Factors Predicting Transitions from Medicare-Only to Medicare-Medicaid Enrollee Status. Data and Methodology


We combined data from multiple sources to examine beneficiary characteristics and service utilization patterns that predict transitions from Medicare-only eligibility to MME status on a monthly basis. We also examined beneficiary characteristics and service utilization patterns that predict entry into nursing home care not covered by the Medicare skilled nursing facility (SNF) benefit during 2009, in which case beneficiaries either pay for the care with private insurance, Medicaid, or out of pocket.

We used the 2009 Medicare Master Beneficiary Summary File (MBSF) to identify all individuals age 22 and older who were ever enrolled as MMEs in 2009.4 We used the Medicare 5 percent sample in the MBSF to provide a group of Medicare beneficiaries who never were MMEs during the year and to estimate the size of the population of Medicare-only beneficiaries in 2009. The MBSF includes summary demographic information about all Medicare beneficiaries, including date of birth, date of death, state of residence, monthly Medicare managed care enrollment, monthly Medicaid-eligibility status, Medicaid benefits, and original reason for Medicare eligibility (disability, aged, or end-stage renal disease [ESRD]). To assess service utilization, we excluded beneficiaries covered by a comprehensive Medicare or Medicaid managed care plan during 2009 as well as beneficiaries not continuously eligible for Medicare during the six months prior to a potential transition, because Medicare and Medicaid claims data required to assess service utilization are not generally available for these beneficiaries. We also excluded beneficiaries who became eligible for Medicare on the basis of ESRD, because these enrollees have unique pathways to Medicare eligibility.5

After identifying our sample of enrollees, we linked MBSF data with other data sources to obtain additional information about the service utilization and demographic characteristics of these beneficiaries.

  • Chronic Conditions Warehouse (CCW) Timeline Files, 2008 and 2009. These files, maintained by Centers for Medicare and Medicaid Services (CMS), contain person-level residence status for all Medicare beneficiaries for each day during a year. These files were used to compute nursing home stays (including Medicare-financed SNF stays and other nursing home stays).

  • Medicare Inpatient, Outpatient, and Hospice Claims Files, 2008 and 2009. These files enabled us to compute inpatient stays, ER visits, and hospice use.

  • MBSF Chronic Conditions Segment, 2009. This file contains CCW flags identifying 27 chronic conditions, including Alzheimer's, related disorders, dementia, and depression.

  • MBSF Cost and Utilization Segment, 2008. This file enabled us to control for aggregated 2008 Medicare Part B services and durable medical equipment (DME) claims.

We used the data to estimate a multivariate logistic model for the likelihood that a beneficiary would become dually eligible, conditional on being Medicare-only in the prior month, as a function of prior service utilization and chronic conditions. To do this, we constructed a sample of pooled person-month observations that, for any given month from February to December, was representative of beneficiaries who were Medicare-only in the prior month.6 This approach resulted in a dataset consisting of 16,807,633 person-month observations based on 1,767,388 individuals.7 To control for underlying month-to-month variation in transition rates, we include an indicator variable for each month. Because the population of Medicare beneficiaries under age 65 differs substantially from those age 65 and above (beneficiaries under age 65 typically qualify for Medicare via a disability versus those who age in at 65), we estimate separate models for these two broad age groups.

Table 1 presents summary statistics for the variables in the regression model. In any given month, 5 per thousand Medicare-only beneficiaries under age 65 become eligible for Medicaid. This rate is much higher than the corresponding rate of 1 per thousand for Medicare-only beneficiaries age 65 and over. Our main measures of service utilization include an indicator of whether a beneficiary had an ER visit resulting in an inpatient admission in the past six months, an indicator of whether a beneficiary had a SNF stay in the past six months, and an indicator of a nursing home stay not financed by Medicare in the past six months. On average, 6 per thousand Medicare-only beneficiaries under age 65 had a SNF stay in the past six months and 3 per thousand had a non-SNF nursing home stay. The corresponding rates for beneficiaries 65 and over are 23 and 10 per thousand, respectively. To account for chronic conditions we include an indicator variable for Alzheimer's, related disorders, or dementia; an indicator variable for depression; and an indicator variable for any of the other (that is, physical) chronic conditions. On average, 170 per thousand beneficiaries under 65 and 90 per thousand beneficiaries 65 and over, have depression. Other individual level controls are number of planned inpatient stays, number of outpatient ER visits, and number of weeks of home health use, all measured within the past six months. We also include 2008 DME and Part B physician payments to control for health status. To account for observed and unobserved state-level characteristics, we included state fixed effects in our regression models.8

TABLE 1. Summary Statistics for Variables Included in Models, by Age Group
Variable Age Under 65 Age 65 and Above
Mean S.D. Mean S.D.
Transitioned from Medicare-only to MME Status   0.005     0.072     0.001     0.035  
Number of Planned Inpatient Stays 0.042 0.237 0.045 0.235
Number of Outpatient ER Visits 0.189 0.816 0.131 0.473
Number of Weeks of Home Health Use 0.260 2.062 0.462 2.613
ER Visit Resulting in Inpatient Stay 0.047 0.213 0.056 0.230
SNF Stay 0.006 0.076 0.023 0.150
Other Nursing Home Stay 0.003 0.056 0.010 0.098
ER Visit Resulting in Inpatient Stay, and SNF Stay 0.003 0.056 0.013 0.113
ER Visit Resulting in Inpatient Stay, and Other NH Stay 0.001 0.027 0.002 0.049
SNF and Other Nursing Home Stay 0.001 0.028 0.004 0.060
ER Visit Resulting in Inpatient Stay, SNF Stay, and Other Nursing Home Stay   0.000 0.020 0.002 0.045
Alzheimer's, Related Disorder, or Dementia 0.018 0.132 0.086 0.280
Depression 0.170 0.376 0.087 0.281
Number of Other Physical Chronic Conditions 1.709 2.126 3.154 2.491
Any Hospice 0.001 0.038 0.005 0.070
Total 2008 DME Payments 252 1,349 203 809
Total 2008 Part B Physician Payments 268 436 434 470
Age 53 9 75 8
Female 0.432 0.495 0.550 0.498
Number of Person-Month Observations 2,480,450 14,327,183
Number of Unique Beneficiaries 312,858 1,454,530
SOURCE: Mathematica analysis of 2008 and 2009 MBSF, CCW Timeline, and Medicare Claims Files.
NOTE: Service utilization is measured within the prior six months, unless described otherwise. Statistics are weighted to be representative of the population.

We then estimated a multivariate logistic model, using the same set of predictors, for the likelihood that a beneficiary would enter nursing home care that was financed by private insurance, Medicaid, or out of pocket. For this analysis, in addition to excluding those already in MME status in a previous month, we also removed person-month observations where a beneficiary was already in a non-SNF nursing home. This restriction reduced the sample size to 16,459,350 person-month observations for 1,730,765 individuals across both age groups.

For all the regression analyses, to make the results meaningful we use the estimates (coefficients and odds ratios from the estimated regression models as presented in the Appendix) to infer predicted transition rates for the sample. We then compare predicted transition rates across different patterns of service utilization, to understand the impact of a particular service utilization pattern on the likelihood of becoming dually eligible.

View full report


"MMTransV2.pdf" (pdf, 233.84Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®