Is Health Care Spending Higher under Medicaid or Private Insurance?

Transcription

1 Jack Hadley John Holahan Is Health Care Spending Higher under Medicaid or Private Insurance? This paper addresses the question of whether Medicaid is in fact a high-cost program after adjusting for the health of the people it covers. We compare and simulate annual per capita medical spending for lower-income people (families with incomes under 200% of poverty) covered for a full year by either Medicaid or private insurance. We first show that low-income privately insured enrollees and Medicaid enrollees have very different socioeconomic and health characteristics. We then present simulated comparisons based on multivariate statistical models that estimate the effects of private and Medicaid coverage on the likelihood of using services, and the level of expenditures, given any use, holding constant demographic, economic, and health status characteristics. The simulations demonstrate that if people with Medicaid coverage with their health status, disability, and chronic conditions were given private coverage, they would cost considerably more than they do today. Conversely, if the privately insured were given Medicaid coverage, spending would be lower. We find no evidence that spending differences between Medicaid and private coverage for low-income people are due to lower service use by Medicaid beneficiaries. We conclude that most of the difference in expenditures is due to differences in provider payment rates. Medicaid spending is a perennial feature of policy, legislative, and budget discussions at both the federal and state levels. Medicaid provides acute and long-term care coverage to more than 40 million low-income Americans, many of whom are very sick and/or disabled; Medicaid spending is expected to reach $278 billion in 2003, surpassing Medicare. The program grew by 10% in 2001, 13% in 2002, and is projected by the Congressional Budget Office to grow at 8.5% per year for the rest of the decade (Congressional Budget Office 2003). Plunging state revenues have placed extraordinary pressure on state budgets. Proposals for cutbacks in Medicaid, as well as education and other state functions, are the subject of vigorous debate in state legislatures nationwide. The Bush administration has proposed a Medicaid reform package that includes establishing fixed allocations for acute and long-term care. Though voluntary, these allocations would limit the amount of money available to states, while giving them increased flexibility to reduce benefits, use cost sharing and employ enrollment caps. Jack Hadley, Ph.D., is principal research associate at the Urban Institute and a senior fellow at the Center for Studying Health System Change. John Holahan, Ph.D., is director of the Health Policy Center at the Urban Institute. Funding for this research came from the Henry J. Kaiser Family Foundation under The Cost of Not Covering the Uninsured Project. Address correspondence to Dr. Hadley at the Urban Institute, 2100 M St., N.W., Washington, DC jhadley@ui.urban.org Inquiry 40: (Winter 2003/2004). Ó 2003 Excellus Health Plan, Inc /03/

2 Inquiry/Volume 40, Winter 2003/2004 Figure 1. Medicaid benefits At the core of the proposal is a response to states assertions that they could lower Medicaid costs with greater flexibility in benefit design and delivery system structure. It is frequently argued that the Medicaid benefit package is extremely rich compared to benefit packages available to low-income working Americans through their employers plans. Medicaid covers a far greater range of benefits, though most of these are optional to the states. Moreover, the Medicaid program has little or no cost sharing. Medicaid benefits, both mandatory and optional, are shown in Figure 1; few private plans cover as broad a range of services. The increased flexibility in the Bush administration s proposal is designed to give states more control over benefit packages, and arguably, to increase their ability to control spending. 324

3 Medicaid vs. Private Insurance This paper addresses the question of whether Medicaid actually is a high-cost program after one adjusts for the health of the people it covers. While the benefit package is extensive, studies have shown that Medicaid payment rates to providers are well below Medicare rates and even further below commercial rates (Norton and Zuckerman 2000; Medicare Payment Advisory Commission 2002). States have had a difficult time retaining participation of managed care plans, in part because of low capitation rates. The question we address is whether Medicaid is in fact more costly than private insurance for comparable populations. Some have argued that Medicaid should enroll its beneficiaries in private plans or give Medicaid beneficiaries a fixeddollar voucher or tax credit (a defined contribution) and allow them to buy into private plans. Would this approach be less costly for Medicaid, assuming that Medicaid would continue to cover beneficiaries out-of-pocket liabilities as well as pay employees share of the private insurance premium? In this paper, we compare and simulate annual per capita medical spending for lower-income people (families with incomes under 200% of poverty) who have either Medicaid coverage or private coverage for a full year (or for the entire time they are in the sample if they are observed for less than a year). 1 We begin with simple comparisons of total medical expenditures for privately insured people and Medicaid enrollees, also looking at the distribution of spending by source of payment and by type of service. We then provide data on differences in demographic and economic characteristics between the low-income population with Medicaid coverage and the low-income population with private coverage, showing that the populations are quite different. We also show that the populations have significantly different health characteristics, measured by self-reported health status and the prevalence of physical limitations and acute and chronic conditions. Given that the populations are so different, we then present simulated expenditure comparisons based on multivariate statistical models. The models estimate the effects of private and Medicaid coverage on the likelihood of using services, and the level of expenditures, given any use, holding constant demographic, economic, and health status characteristics. We use these expenditure models to simulate what medical spending would be if the Medicaid enrollees had private coverage and, conversely, if lower-income people with private coverage were shifted into Medicaid, both for all people and by self-reported health status. Because of the current interest in premium assistance programs, which help lowincome working families pay the employee contribution to enroll in employer-sponsored insurance (Neuschler and Curtis 2003), we then simulate the effect of giving working Medicaid beneficiaries private coverage. Finally, we examine whether differences in expenditures reflect differences in utilization. Data All data for this analysis are from the Medical Expenditure Panel Survey (MEPS) for the years 1996, 1997, 1998, and The MEPS surveys a nationally representative sample of the noninstitutionalized population (Cohen 1997). It contains detailed information on annual total charges and payments for health care used, monthly information on insurance coverage, and detailed demographic and health characteristics. The analysis sample used in this paper excludes people 65 or older and nonelderly people covered only by Medicare, and includes only people who have full-time coverage from either private insurance or Medicaid. 2 To increase the comparability of the privately insured and Medicaid samples, we also limit the analysis to people in families with low incomes, defined as less than twice the federal poverty level. Data from the four surveys were pooled to form the single analysis file. Expenditures were inflated to 2001 dollars using the annual percentage increase in the National Health Accounts (NHA). The MEPS defines expenditures as payments made for health care services. This definition excludes the cost of services for which no explicit and identifiable payment is linked to a specific patient (except for services provided by public hospitals and clinics). For example, MEPS does not count provider revenues from general government appropriations to hospitals and Medicare and Medicaid disproportionate share (DSH) payments since they are not payments for specific patients. However, the MEPS measure of total expenditures includes all payments for a broad range of health care services, not just those made by insurance plans for covered services. 325

4 Inquiry/Volume 40, Winter 2003/2004 In particular, it includes people s out-of-pocket payments for cost-sharing obligations and for services not covered by their insurance plan, as well as payments made by other sources, such as workers compensation, the Department of Veterans Affairs, or state or local medical assistance programs. MEPS unique definitions and methods for measuring expenditures lead to significantly lower national health expenditure estimates than reported by the NHA. To correct for MEPS systematic underreporting, we used information from a detailed comparison of the MEPS and NHA estimates to develop an adjustment factor of 1.25 to align the MEPS estimates with the NHA (Selden et al. 2001). The adjustment factor is the ratio of the comparable NHA and MEPS expenditure levels (Hadley and Holahan 2003). The analysis by Selden et al. (2001) also shows that the aggregate Medicaid expenditure in MEPS is lower than Medicaid spending in the National Health Accounts by about 24%. MEPS data on aggregate Medicaid spending also are considerably lower than Medicaid administrative data. There are several possible reasons for this. One is that administrative data may include upper payment limit expenditures that are attributed to hospital care. Administrative data also include the overhead expenses associated with capitated health plans. Inaccuracies in reporting coverage or sources of payment also can affect the estimates of Medicaid spending. MEPS experts believe that some people fail to report Medicaid coverage or misreport Medicaid coverage as private coverage, a problem MEPS shares with other surveys (e.g., the Current Population Survey). Another part of the shortfall has been attributed to confusion over the true source of payment for those in managed care plans. For example, providers may believe they are billing a private insurance plan when, in fact, that private insurance plan has contracted with Medicaid. These expenditures often are reported as private spending, thus overstating aggregate private spending on the MEPS and understating aggregate Medicaid spending. The factors discussed previously primarily affect estimates of aggregate Medicaid spending. Since our analysis focuses on per person spending, these limitations should not raise major concerns about bias in our estimates. However, the MEPS also understates the expenditures on all groups because it does not collect data on over-the-counter pharmaceuticals and is believed to understate spending on laboratory services. The MEPS also may have issues of recall (i.e. respondents forgetting some utilization). Since these possible limitations are not unique to Medicaid, we conclude that there is no reason to believe that Medicaid spending is systematically underrepresented any more so than private spending. While both may be underreported, this should not bias comparisons of per person spending differences between Medicaid enrollees and the privately insured. MEPS estimates of Medicaid spending per person still may seem low compared to estimates from Medicaid administrative data. In the subsequent tables, we use data only for full-time enrollees. Administrative data show that expenditures for those in the program for a full year are significantly lower than those in the program for a portion of the year, in part because medical spend-down or a severe illness can lead to Medicaid enrollment for eligible people who are uninsured. We exclude from our analysis those with joint Medicaid-Medicare coverage, who tend to be more costly than those covered only by Medicaid. Finally, the MEPS does not survey the institutionalized, including people in the hospital for more than 30 days. Methods We used a standard two-part approach to estimating the multivariate statistical model for simulating expenditures under different assumptions about insurance coverage (Jones 2000; Manning and Mullahy 2001). The first part employs a logistic model to estimate the probability of having any expenditure during the year. The second part estimates the effects of insurance and other characteristics on spending, given that the person has incurred some expenses. All data were weighted using the MEPS person weights for both estimating the models coefficients and simulating predicted spending. We used the STATA software program for statistical estimation and computations. 3 We estimated two sets of models. The first set combines the data for the privately insured and Medicaid enrollees into a single sample and includes a dichotomous variable to indicate which type of coverage a person has. The coefficient 326

5 Medicaid vs. Private Insurance on the insurance coverage variable indicates whether being covered by Medicaid or private insurance has a significant effect on either the probability of any spending or on total spending, holding constant the effects of sociodemographic and health characteristics. 4 The coefficients of the sociodemographic and health characteristics variables are the same for both populations. To predict spending for the privately insured under the assumption of Medicaid coverage, we set the coverage variable to indicate Medicaid coverage and then applied the model s other coefficients to the privately insured population s characteristics. For the converse, predicted spending for Medicaid enrollees under the assumption of private coverage, we set the coverage variable to indicate private coverage and then applied the model s coefficients to the characteristics of the Medicaid enrollees. These predictions assume that underlying behavior is the same in the two populations and that differences in predicted spending are due to differences in the characteristics of the populations and to any difference associated with type of coverage, as measured by the coefficient of the coverage variable. The second set of models was estimated separately for the privately insured and Medicaid enrollees. We used these models, which allow the effects (parameter estimates) of sociodemographic and health characteristics to vary across the two populations, to predict spending for a set of hypothetical people with identical characteristics. 5 The characteristics of the hypothetical people were set equal to the average characteristics of subsets of the MEPS sample observations divided into groups based on type of coverage (private insurance or Medicaid) and self-reported health status. We then combined the parameters from the separate private and Medicaid expenditure models with the characteristics of the hypothetical people to predict what expenditures would be for the same set of people, first assuming private insurance coverage and then assuming Medicaid coverage. Overall, we used the multivariate models to address three questions: h First, controlling for the effects of sociodemographic and health characteristics, is Medicaid coverage significantly more costly than private coverage? We addressed this question by testing whether the coefficient of the Medicaid coverage variable is positive and statistically significant in the expenditure models estimated from the combined samples of the privately insured and Medicaid enrollees. h Second, how would spending change if people with private insurance had Medicaid coverage or if Medicaid enrollees had private coverage? To address this question, we first used the model estimated with the combined sample of people, but made separate predictions based on their different characteristics. We then used the separate private and Medicaid models and made a second set of predictions by applying the characteristics of those covered by Medicaid to the private model and vice versa. h Third, how would spending vary for people with different health characteristics? We answered this question by applying the separate Medicaid and private insurance models to people with different health characteristics who were simulated to have either Medicaid or private coverage, respectively. Independent Variables in the Statistical Models 6 Sociodemographic Characteristics To capture the effects of sociodemographic factors on spending, the statistical models for adults include sets of dichotomous variables for gender, age, race and ethnicity, education, family income relative to poverty, and marital status. The models for children control for gender, age, race and ethnicity, family income relative to poverty, and parents education and marital status. All models also include controls for census region. Health Characteristics Although the MEPS contains detailed information on the presence of both acute and chronic conditions, there is some concern that these factors may be under-reported for the uninsured because people without coverage tend to have fewer contacts with medical providers. Therefore, for adults, we use a combination of selfreported general health, mental health and functional status measures, along with measures of acute and chronic conditions derived from contacts with medical providers. We use a smaller 327

6 Inquiry/Volume 40, Winter 2003/2004 Table 1. Annual per capita medical expenditures by low-income Americans, by type of coverage Population and type of coverage All people Excluding people with limitations b Age and gender Private Medicaid Private Medicaid Adults N a 3,499 2,060 3,119 1,111 All ($) 2,843 4,877* 2,253 1,752* Male ($) 2,435 5,504* 1, * Female ($) 3,161 4,617* 2,499 1,944* Age ($) 1,807 3,143* 1,712 1,839 Age ($) 3,425 6,374* 2,584 1,607* Children N a 2,235 4,470 1,979 3,619 All ($) 1, ** 1, Male ($) 1, Female ($) 1, , Age 0 5 ($) 2, ** 2, Age 6 18 ($) 1, Source: Authors tabulations, Medical Expenditure Panel Survey. a Unweighted number of cases. b See Tables 6 and 7 for specific measures used to identify limitations; also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. ** Significantly different from private coverage at the 10% level. and different set of health measures for children because they have a much lower incidence of specific medical conditions. Finally, we include a dichotomous indicator of whether the person died or became institutionalized during the year. Results Actual Expenditures Actual expenditures for adults and children by type of coverage are shown in Table 1. Columns 1 and 2 show annual per capita spending for all people, while columns 3 and 4 show spending after excluding those who report fair or poor mental health, any disability, any physical or functional limitation, or receipt of Supplemental Security Income (SSI). (As will be shown later, the privately insured and Medicaid populations have very different proportions of people on SSI or with physical limitations.) Expenditures are greater for adults with Medicaid coverage than for those with private coverage when all adults are included (columns 1 and 2); all of the differences are statistically significant. However, when we exclude adults with fair or poor mental health or any physical limitations, spending per adult is significantly lower for Medicaid than for the privately insured, even though excluding people with limitations or on SSI increases the income disparity between the Medicaid and privately insured samples (see Tables 4 and 5 later). Presumably, spending would be even greater for higher-income people because they likely would have more generous plans and would be less likely to be deterred from seeking care by cost-sharing obligations. Again, these findings hold for males and females and for older adults, with a statistically insignificant difference for younger adults. Table 1 also shows that spending is greater for children with private coverage than for children covered by Medicaid, whether or not we exclude those with limitations or disabilities. However, none of these differences is statistically significant at the 5% level. Higher but statistically equivalent spending for privately insured children persists even when controlling for children s gender and age. Table 2 provides data on expenditures broken out by source of payment. Not surprisingly, private insurance pays for the bulk of the expenditures of those covered privately, while Medicaid pays most of the costs of the people it covers. The most notable difference is in out-of-pocket 328

7 Medicaid vs. Private Insurance Table 2. Annual per capita medical expenditures for low-income Americans with private insurance or Medicaid coverage, by type of coverage and source of payment Population and type of coverage Age and source All people Excluding people with limitations b of payment Private ($) Medicaid ($) Private ($) Medicaid ($) Adults Total expenditures 2,843 4,877* 2,253 1,752* Private 2,051 2* 1,617 0* Medicaid 17 4,003* 8 1,540* Out of pocket * * Other a * Children Total expenditures 1, ** 1, Private 1,004 0* 853 0* Medicaid * 4 645* Out of pocket * * Other 19 81* 13 71* Source: Authors tabulations, Medical Expenditure Panel Survey. a Other includes Medicare, VA, Champus, other federal, other state and local, workers compensation, other public, other private, and other sources. b See Tables 6 and 7 for specific measures used to identify limitations; also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. ** Significantly different from private coverage at the 10% level. spending. For all adults, including those with any limitations, out-of-pocket expenditures are more than twice as high for those with private coverage than for those with Medicaid. This is despite the fact that the overall level of expenditures is much higher for those with Medicaid coverage. For adults without any limitations, as well as for all children, out-of-pocket spending is 5.5 to 7.5 times higher for those with private coverage than for those with Medicaid. While these data show that out-of-pocket spending for Medicaid beneficiaries is much lower than out-of-pocket spending by low-income people with private coverage, other data (from Table 4) show that Medicaid enrollees have significantly lower incomes than the privately insured. This raises the question of whether Medicaid enrollees would be able or should be expected to bear the same level of cost sharing as the privately insured. Table 3 provides data on spending by type of service. For both groups, most expenditures are for inpatient care, office-based services, outpatient and emergency department care, and prescription drugs. Only a small share of expenditures (generally under 15%) is for dental and other services. This is significant because it is these services that usually are associated with the broad Medicaid benefit package. Not only are these expenditures small as a share of the total, but they are significantly higher for those with private coverage than for those with Medicaid coverage in three of the four comparisons. Differences in spending between people with private and Medicaid coverage reflect differences in health status and sociodemographic characteristics, which affect care-seeking behavior, as well as differences in benefit packages and provider payment rates. We next describe the differences in demographic and health characteristics, and then use the multivariate models to analyze how spending differs under Medicaid and private coverage when controlling for these factors. Demographic Characteristics Tables 4 and 5 show the differences in demographic and economic characteristics for adults and children, respectively. (Recall that the data are limited to children and nonelderly adults who are covered for the entire observation period by either Medicaid or private insurance and have family incomes below 200% of the poverty level.) As Table 4 shows, members of the adult population with Medicaid coverage are much more likely to be female, and more likely to be 329

8 Inquiry/Volume 40, Winter 2003/2004 Table 3. Annual per capita medical expenditures for low-income Americans with private insurance or Medicaid coverage, by type of coverage and type of service Population and type of coverage Age and type All people Excluding people with limitations a of service Private ($) Medicaid ($) Private ($) Medicaid ($) Adults Total expenditures 2,843 4,877* 2,253 1,752* Type of service Inpatient 1,024 2,109* Office-based doctor * Outpatient/ER * Prescription * * Home health * 1 27 Dental/other * Children Total expenditures 1, ** 1, Type of service Inpatient Office-based doctor * Outpatient/ER ** * Prescription * Home health Dental/other * * Source: Authors tabulations, Medical Expenditure Panel Survey. a See Tables 6 and 7 for specific measures used to identify limitations; also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. ** Significantly different from private coverage at the 10% level. young, with higher proportions between ages 19 and 29 and a lower proportion older than 40. Compared to those with private insurance, they are much more likely to be African American or Hispanic and much less likely to be white. They are more likely to have had less than a high school education. Average family incomes are substantially lower, and those with Medicaid coverage are much more likely to be widowed, divorced, separated, or never married. Table 5 shows that children with Medicaid coverage are more likely to be under five years of age, and less likely to be teenagers. Children with Medicaid coverage are less likely to be white, and more likely to be Hispanic or African American. Their parents are much more likely not to have graduated from high school and, correspondingly, much less likely to have attended college or to be college graduates. Family incomes are much lower for those with Medicaid coverage: $22,735 compared to $38,990 for the privately insured. More than 70% of children with Medicaid coverage have family incomes below the federal poverty line (FPL), while about 80% of those with private coverage have family incomes between 100% FPL and 200% FPL. Finally, children with Medicaid coverage are much more likely to live in a single-parent family. Health Status Tables 6 (adults) and 7 (children) show significant health status differences between the privately insured and Medicaid enrollees. The adult population covered by Medicaid is much less likely to report excellent or very good health and more likely to report fair or poor health. For all adults, the differences are quite striking. Just under 60% of those with private coverage report being in excellent or very good health versus 34% of those with Medicaid coverage. About 12% of those with private coverage report being in fair or poor health versus 36% for those with Medicaid coverage. Underlying the substantial differences in selfreported health status among all adults, those with Medicaid coverage are much more likely to report having some type of limitation, for each 330

9 Medicaid vs. Private Insurance Table 4. coverage Characteristics of low-income adults with private insurance or Medicaid Population and type of coverage All adults Excluding adults with limitations a Private Medicaid Private Medicaid Region (%) West * * Northeast * * Midwest * * South * * Gender (%) Male * * Female * * Age (%) * * * * * * Race/ethnicity (%) White * * Black * * White Hispanic * Other * * Education (%) Less than high school * * High school * Some college * * College or more * * Family income (%, by poverty status) 0 100% of FPL * * % of FPL * * Family income ($) 32,677 18,614* 33,111 19,178* Marital status (%) Married * * Widowed, divorced, separated * * Never married * * Source: Authors tabulations, Medical Expenditure Panel Survey. a See Tables 6 and 7 for specific measures used to identify limitations; also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. of the measures that we looked at. Overall, 43.1% of Medicaid enrollees report at least one of the specific limitations, compared to just 10.8% of the low-income privately insured. They are far more likely to have fair or poor mental health, to report an activity of daily living (ADL) or instrumental activity of daily living (IADL) impairment, to use assistive devices, to have difficulty lifting, walking, or climbing stairs, and to have social and cognitive limitations. They also are more likely to have died or been institutionalized during the year. Including those with limitations, Medicaidcovered adults are also more likely to have chronic and acute medical conditions, including diabetes, hypertension, asthma, back disorders, heart or cerebrovascular conditions, digestive conditions, and muscular/skeletal conditions. Excluding those with limitations, Medicaid adults still are more likely to report fair or poor health, and to have asthma or digestive conditions, and given the relatively generous eligibility criteria for covering pregnant women are more likely to be pregnant. There are generally 331

10 Inquiry/Volume 40, Winter 2003/2004 Table 5. coverage Characteristics of low-income children with private insurance or Medicaid Population and type of coverage All children Excluding children with limitations a Private Medicaid Private Medicaid Region (%) West Northeast Midwest South Gender (%) Female Male Age (%) * * * * Race/ethnicity (%) White * * White Hispanic * * Black * * Other * Parents education (%) Less than high school * * High school Some college * * College graduate * * Family income (%, by poverty status) 0 100% of FPL * * % of FPL * * Family income ($) 38,990 22,735* 39,473 22,372* Parents marital status (%) Married * * Single female parent * * Single male parent Source: Authors tabulations, Medical Expenditure Panel Survey. a See Tables 6 and 7 for specific measures used to identify limitations; also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. no significant differences in the other acute conditions. Children with Medicaid coverage are much less likely to be in excellent health and much more likely to be in fair or poor health. 7 They are also more likely to be in fair or poor mental health and to have ADL or IADL impairments. Overall, 18.7% of children on Medicaid report having some type of limitation, compared to 12.5% of low-income children covered by private insurance. Children with Medicaid coverage are also more likely to have asthma and less likely to have infectious diseases than privately insured low-income children. Medicaid is Less Costly When Controlling for Differences in Population Characteristics Tables 8 and 9 report the coefficient estimates of the two-part expenditure models for the combined samples of privately insured and Medicaid-covered adults and children, respectively. (Similar models also were estimated excluding people who either report having a physical or mental health limitation or receiving SSI. 8 ) The 332

11 Medicaid vs. Private Insurance Table 6. Health status and health conditions of low-income adults with private insurance or Medicaid coverage Population and type of coverage All adults Excluding adults with limitations a Private Medicaid Private Medicaid Self-reported health status (%) Excellent * * Very good * Good Fair * * Poor *.7 2.9* Limitations (%) Fair or poor mental health *.0.0 ADL/IADL screener *.0.0 Uses assistive devices *.0.0 Difficulty lifting, walking, or with steps *.0.0 Social or cognitive limitations *.0.0 Work/housework/school limitations *.0.0 Unable to perform activity *.0.0 Deceased or institutionalized.5 2.4*.0.0 Any limitations * Medical conditions (%) Diabetes * Otitis media * Hypertension * Asthma * * Back disorder * Infectious disease Malignant neoplasms Endocrine * Blood * Heart or cerebrovascular * * Bronchitis Digestive * * Genitourinary * Skin Musculoskeletal * * Fracture Pregnancy * * Source: Authors tabulations, Medical Expenditure Panel Survey. a Also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. first row of each table shows the effect of Medicaid coverage, relative to private insurance, on: 1) the relative odds of having any expenditure during the year, and 2) total expenditures given that the person has incurred some expenses. Both tables hold constant the effects of sociodemographic and health characteristics. For adults, there was no significant effect of Medicaid coverage on the likelihood of having any expense (i.e., those with Medicaid coverage are no more or less likely to have received a service). But for those with an expense, total expenditures were significantly lower, by approximately 17%. For children, those with Medicaid coverage were more likely to incur an expense (to have had at least one service). However, for those with an expense, the estimated coefficient indicated a 9% lower (though not statistically significant) expenditure level for those with Medicaid coverage. When the models were estimated excluding people who have limitations or receive SSI, the estimates of the relative odds of having any expenditure were identical for adults and only slightly lower for children. However, the 333

12 Inquiry/Volume 40, Winter 2003/2004 Table 7. Health status and health conditions of low-income children with private insurance or Medicaid coverage Population and type of coverage All children Excluding children with limitations a Private Medicaid Private Medicaid Self-reported health status (%) Excellent * * Very good Good * * Fair * * Poor.6 1.3*.4.5 Limitations (%) Fair or poor mental health *.0.0 ADL screener.5 1.4*.0.0 Special group (age,5).3 1.1*.0.0 IADL screener Limited in any activity (age,5) Deceased or institutionalized Any limitations * Medical conditions (%) Asthma * * Infections * * Bronchitis * Digestive Genitourinary Skin Muscoskeletal Fracture Source: Authors tabulations, Medical Expenditure Panel Survey. a Also excludes people receiving SSI payments. * Significantly different from private coverage at the 5% level. differences in spending, conditional on having any expense, increased about twofold, to 38% lower expenditures for adults on Medicaid and 18% lower for children on Medicaid. 9 Other results from the expenditure models showed that being white and having at least some college education were significant positive predictors of expenditures. Since those with private coverage were much more likely to be white and to have more education, these effects contribute to the finding (from Table 1) that private expenditures exceed Medicaid expenditures when people with physical or mental health limitations are excluded from the data. Health characteristics also were strongly related to both the likelihood of using services and to total expenditures given an expense. Self-reported health status and several measures of disability or physical limitations were highly significant. Most of the measures of medical conditions also affected the likelihood of having a service and total expenditures, given any service use. Diabetes, hypertension, malignancies, heart or cerebrovascular conditions, digestive conditions, and pregnancy were most strongly related to expenditures. Simulating the Effects of Changes in Coverage of Current Medicaid Enrollees and the Privately Insured To simulate the combined effects of differences by type of coverage in the likelihood of using any services and in payment levels, given service use, we predicted expenditures for hypothetical people representing the average person covered by either private insurance or Medicaid. The predictions were made under two alternative assumptions about the effects of personal and health characteristics on spending. The first is that the privately insured and Medicaid beneficiaries have identical behaviors (i.e., the coefficients of the spending model are the same for both 334

13 Medicaid vs. Private Insurance Table 8. Two-part expenditure model, all adults Probability of any expense (logit) Log of total expenses Variable Relative odds P-statistic Coefficient P-statistic Medicaid coverage ÿ Northeast Midwest South Male ÿ Age ÿ Age ÿ Age ÿ Age ÿ Age ÿ Age ÿ Age Age ÿ Black ÿ White Hispanic ÿ Asian ÿ American Indian ÿ Less than high school ÿ High school ÿ Some college % of FPL ÿ % of FPL ÿ Widowed, divorced or separated Never married Very good health Good health Fair health Poor health Fair/poor mental health ADL/IADL screener Uses assistive devices ÿ Difficulty walking, lifting or with steps ÿ Social or cognitive limitations Any limitations Unable to perform activity Deceased or institutionalized Diabetes Otitis media Hypertension Asthma Back disorder Infectious disease Malignant neoplasms Endocrine Blood Heart or cerebrovascular Bronchitis Digestive Genitourinary Skin Musculoskeletal Fracture Pregnancy Constant

14 Inquiry/Volume 40, Winter 2003/2004 Table 9. Two-part expenditure model, all children Probability of any expense (logit) Log of total expenses Variable Relative odds P-statistic Coefficient P-statistic Medicaid coverage ÿ Northeast Midwest South Male Age Age ÿ Age ÿ Age ÿ Black ÿ White Hispanic ÿ Asian ÿ American Indian ÿ Less than high school ÿ High school ÿ Some college ÿ % of FPL % of FPL ÿ Single female parent ÿ Single male parent Very good health Good health Fair health Poor health Fair/poor mental health ADL screener Deceased or institutionalized ÿ IADL screener ÿ Limited in any activity Special group Asthma Infectious disease Bronchitis Digestive Genitourinary Skin Musculoskeletal Fracture Constant samples). The alternative assumption is less restrictive, in that it allows behavior to vary across the two populations. This is represented by the coefficients of the expenditure models estimated separately for the privately insured and Medicaid beneficiaries. The results of these simulations are shown in Table 10. The upper half of the table reports predicted spending for the average adult and average child with current Medicaid coverage, comparing their baseline (Medicaid) spending to what it would be if they had private coverage. (Note that baseline spending is predicted because we are comparing spending for hypothetical average people. Actual average spending, especially for the Medicaid population, is prone to distortion because of the effects of outliers that is, a small number of cases with very high spending because of multiple limitations and conditions.) Similarly, the bottom half of the table reports predicted spending for the average adult and child currently with private insurance, again comparing baseline (private) spending to what it would be if they were covered by Medicaid. For each of these hypothetical people, we show the predictions under the two assumptions about identical behavior (combined expenditure model for Medicaid and privately insured) versus 336

15 Medicaid vs. Private Insurance different behavior (separate expenditure models for Medicaid and privately insured). The predictions based on the first assumption highlight the effects of the differences in characteristics between the privately insured and Medicaid populations. The second assumption allows both characteristics and underlying care-seeking behaviors to vary. The results in Table 10 show that if those with Medicaid coverage were given private coverage, expenditures would be higher. Under the assumption of similar underlying behaviors (rows 1a and 1b), if adults with Medicaid coverage were given private coverage, predicted medical spending would increase from $3,250 to $3, Spending for children would be virtually unchanged, increasing only slightly to $738 from $718. Allowing for underlying behaviors to differ (rows 2a and 2b) suggests that predicted spending would increase even more if Medicaid beneficiaries had private coverage: for adults, spending would increase from $3,145 to $4,410 per person; for children, it would increase from $719 to $795. Table 1 shows that spending for nondisabled adults and for children with private coverage is higher than for those with Medicaid coverage, despite the fact that those with private coverage seem to be in significantly better health. This seems to reflect the differences in racial composition, education and to some extent income, as well as the higher payment rates and possibly better access associated with private coverage. The estimates in Table 10 address the question: what if the same people were given different coverage? The spending simulations suggest that the Medicaid population, particularly adults, would cost considerably more if they had private coverage. The bottom half of Table 10 (rows 3 and 4) shows that if the average low-income adult or child currently with private coverage were given Medicaid coverage, each would have lower spending. Assuming identical underlying behavior (rows 3a and 3b), predicted spending on adults would fall from $2,230 to $1,884; spending on children would decline from $1,046 to $1,006. The separate expenditure models that allow underlying behavior to vary (rows 4a and 4b) also predict that spending would fall if the privately insured were covered by Medicaid, from $2,221 to $2,056 for adults and from $966 to $712 for children. Table 10. Simulations of the impact on medical expenditures of changes between private insurance and Medicaid coverage for low-income Americans Predicted spending ($) Baseline Alternative coverage a coverage b If average person with Medicaid coverage were given private coverage 1. Combined expenditure model c a. Adults 3,250 3,848 b. Children Separate expenditure model d a. Adults 3,145 4,410 b. Children If average person with private coverage were given Medicaid coverage 3. Combined expenditure model c a. Adults 2,230 1,884 b. Children 1,046 1, Separate expenditure model d a. Adults 2,221 2,056 b. Children a Baseline coverage is Medicaid in panels 1 and 2 and private insurance in panels 3 and 4. b Alternative coverage is private insurance in panels 1 and 2 and Medicaid in panels 3 and 4. c Assumes same expenditure model coefficients for privately insured and Medicaid enrollees. d Assumes different expenditure model coefficients for privately insured and Medicaid enrollees. Simulating the Effects of Differences in Type of Coverage for People with Different Health Status Characteristics The next set of simulations explores how spending would change for people with different health characteristics. Again, we constructed two sets of hypothetical people grouped by health status, one set based on the average characteristics of the privately insured and the other set on the average characteristics of Medicaid beneficiaries. We then simulated predicted spending for these hypothetical people first assuming their current or baseline coverage, either Medicaid or private, and then assuming the alternative type of coverage. These simulations, which were based on statistical models estimated separately for Medicaid enrollees and the privately insured, indicate how spending would change for people at each level of health status if their insurance coverage changed from Medicaid to private and vice versa. 337

16 Inquiry/Volume 40, Winter 2003/2004 Table 11. Simulations of the impact on medical expenditures of changes between private insurance and Medicaid coverage for low-income adults and children by health status Health status Predicted spending ($) a Baseline coverage b Alternative coverage c If average person with Medicaid coverage were given private coverage Adults Excellent 1,112 1,787 Very good 1,775 2,404 Good 2,704 3,163 Fair 4,940 7,622 Poor 9,615 14,785 Children Excellent Very good Good 821 1,164 Fair/poor d 2,133 2,313 If average person with private coverage were given Medicaid coverage Adults Excellent 1,341 1,065 Very good 2,091 1,974 Good 2,427 2,604 Fair 5,080 4,131 Poor 9,265 8,327 Children Excellent Very good Good 1, Fair/poor d 2,702 2,433 a Predicted spending based on separate expenditure models for privately insured and Medicaid beneficiaries. b Baseline coverage is Medicaid in the upper panel and private insurance in the lower panel. c Alternative coverage is private insurance in the upper panel and Medicaid in the lower panel. d Fair and poor categories for children are combined due to small sample size. The results of these simulations in Table 11 show that both private and Medicaid spending increase as health status worsens. The simulations also show that if those with Medicaid coverage were given private coverage, spending would increase for every level of health status. The differences are particularly large for the sickest people adults in fair or poor health. While these individuals are very costly to Medicaid programs, they would be even more costly if they had private insurance. For example, if Medicaid-covered adults in excellent health were given private coverage, expenditures would increase from $1,112 to $1,787. For adults in fair health, expenditures would increase from $4,940 to $7,622; for those in poor health, spending would increase from $9,615 to $14,785. The results show similar but less dramatic changes if children with Medicaid coverage were given private coverage. Similarly, as shown in the lower half of Table 11, if those with private coverage were given Medicaid coverage, spending would fall for all but those in good health. Simulating the Effects on Workers There is considerable interest in premium assistance programs that might allow states to use Medicaid funds to enroll working Medicaid beneficiaries with access to employer-sponsored insurance into their employers plans. The Medicaid funds would pay the employees share of the premium cost required by many employers who sponsor group insurance, and presumably also would pay other cost-sharing obligations (coinsurance and deductibles). In response to this interest, we simulated the effect on total expenditures of giving private coverage to adult Medicaid enrollees who work. Data on demographic and health characteristics (not shown) indicate that even among those who work, Medicaid beneficiaries are more likely to be African American, are less well educated, have lower incomes, are in worse health, and are more likely to report limitations. Total expenditures without controlling for these characteristics are about equal, $2,429 for those with private coverage and $2,477 for those on Medicaid. We used the regression results from the expenditure models estimated separately for Medicaid and the privately insured to estimate how spending would change if workers covered by Medicaid were given private insurance and if privately insured low-income workers were covered by Medicaid. The results shown in Table 12 indicate that if those with Medicaid coverage were given private coverage, spending would increase from $1,850 per adult to $2,395. States might be able to save under premium assistance programs if they could capture the employer contribution. However, by the same token, employers would face new costs and the affected low-income workers would face higher cost-sharing burdens, unless states also covered 338

17 Medicaid vs. Private Insurance Table 12. Simulations of the impact on medical expenditures of changes between private insurance and Medicaid coverage for low-income working adults Predicted spending ($) a Baseline coverage b Alternative coverage c If average worker with Medicaid coverage were given private coverage 1,850 2,395 If average worker with private coverage were given Medicaid coverage 1,906 1,815 a Predicted spending based on separate expenditure models for privately insured and Medicaid beneficiaries. b Baseline coverage is Medicaid in the upper panel and private insurance in the lower panel. c Alternative coverage is private insurance in the upper panel and Medicaid in the lower panel. cost sharing as well as workers premium contributions. Although total costs would be shared between Medicaid and employers, total health care spending would increase if Medicaid-covered working adults were switched to private coverage. In addition, some observers have recommended allowing firms to have the option of buying their low-income workers into Medicaid. If low-income workers with private coverage were given Medicaid coverage, total spending per worker would fall slightly from $1,906 to $1,815. Are Expenditure Differences Due to Payment Rate Differences or Utilization Differences? The preceding simulations suggest that expenditures of people with private coverage exceed those of people with Medicaid coverage for both adults and children, when the effects of differences in characteristics are held constant. Finding that private coverage is actually more costly than Medicaid coverage does not answer the questions of whether individuals with private coverage are obtaining more services, whether providers are paid at higher rates, or both. To address these questions, we estimated multivariate statistical models for office visits, doctor visits, and hospital days for both adults and children. (These models, which used the same methods and independent variables as the expenditure models, are available on request.) As before, we estimated separate Table 13. Simulations of the impact on utilization of changes between private insurance and Medicaid coverage for low-income adults and children Actual coverage a Utilization Predicted coverage b If those with Medicaid coverage were given private coverage Adults Office visits Doctor visits Hospital days Children Office visits * Doctor visits * Hospital days If those with private coverage were given Medicaid coverage Adults Office visits * Doctor visits Hospital days Children Office visits Doctor visits Hospital days.18.21* a Actual coverage is Medicaid in the upper panel and private insurance in the lower panel. b Predicted coverage is private insurance in the upper panel and Medicaid in the lower panel. * Significantly different from actual coverage at 5% level. models for adults and for children, using the full samples that include people with limitations. We used these models to predict utilization levels if those with Medicaid coverage were given private coverage and, alternatively, if those with private coverage were given Medicaid coverage. We compared the predicted levels of office and doctor visits and hospital days with the actual levels. In general, the differences between predicted and actual utilization were very small (Table 13). In the cases where the differences were significant, use under Medicaid was greater than use under private coverage. This result implies that if those with Medicaid coverage were given private coverage, the predicted numbers of annual office visits, doctor visits, and hospital days would be very similar or might even fall. Conversely, if the privately insured were covered by Medicaid, their service use 339

18 Inquiry/Volume 40, Winter 2003/2004 would be the same or increase slightly. Marquis and Long (1996) reported very similar results in an earlier analysis of the effects on service use of switching from Medicaid to private coverage. The available data report fairly basic measures of utilization. They do not capture access to specialists or the use of high-cost diagnostic and therapeutic procedures. The fact that the expenditure differences for adults in fair and poor health were so much higher for those with private coverage may imply that sicker people with private coverage have greater access to specialists and advanced/expensive technology that is, the intensity of services per office visit or per hospital day are greater for sicker people with private insurance. While service intensity may be greater for sicker people with private insurance, these results imply that the differences between private and Medicaid spending are primarily due to differences in payment rates, not to lower utilization by Medicaid enrollees. This conclusion is consistent with evidence that provider payment rates under private insurance are substantially higher than those under Medicaid. For example, Norton and Zuckerman (2000, p. 226) estimated that Medicare fees, which are generally lower than private fees, were about 35% higher than Medicaid fees in 1998; data from the Medicare Payment Advisory Commission (2002, p. 156) indicated that in 2000, private payments for hospital care were almost 6% higher relative to costs than Medicaid payments. Conclusions and Policy Implications In this paper, we have compared expenditures for low-income adults and children with Medicaid coverage to spending for low-income people with private coverage. We have shown that private expenditures exceed Medicaid spending, except for the adult population when including those with disabilities. While this may reflect differences in payment rates and access to certain providers, it also seems to reflect the fact that the privately insured population is more likely to be white, to be better educated, and to have higher incomes, all factors which are positively associated with expenditures. The health status of those with Medicaid coverage is clearly worse whether measured as self-reported health status, the presence of disabilities and functional limitations, or the presence of chronic conditions. If those with Medicaid coverage, with their health status, limitations, and chronic conditions, were given private coverage, they would cost considerably more than they do today under Medicaid. If those with private coverage were given Medicaid coverage, spending would be lower, though the differences are not as striking because this is a generally healthier group. We found no evidence that spending differences between Medicaid and private coverage for low-income people are due to Medicaid beneficiaries lower service use of basic measures of utilization office visits, doctor visits, and hospital days. We conclude that most of the difference in expenditures is due to differences in provider payment rates. There also may be differences in access to specialists and technology/intensive care for those in fair or poor health. These results strongly imply that Medicaid is not a high-cost program on a risk-adjusted basis. The program s expenditures are in fact considerable, but it is primarily because of the population it serves a population in significantly worse health than the low-income population with private coverage. The Medicaid program has held down costs successfully by limiting provider payment rates, and perhaps by limiting access to specialists and technology/intensive procedures. The notion that Medicaid is a Cadillac program because of its long list of benefits is a myth. Most of the benefits in the lengthy list of optional services that the states can choose to cover do not add much to program costs. Data presented in this paper show that only a small share of spending for those with Medicaid coverage is for services considered optional; expenditures on those services for people with private coverage are actually higher. These results further show that moving those who are now on Medicaid into private coverage could increase outlays significantly. While access to some services, particularly for those with serious health conditions, might be improved, proponents of such shifts should be aware that there could be serious fiscal consequences of such a policy change. Even if one accepts that Medicaid is in fact a very frugal form of insurance coverage, it is nevertheless true that its costs are high in an absolute sense, and that rising Medicaid costs cause severe fiscal strain for states, especially during economic downturns when states revenue growth slows or even falls. Our analysis suggests that reducing 340

19 Medicaid vs. Private Insurance optional benefits will have little impact on Medicaid costs, since high Medicaid costs are fundamentally due to the poor health of many Medicaid enrollees. Unfortunately, as long as most states are unwilling to raise taxes and unable to use deficit financing as a way of smoothing out cyclical fluctuations in revenues and Medicaid costs, they inevitably will face periodic fiscal crises that they can solve only by drastic cuts in Medicaid enrollments and/or benefits. Other research suggests that reducing the destabilizing effects of Medicaid on states finances is likely to come about only with a greater federal role in financing Medicaid costs (Holahan, Weil, and Wiener 2003). Notes Diane Rowland, Barbara Lyons, Catherine Hoffman, David Rousseau, Stephen Zuckerman and Timothy Waidmann provided useful comments on earlier drafts of this report. We also gratefully acknowledge Marc Rockmore, who provided research and statistical assistance. 1 Just over 4% of the people in the analysis are observed for less than a full year, primarily due to birth, death, institutionalization, or military service. However, their insurance coverage for the entire period they are observed is either Medicaid or private insurance. 2 Full time refers to people observed for a full year plus those included in MEPS for less than a full year due to birth, death, institutionalization, military service, or other factors that make someone ineligible for the survey. 3 We did not adjust the estimates of the standard errors of the regression coefficients for the effects of MEPS complex sampling design because testing hypotheses about the statistical significance of the parameter estimates is not a primary goal of the analysis. However, the complex sampling design was taken into account in calculating t-statistics for differences in spending and other characteristics between the Medicaid and privately insured populations. 4 Ideally, one would like to treat type of insurance coverage as endogenous, since people s decisions about whether to have coverage presumably depend on their expectations about the need for medical care. In this analysis, however, all people have coverage, although it still may be the case that some of those with Medicaid coverage have Medicaid because they have experienced high medical spending. This suggests that if there is a bias due to selection of people into Medicaid, rather than private coverage or no coverage, it most likely would tend to be a positive bias (i.e., overstating the true effect of Medicaid coverage on medical spending). Unfortunately, the public use version of the MEPS data does not readily allow one to adjust for possible endogeneity bias. 5 We use hypothetical people rather than data for individuals from the samples because the twopart expenditure models tend to produce unreliable estimates for particular people with multiple expensive medical conditions or disabilities (personal communication, W. Manning, May 12, 2003). Predicting expenditures for hypothetical people with average characteristics produces more stable estimates of central tendencies compared to calculating the mean of predicted expenditures based on a sample of individuals with widely different health conditions. 6 Mean values of independent variables by age and coverage are available in Hadley and Holahan (2003), Tables Children s health status is reported by the survey respondent (typically a parent) rather than the child. 8 Complete regression results for the pooled and separate Medicaid and private insurance samples by age are available in Hadley and Holahan (2003), Tables Ibid. 10 These predictions do not have standard errors because each prediction is based on the characteristics of an individual hypothetical person. To develop a sense of the range of these predictions, we estimated the expenditure models 200 times using repeated randomly drawn half samples of the privately insured and Medicaid-covered adults for each estimation of the model. We then used the 200 sets of parameter estimates to predict expenditures for the same hypothetical people, and calculated means and standard deviations for the set of 200 predictions. This exercise yielded the following mean predictions and standard deviations (in parentheses) for the hypothetical adult currently covered by Medicaid (as in the upper half of Table 10): Predicted spending, all adults ($) Baseline coverage (Medicaid) Alternative coverage (private) Combined expenditure model 3,270 (169) 3,707 (215) Separate expenditure models 3,184 (188) 4,329 (365) The relatively small standard deviations suggest that the predicted differences are significantly different from each other. 341

20 Inquiry/Volume 40, Winter 2003/2004 References Cohen, J Design and Methods of the Medical Expenditure Panel Survey Household Component. MEPS Methodology Report No. 1, Pub. No Rockville, Md.: Agency for Healthcare Research and Quality. Congressional Budget Office Medicaid Baseline. March. Hadley, J., and J. Holahan Is Healthcare Spending Higher under Medicaid or Private Insurance: Statistical Appendix. Washington, D.C.: Kaiser Family Foundation. Available at: Holahan, J., A. Weil, and J. Wiener, eds Federalism and Health Policy. Washington, D.C.: The Urban Institute Press. Jones, A Health Econometrics. In Handbook of Health Economics, A. Culyer and J. Newhouse, eds. Amsterdam: Elsevier. Manning, W., and J. Mullahy Estimating Log Models: To Transform or Not to Transform? Journal of Health Economics 20(4): Marquis, M.S., and S. Long Reconsidering the Effect of Medicaid on Health Care Services Use. Health Services Research 30(6): Medicare Payment Advisory Commission (Med- PAC) Report to the Congress: Medicare Payment Policy. (March):156. Washington, D.C.: Medicare Payment Advisory Commission. Neuschler, E., and R. Curtis Premium Assistance: What Works? What Doesn t? Washington, D.C.: Institute for Health Policy Solutions. Norton, S., and S. Zuckerman Trends in Medicaid Physician Fees, Health Affairs 19(4): Selden, T., K. Levit, and J. Cohen, et al Reconciling Medical Expenditure Estimates from the MEPS and the NHA, Health Care Financing Review 23(1):