respectively. Diagnoses were coded with ICD-9 codes, and most procedures were coded with CPT-4 codes. Member identification numbers were encrypted to preserve confidentiality, and no identifying information (such as name, address, or telephone number) was provided.

B. Study Sample

The analysis was restricted to those who were continuously enrolled in the health plan during a given calendar year. The study group excluded those who were insured under Medicaid or Federal or State employee plan contracts because they were subject to different coverage provisions. Also excluded were members residing outside of Vermont and those over age 64 because their primary coverage was through Medicare.

The study group also was restricted according to plan or group type. For Kaiser/CHP, the analytic sample was limited to those with commercial group coverage because they dominated the Kaiser/CHP membership. In addition, the Kaiser/CHP analytic sample excluded members in self-insured groups because they were not subject to the Vermont parity law. In contrast, the BCBSVT sample included members in self-insured plans (known as "cost plus"); these groups were subject to the parity law because an insurance certificate was provided to each subscriber. BCBSVT members enrolled in products that relied on managed care for MH/SA services prior to parity were excluded because their claims data were incomplete. The BCBSVT analysis, therefore, focused on the three products that shifted large shares of their members from indemnity to managed care for MH/SA services following the implementation of parity.

C. File Construction

A person-quarter utilization file was constructed for each health plan for a 4-year period (1996 through 1999). Considerable effort was devoted to identifying MH/SA claims using criteria defined by the two health plans. The goal was to follow - as closely as possible - the procedures used by each plan to adjudicate MH/SA claims and to accumulate the claims against the preparity benefit limits. Health plan officials assisted in developing plan-specific algorithms that could be applied to their respective claims databases. Each health plan used some plan-specific procedure codes for MH/SA services that were incorporated in the algorithms.

To identify inpatient MH/SA claims, both plans relied on revenue and diagnosis codes. In addition, Kaiser/CHP used admission type and procedure codes, while BCBSVT used provider type. Inpatient claims that met the plan-specific criteria were flagged and classified as mental health or substance abuse admissions, based on their primary diagnosis.

For outpatient facility and professional claims, a combination of procedure codes and revenue codes were used, as well as specialty provider type for BCBSVT and billing area for Kaiser/CHP. Partial hospitalization claims were flagged separately based on revenue codes. Both health plans counted two "days" of such treatment as equivalent to one day of inpatient treatment. Claims for professional services were also differentiated according to whether they were provided in an inpatient setting: BCBSVT counted these services against the annual and lifetime dollar limit, whereas Kaiser/CHP excluded these services from the pre-parity visit limit. As with inpatient claims, all claims that met the selection criteria as mental health or substance abuse visits were classified according to their primary diagnosis.

Each type of use was quantified in terms of a dichotomous measure of no use/any use (0,1) and a continuous measure of the level of use (visits, days). For BCBSVT, spending was measured for each type of use in three ways: "total spending" was defined as the allowed charge, which included the health plan payment plus the member payment (that is, deductible, coinsurance, or copayment); "health plan payment" was defined as the actual payment by the health plan, net of member cost sharing; and "patient copayment" was defined as the member payment.

Reliable spending data were not available at the claim level for Kaiser/CHP because much of the care was provided in a staff-model HMO where providers were salaried or in a group-model HMO where providers were capitated. However, aggregate measures of MH/SA spending were imputed for Kaiser/CHP based on BCBSVT unit costs.

D. Approach to Descriptive and Multivariate Analysis

The descriptive analysis provided a snapshot of pre-versus post-parity levels of access, use, and spending. Analyses were conducted separately for mental health and substance abuse treatment. PROC DESCRIPT in SUDAAN was used to produce standardized measures, which controlled for age, gender, and subscriber status (Shah, Barnwell, & Bieler, 1997). Frequency distributions of MH/SA utilization and spending were also produced over the 4-year period to track shifts in the level of annual use following implementation of parity. This approach enabled an assessment of the extent to which health plan members were receiving services that would have exceeded the pre-parity benefit limits. Analyses were conducted for all members, with separate analyses for those with serious mental conditions (major depression, bipolar disorder, or schizophrenia).

The multivariate analysis provided a more rigorous test of the effect of parity. PROC LOGISTIC was used to test the effect of parity on the probability of use, while PROC REG was used to examine the effect of parity on the level of use among those with any use (SAS Institute Inc., 1999). The multivariate analysis controlled for demographic characteristics, including age, gender, subscriber status, and county of residence (a proxy for such local factors as public and private provider supply). Due to the small number of observations in seven counties, adjacent counties were grouped - Caledonia/Essex/Orleans, Franklin/Grand Isle, and Windham/Windsor - similar to the catchment areas used for publicly funded services.

The volume-of-use analyses controlled for type of MH/SA diagnosis. The MH analyses included four diagnosis variables: major depression/bipolar disorder/schizophrenia, mild/moderate depression, adjustment reaction, and dual MH/SA diagnosis. The SA analyses included an indicator of dual MH/SA diagnosis but did not specify the type of MH diagnosis due to the limited number of observations.

The multivariate analysis included a "quarter counter," ranging from 1 to 16, to control for secular trends independent of parity. The BCBSVT analyses also controlled for the type of plan (Basic, Comp, or VFP) and whether MH/SA benefits were managed or unmanaged during the quarter.

The variable of primary interest was the parity indicator, which had a value of 1 in post-parity quarters and a value of 0 in pre-parity quarters. The coefficient estimates associated with this variable indicated the direction and magnitude of the effect of parity on access, use, and spending (controlling for individual characteristics, geographic location, and the secular trend). In addition to examining the sign and significance of the parity coefficient, odds ratios were obtained from the logistic regressions.³⁰ Predicted levels of use were also computed for selected dependent variables related to utilization and spending, where the parity coefficient was statistically significant. Selected results of the multivariate analysis were incorporated into the discussion of the descriptive analysis to highlight the independent effect of parity.

The complete results of the regression analyses are presented in this appendix. The determinants of mental health access and use are presented first, followed by the determinants of substance abuse treatment.

Previous | Table of Contents | Next