integrated health care survey designs: analytical enhancements achieved through linkage of surveys...

Integrated Health Care Survey Designs: Integrated Health Care Survey Designs: Analytical Enhancements Achieved Through Linkage of Analytical Enhancements Achieved Through Linkage of

Surveys and Administrative DataSurveys and Administrative Data

2008 European Conference on Quality in 2008 European Conference on Quality in Official Statistics (Q2008) Official Statistics (Q2008)

Steven B. Cohen, Ph.D.Steven B. Cohen, Ph.D.

Purpose of DiscussionPurpose of Discussion

Integrated survey design featuresIntegrated survey design features Related enhancements to data quality and analytical capacityRelated enhancements to data quality and analytical capacity Capacity to reduce bias attributable to survey nonresponseCapacity to reduce bias attributable to survey nonresponse Applications to AHRQ Data Portfolio and Research Initiatives Applications to AHRQ Data Portfolio and Research Initiatives

to inform to inform health outcomeshealth outcomes Limitations Limitations Future model for considerationFuture model for consideration

Advancing Advancing Excellence in Excellence in Health CareHealth Care

Health Outcomes FocusHealth Outcomes FocusPart of AHRQ’s MissionPart of AHRQ’s Mission

AHRQ Mission: To improve the AHRQ Mission: To improve the quality, safety, quality, safety, efficiency,efficiency, and and effectiveness of health care for all effectiveness of health care for all AmericansAmericans


Integrated survey design featuresIntegrated survey design features

Direct linkage between sample members in core Direct linkage between sample members in core survey with larger host survey; administrative survey with larger host survey; administrative records; or follow-up surveysrecords; or follow-up surveys

Use of secondary data (e.g. aggregate data at the Use of secondary data (e.g. aggregate data at the county/state level) as core component of surveycounty/state level) as core component of survey

Prior survey record of call data informs data Prior survey record of call data informs data collection strategiescollection strategies

Informs sample design, nonresponse and Informs sample design, nonresponse and poststratification adjustments, imputation and data poststratification adjustments, imputation and data supplement for item nonresponsesupplement for item nonresponse

Need for greater attention to ensuring confidentiality: Need for greater attention to ensuring confidentiality: limitations in public use data limitations in public use data


Capacity to reduce bias attributable Capacity to reduce bias attributable to survey nonresponseto survey nonresponse

Adjustments for unit nonresponseAdjustments for unit nonresponse Detailed information available on demographic/socio-Detailed information available on demographic/socio-

economic characteristics of both respondents/and economic characteristics of both respondents/and nonrespondents from sample frame of host survey nonrespondents from sample frame of host survey administrative records administrative records

Incorporation of secondary dataIncorporation of secondary data

Adjustments for item nonresponseAdjustments for item nonresponse Data replacementData replacement Cold deck imputationCold deck imputation

Agency for Healthcare Research and QualityAgency for Healthcare Research and QualityAdvancing Excellence in Health CareAdvancing Excellence in Health Care • • www.ahrq.govwww.ahrq.gov


Medical Expenditure Panel SurveyMedical Expenditure Panel Survey(MEPS)(MEPS)

Annual Survey of 15,000 households:Annual Survey of 15,000 households: provides national estimates of health care use, expenditures, insurance provides national estimates of health care use, expenditures, insurance coverage, sources of payment, access to care and health care qualitycoverage, sources of payment, access to care and health care quality

Permits studies of:Permits studies of: Distribution of expenditures and sources of paymentDistribution of expenditures and sources of payment Role of demographics, family structure, insuranceRole of demographics, family structure, insurance Measurement of expenditures in managed careMeasurement of expenditures in managed care Expenditures for specific conditionsExpenditures for specific conditions Trends over timeTrends over time

Advancing Advancing Excellence in Excellence in Health CareHealth Care Design SpecificationsDesign Specifications

Target Precision Target Precision Specifications for national Specifications for national and regional estimates; policy and regional estimates; policy relevant subgroupsrelevant subgroups

Overall Design effect of 1.6Overall Design effect of 1.6 200 PSU design (Max)200 PSU design (Max) Overall/round specific survey Overall/round specific survey

response rate requirementsresponse rate requirements

Linkage to NHISLinkage to NHIS Multistage designMultistage design Disproportional samplingDisproportional sampling Longitudinal designLongitudinal design Minimize survey cost for Minimize survey cost for

fixed precisionfixed precision


Key Features of MEPS-HCKey Features of MEPS-HC

Survey of U.S. civilian noninstitutionalized populationSurvey of U.S. civilian noninstitutionalized population Sub-sample of respondents to the National Health Interview Survey (NHIS)Sub-sample of respondents to the National Health Interview Survey (NHIS) Oversample of minorities and other target groupsOversample of minorities and other target groups Panel Survey – new panel introduced each yearPanel Survey – new panel introduced each year

– Continuous data collection over 2 ½ year periodContinuous data collection over 2 ½ year period

– 5 in-person interviews (CAPI) 5 in-person interviews (CAPI)

– Data from 1st year of new panel combined with data from 2nd year of Data from 1st year of new panel combined with data from 2nd year of previous panelprevious panel

MEPS Overlapping PanelsMEPS Overlapping Panels(Panels 8 and 9)(Panels 8 and 9)

MEPS Household Component MEPS Panel 8 2003-

2004

Round 2 Round 3 Round 4 Round 5

Round 1 Round 2 Round 3

MEPS Panel 92004-2005

1/1/2003 1/1/2004

Round 1NHISNHIS20022002

NHISNHIS20032003 Round 4 Round 5


MEPS Household ComponentMEPS Household ComponentSample DesignSample Design

Oversampling of policy relevant domainsOversampling of policy relevant domains19961996 Minorities (Blacks & Hispanics)Minorities (Blacks & Hispanics)

19971997 MinoritiesMinoritiesLow income Low income

Children with activity limitationsChildren with activity limitations

Adults with functional limitationsAdults with functional limitations

Predicted high expenditure casesPredicted high expenditure cases

ElderlyElderly

1998-20011998-2001 MinoritiesMinorities

2002+2002+ Minorities, Asians, Low IncomeMinorities, Asians, Low Income

Advancing Advancing Excellence in Excellence in Health CareHealth Care MEPS ComponentsMEPS Components

Household Component (HC) - 15, 000 households, Household Component (HC) - 15, 000 households, 37, 000 individuals37, 000 individuals

Medical Provider Component (MPC) - designed to Medical Provider Component (MPC) - designed to supplement /replace household reported expenditure supplement /replace household reported expenditure data data

Insurance Component (IC) - 30,000 establishments; Insurance Component (IC) - 30,000 establishments; elicits insurance availability, premium contribution, elicits insurance availability, premium contribution, and benefit provision information; can be used to and benefit provision information; can be used to generate estimates at the state levelgenerate estimates at the state level

IC sample linked to HC designed to supplement or IC sample linked to HC designed to supplement or replace household reported health coverage data replace household reported health coverage data


MEPS - Integrated Survey Design MEPS - Integrated Survey Design FeaturesFeatures

National Health Interview Survey serves as sample National Health Interview Survey serves as sample frame for Household Componentframe for Household Component

Census Bureau Business Register serves as Census Bureau Business Register serves as Insurance Component sample frame Insurance Component sample frame

Secondary data on health care measures supplement Secondary data on health care measures supplement surveys Linked survey of medical providerssurveys Linked survey of medical providers

Linked survey of employersLinked survey of employers

Distinct data sources linked for longitudinal analysesDistinct data sources linked for longitudinal analyses


DemographicsDemographics

(ref. person)(ref. person)

[9][9]

Household Household CharacteristicsCharacteristics

[5][5]

Socio-Economic Socio-Economic StatusStatus

[6][6]

GeographicGeographic

[4][4]

HealthHealth

[5][5]

Age Age DU sizeDU size Poverty statusPoverty status Census regionCensus region Health statusHealth status

Race/ethnicityRace/ethnicity Refused phone #Refused phone # EducationEducation MSA sizeMSA size Need help -Need help -personal care personal care

Marital statusMarital status IncomeIncome MSA/central city MSA/central city Not working -Not working -health reasonshealth reasons

GenderGender Type of PSUType of PSU Employment statusEmployment status Urban/ruralUrban/rural

Any Asian in HHAny Asian in HH

Any Black in HHAny Black in HH

Interview language Interview language Type of home Type of home (house, apt., etc.)(house, apt., etc.)

Family medical Family medical expense categoryexpense category

No. of nights in No. of nights in hospitalhospital

U.S. CitizenU.S. Citizen Time w/out phoneTime w/out phone Home ownershipHome ownership Health care Health care coveragecoverage

Born in U.S.Born in U.S.

29 potential predictors (HH or reference person 29 potential predictors (HH or reference person characteristics) of DU level NR based on NHIS data (all characteristics) of DU level NR based on NHIS data (all

eligible MEPS sample persons)eligible MEPS sample persons)


Testing for Panel EffectTesting for Panel Effect

Advancing Advancing Excellence in Excellence in Health CareHealth Care Medical Provider Component Medical Provider Component

PurposePurpose Compensate for household item nonresponseCompensate for household item nonresponse Gold standard for expenditure estimatesGold standard for expenditure estimates Greater accuracy and detailGreater accuracy and detail Imputation sourceImputation source Supports methodological studiesSupports methodological studies

Advancing Advancing Excellence in Excellence in Health CareHealth Care Medical Provider Component Medical Provider Component

Targeted SampleTargeted Sample All associated hospitals and associated physiciansAll associated hospitals and associated physicians All associated office-based physiciansAll associated office-based physicians All associated home health agenciesAll associated home health agencies All associated pharmaciesAll associated pharmacies

Data CollectedData Collected Dates of visitDates of visit Diagnosis and procedure codesDiagnosis and procedure codes Charges and paymentsCharges and payments


MPC: Correction Source for MPC: Correction Source for Item NonresponseItem Nonresponse

YYijij = Imputed $ = Imputed $ijijnonresponsenonresponsenonresponsenonresponseYYijij = Household $ = Household $ijij 11nonresponsenonresponsereportedreportedYYijij = Provider $ = Provider $ijij

reportedreportedNonresponseNonresponseYYijij = Provider $ = Provider $ijij reportedreportedReportedReportedMEPS value - YMEPS value - Yijij ProviderProviderHouseholdHousehold

Source for event level expendituresSource for event level expenditures

1Recalibrated as necessary based on analyses of concordance between sources


Determination of Factors for Determination of Factors for Expenditure ImputationExpenditure Imputation

Factors associated with

predicting medical

expenditures

Factors associated with

item nonresponse

Hot Deck Imputation:

Classification Variables for Donors and Recipients

Agency for Healthcare Research and QualityAgency for Healthcare Research and QualityAdvancing Excellence in Health CareAdvancing Excellence in Health Care • • www.ahrq.govwww.ahrq.gov

The Utility of Extended Longitudinal Profiles

in in Predicting Future Health Care Predicting Future Health Care

ExpendituresExpenditures

Advancing Advancing Excellence in Excellence in Health CareHealth Care Predictive Models Predictive Models

Model 1: Logistic Model with prior year’s medical Model 1: Logistic Model with prior year’s medical expenditures and precursor information (t-1).expenditures and precursor information (t-1).

(Y=1 top 10% in $s; 0 otherwise)(Y=1 top 10% in $s; 0 otherwise) Model 2: Logistic Model with prior year’s medical Model 2: Logistic Model with prior year’s medical

expenditures (t-1) and precursor information (t-1 expenditures (t-1) and precursor information (t-1 and t-2).and t-2).

Advancing Advancing Excellence in Excellence in Health CareHealth Care Evaluation of Model PerformanceEvaluation of Model Performance

Develop Model on 2004-2005 MEPS Panel, 2003 NHISDevelop Model on 2004-2005 MEPS Panel, 2003 NHIS

Validate Model on 2006-2007 MEPS Panel, 2005 NHISValidate Model on 2006-2007 MEPS Panel, 2005 NHIS

Model 1: Logistic Model with prior year’s medical Model 1: Logistic Model with prior year’s medical expenditures and precursor information (t-1).expenditures and precursor information (t-1).

Model 2: Logistic Model with prior year’s medical Model 2: Logistic Model with prior year’s medical expenditures (t-1) and precursor information (t-1 and expenditures (t-1) and precursor information (t-1 and t-2).t-2).

Advancing Advancing Excellence in Excellence in Health CareHealth Care Insurance Component - PurposeInsurance Component - Purpose

Availability of health insuranceAvailability of health insurance Access to health insuranceAccess to health insurance Cost of health insuranceCost of health insurance Benefit and payment provisions of private Benefit and payment provisions of private

health insurancehealth insurance

Advancing Advancing Excellence in Excellence in Health CareHealth Care Insurance Component - SampleInsurance Component - Sample

30,000 establishments: derived from Census 30,000 establishments: derived from Census Bureau frameBureau frame

Supports national and state estimates Supports national and state estimates Employers linked to HC sampleEmployers linked to HC sample Data released in tabular form on MEPS Data released in tabular form on MEPS

websitewebsite


Key Administrative Data Available for MEPS Key Administrative Data Available for MEPS Insurance ComponentInsurance Component

1.1. IndustryIndustry

2.2. PayrollPayroll

3.3. Age of FirmAge of Firm

4.4. Establishment SizeEstablishment Size

5.5. Enterprise SizeEnterprise Size

6.6. LocationLocation

7.7. Multi/Single Unit FirmMulti/Single Unit Firm

8.8. Form of OrganizationForm of Organization


Uses of Administrative Data in the Insurance Uses of Administrative Data in the Insurance ComponentComponent

1.1. SamplingSampling2.2. ImputationImputation3.3. EditingEditing4.4. ModelingModeling5.5. Table ProductionTable Production6.6. Weight Adjustment for Non Response and Control TotalsWeight Adjustment for Non Response and Control Totals


Key Improvements Due to Use of Administrative Data Key Improvements Due to Use of Administrative Data in the Insurance Componentin the Insurance Component

1.1. Reduces Respondent BurdenReduces Respondent Burden2.2. Improves Sampling PrecisionImproves Sampling Precision3.3. Helps Find Respondent ErrorsHelps Find Respondent Errors4.4. Improves Weight AdjustmentImproves Weight Adjustment5.5. Allows Estimates Be Made for Numerous Key CategoriesAllows Estimates Be Made for Numerous Key Categories6.6. Essential for Modeling and ResearchEssential for Modeling and Research


Health Workforce Analysis: Area Health Workforce Analysis: Area Resource FileResource File

Area Resource File (ARF) is a health resource information Area Resource File (ARF) is a health resource information system that enables policymakers, researchers, planners system that enables policymakers, researchers, planners and others to analyze the current state of health care and others to analyze the current state of health care access at the county level.access at the county level.

Content includes geographic codes and classifications; Content includes geographic codes and classifications; health professions supply and detailed demographics; health professions supply and detailed demographics; health facility numbers and types; hospital utilization; health facility numbers and types; hospital utilization; population characteristics and economic data; population characteristics and economic data; environment; and health professions training resources.environment; and health professions training resources.

Sponsored by HRSASponsored by HRSA

Advancing Advancing Excellence in Excellence in Health CareHealth Care AHRQ Data CenterAHRQ Data Center

Provides researchers access to non-public use MEPS Provides researchers access to non-public use MEPS data (except directly identifiable information) and other data (except directly identifiable information) and other restricted data sets;restricted data sets;

Mode of data analysis Mode of data analysis – on a secure LAN at AHRQ, Rockville on a secure LAN at AHRQ, Rockville

– task order agreement with data contractor task order agreement with data contractor

– combinations of both.combinations of both.

Advancing Advancing Excellence in Excellence in Health CareHealth Care User Supplied Secondary DataUser Supplied Secondary Data

Data Type and/or SourceData Type and/or Source– Area Resource File Area Resource File – Health Care Market Variables @ zip code levelHealth Care Market Variables @ zip code level– Proprietary county level HMO variablesProprietary county level HMO variables– State and MSA level data from Interstudy PublicationsState and MSA level data from Interstudy Publications– State level Medicaid and poverty variablesState level Medicaid and poverty variables– County level unemployment ratesCounty level unemployment rates– State level data from BLSState level data from BLS– NHISNHIS– Urban InstituteUrban Institute– Academy for Health Services Research and PolicyAcademy for Health Services Research and Policy– Census BureauCensus Bureau– HCFAHCFA– Proprietary state level dataProprietary state level data– State income tax rates State income tax rates – Centers for Medicare and Medicaid ServicesCenters for Medicare and Medicaid Services

Research FocusResearch Focus– Changes in Medicaid and SCHIPChanges in Medicaid and SCHIP– Access to Care IssuesAccess to Care Issues– Changes in Health Insurance CoverageChanges in Health Insurance Coverage– Disparities in Health Care Expenditures for FamiliesDisparities in Health Care Expenditures for Families– State Level Health Care ExpendituresState Level Health Care Expenditures

Advancing Advancing Excellence in Excellence in Health CareHealth Care LimitationsLimitations

Greater restrictions in data access for public use

Competing demands on host sample frames More frequent survey contacts reduce overall

response rate Requires greater coordination across data

sources and organizations


SummarySummary

Capacity of integrated survey designs to serve as Capacity of integrated survey designs to serve as cost efficient sampling framescost efficient sampling frames

Capacity of integrated survey designs to reduce bias Capacity of integrated survey designs to reduce bias attributable to nonresponseattributable to nonresponse

Related enhancements to data quality and analytical Related enhancements to data quality and analytical capacitycapacity

MEPS applications MEPS applications Limitations Limitations Discussion questionsDiscussion questions

integrated health care survey designs: analytical enhancements achieved through linkage of surveys...

Documents