5

Data Sources

The development and production of a measure of medical care economic risk (MCER) depend on the available data. This chapter reviews the data sources that might be used to construct a measure of MCER. Our discussion covers both the medical expenditure risk and financial resources components of a potential measure. In the end, the choice of a survey to serve as home to the measure is tightly constrained by the available options. Nevertheless, it is important to understand the strengths and limitations of available data on all elements that are relevant in determining the cost of addressing potential medical care needs and the ability of families and individuals to pay for those costs.

OVERVIEW OF SURVEY DATA ON MEDICAL CARE COSTS AND FINANCIAL RESOURCES

In reviewing potential data sources, we distinguish between development and production applications. The data requirements for developing a measure are not the same as the requirements for producing a measure on a recurring basis. Development has more extensive data needs than production, but on a number of dimensions the requirements are less demanding. For production, the survey must be (1) annual, (2) representative of the civilian noninstitutionalized population, (3) released on a timely basis, and (4) have a sample sufficiently large to provide precise measures of change in MCER over time.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 89
5 Data Sources The development and production of a measure of medical care eco- nomic risk (MCER) depend on the available data. This chapter reviews the data sources that might be used to construct a measure of MCER. Our discussion covers both the medical expenditure risk and financial resources components of a potential measure. In the end, the choice of a survey to serve as home to the measure is tightly constrained by the available options. Nevertheless, it is important to understand the strengths and limitations of available data on all elements that are relevant in determining the cost of addressing potential medical care needs and the ability of families and individuals to pay for those costs. OVERVIEW OF SURVEY DATA ON MEDICAL CARE COSTS AND FINANCIAL RESOURCES In reviewing potential data sources, we distinguish between develop- ment and production applications. The data requirements for developing a measure are not the same as the requirements for producing a measure on a recurring basis. Development has more extensive data needs than produc- tion, but on a number of dimensions the requirements are less demanding. For production, the survey must be (1) annual, (2) representative of the civilian noninstitutionalized population, (3) released on a timely basis, and (4) have a sample sufficiently large to provide precise measures of change in MCER over time. 89

OCR for page 89
90 MEDICAL CARE ECONOMIC RISK Data Requirements for the Measurement of MCER To develop a prospective measure of MCER requires longitudinal data, so that medical care expenditures observed prospectively over the course of a period—ideally a year—can be related to characteristics observed at the start of that period.1 Relevant baseline characteristics include those that are potentially predictive of medical expenditures. These include the following: · General health status ·  hronic conditions—in particular, conditions that are associated C with actual or potential expenditures · Health insurance coverage · Breadth of services/treatments covered ·  otential liability for out-of-pocket costs—copays, deductibles, and P caps on personal expenditures · Current health insurance premiums Actual out-of-pocket expenditures for medical care in the prior year may be the strongest predictor of expenditures during the current year, and although they are not a baseline characteristic per se, these expenditures ought to be included in the development of a predictive model of prospective risk. Both premiums and other out-of-pocket expenditures should be included. With longitudinal data, out-of-pocket medical expenditures and pre- miums over the course of the next year would be compared with resources over the same period to determine the economic burden imposed by medical expenditures. This burden measure would become the dependent variable in a model predicting economic risk in the second year from the set of baseline characteristics listed above. This model would then be applied to the data set used to estimate MCER on an annual basis. Requirements for the resources component include · Earned income ·  nearned income, equivalent to the unearned component of Cen- U sus money income ·  ash value of in-kind benefits, such as the Supplemental Nutrition C Assistance Program, school free and reduced-price breakfast and lunch programs, and housing assistance · Taxes paid—federal, state, and payroll · Work-related expenses, including child care and commuting · Liquid assets 1  Chapter 4 also discusses the calculation of a retrospective measure of MCER using CPS ASEC data. We focus here on the preferred prospective measure, which requires longitudinal data.

OCR for page 89
DATA SOURCES 91 The in-kind benefits, taxes, and work-related expenses items are needed in conjunction with earned and unearned money income to construct dispos- able income. We recommend in Chapter 3 that a portion of liquid assets be included in family resources as well. Finally, both the risk variables and the resources variables must be recorded at the person level, so that the variables in each case can be ag- gregated to the health insurance unit level (for aspects of modeling risk) and family level (for comparing risk with resources). Sufficient information on family relationships must be included to enable the membership of each health insurance unit and family in a household to be identified. Data Sources for Development of a Measure The panel looked closely at three longitudinal surveys: the Medical Ex- penditure Panel Survey (MEPS) (see http://www.meps.ahrq.gov), the Survey of Income and Program Participation (SIPP) (see http://www.census.gov/ SIPP), and the Health and Retirement Study (HRS) (see http://hrsonline.isr. umich.edu). The HRS has the immediate disadvantage of not covering the entire population; its universe is persons over 50 at its broadest. In addi- tion, the interview frequency for the HRS is every 2 years whereas MEPS and SIPP currently conduct two to three interviews per year. (SIPP is being redesigned to conduct one interview per year.) With MEPS and SIPP, then, one can observe expenditures in the year following the measurement of baseline characteristics. That the HRS collects expenditure data covering the time since the last interview (typically 2 years) could affect their quality, although comparisons suggest that means and distributions of expenditures are generally similar to MEPS outside the upper tail (Hurd and Rohwedder, 2009). The quality of the income and asset information collected in the HRS—especially for retired persons—is a particular strength (on the com- parison of Current Population Survey and HRS income data, see Hurd and Rohwedder, 2006). Table 5-1 summarizes the collection of relevant variables in the three surveys. None of the three surveys collects all the variables that would be required to develop a prospective measure of MCER as described in Chap- ter 4. Most notably, none of the three surveys collects a description of the services and treatments covered by each person’s health insurance plan, and none of the surveys collects sufficient information with which to assess each sample member’s potential liability for out-of-pocket medical costs, although MEPS and the HRS do collect limited information: participation in health maintenance organizations in MEPS and the HRS and whether coverage for a preexisting condition is limited in the HRS. As noted by Czajka (in Part III of this report), MEPS collected detailed information on the health insurance plans of sample members in 1996 but has not done

OCR for page 89
92 MEDICAL CARE ECONOMIC RISK TABLE 5-1  Collection of Variables Needed to Develop a Prospective Measure of MCER in Three Longitudinal Surveys Variable MEPS SIPP HRS Measures of Health General health status (poor to excellent) X X X Activity limitations X X X Functional limitations X X X Chronic medical conditions X X Measures of Health Insurance Coverage Current health insurance coverage X X X Services/treatments covered Potential liability for out-of-pocket costs a a,b Current health insurance premiums X Measures of Medical Expenditures Prior year insurance premiums Xc X Prior year out-of-pocket expenditures X X X Insurance premiums during the next year Xc X X Out-of-pocket expenditures during the next year X X X Measures of Resources Earned income X X X Unearned income X X X In-kind benefits d X d Taxes paid e Commuting and child care expenses X Liquid assets Xf X X Measures of Family Relationships Relationship to the householder X X g All parent-child relationships X X All marital relationships X X NOTES: HRS = Health and Retirement Study; MEPS = Medical Expenditure Panel Survey; SIPP = Survey of Income and Program Participation.  aData include participation in health maintenance organizations.  bData include whether coverage for a preexisting condition is limited.  cCollected for private health insurance but not for Medicare.  dSupplemental Nutrition Assistance Program (formerly Food Stamp Program) benefits are the only in-kind benefits collected.  eTaxes paid are collected in a topical module once per panel, but nonresponse is very high and missing data are not imputed.  fAssets are collected only once per panel, so they will be present for only half of an annual sample. Asset data are not included in the public use file.  gPerson-level data are collected on just the sample member (over 50) and spouse. SOURCE: Developed by the panel from published questionnaires and codebooks.

OCR for page 89
DATA SOURCES 93 so again.2 Thus, the development of a prospective measure of MCER will have to be accomplished without accounting fully for the impact of health insurance coverage on potential out-of-pocket costs. Other survey-specific data gaps exist, which limit how fully each survey could support the modeling of MCER. MEPS does not collect Medicare premium payments (although these might be imputed based on income) or most of what distinguishes disposable income from money income. MEPS collects data on assets only once, in the final interview for each panel. Such data would therefore not be usable as a baseline characteristic. Liquid assets could be projected backward, however, and that might be acceptable even though some assets may have been used in paying for exceptional medical costs. With only a portion of liquid assets being included in resources, fol- lowing the panel’s recommendation in Chapter 2, the asset component of resources is relatively insensitive to this type of error. SIPP collects no data on chronic medical conditions, which is likely to be one of the most important predictors of subsequent medical expenditures, given that none of the surveys collects information on the details of health insurance coverage. SIPP does collect information on in-kind benefits and commuting and child care costs, but its data on taxes paid do not appear to be useful. The HRS lacks information on insurance premiums paid since the prior interview, although it obtains current premiums, which may provide a reasonably good proxy. The HRS has the same limitations as MEPS with respect to disposable versus money income, but, unlike MEPS, liquid assets are available as a baseline characteristic. The HRS captures some additional information that could be useful in modeling the economic burden of medical expenditures. The survey asks respondents with large out-of-pocket medical expenses how they financed these expenditures (although the re- sponse categories combine earnings and savings, which would be useful to separate), and it collects information on assistance that children may have provided with payments. In addition, Medicare claims data have been linked to the HRS, expanding the survey’s available data on expenditures— primarily for the population ages 65 and older.3 2  MEPS has added a few more questions on types of health insurance coverage to support analysis of the ACA and is considering what additional questions and content might be tested and added to the Insurance Component. Interest centers on employer plans and offerings, firm size, actuarial value, stop-loss policies, wellness programs, and additional detail on the characteristics of self-insured plans and small employer anticipated exchange participation. (See http://aspe.hhs.gov/hsp/12/surveyenhancements/ib.shtml.) 3  The HRS also serves as the central data source for the Future Elderly Model (FEM), a mi- crosimulation model developed by the University of Southern California and RAND Roybal Center for Health Policy Simulation. The FEM combines data from the HRS, MEPS, the Medi- care Current Beneficiary Survey, and the National Health Interview Survey and can be used to predict health status and economic outcomes for individuals 51 and older. For an example of an application of this model see Lakdawalla, Goldman, and Shang (2005).

OCR for page 89
94 MEDICAL CARE ECONOMIC RISK It is possible to exploit the panel design features of the CPS sample (see Chapter 4) to conduct longitudinal analyses with successive annual supple- ments. Half of the addresses that are included in the CPS ASEC sample in a given year were included in the sample the prior year. Such analyses encounter serious obstacles, however. The sample units are addresses, not the persons living at those locations. Persons who move during the year (about 14 percent of the population, based on recent estimates) would be excluded from any longitudinal analysis, introducing an obvious bias, as moving may be related to changes in circumstances that are relevant to medical care expenditures, resources, or family composition. Nonresponse to the supplement (about 15 percent currently) may introduce further bias in addition to reducing the number of sample households present in 2 con- secutive years. On top of these concerns, the CPS is weaker than the other three surveys in its collection of data elements needed to model medical care expenditure risk. In summary, none of the surveys is nearly as strong as we would like in its measurement of key baseline characteristics. With its strong measures of chronic conditions and very high-quality expenditure data, MEPS is clearly superior to SIPP. The HRS could provide a supplemental data source for the one-fifth of households that fall into the HRS universe. Estimates from the HRS could be used to validate the model estimates from MEPS for this segment of the population (or perhaps just the elderly), although differences in the variables available to serve as predictors would have to be taken into account. Data Sources for Production of a Measure Once a model of MCER has been developed, the estimates could be used directly (in MEPS), or the predictive model could be applied to another data set that provides measures of the relevant baseline characteristics. The latter approach offers a way to make the measurement of MCER more timely and to extend the measure to a larger and possibly more representa- tive sample. Because longitudinal data would not be required but production would impose other requirements, the set of surveys that could potentially serve this purpose is not the same as the three evaluated above. We include MEPS—but the full-year consolidated file rather than the longitudinal file. The consolidated file has an annual reference period and combines two successive, overlapping panels. We do not include the HRS or SIPP. The HRS represents too little of the population to be used for produc- tion. Although there are a number of issues with the use of SIPP for this purpose (see Czajka, in Part III), the survey’s abutted panel design has the

OCR for page 89
DATA SOURCES 95 most serious implications for measurement.4 It has been shown that the measured poverty rate declines over the life of a SIPP panel independently of the true trend, and there is an especially steep decline over the first two to three waves (Anderson and Fields, 2010; Czajka, Mabli, and Cody, 2008; Weinberg, 2003). When the 1995 National Research Council panel recommended a new poverty measure that would be produced from SIPP (National Research Council, 1995), the Census Bureau developed plans to restore the overlapping panel design, whereby a new panel was started each year; however, this was not done (Weinberg, 1999). Overlapping panels ensure a uniform bias for cross-sectional, annual estimates, which is why MEPS does not have the same problem as SIPP. In considering surveys that meet the requirements for production out- lined earlier, we restrict our attention to surveys conducted by the federal government. If a measure of MCER is to be produced by the federal gov- ernment on an annual basis, closely tied to the release of the SPM, the data used to construct that measure must be obtained from a federal survey or surveys. The data collection schedule and the quality and consistency of the data that are collected are critical elements in the production of an an- nual measure that can be used to track changes in medical care financial risk over time. Although there are serious limitations to the relevant data being collected in federal surveys at the present time, full federal authority over all of the processes that contribute to the production of a measure of MCER is essential to ensuring the integrity and viability of the mea- sure. In addition to MEPS, then, we consider the CPS ASEC (see http:// www.census.gov/hhes/www/hlthins/data/index.html), the National Health Interview Survey (NHIS) (see http://www.cdc.gov/NCHS//NHIS.htm), the American Community Survey (ACS), and the Consumer Expenditure (CE) series quarterly survey (see http://www.bls.gov/cex). The responsible agen- cies for these surveys are the U.S. Census Bureau for the CPS ASEC, SIPP, and ACS, the National Center for Health Statistics for NHIS, the Agency for Healthcare Research and Quality (AHRQ) for MEPS, and the Bureau of Labor Statistics for the CE. The CPS ASEC provides a standard for statistical precision because of its role as the official source of monthly unemployment estimates and annual poverty rates and its widespread use for estimating the percentage of the population without health insurance coverage. The CPS ASEC col- lects interviews from about 80,000 households each year. MEPS and NHIS samples vary in size over time. The largest recent MEPS sample, for calen- 4  With the abutted panel design, which was introduced in 1996, successive panels are end to end—that is, the start of one panel coincides with the completion of the preceding panel. Previously, a new panel was started each year, as is the case with MEPS.

OCR for page 89
96 MEDICAL CARE ECONOMIC RISK dar year 2009, had fewer than 15,000 households, whereas the 2011 NHIS was projected to have a final interviewed sample of 35,000 households (U.S. Government Accountability Office, 2012). Like MEPS, the annual sample for the CE quarterly survey is under 15,000 households. The ACS, in con- trast, collects data from about 2 million households each year. To put this in perspective, the mean state sample in the ACS is larger than the NHIS national sample (the median ACS state sample is considerably smaller). The ability of the ACS to support estimates for states and large metropolitan areas at levels of precision comparable to some of these national samples is appealing, but as we show, the data elements are too limited for our needs. Of the five surveys, the CPS ASEC has the most timely release, just 6 months after the completion of data collection and 9 months after the end of the survey reference period. The CPS is also the source of both the official poverty measure and the SPM, to which the MCER measure is intended as a companion (Czajka, in Part III). Producing the two measures from the same survey would enable more direct comparisons than if the two were based on different surveys. Table 5-2 summarizes the collection of variables needed to produce an annual prospective measure of MCER. Measures of medical expenditures are required only for the prior year—where they are used as baseline char- acteristics. The model will predict medical expenditures during the next year as a function of the baseline characteristics. Although model development will focus on the fullest set of baseline characteristics, the model will have to be reestimated using just those baseline characteristics that are available for a particular survey. Fewer baseline characteristics imply a weaker model unless the baseline characteristics that are omitted have no impact. Because none of the longitudinal surveys provides detailed information on what is actually included in health insurance coverage, such variables will not be included in the predictive model, so the absence of such vari- ables from all five surveys, although a major limitation for modeling, is be- side the point. Only MEPS and the NHIS provide information on functional limitations and chronic medical conditions, which are likely to be important predictors. The NHIS lacks information on prior year premiums or out-of- pocket expenditures, however. What it does include are several questions relating to the financial burden posed by medical care. In 2011, the NHIS added three new questions that asked whether the family had problems paying its medical bills in the past 12 months, whether there were medical bills that were being paid over time, and whether there were medical bills that the family was unable to pay at all. The NHIS is also very weak on resources. MEPS collects much more information on resources but lacks the components that differentiate money income from disposable income. The CPS ASEC is the only one of the surveys that can estimate dispos- able income currently, but it lacks a measure of liquid assets. Given that

OCR for page 89
DATA SOURCES 97 TABLE 5-2  Collection of Variables Needed to Produce an Annual Prospective Measure of Medical Care Economic Risk CPS Variable ASEC MEPS NHIS ACS CE Measures of Health General health status (poor to excellent) X X X X Activity limitations Xa X X Xa Functional limitations X X Chronic medical conditions X X Measures of Health Insurance Coverage Current health insurance coverage Xb X X X Services/treatments covered Potential liability for out-of-pocket costs Current health insurance premiums Measures of Medical Expenditures Prior year insurance premiums X Xc Xd Prior year out-of-pocket expenditures X X e Xd Measures of Resources Earned income X X X X X Unearned income X X X X In-kind benefits Xf g g Taxes paid Xh i Child support paid X Commuting expenses Xh Child care expenses X X Liquid assets Xj Xk Measures of Family Relationships Relationship to the householder X X X X l All parent-child relationships X X X All marital relationships X X X NOTES: ACS = American Community Survey; CE = Consumer Expenditure Survey; CPS ASEC = Current Population Survey Annual Social and Economic Supplement; MEPS = Medi- cal Expenditure Panel Survey; NHIS = National Health Interview Survey.  aBlindness, deafness, and limitations in four activities of daily living are the only items collected.  bCoverage by type is ever in the prior calendar year rather than at the time of the survey.  cCollected for private health insurance but not for Medicare.  dExpenditures during the past quarter are collected in each quarterly interview.  eIncludes questions on the financial burden posed by medical care.  fFood stamp (SNAP) benefits are reported, as is the receipt of other in-kind benefits, but the value of these other benefits is simulated.  gFood stamp (SNAP) benefits are the only in-kind benefits collected.  hAmounts are simulated rather than reported.  iIncome taxes withheld from earnings are collected.  jAssets are collected only once per panel, so they will be present for only half of an annual sample. Asset data are not included in the public use file.  kFinancial assets are collected, but retirement accounts are not separated from other accounts.  lData are collected for the consumer unit rather than the family or household. It is not pos- sible to reconstruct data for health insurance units when they differ from consumer units. SOURCE: Developed by the panel from published questionnaires and codebooks.

OCR for page 89
98 MEDICAL CARE ECONOMIC RISK components of disposable income are or were previously imputed in the CPS ASEC to create a number of alternative poverty measures (taxes and commuting expenses are currently simulated or imputed, and child care expenses and child support payments were previously imputed), similar imputations could be performed with MEPS as well—or the additional variables could be added to the MEPS questionnaire in the future. Both the ACS and the CE fall short in a number of other ways. The ACS collects only limited measures of health, and the CE collects none. The ACS collects no medical expenditure data, nor does it collect from persons unrelated to the householder the relationship information needed to con- struct health insurance units or families. The CE does collect expenditure data, but this is done for “consumer units,” and when these do not align with health insurance units or families, the latter cannot be constructed. Given the data limitations shown here and the possibility that the research to develop a prospective measure may show little gain over a retrospective measure, production of a retrospective measure of MCER remains an option. Ideally, such a measure would take account of measures of chronic health conditions and functional limitations, which are avail- able in MEPS and the NHIS, and it would also take account of features of health insurance coverage, which are not available in any federal household survey. Minimally, however, it would require only measures of prior year premiums and out-of-pocket expenditures, along with prior year measures of all of the resources that would be needed for a prospective measure, assuming that disposable income as defined for the SPM, plus a portion of liquid assets, would be used as the measure of resources. In essence, a retrospective measure of medical care economic risk of this kind would be similar to a retrospective measure of financial burden, as described in Chapter 2, with the exception of including a portion of liquid assets in the former measure. The CPS ASEC falls short only on liquid assets although it relies on simulation and imputation for a number of the components that distinguish disposable income from Census money income. MEPS has liquid assets, but for only half the sample, and it lacks most of the components that distinguish disposable income from Census money income. MEPS also lacks prior year insurance premiums for Medicare, but otherwise meets the minimal requirements for a retrospective measure. CONCLUSIONS AND RECOMMENDATIONS Although the panel favors a prospective measure of MCER over a ret- rospective measure, the more substantial data requirements for developing a model with which to estimate the prospective measure cannot be fully met with an existing survey. The MEPS longitudinal file comes closest to meeting these requirements.

OCR for page 89
DATA SOURCES 99 Recommendation 5-1: The panel recommends that the development of a model for estimating a prospective measure of medical care eco- nomic risk be carried out with the Medical Expenditure Panel Survey (MEPS) longitudinal file. The panel also recommends that the Health and Retirement Study (HRS) be used to validate the results of the MEPS modeling for at least the elderly, if not the entire population over age 50, which the HRS sample represents. For production, the choice is less clear. The MEPS full-year consoli- dated file and the CPS ASEC have different strengths and different limi- tations with respect to required data elements, making them difficult to compare. On other points of comparison, the CPS ASEC is more timely than MEPS, has five times the sample size, and serves as the source of the complementary SPM. On balance, these considerations favor the CPS ASEC if it can be shown that a predictive model of MCER can be transported successfully from MEPS to the CPS ASEC. If MCER depends too heavily on the measures of chronic medical conditions and functional limitations that are present in the MEPS but not the CPS ASEC, then the CPS ASEC would not be a satisfactory choice. If the development effort should demonstrate that a prospective mea- sure is itself not viable at present or not sufficiently different from a ret- rospective measure, then the CPS ASEC would be a stronger choice for this alternative measure. In this case, however, there may be value in using MEPS to create a supplemental measure, in which MCER could be asso- ciated with the chronic conditions and functional limitations that MEPS measures but the CPS ASEC does not. A principal limitation of using either MEPS or the CPS ASEC for either a prospective or retrospective measure of MCER is the lack of information on insurance coverage beyond the general categories of employer-sponsored, Medicare, Medicaid, individually purchased, or other sources. After 2014 when the major insurance reforms of the Patient Protection and Affordable Care Act (ACA) take place, the Census Bureau or AHRQ could consider adding a question about the level of coverage in terms of bronze, silver, or gold levels of actuarial value. This information could serve as a proxy for plan type and cost-sharing for insured families in the lower income ranges. Individuals and families may also become more aware of out-of-pocket costs for premiums because ACA requires that this information be included in tax filing. For the CPS ASEC, the value of its health insurance data for measuring MCER could be enhanced by capturing coverage at the time of the interview in addition to or instead of the prior calendar year. Other items to consider adding to the CPS ASEC include functional limitations, chronic medical conditions, and liquid assets. Items to consider adding to MEPS include additional in-kind benefits, federal and state income taxes,

OCR for page 89
100 MEDICAL CARE ECONOMIC RISK and commuting and child care expenses. These items would enable MEPS to replicate the SPM more effectively. Recommendation 5-2: The panel recommends that the Census Bureau and the Agency for Healthcare Research and Quality assess the merits of adding items to both the Current Population Survey Annual Social and Economic Supplement and the Medical Expenditure Panel Survey to at least partially address the most critical data limitations identified for measuring medical care economic risk.