HCBS Waiver: Economic Utility and Related Health Outcomes: Preliminary Study Design

Participants
Cases. Individuals with physical disabilities who (a) resided in an institutional setting supported by Medicaid funding for at least 12 months prior to deinstitutionalization, (b) moved into a community-based setting during fiscal year 2004 or 2005, and (c) used HCBS waiver funding to do so.

Comparison group. Individuals with physical disabilities who have (a) never lived in an institutional setting, and (b) enrolled in the HCBS waiver program in 2004 or 2005, and (c) received HCBS waiver funds for at least 12 months before and after enrolling in HCBS. Note: These participants all receive support through Medicaid, and by definition, these individuals typically have low incomes. In addition, there is typically an overrepresentation of minority groups as a result of the correlation between poverty and race/ethnicity.

Phases
PHASE 1
Method
During Phase 1, investigators will obtain data from the Kansas Medicaid Program. States maintain more current data through the Medicaid Management and Information System (MMIS), managed by their Medicaid programs (Westmoreland, T., 2000). The State of Kansas has a progressive history of using policy and financing structures to support a reduction in institutional services and an expansion of community-based services. Even before the Olmstead v. L. C. decision, Kansas reflected a commitment to implementing and modifying waivers to achieve outcomes such as those promulgated by the Olmstead decision. This is reflected in Kansas’s pursuit of population-specific, instead of combined population waivers. Thus, the state of Kansas provides an excellent context for developing the assessment strategy model for this project. 

Our research plan calls for obtaining relevant Kansas Medicaid data at the individual level. We will analyze data to determine components that will be used to develop measures for the assessment model. These measures will include:

To measure avoidable hospital conditions and secondary health conditions, we will construct episodes of care for individuals. These constructs will summarize claims for periods of time (i.e., 6 months) preceding the onset of a specific condition, as identified by ICD-9 codes in the data. For example, to examine urinary tract infections (UTIs), we will assess the overall incidence during each year by summing the instances per person per year. This would allow us to compare rates over time; we would expect the conditions to decline under optimal care conditions. In addition, we will flag medical visit(s) associated with UTIs within a six-month period. This will demonstrate the possible connection between medical treatment and onset of secondary conditions, in this case UTI, which could be associated with the treatment.

PHASE 2
Data Analysis
In Phase 2, we will use the model to conduct analyses to evaluate the success of the Kansas HCBS program in maintaining or minimizing (a) the use of healthcare services, and (b) secondary health conditions. The design for these analyses will be a retrospective cohort analysis that reflects four sequential one-year time periods for individuals with physical disabilities living in the community who have never lived in an institution and individuals with physical disabilities who have moved into the community from an institution. Thus, in total, we will have four years of health expenditure and utilization data for each individual: one year before the person received waiver support and 3 years after receiving waiver support. This longitudinal perspective will allow us to assess how health expenditures and utilization change over time.
Analyses will include descriptive statistics, Chi2 tests, and t-tests as appropriate. In addition, staff will develop multivariate models to predict relevant contributions of independent variables to dependent variables. 

The assumptions underlying this model, summarized in Winer, Brown, and Michels (1991) are that specific measurements can be expressed as the sum of five components:
yp = C + axd + bxu + cxex + dxo + st, where for time: t = 1,....,8

In this model, y represents the measurement for preventive and performance measure with respect to demographics, healthcare utilization and expenditures, time and other factors; C is a component which is constant for all treatments and elements; axd is a component that is constant over all elements for all associated demographics; bxu is a component that is constant over all elements for healthcare utilization; cxex is a component that is constant over all elements for healthcare expenditures; dxo is a random component, depending upon uncontrolled sources of variation, that is assumed to be normally distributed; st is a component which is constant for all elements in a given time period but may differ for different populations over time.

Unit of analysis.Person level healthcare utilization andexpenditures as well as secondary health condition outcomes.
Participants.(1)All Kansas HCBS beneficiaries with physical disabilities enrolled by Medicaid who moved from an institution to a community setting in fiscal year 2004 or 2005.  his is a sample of n = 68 (Kansas Department on Aging, 2005). (2) All Kansas HCBS beneficiaries with physical disabilities enrolled in Medicaid who in 2004 or 2005 enrolled in the HCBS PD waiver but never lived in an institution. This is a population of 1,200.  We will match one person in sample A (described above) with five persons in sample B based on their risk assessment score on the Uniform Assessment Instrument (UAI) (Community Supports and Services, 2005). The result would be a 1:5 ratio for purposes of this analysis. This will result in a total sample size of N = 340 and will reduce the overall variation in the analyses by increasing the specificity of the sample. 

Sample group A is defined and limited by the number of individuals who are moving out of institutions into community settings. While the sample size (n = 68) may appear small, it is necessary to consider it in context. The total population of people with physical disabilities who live in institutional settings is quite small (LaPlante, M., 1992).  In addition, the rate at which individuals are moving into community settings in Kansas is consistent with the national rate (Eiken, Hatzmann, & Ascuitto, 2003; Holtz & Eiken, 2003; Schaeffer & Eiken, 2003). For this study, staff will follow up this analysis with a more detailed data collection for in-depth information that would not be achievable with a much larger sample. 

Dependent variables. What are the overall patterns of utilization for major categories of health services (hospital, home health, medical, pharmacy) for each group? How much did Kansas Medicaid spend to provide direct medical care (hospital, home health, medical, pharmacy) for each group? How did expenditures change for each group after moving into the community or enrolling in the HCBS waiver? What primary and secondary health outcomes did each group experience? How did these incidences change for each group moving into the community or beginning receipt of the HCBS waiver? How did the groups compare to each other in overall pattern of utilization, expenditures for direct medical care, and primary and secondary health condition incidences? 
Independent variables.The primary independent variable will be whether the individual lived in an institution or the community before receiving HCBS services.  Other variables of interest will include co-morbidity, type of community living arrangement (individual, congregate, family, other), age, gender, race, and the extent of personal assistance.
Reliability/validity.Existing literature documents the strengths and limitations of Medicaid data (Carson, Ray, & Strom, 2000; Iezzoni, 2002; Steinwachs, Stuart, Scholle, Starfield, Fox, & Weiner, 1998). Strengths include: data collected on a regular basis; comprehensive claims data; and detailed data available for individual level analysis.  Weaknesses include poor validity and completeness of primary diagnoses, a time lag for availability, and data are sometimes cumbersome to handle. We are not concerned about the poor validity of the primary diagnoses, because we know that the primary diagnosis is also available in the HCBS data collected on everyone included in our sample.
We will report and discuss the meaning of these findings with relevant staff in Kansas Medicaid and HCBS programs. In addition, we will disseminate process control information and integrity of procedures so that other state Medicaid programs can systematically replicate the model for their state HCBS and Medicaid programs. To do this, we will carefully document every step, including data management, logic, and decision making for variables included and types of analysis chosen, data analysis, computer syntax, and barriers and facilitators encountered throughout the project. 

PHASE 3
For Phase 3, we will examine the relationship between community participation and healthcare utilization and expenditures for individuals with physical disabilities. To do this, we will administer the PARTS/FABS measure to a subset of each group. Using these scores, we will analyze the relationship between community participation and healthcare utilization and expenditures for the entire sample, as well as for the two population subsets individually. We will compare PARTS/FABS scores to help understand how these groups differ in community participation. For this measure, we will collect person x environment data for two points in time to better assess participation trends: at the point of administration and (retrospectively) six months before the point of administration. We will also use the scores to further analyze a subset of covariates related to risk of institutionalization, health utilization, and health expenditures. The Kansas Director of Community Supports and Services (including HCBS) will help us recruit participants. To improve the response rate and to compensate respondents for their time and information we will offer a participant $20 per person.

Data Analysis
Analyses for Phase 3 will include descriptive statistics and bi-variate analyses, including Chi2 tests and t-tests. In addition, staff will develop multivariate models to predict relevant contributions of independent variables to dependent variables. 
Unit of analysis. Person level community participation, healthcare utilization andexpenditures, as well as primary and secondary health condition outcomes.
Participants/sample size/power.Participants will be a subset of the two groups analyzed in Phase 2.  Sample size will be approximately 51 for each group.  Assuming this sample size and a medium (.5) effect size, the power will be .8. 
Dependent variables.What is the level of community participation, as measured by PARTs/FABs?  What are each group’s overall patterns of utilization for major categories of health services (hospital, home health, medical, pharmacy)? How much did Kansas Medicaid spend to provide direct medical care (hospital, home health, medical, pharmacy) for each group? How did expenditures change for each group after moving into the community or beginning receipt of the HCBS waiver? What primary and secondary health outcomes did each group experience? How did these incidences change for each group moving into the community or beginning receipt of the HCBS waiver? How did the groups compare to each other in overall pattern of utilization? Expenditures for direct medical care? Primary and secondary health condition incidences? 
Independent variables.The primary independent variable will be whether the individual lived in an institution or community setting before receiving HCBS services.  Other variables of interest will include, co-morbidity, living arrangement (individual, congregate, family, other), age, gender, race, PARTS/FABS scores, and extent of personal assistance services.

References
Carson, J., Ray, W., & Strom, B. (2000). Medicaid databases. In B. L. Strom (Ed.). Pharmacoepidemiology. (3rd ed.). (Chapter 19). West Sussex, England: Wiley.

Community Supports and Services, Social and Rehabilitation Services. (2005). Assessment (Chapter 4). PD waiver policy & procedure manual. Available online: http://www.srskansas.org/hcp/css/PDWaiver.htm

Eiken, S., Hatzmann, M., & Asciutto, A. (2003). One-to-one: Vermont’s nursing home transition plan. U.S. Department of Health and Human Services. Available online:  http://aspe.hhs.gov/daltcp/reports/VTtrans.htm#section3

Holtz, D. & Eiken, S. (2003). Fast Track and other nursing home diversion initiatives: Colorado’s nursing home transition grant. Department of Health and Human Services. Available online: http://aspe.hhs.gov/daltcp/reports/COtrans.htm

Iezzoni, L. I. (2002). Using administrative data to study persons with disabilities. Milbank Quarterly, 80(2), 347-79.

LaPlante, M., Kaye, S., Kang, T., & Harrington, C. (2004). Unmet need for personal assistance services: Estimating the shortfall in hours of help and adverse consequences. Journal of Gerontology, 59B, S98-S108.

Schaefer, M. & Eiken, S. (2003). Passages: Arkansas’s nursing home transition program.  U.S. Department of Health and Human Services. Available online: http://aspe.hhs.gov/daltcp/reports/ARtrans.htm

Steinwachs, D., Stuart, M., Scholle, S., Starfield, B., Fox, M., & Weiner, J. (1998). A comparison of ambulatory Medicaid claims to medical records: A reliability assessment. American Journal of Medical Quality, 13(2), 63-69.

Wiener, J. M., Brown, D., Gage, B., Khatutsky, G., Moore, A., &  Osber, D. (2004). Home and community-based services: A synthesis of the literature. Research Triangle Park, NC: Research Triangle Institute.