MassHealth SCO Program Evaluation Pre-SCO Enrollment Period CY2004 and Post-SCO Enrollment Period CY2005 Nursing Home Entry Rate and Frailty Level Comparisons Prepared by JEN Associates, Incorporated June 6, 2008 Table of Contents Overview 3 Study Methods 3 Data Sources 4 Elderly SCO Enrollee and All Medicaid Elderly Population Profiles 4 Demographics 5 Long Term Care (LTC) Utilization Status 5 Chronic Disease Prevalence 6 Utilization Rates 6 Comparison Population 7 Outcome Measures 8 Key Findings 9 Conclusion 11 Overview For many years, the MassHealth Office of Long Term Care has contracted with JEN Associates, Inc. (JAI) of Cambridge to provide analytical and statistical consulting support for the MassHealth Senior Care Options (SCO) program. Specifically, JAI has linked the administrative data for Medicare and Medicaid, and has developed and provided critical descriptive and financial information needed to structure and implement the SCO model SCO is an integrated Medicare and Medicaid managed care program that has been offered to elderly Medicaid eligible beneficiaries since 2004. Massachusetts Medicaid and the federal Centers for Medicare & Medicaid (CMS) jointly contract with qualified managed care plans (SCOs) to provide a complete benefit package that includes the full range of Medicaid and Medicare services for enrollees. The SCO plans are open to elderly at all levels of disability. The integration of acute and long term care benefits and a Medicaid capitation rate that is responsive to changing levels of frailty make SCO plans especially well suited for providing flexible and extended community care to enrollees. A capitation structure that is responsive to beneficiary health status independent of the setting of care is a significant component in the SCO design. The intended result of this component of the program design is to provide enhanced financing for community care, and ultimately reduce long term nursing home stays. While frail elderly with minimal to moderate disabilities will receive sufficient services to be able to remain in the community, only the most frail elderly of the SCO population will require nursing home care. In other words, the level of impairment in SCO enrollees entering the nursing home population will be higher than in non-SCO populations. Also, the impairment level of the community care population will rise as frail elderly are diverted from nursing home entry. Consequently, the state asked JAI to compare SCO populations to other like populations (i.e. matched cohorts) for outcome measures related to reduced rates of nursing home entry and changes in the frailty level in the SCO new-to-nursing and community care populations. Study Methods JAI based its evaluation methods on a comparison between Medicaid beneficiaries who enroll in a SCO plan and a matched cohort of Massachusetts Medicaid enrollees who remain under fee-for-service Medicaid and Medicare benefits. A propensity-matched case/control cohort design statistically adjusts for differences between the case and comparison population both in the selection of a comparable control population and in the multivariate effect measurement analyses. The study population consists of elderly populations with concurrent eligibility for Medicare and Medicaid benefits at the time of SCO entry. Analysis is based on patterns of nursing home utilization subsequent to the SCO enrollment date. The controls are analyzed on utilization patterns following a date matched to the cases’ plan enrollment dates (the control’s index date). The differential in Nursing Home use measured in the follow-up period among the cases and controls is the basis for the program’s effect estimate. Data Sources The baseline data for both the case and comparison populations was derived from Medicare and Medicaid administrative data. The data included program enrollment records with beneficiary demographics and health services claims data with detailed information on pre-enrollment patterns of procedures, diagnoses and episodes of acute hospital and long term care utilization. In the post-enrollment period, health services claims data is no longer available since the SCO plans are paid on a monthly capitated basis. SCO capitation payments are related to the assessed need of the beneficiary. In order to support a time varying capitation rate for a beneficiary, the monthly enrollment records include information on both nursing home status and need for community long term care services. Health services utilization data in the post-index date period is available for the comparison population. To implement a fair comparison between outcomes in the SCO enrollees and the comparison population a data source should be employed that provides equal information for the complete study population. The CMS Nursing Home Minimum Data Set (MDS) contains records of all stays in certified nursing facilities. The records include dates of stay, morbidity flags, activity of daily living assessments and other supporting data. Similarly the CMS Outcome and Assessment Information Set (OASIS) database includes records of utilization and assessment status for beneficiaries using licensed home health agencies, regardless of SCO enrollment. In summary the pre-enrollment period for study cases and controls can be profiled in depth using Medicaid and Medicare claims and enrollment data and the post enrollment period can be analyzed using SCO and Medicaid/Medicare enrollment data and the CMS MDS and OASIS sources. Table 1-Available Data Sources for SCO Enrollees and Comparison Population Data Sources SCO Enrollees Control Population SCO Enrollment Records Yes N/A SCO Assessment Records Yes N/A MassHealth SCO Enrollee Rate Payment File Yes N/A CMS Medicare Beneficiary Denominator Yes Yes OASIS Yes Yes MDS (NF/Certified Only) Yes Yes FFS Claims (MCR MCD) Pre-Enrollment Yes Elderly SCO Enrollee and All Medicaid Elderly Population Profiles The SCO program is designed to provide integrated care for beneficiaries regardless of setting of care and frailty status. The program is expected to enroll beneficiaries that are community-dwelling and relatively healthy all the way through permanent nursing home residents. The program is voluntary and actual enrollment patterns will rely on a number of factors. In order to be able to fairly construct a control population, the key characteristics of SCO enrollees and other dually eligible Medicaid beneficiaries must be understood. The tables below contain specific measures of comparison between CY 2004 SCO enrollees and non SCO-enrollee Medicaid dually eligible elderly. In order to profile pre-enrollee characteristics the descriptive tables include only Medicaid recipients with dual eligibility in 2004 and with a history of fee-for-service dual eligibility in 2003. As a result of this restriction approximately 5% of CY 2004 SCO enrollees are not included in the descriptive tables. Tables 2-5 provide demographic, long term care status and morbidity statistics for the Medicaid beneficiaries who enrolled in a SCO plan between March and December 2004 as well as the total Massachusetts elderly dual eligible beneficiaries. The data on pre- enrollment utilization, LTC status and diseases is from CY 2003 claims and enrollment records. Demographics The dually eligible elderly are a heterogeneous population. The key beneficiary groups are low income seniors who are categorically needy for Medicaid, and other seniors who qualify as medically needy for Medicaid once they become very frail and require nursing facility level of care. Table 2 SCO and All Dually Eligible Medicaid Elderly Demographic Profile CY 2004 SCO Cases All MA Elderly Duals N=635 N=96,313 Gender Male 27% 28% Female 73% 72% Age Categories Age 65-74 53% 42% Age 75-84 33% 35% Age 85+ 13% 23% On average, SCO enrollees in 2004 were younger than the total Medicaid elderly population, with fewer “very old” elderly enrollees and more “younger” elderly. Long Term Care (LTC) Utilization Status The highest frequency age group for long stay nursing home enrollees is over age 80. The younger age of SCO enrollees suggests that fewer nursing home residents are in the program than in the overall Medicaid elderly population. In Table 3, the SCO and Medicaid elderly populations are categorized according to their LTC status at the end of CY 2003. The categories are specifically designed for persistent users of community long term care services with episodes that are greater than 3 months in duration. The episode algorithm used for the stratification does allow for off-utilization periods related to inpatient hospital care. Table 3 CY 2003 Baseline LTC Utilization Profile CY 2004 SCO Cases All MA Elderly Duals N=635 N=96,313 LTC Setting Nursing Home Residents 12% 25% Community LTC Users 24% 12% Community Well 64% 63% On average, SCO enrollees were less likely to be long term nursing home residents than the total elderly dual eligible population and more likely to have a history of community focused care. Community s care includes Medicaid personal care, adult foster care, adult day health services and Medicare home health care. Chronic Disease Prevalence The SCO program is designed for the integration of Medicaid financing of community focused care with Medicare financing for medical services. The integration of benefits is attractive to beneficiaries with chronic disease and disability. Table 4 includes prevalence statistics in the pre-SCO enrollment period for selected chronic diseases that have a significant impact on the elderly. Table 4 CY 2003 Baseline Chronic Disease Profile CY 2004 SCO Cases All MA Elderly Duals N=635 N=96,313 CY 2003 Chronic Disease Diabetes 49% 34% Indicators CHD 39% 41% CVD 16% 20% CRD 27% 27% Arthritis 35% 30% CHF 19% 21% Chronic Disease Count Avg. 1.86 1.72 The SCO population does have a substantially higher rate of diabetes. However the SCO prevalence rate for the conditions in Table 4 is not remarkably higher than the rate for the overall elderly dually eligible. Utilization Rates The SCO program should provide easier access to care with improved coordination between supportive care providers and physicians. Limitations in access or difficulties in cross program benefit management may motivate SCO enrollment. Table 5 presents specific measures of utilization in the pre-SCO enrollment period. Table 5 CY 2003 Baseline Selected Utilization Measures CY 2004 SCO Cases All MA Elderly Duals N=635 N=96,313 Hospitalization in Year 35% 38% Re hospitalization within 30 Days 6% 7% General Practitioner Visit in Year 79% 87% Outpatient ER Visit in Year 51% 42% The patterns of acute care admission and readmission between SCO enrollees and the comparison are not substantially different; however there are some indications of both fewer general practice doctor visits and higher levels of outpatient emergency room utilization. Comparison Population The information in Tables 2-5 indicate a number of significant differences exist between the SCO population and the general Medicaid elderly population. The analytic challenge is to adjust for these differences such that a fair comparison can be made between the SCO enrollees and a like control population. JAI considered multivariate regression models to implement complex statistical adjustments to account for population differences. However, JAI chose a more sophisticated approach employing a two stage process: 1) selecting controls that are matched to cases using direct matching and statistical matching; 2) measuring effects using a multivariate regression model that further adjusts for the remaining differences in population characteristics. Tables 2-5 suggest the specifications for the control selection “propensity” model. The control selection process identifies a population that has the same balance of characteristics as observed in SCO enrollees. Population demographics, history of chronic disease, history of utilization of acute and long term care services, Medicaid and Medicare status are taken into account in the selection process. A combination of direct matching and propensity matching are used to select controls that are similar to the case population. Table 6 includes a complete list of the factors used for control selection. Table 6 Control Selection Factors Study Member Characteristic Match Type Gender Male Female Direct Match Index Age Categories Age <65 Age 65-74 Age 75-84 Age 85+ Direct Match Race/Ethnicity White Black Hispanic Other/Unknown Race Direct Match County Direct Match SCO Enrollment Month Direct Match MCR Status in Index Month Part A Only Part B Only Parts A & B Part A Only/State Paid Premium Part B Only/State Paid Premium Parts A & B/State Paid Premium Direct Match MA Risk Status Month prior to Index Community/Other Community LTC Nursing/Institutional LTC Direct Match MCD NH Case Mix Status Month prior to Index Direct Match Alzheimer's/CMI Indication in Month prior to Index Direct Match Base Period (1-3 months pre-index) Medical Utilization Inpatient Utilization Home Health Utilization Adult Foster Care Day Habilitation Utilization MCD Waiver Utilization Propensity Base Period (1-3 months pre-index) LTC Setting Hierarchy Long Stay Nursing Home Post Acute Care SNF Community Propensity FFS Dual Eligible in 2003 Direct Match Count of CY 2003 JAI Frailty/Impairment Groups 0 1-3 4-6 7+ Propensity CY2000 Chronic Disease Indicators Diabetes CHD CVD CRD Arthritis CHF Propensity The control selection specification aimed for 3 controls to be selected for each case. The result of the control selection process was the selection of 1,898 controls. A total of 24 cases could not be matched to a full set of 3 controls. With the selection of a comparison population, the program effectiveness analysis proceeds through the analysis of differential outcomes among SCO enrollees and controls in the post-enrollment/index date periods. Outcome Measures The impact of the SCO model is based on the dynamic nature of the Medicaid capitation rate and the requirement to perform regular assessments. There is a financial incentive to recognize a high risk case and to manage both the Medicaid and Medicare benefits to reduce adverse events. One of the most costly events to Medicaid is long term entry into a nursing facility. By increasing access to community long term care in a timely way, it is hypothesized that nursing home entry will be reduced. There are three major types of nursing home episodes: 1) extended rehabilitation care following a hospitalization; 2) end-of-life care; 3) long term placement in an institution. The first NH modality is perhaps a natural outcome of an acute care episode. The second model depends on the availability and quality of end-of-life community care. The third nursing home modality, long term placement, is the most costly to the Medicaid program and perhaps the most difficult to reduce since it is attributable in many cases to a history of functional decline. The outcome measure of interest is measured using multivariate proportional hazards model measuring the SCO program’s effect on average time to a NH entry. A supporting analysis examines the characteristics of SCO members who do enter nursing care for indications of higher risk. The analyses are restricted to SCO enrollees in 2004 and only examine outcomes through the end of CY 2005. [As more data becomes available, the analyses will be extended to later years]. Key Findings SCO enrollment began in March 2004. However, the median enrollment point in 2004 was September. Between SCO entry and the end of the study period a total of 52 SCO enrollees and 213 controls used a nursing facility. Most nursing facilities include both Medicare short term SNF and Medicaid long term care services. Both financing streams are included since SNF utilization is frequently a prelude to entry into Medicaid financed nursing home care. Table 7 provides descriptive detail on the number of study members entering a nursing facility utilization episode in the follow-up period as stratified by the type of episode. The first episode type, extended rehabilitation, is primarily driven by Medicare SNF stays. Other episode types are related to episodes that go beyond the Medicare SNF benefit and are primarily Medicaid financed. The episodes are derived from a mixture of NH MDS admission and assessment records, SCO rate cell in the month, and Medicare/Medicaid fee-for-service records for nursing home use. Virtually all the episodes are detectable in the MDS data source, however supporting information, including a more accurate admission date, was obtained from the other sources. The analysis of elapsed time to entry was based on a count of whole months from the index date to the month of the nursing facility utilization episode start. Table 7 Nursing Home Entry Rate by Type NH Entry and Type of Stay SCO Enrollees Control Population Nursing Home Entry Person Count 52 8.7% 213 12.0% Average months to NH Episode 7.27 6.16 Average episode months in NH 2.65 3.94 Discharge from Nursing Home Count 29 56.0% 111 52.0% Type of NH Stay Rehab (<4 months) 22 42.3% 96 45.1% End of Life (<6 months w/ Death) 9 17.3% 25 11.7% Long Term Residency (>=4 months) 8 15.4% 64 30.0% Undetermined (Limited Follow-up) 13 25.0% 28 13.1% The descriptive statistics demonstrate that SCO enrollees in comparison to the control population enter nursing facilities at a lower rate. In addition the time to first nursing utilization is greater and the time spent in a nursing facility episode is less than in the control population. For SCO enrollees that do use a nursing facility there is substantially lower frequency of long term residency. The one exception to the pattern is a higher tendency for nursing facility episodes for SCO end-of-life care. Due to the longer time to first utilization of a nursing facility by SCO enrollees, the number of users without sufficient follow-up time for a classification is elevated. For complete classification, four-six months of follow-up data is required to classify a nursing facility utilization episode. A multivariate proportional hazards model was employed to measure the effect of SCO enrollment on first use of nursing facilities. The specification of the model followed a step-wise selection to identify significant effect correlates. The results of the model are presented in Table 8. Table 8 Proportional Hazards Model for Nursing Home Entry Analysis of Maximum Likelihood Estimates Variable Para meter Estimate Std. Error Chi Sq Pr > Chi Sq Hazard Ratio 95% Confidence Limits Inpatient Utilization 1-3 Months Pre-index 0.62210 0.19333 10.35 0.0013 1.86 1.28 2.72 Home Health Utilization 1-3 Months Pre-index 0.61237 0.14258 18.45 <.0001 1.85 1.40 2.44 Assisted Living Utilization 1-3 Months Pre-index 1.64766 0.20843 62.49 <.0001 5.20 3.45 7.82 MCD Waiver Utilization 1-3 Months Pre-index 0.61855 0.21108 8.59 0.0034 1.86 1.23 2.81 NH Utilization 1-3 Months Pre-index 1.73018 0.27078 40.83 <.0001 5.64 3.32 9.59 SNF Utilization 1-3 Months Pre-index 1.18914 0.43427 7.50 0.0062 3.28 1.40 7.69 FFS-Dual Eligible from July 2003 -0.92930 0.23555 15.57 <.0001 0.40 0.25 0.63 2003 Eligibility/JAI Impairment Count 4-6 0.34968 0.15804 4.90 0.0269 1.42 1.04 1.93 2003 Eligibility/JAI Impairment Count 7+ 0.71455 0.20649 11.97 0.0005 2.04 1.36 3.06 2003 Eligibility/COPD 0.41219 0.13650 9.12 0.0025 1.51 1.16 1.97 SCO Enrollment -0.54218 0.16308 11.05 0.0009 0.58 0.42 0.80 The adjusted results demonstrate a strong and significant protective effect of the SCO program against nursing home entry after adjusting for selected pre-index measures of patient risk. The effect measure is a measure of time to entry during the period from SCO enrollment to the end of the data. The measure is a composite of nursing home utilization deferral and diversion. The Hazard Ratio can be directly interpreted as a 42% reduction in nursing home utilization risk. The second outcome indicator is the profile of functional ADL scores for the SCO and control population that start a nursing facility use episode. Not all MDS records include ADL information and only those episodes with the information and with MDS dates that correspond to the NF entry are included in the analysis. Table 9 is a univariate tabulation of scores for each major ADL category and a summary record across ADLs. The summary ADLs are computed based on all functional scores, the higher the score the greater the level of required assistance, the final tabulation in the table is based on a selection of ADLs. Table 9 ADL Levels at NF Episode Initiation TYPE OF NH STAY = ALL SCO Matched Cases Matched Controls N = 52 N = 213 Individual and Cumulative MDS ADL Scores Variable Description N Mean 25th Pctl 50th Pctl 75th Pctl N Mean 25th Pctl 50th Pctl 75th Pctl Asmtf MDS Assessment = No 13 25.0% 25 11.7% Asmtf MDS Assessment = Yes 39 75.0% 188 88.3% ADLs g1av BED* 39 2.18 0 3 3 188 2.08 1 2 3 g1bv TRANS* 39 2.28 1 3 3 188 2.46 2 3 3 g1cv WLK RM* 39 3.64 2 3 8 187 3.37 2 2 3 g1dv WLK CORR* 39 4.49 2 3 8 187 3.65 2 2 8 g1ev LOC UNIT 39 2.87 2 3 4 187 2.74 2 2 4 g1fv LOC OFFU 39 3.28 2 3 4 187 3.93 2 3 4 g1gv DRESS* 39 2.77 2 3 4 187 2.58 2 3 3 g1hv EAT* 39 0.77 0 0 1 188 0.69 0 0 1 g1iv TOLIET* 39 2.51 2 3 4 188 2.54 2 3 3 g1jv HYGIENE* 39 2.62 2 3 3 187 2.41 2 2 3 g2v BATH 39 3.05 3 3 4 187 2.86 3 3 3 ADLscore Sum Above ADLs 39 30.46 19 30 42 187 29.30 21 28 40 ADLscore8 Sum of Indicated (*) ADLs 39 21.36 13 21 32 187 19.93 13 18 28 The unadjusted data in Table 9 suggests that SCO enrollees entering a nursing facility episode are assessed with more functional impairment than the controls. The differential is not large, but as seen in Table 10 it is higher for the subset of episodes that are classified as long stay. The results are intriguing since the underlying hypothesis predicts that SCO community care will lead to decreased NH episodes, and the SCO enrollees who do enter a nursing facility will represent a more frail population. The adjusted analysis of nursing facility episode initiation shows a decrease from the number of expected nursing home entries as well as decrease in the length of stay. In addition, the unadjusted first year results of frailty status of SCO and controls suggest that the hypothesis of greater frailty among those SCO cases that do enter facility care is true. Table 10 ADL Levels at Long Stay NF Episode Initiation TYPE OF NH STAY = LTC (>=4 MONTHS) SCO Matched Cases Matched Controls N = 10 N = 78 Individual and Cumulative ADL Scores Variable Description N Mean 25th Pctl 50th Pctl 75th Pctl N Mean 25th Pctl 50th Pctl 75th Pctl Asmtf MDS Assessment = No 0 0.0% 3 3.8% Asmtf MDS Assessment = Yes 10 100.0% 75 96.2% ADLs g1av BED* 10 2.10 0 2.5 3 75 1.93 0 2 3 g1bv TRANS* 10 2.20 2 2.5 3 75 2.52 2 3 3 g1cv WLK RM* 10 4.20 2 3 8 75 3.33 2 3 3 g1dv WLK CORR* 10 5.20 2 8 8 75 3.52 2 2 8 g1ev LOC UNIT 10 2.70 2 3 4 75 2.75 2 3 4 g1fv LOC OFFU 10 3.50 2 3.5 4 75 3.75 2 3 4 g1gv DRESS* 10 3.00 2 3 4 75 2.61 2 3 3 g1hv EAT* 10 1.10 0 0.5 1 75 0.57 0 0 1 g1iv TOLIET* 10 2.70 1 3 4 75 2.67 2 3 3 g1jv HYGIENE* 10 2.80 2 3 4 75 2.60 2 3 3 g2v BATH 10 3.30 3 3 4 75 3.03 3 3 4 ADLscore Sum Above ADLs 10 32.80 19 37 45 75 29.28 21 30 40 ADLscore8 Sum of Indicated (*) ADLs 10 23.00 13 27 33 75 19.89 13 19 27 Conclusion The analysis of data from SCO enrollees and control subjects in CY2004-2005 provide an early glimpse of the effectiveness of the SCO program in reducing nursing home utilization. The data is limited by the amount of follow-up time available. Data from CY 2006 will be used to extend the analyses. The new analyses will both increase the study population to include CY 2005 SCO enrollees and their control counterparts and will increase the amount of follow-up time. The analyses of the data strongly indicate an early SCO impact in reducing nursing facility use. The analysis of ADL levels among new nursing home users suggests that the SCO population that does use facility care is frailer than the level seen in comparative non-SCO populations. The corollary that SCO financed community care is effective in maintaining more frail populations in the community is supported by these early analyses. CHD=Chronic Heart Disease, CVD= Cerebrovascular Disease, CRD= Chronic Respiratory Disease, CHF=Heart Failure JEN Associates, Inc. Confidential Page 12 7/21/2008