Analytical Issues: Using Linked MAX/ - PowerPoint Presentation

1. Analytical Issues: Using Linked MAX/NHANES Data to Study Obesity CostsAugust 7, 2012Presentation to the National Conference on Health StatisticsAllison Hedley Dodd, PhD

2. Medicaid Analytic eXtract (MAX) backgroundAnalytical issues for MAX cost dataLinking NHANES to multiple MAX recordsMulti-year Medicaid enrollmentInherent MAX issuesFactors that affect Medicaid cost dataDual statusBenefit statusS-CHIP statusManaged care enrollmentState variation in Medicaid dataOverview2

3. Background3

4. Generated from quarterly Medicaid Statistical Information System (MSIS) files submitted by stateSeven quarters processed to create annual files with adjudicated claims and reconciled enrollment recordsGoes through validation process to identify (and possibly) fix data issuesData issues summarized in anomaly tables (available on CMS website)MAX Background4

5. Annual set of files:Claims filesInpatient (IP)Long-term care (LT)Prescription drug (RX)Other (outpatient services/home-health care/medical equipment ) (OT)Person Summary file (PS)Enrollment informationSummary claims informationPS file sufficient for most analysesMAX Background5

6. Basis of Eligibility (BOE) categoryChildAdultDisabled (should be non-aged)AgedBOE provided by stateEligibility coded monthlyEligibility=EnrollmentEligibility≠UtilizationMedicaid Eligibility Data6

7. Fee-for-service data – reports cost of services providedCapitation data – reports monthly fee paid by states for enrollment in managed careEncounter data – reports no Medicaid cost but reports utilization of managed careTypes of Medicaid Claims7

8. Analytical Issues8

9. NHANES record matches to same Medicaid enrollee in multiple MAX yearsMedicaid enrollees are likely to be enrolled in Medicaid for more than one yearRecords that match to same NHANES participant will be highly correlatedAdjust analysis planSuggestion: Only keep records that match in same year that NHANES data collected (Note: NHANES survey year is a restricted variable)Linking to Multiple MAX Records – Part 19

10. NHANES record matches to same Medicaid enrollee within the same MAX yearEnrolled in Medicaid in more than one stateLikely okay to leave as isData issueAssigned more than one Medicaid ID within the same stateMay consider combining recordsLinking to Multiple MAX Records – Part 210

11. Including certain enrollees may underestimate average costs:DualsEnrollees with restricted benefitsS-CHIP only enrollees can muddy cost dataManaged care enrolleesEffect on costs unclearMay weaken associations with costMAX Classifications That Affect Cost Data11

12. Dual Status – Enrollees who qualify for both Medicaid and Medicare benefitsMedicare first payer for:Inpatient servicesOutpatient and physician servicesSome home healthPrescription Drugs (starting in 2006 with Part D)Limited Skilled Nursing Facility (SNF) servicesMedicaid pays for:Medicare premiums and cost-sharingLong-term careOptional services: dental, hearing, vision, home- and community-based servicesDual Status – Part 112

13. Dual enrollees likely to have lower Medicaid costs than non-duals due to cost-sharing39% disabled enrollees are duals (2008)92% aged enrollees are duals (2008)Suggestion: Remove duals from cost analysesNote: If remove duals, will be difficult to do analysis of ages 65+Dual Status – Part 213Source: Kaiser Commission on Medicaid and the Uninsured and Urban Institute estimates based on data from FY 2008 MSIS, 2012.

14. Restricted-Benefit – Medicaid enrollees eligible for only limited coverageThree main categories:Aliens eligible only for emergency servicesDualsEnrollees in 1115 waivers that provide only family planning services11% enrollees have restricted benefits (2008)Suggestion: Remove restricted-benefit enrollees from cost analyses because likely to have lower costs than full-benefit enrolleesRestricted Benefit Status14Source: Borck et al. "The Medicaid Analytic Extract 2008 Chartbook." 2012.

15. States cover children in Separate Children’s Health Insurance Program (S-CHIP) or CHIPCHIP data included in MSIS/MAXStates not required to submit S-CHIP data but some doSuggestion: Exclude S-CHIP only enrollees because enrollment and services are not fully reportedS-CHIP Only 15

16. Primary Care Case Management (PCCM) Capitation data for case management only (typically small fees)Cost for services captured in FFS data Vast majority of costs in FFS dataSuggestion: Keep PCCM enrollees in analysisComprehensive Managed Care (HMO/HIO/PACE) (CMC)Most care provided through programCosts captured in capitation data (not reflective of service use)Types of Managed Care – Part 116

17. CMC cont’d Possible for CMC enrollees to have FFS claimsPartial-year enrollmentCarve out servicesMajority of costs in capitation dataNot clear whether costs lower than FFS enrolleesDifficult to detect cost association because capitation payments do not vary by service useSuggestion: Remove CMC enrollees from analysis if looking for associationTypes of Managed Care – Part 217

18. Prepaid Health Plan (PHP) Specific ServicesBehavioral HealthDentalLong-term CareOtherCosts captured in capitation data (not reflective of service use)Used in conjunction with FFS or CMCIn some states, all enrollees in PHPTypically only covers carve out servicesSuggestion: Leave PHP enrollees in analysisTypes of Managed Care – Part 318

19. Analysis plan needs to define eligibility/enrollment for study population:Last month status Typically used for BOE categoryStatus occurred for all months in yearYields smallest but cleanest data set (rarely used)Status occurred for all months eligibleVery restrictive for CMC enrollmentStatus occurred for at least one month Typically used for CMC enrollmentDefining Eligibility/Enrollment19

20. Variation inherent in state-submitted dataPrograms differClassification of programs may differPHP vs CMCQuality of dataMaine 2005-present: only enrollment and RX dataMost Medicaid research is state-based but NCHS hesitant to reveal respondent’s state even in RDCMedicaid Data are State Data20

21. MAX variables can control for some differences but not allSuggestion: Use masked state variable to control for state differences without actually knowing the stateIssue: Cannot interpret data from MAX-NCHS merged data sets by stateControlling for State Variation21

22. Adjust costs to same yearAnalytical decisionsThink about whether total costs are being affected by/masking important informationLong-term care populationParticular service categoriesAnnual costs or per member per month (PMPM)?All enrollees or only service recipients?MAX Cost Analyses22

23. Analysis of enrollees 20+ years3,227 records - NHANES 1999-2004 and feasibility file9,809 records matched in NHANES and MAX 1999-2004Of those, 3,847 records matched with MAX year in NHANES data release years Of those,1,061 records matched who were full-benefit, non-duals who were not in S-CHIP only and never in CMCRestricted Benefits – 439 (11.4% of 3,847 records)Duals – 1,818 (47.3%)S-CHIP Only – 6 (0.2%)CMC enrollees – 1,128 (29.3%)Obesity Analysis Example23

24. Analysis cont’dAfter implementing MAX cost restrictions, only 50 elderly records (ages 65+)Final sample size of enrollees ages 20-64 years: 603 recordsImplemented NHANES restrictionsNot pregnantMeasured BMI data availableLimited to 1 match within NHANES data release years (used survey year) (n=329)Resolved duplicates (n=4)Obesity Analysis Example cont’d24

25. NCHS feasibility files useful but can overestimate sample sizeFeasibility file: 3,227 recordsFinal sample: 603 recordsRestricted matches to same year Implemented MAX cost restrictionsRemoved elderlyResolved duplicatesMAX/NHANES cost analyses may be limitedMAX/NHANES enrollment analyses would not be affected as greatly due to fewer MAX restrictionsConclusions25

26. Allison Hedley Doddadodd@mathematica-mpr.comMAX datahttp://www.cms.gov/Research-Statistics-Data-and-Systems/Computer-Data-and-Systems/MedicaidDataSourcesGenInfo/MAXGeneralInformation.htmlQuestions?26