Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
2 Mechanisms and Methods: Looking At the Impact of Health Insurance on Health The relationships between health insurance and access to health care, and health insurance and care received, have been the subject of hundreds of studies over the past several decades. More recently, the relationship between health insurance and health outcomes has also been examined. This chapter describes the Committeeâs analytic approach to its critical review of this research to inform the understanding of the relationships between health insurance, health care, and health outcomes for adults. The chapter is organized in three sections. First, it outlines the mechanisms by which the Committee postulates that health insurance affects health-related out- comes. Whether one has health insurance, a regular source of care and, if one is uninsured, the length of time that one is without coverage all influence access to care and affect health-related outcomes. The second section discusses issues related to the measurement of health insurance effects and considerations of research design that affect the inferences that can be drawn. It explores analytic strategies to distinguish the effects of health insurance status from those of personal attributes that are correlated with health insurance, including health status, race and ethnicity, and socioeconomic status, which may confound1 the results of studies that relate health insurance to health outcomes. The section gives particular attention to the two-way causal relation- ship between health status and insurance status.2 It also describes the major popu- 1See Appendix C for definitions of technical terms such as âconfound.â 2Perhaps most problematic for determining the effects of health insurance on personal health is that an individualâs health condition may be directly related to enrollment in particular kinds of insurance 25
26 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE lation surveys and databases that provide information about Americansâ use of health care and epidemiological information about health status and disease preva- lence. These sources provide the data for many of the most informative studies reviewed. The final section of this chapter presents the Committeeâs methods for sys- tematic review and synthesis of the research on health insurance status effects. This section describes how studies were identified for inclusion in the review, the criteria for evaluating the methodological quality of studies and study results, and how the Committeeâs findings are presented in Chapter 3. MECHANISMS AND MEASURES OF ACCESS TO HEALTH CARE Health insurance facilitates access to health care by removing or diminishing financial barriers to obtaining care. Among people who have insurance, the extent of cost sharing also influences the use of health care (Newhouse et al., 1993; Zweifel and Manning, 2000). An extensive body of research consistently finds a strong and positive relationship between health insurance and access to care, even as the definitions and measures of access have been strengthened. Population- based surveys conducted over the past three decades have evaluated access to primary care in relation to health insurance status with measures such as any physician visit within a year, the number of physician visits per year, having a regular source of care, and the ability to obtain care when needed (Freeman and Corey, 1993; Hafner-Eaton, 1993; Newacheck et al., 1998; Nelson et al., 1999; Zuvekas and Weinick, 1999; Haley and Zuckerman, 2000; Kasper et al., 2000; Shi, 2000; Weinick et al., 2000; Hoffman et al., 2001). Public policy and health care industry interests in high-quality and efficient health care have developed in tandem with the progress of clinical effectiveness research over the past decade. The standards of evidence for the efficacy of health insurance in promoting better health outcomes have evolved from enumerating physician visits to measurable improvements in effective processes of care. The notion of âaccessâ itself has shifted from a simple measure of utilization to mea- sures that incorporate the quality of care and health outcomes. In 1993, the Institute of Medicine (IOM) Committee on Monitoring Access to Personal Health Care Services reconceptualized access as âthe timely use of personal health services to achieve the best possible health outcomesâ and recommended a set of health outcome measures that could serve to monitor populations over time for access to basic health services (Millman, 1993). programs (e.g., as a medically needy or disabled Medicaid beneficiary and other Medicaid enrollees who join the program only when they incur a hospitalization or need other expensive care; in the case of nongroup private health insurance, only persons in good health may be accepted for coverage). See Box 2-1 for further discussion.
MECHANISMS AND METHODS 27 In studies of access to care and health outcomes, several factors mediate the relationship between health insurance and health-related outcomes. These include being able to see a provider when one believes care is needed, having a regular source of health care, having continuity of coverage, and the duration of periods without health insurance. Measures for each of these factors provide some infor- mation about an individualâs or populationâs access to health services that supple- ments the measurement of health insurance status at a given point in time. These measures are discussed below. Getting Care When Needed The ability to see a physician or other health care provider when one believes medical attention is needed is a fundamental and intuitive measure of access to health care. Most Americans mistakenly believe that people without health insur- ance have this level of access (IOM, 2001a). Although the lack of health insurance is not the only reason someone might not be able to see a health care practitioner when needed, it is a major one.3 Adults without health insurance are far more likely to go without health care that they believe they need than are adults with health insurance of any kind (Lurie et al., 1984, 1986; Berk and Schur, 1998; Burstin et al., 1998; Baker et al., 2000; Kasper et al., 2000; Schoen and DesRoches, 2000; Davidoff et al., 2001; Holahan and Spillman, 2002). While the overall percentage of adults who reported that cost prevented them from seeing a doctor in the previous 12 months increased only slightly from 10 percent to 11 percent between 1991 and 1996, the proportion of uninsured adults who reported this barrier to care increased from 28 to 35 percent, and the fraction of insured adults reporting this barrier decreased slightly from 8 to 7 percent (Nelson et al., 1999).4 In 1998, nearly 70 percent of uninsured adults in poor health could not see a doctor at some time during the year because of cost (Ayanian et al., 2000). A study that polled 1,100 patients four months after their initial visit to an emergency department found that patients who lost their health insurance were more than twice as likely as those who maintained their coverage to have delayed seeking care in the four-month interval (Burstin et al., 1998). Evaluations based on professional judgment confirm findings based on a sub- jectively determined need for care. In one study with a national probability sample of almost 3,500 adult respondents, a physician panel identified 15 serious condi- tions for which they deemed medical attention necessary (Baker et al., 2000). In 3Other reasons include not being able to find a provider within traveling distance and not being able to find a practitioner that participates in oneâs health plan (particularly a problem for Medicaid enrollees in states with low provider payment levels), language or cultural barriers, and understanding of symptoms and health conditions as requiring professional attention. See the IOM report, Access to Health Care in America (Millman, 1993) for a discussion of barriers to access. 4All of these differences are statistically significant.
28 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE an analysis that adjusted for demographic and economic characteristics and also for health status and having a regular source of care, the authors found that an uninsured adult was much less likely than an insured adult to get care for a reported symptom (odds ratio [OR] = 0.43). Examining only those symptoms for which the respondent thought care was needed, those without insurance were even less likely to have received care (OR = 0.28). Among those who did not receive needed care, the uninsured were far more likely than those with insurance to report that they did not get care because of cost (95 percent and 23 percent, respectively) (Baker et al., 2000).5 A lack of health insurance acts not only as an initial barrier to care but may continue to impede the receipt of appropriate, effective care. Even if uninsured patients receive primary care, referrals to specialists, ancillary diagnostic and treat- ment services and medications are more difficult to obtain. Primary care providers who treat uninsured and other low-income patients report greater difficulty in arranging for referrals and services that they cannot directly provide for their uninsured patients than for those who are insured (Fairbrother et al., 2002). Persons who never present themselves to a health care provider are not accounted for in health services research that documents and measures utilization and outcomes with hospital administrative records, patient chart reviews, and clinic encounter forms. This is a âblind spotâ and source of bias in studies of health insurance effects because overall, persons without health insurance are estimated to use roughly two-thirds of the services that those who do have insurance use (Marquis and Long, 1995). Because those without health insurance are less likely to see a provider than are others with insurance and thus are less likely to be included in research documentation, studies that rely on health care records to compare groups who received some care may overstate utilization by uninsured populations. Having a Regular Source of Care In addition to supplying the financial resources that enable one to obtain health care when needed, insurance coverage also improves receipt of appropriate care by facilitating the use of a regular source of care or primary care provider.6 5Previously reported in IOM (2001a). 6Measurement of a usual or regular source of care is frequently based on survey participant re- sponses to a single question such as âIs there a place or health care practitioner that you routinely go to for medical care?â However, some studies use more restrictive definitions of a regular source of care (e.g., they exclude hospital emergency departments as a positive response to the question.) The studies included in this review that document a regular source of care may use different operational definitions. One problem with evaluating utilization outcomes in terms of having a regular source of care is that frequent users of services are more likely to report having a regular source; the direction of causality is not clear. As the concept of primary care has been further conceptualized, the attributes associated with a regular source of care have become more specific (Starfield, 1992, 1998; Shi, 2000).
MECHANISMS AND METHODS 29 Both health insurance and having a regular source of care contribute indepen- dently to the utilization of health services (Solis et al., 1990; Mosen et al, 1998; Mandelblatt et al., 1999; Zuvekas and Weinick, 1999; Cummings et al., 2000; Breen et al. 2001). Having a regular source of care enhances the appropriate use of ambulatory care as measured by receipt of preventive services, management of chronic conditions, and population rates of avoidable hospitalizations (Bindman et al., 1995; Starfield, 1995; Pappas et al., 1997; Kozak et al., 2001). The independent contribution that having a regular source of care makes to the receipt of appropriate care reinforces rather than diminishes the importance of health insurance, because health insurance is an important determinant of obtain- ing and maintaining an ongoing relationship with a health care provider. Adults with health insurance are much more likely than those who are uninsured to have a regular source of care, a consistent finding across states with very different health care resources and provider configurations (IOM, 2001a; Holahan and Spillman, 2002). An analysis based on the 1997 National Health Interview Survey found that among adults eligible for Medicaid, 42 percent of those not enrolled in the program did not have a regular source of care, whereas only 12 percent of those with Medicaid coverage lacked one (Davidoff et al., 2001). Even those uninsured adults who have chronic conditions are substantially more likely to lack a regular source of care than are chronically ill adults with health insurance. Among unin- sured adults, 19 percent with heart disease, 14 percent with hypertension, and 26 percent with arthritis do not have a regular source of care, compared with 8, 4, and 7 percent, respectively, of their insured counterparts (Fish-Parcham, 2001). Someone without health insurance who can identify a regular source of care may still face difficulties in obtaining recommended and effective health care services that are outside the scope of practice of their regular provider, such as referrals to specialists, ancillary services, and hospital-based care. Continuity of Care and Coverage Not only is continuity of care, as measured by having a regular source of care, important, continuity of insurance coverage is also critical to the receipt of appro- priate care. Breaks in coverage can disrupt care relationships to the detriment of quality health care. Being uninsured for longer periods of time can be expected to have larger effects on utilization of services (and consequently on health) than being uninsured for shorter periods. Although only a few studies have examined health outcomes by length of time uninsured (Lurie et al., 1984, 1986; Ayanian et al., 2000; Kasper et al., 2000; Baker et al., 2001), a number of studies have looked at intermediate measures of access to care and volume of services (e.g., any physician visit or number of visits within a year) and find strong relationships between breaks in coverage and length of time uninsured, on the one hand, and reduced access to and utilization of services, on the other (Burstin et al, 1998;
30 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE Ayanian et al., 2000; Kasper et al., 2000; Schoen and DesRoches, 2000; Hoffman et al., 2001).7 One study examined clinical outcomes in a small cohort of patients with high blood pressure who were followed after some of them lost Medi-Cal coverage (Californiaâs Medicaid program). Blood pressure control for those who lost public insurance was worse at six months than at one year following the loss of coverage. The authors hypothesize that the deterioration in blood pressure control was reversed to some extent after new care arrangements were established following the termination of benefits (Lurie et al., 1986). In another study that compared transitions among health care plans to the complete loss of coverage, researchers found evidence of a temporary weakening of access to care among those who changed plans and a more pronounced loss of access among those who lost health insurance (Burstin et al., 1998). The experience of intermittently insured adults, in terms of both access and outcome measures, falls between that of continuously insured adults and continuously uninsured adults but is more similar to the latter than the former (Schoen and DesRoches, 2000; Baker et al., 2001). These results suggest that continuity and stability in health insurance coverage contribute to reliable access and effective health care. Finally, in a large-scale survey comparing persons uninsured for shorter or longer time periods, measures of access, appropriate care, and health status were worse for those uninsured for longer periods (Ayanian et al., 2000). These findings are particularly relevant to the interpretation of results for Medicaid enrollees, whose coverage tends to be of limited duration and who may have been uninsured before they were enrolled in Medicaid (IOM, 2001a; Perkins et al., 2001). In fact, more than half of single women with Medicaid at the beginning of a year have lost their coverage before the end of the year, and between one-third and one-half of them are uninsured a year after they lose Medicaid coverage (Short and Freedman, 1998; Garrett and Holahan, 2000). METHODOLOGY AND MEASUREMENT OF HEALTH INSURANCE EFFECTS Chapter 1 poses the problem of how to determine whether a âdoseâ of health insurance is effective in improving an individualâs health. Experimental studies provide a stronger research design than observational studies for concluding that insurance itself affects health. However, most of the research examining the im- pact of health insurance on health-related outcomes is based on observational 7Although access to health care services measured at this most general level (i.e., in terms of the number of visits to physicians and self-reported delays in seeking care) is not the focus of the Committeeâs review, it can be postulated that longer times without coverage that result in more greatly reduced utilization and delayed care would similarly affect the specific health-related outcomes examined in this report.
MECHANISMS AND METHODS 31 (nonexperimental) and, with several important exceptions, cross-sectional (point- in-time, nonlongitudinal) studies. Indeed, earlier reviews of the research on health insurance status and health outcomes have highlighted the need for longitudinal studies and close examination of natural experiments, for example, when public coverage programs are expanded to new population groups or cut back or among state-level programs with differing eligibility standards (Weissman and Epstein, 1994; Brown et al., 1998). The one major randomized trial of health insurance, the RAND Health Insurance Experiment, did not have an uninsured group, although it did include a group with 95 percent cost sharing and a very high deductible, which approximated no coverage for routine care (Newhouse et al., 1993).8 Because of the great potential for biases in observational studies, the associa- tions observed between health insurance status and health may reflect the effects of factors other than health insurance, such as income, social status, or education. Much of the observational research, especially that conducted in the past decade, has included extensive statistical adjustment for some of these potentially con- founding factors. Such adjustment strengthens the analytic design of observational studies and increases the likelihood that the observed associations between health insurance and health outcomes represent a causal relationship. However, addi- tional, less readily measured personal characteristics covary with health insurance status and may also affect health outcomes. This section examines the sources of potential bias and the analytic strategies used to address them in observational studies of health insurance effects. It also reviews the personal characteristics that are most likely to covary with health insurance statusânamely, health status, race and ethnicity, and socioeconomic status (SES) and considers how they may confound findings of health insurance effects in observational studies. 9 Limits of Observational Studies Sources of Bias In relying on observational studies to determine whether and how health insurance affects health-related outcomes, the first limitation is the wide variability 8See footnote 8 in Chapter 1. 9See the Committeeâs earlier report, Coverage Matters, for an analysis of personal characteristics related to health insurance (IOM, 2001a). In this report âraceâethnicityâ is used to refer to the classification of population groups according to U.S. Census categories unless otherwise indicated. âSocioeconomic statusâ encompasses a variety of indicators, including income, educational attain- ment, type of employment or job classification, and residential area. Health insurance status itself has often been used as a measure of SES. Most studies of health insurance status effects include at most two or three SES indicators in addition to raceâethnicity as control variables.
32 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE in how the term âhealth insuranceâ is defined and measured. In cross-sectional studies, a participantâs health insurance status is measured at one point in time, when other personal information is collected or when health care services are used and utilization is documented in administrative or clinical records. Even longitu- dinal studies that follow a study participant across time often measure health insurance status only at the beginning of the study period. Because health insur- ance status may change over time (with the important exception of those who qualify, on any basis, for Medicare), the classification of study participants as âinsuredâ includes not only those who always had insurance, but also those who may have recently been uninsured and now have coverage. Likewise, those clas- sified as uninsured include not only those who have always been without health insurance, but also those who may have had coverage until recently. The result of the point-in-time measurement of health insurance status is to diminish the ability to measure any effect of health insurance on health outcomes because the groups being compared have overlapping membership. Therefore, the actual effect of health insurance, if it were to be measured in terms of its duration, may be greater than found in observational studies that look at insurance status at a single point in time. Most studies of health insurance effects do not fully account for a second kind of bias in measuring health insurance coverage, namely, the wide range of benefit packages and of cost-sharing and provider participation arrangements subsumed under the category of general health insurance. Health insurance plans that cover a wide array of benefits or that require no or limited cost sharing (deductibles and coinsurance) from enrollees can be expected to affect patient and provider behav- ior differently from plans that cover fewer benefits and require more cost sharing (Zweifel and Manning, 2000). Finally, health insurance plans differ in terms of provider payment and participation rates and arrangements, affecting enrolleesâ ease of access to care and patterns of utilization. Another potential source of bias in observational studies is the nonrandom distribution or selection of study participants among health insurance status cat- egories. Health status is itself a determinant of health insurance coverage. It is closely related to the likelihood that someone has insurance and to the kind of coverage he or she has (IOM, 2001a). Working-age adults in better health are more likely to work full time and at higher paying jobs, thus increasing their chances of having employment-based health insurance. At the same time, those with access to employment-based coverage who anticipate needing health care are somewhat more likely to take up the offer of coverage than similar persons in good health with no expectation of significant health care use and expense. Adults in poor health without access to employment-based coverage are more likely to seek individually purchased coverage but may find that they cannot obtain health insurance because of preexisting conditions. Those in the poorest health who are unable to work (i.e., who are recognized as permanently and totally disabled) or who have very low incomes or high medical expenses may qualify for Medicare or Medicaid. Box 2-1 reviews the eligibility requirements for Medicaid and Medi-
MECHANISMS AND METHODS 33 care that contribute to the distinctive health status profile of working-age adults with public health insurance. Thus, health status differs systematically among persons grouped by health insurance categories because health status is one criterion by which people qualify for coverage, and analytic adjustments for underlying health status tend to be incomplete at best. Health-related behaviors such as smoking, exercise, and diet also affect individual health and are sometimes but not usually included as covariates in analyses of health insurance effects.10 These health behaviors are strongly re- lated to educational attainment, which itself is correlated with health insurance status (IOM, 2001a). Finally, additional personal characteristics, such as the will- ingness to live with risk and the value placed on good health or health care, may bias the selection of enrollees in private health insurance plans, including those who take up the offer of employment-based coverage and those who seek to purchase individual coverage. Because most of those offered employment-based coverage do accept it and relatively few purchase individual coverage, this source of selection bias is not likely to substantially affect the overall comparisons be- tween insured and uninsured adults. Major Covariates Personal characteristics that vary with health insurance status may confound analyses of the effects of health insurance on health-related outcomes because they are independently associated with these outcomes. In addition to health status, which has just been discussed, the most important among these characteristics are race and ethnicity and SES. Race and ethnicity are frequently included in analyses that examine the effects of health insurance status on health outcomes. SES, however, is more difficult to separate completely from health insurance status, even when it is represented in multivariable statistical analyses by employment, educational attainment, or income. In addition, distinctive ethnic and cultural population groups with different economic and behavioral characteristics are sub- sumed under the broad racial and ethnic categories most commonly used in research. For example, Mexican and Cuban Americans and Puerto Ricans are all categorized as Hispanic; likewise, African and Caribbean immigrants may be categorized with African Americans as Black. Racial and ethnic minority groups and persons with lower SES often face barriers to obtaining health care, such as lack of transportation, concerns with personal safety, and provider shortages, that extend beyond those related to being uninsured. Even with insurance and health care, disadvantaged groups often have worse health on average than do socially privileged groups. 10See the report Promoting Health: Intervention Strategies from Social and Behavioral Research, (IOM, 2000b) for a review and analysis of behavioral factors in health.
34 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE BOX 2.1 Health Insurance and Health Status The average health status of adults with private (employment-sponsored or individually purchased) health insurance and those with public health insurance differs, and both differ from the average health status of uninsured adults. Taken together, adults under age 65 with private health insurance are healthier than adults without any insurance, who are in turn healthier than adults with public insurance (Sorlie et al., 1994; Hahn and Flood, 1995). Privately Insured Adults Both adults who have employment-sponsored group insurance and those who have individually purchased policies have better-than-average health status. In the case of employment-sponsored group coverage, insured adults either are working or are the dependents of workers. Workers have a better health status than nonworkers, and insured workers have higher family incomes than uninsured workers. Adults with individual coverage (e.g., the self-employed and early retir- ees) face health screening and risk-rated premiums, and thus tend to be both healthier and wealthier than counterparts who do not obtain individual coverage. Publicly Insured Adults Adults under age 65 obtain health coverage from a public program, either Medi- care or Medicaid, because they are sick, poor, or medically needy. Adults with public program coverage tend to be in substantially worse health than those with private insurance and in somewhat worse health than uninsured adults. Medicare includes people under 65 who have been covered by the Social Se- Rates of health insurance coverage differ for various racial and ethnic groups: African Americans, Asian Americans and Pacific Islanders, and Hispanics are much more likely to be uninsured than are non-Hispanic whites. Eighteen and a half (18.5) percent of African Americans, 18 percent of Asian Americans and Pacific Islanders, and 32 percent of Hispanics are uninsured, compared with 10 percent of whites (Mills, 2001). Ethnic differences in insurance status partially reflect differ- ences in the rates of employment-sponsored insurance coverage. Public insurance programs (e.g., Medicaid) only partially make up for these disparities in employ- ment-sponsored coverage (Gabel, 1999; Monheit and Vistnes, 2000; IOM, 2001a). Insurance is also correlated with socioeconomic status. Forty (40) percent of adults who live in lower-income families (defined as having incomes less than 200 percent of the federal poverty level and 40 percent of adults without a high school diploma are uninsured, compared with 10 percent of adults with at least a college degree (Hoffman and Pohl, 2000; Monheit and Vistnes, 2000). In addition, the factors of raceâethnicity and SES are intertwined. In the United States, ethnic minorities face more limited educational and occupational opportunities. Many ethnic minorities live in highly segregated communities (Massey and Denton, 1989; Cutler et al., 1999). These neighborhoods may be
MECHANISMS AND METHODS 35 curity Disability Insurance program for at least two years and those certified as having end-stage renal disease. Medicaid eligibility is based on income and on additional categorical criteria, such as being pregnant, being disabled or blind, or being a parent with dependent children. If others with incomes higher than the state Medicaid eligibility limit have sufficiently high medical expenses, they are considered âmedically needyâ and qualify. In most states, beneficiaries must periodically re-enroll, making periods without insurance relatively common. The median length of time that an adult un- der age 65 remains on Medicaid is 5 months (Tin and Castro, 2001). There is a greater likelihood that a Medicaid enrollee has just been or is about to become uninsured than is the case for privately insured individuals. Last, many eligible people do not enroll until they seek medical care and the hospital or clinic signs them up in order to receive payment for services provided (Davidoff et al., 2001). Thus, the healthier among those eligible for Medicaid are less likely to be enrolled than those who are sicker. Uninsured Adults Some adults who are uninsured are eligible for Medicaid but not enrolled. Adults who meet income and other eligibility standards may not feel the need to enroll if they do not have unmet health care needs. These adults who are eligible but not enrolled in Medicaid have better overall health status and fewer activity limitations or chronic conditions than otherwise comparable adults enrolled in Med- icaid. Since the enactment of federal welfare legislation in 1996, some adults have mistakenly believed that they are not eligible (Garrett and Holahan, 2000). Other uninsured adults with social and economic characteristics similar to those of per- sons enrolled in Medicaid do not meet other eligibility requirements. more crowded, offer fewer economic opportunities, be more geographically iso- lated from health care providers, and have more environmental health hazards. Each of these factors may adversely affect health. Boxes 2.2 and 2.3 illustrate how health insurance is just one of several sources of health disparities among racial and ethnic groups and persons of different SES. Context of Care Every study that examines the care and outcomes for people without health insurance does so within a particular economic and institutional context of health care for the dominant population, namely, those with health insurance. Nationally, one out of every six or seven adults receiving health care is uninsured. Among states, the ratio of uninsured to insured ranges from one in ten (Minnesota and Rhode Island) to one in four (Texas, Arizona, New Mexico) (IOM, 2001a). Two points follow from this observation. First, national studies that examine the health care and outcomes of uninsured adults mask a wide range of health care financing contexts within which care is rendered. Second, any evaluation of the care of uninsured adults reflects the financing and resource environment that
36 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE BOX 2.2 Race and Ethnicity The effect of race and ethnicity on access to care and health outcomes oper- ates through a number of mechanisms, many not directly linked to financial barri- ers or health insurance status. Some of the racial and ethnic differences in mortal- ity are reflected in differences in the prevalence of specific illnesses and events, such as cancer, cardiovascular disease, substance use, diabetes, infant mortality, and homicide. Even when mortality rates are adjusted to reflect both the preva- lence of these conditions and for income, however, one-third of the difference in the mortality rate between blacks and whites remains (Mutchler and Burr, 1991). Some have suggested that racial and ethnic differences in health are due largely to differences in SES (Sorlie et al., 1994). Yet, at each level of income, African Amer- icans have higher mortality than whites, suggesting that racial and ethnic dispari- ties are not explained solely by differences in socioeconomic status (Pamuk et al., 1998). Last, access to health care may be impeded by language barriers for ethnic minorities, especially but not exclusively among immigrants. Among Hispanics in the United States a greater degree of acculturation (and thus higher income) appears to be associated with a decline in health status (Vega and Amaro, 1994). Since acculturation may be associated with greater educational and employment opportunities, this also suggests that the effect of ethnicity is independent of SES and may be related to social structure, diet, life style, and other health practices and beliefs. Race or ethnicity may be a marker for other cultural beliefs that go beyond the commonly examined labels used in health ser- vices research. For example, beliefs about the causes of cancer may differ be- tween African Americans and whites (Maynard et al., 1986). Cultural differences between patients and providers may result in poor commu- nication that undermines effective care and patientsâ adherence to treatment reg- imens (IOM, 2002). Cultural similarity between patients and providers may facili- tate communication and decision making (Komaromy et al., 1996; Saha et al., 1999; Schulman et al., 1999). Because African Americans and Hispanics are un- derrepresented in the health care professions, this concordance is often not real- ized (Collins et al., 1999; Saha et al., 1999, IOM, 2002). Racial bias in diagnosis and treatment is yet another potential explanation of the observed disparities (Reisch et al., 2000; IOM, 2002). Further, beyond implicit bias, the experience of overt discrimination may directly affect both access to care and health. Discrimination may affect trust in the patientâdoctor relationship, which in turn may also affect the use of health care and health outcomes (Maynard et al, 1986; Doescher et al., 2000). Discrimination may restrict the choice of health care providers, and therefore the quality of care available (Krieger et al., 1993). Finally, the experience of racism may directly result in psychological distress that could adversely affect health (Broman, 1996; Carrasquillo et al., 2000). prevails at that time and, if the balance of uninsured to insured populations changes, the general processes and outcomes of care may also change, for both those who have and those who lack health insurance. This phenomenon will be addressed in a subsequent Committee report on community impacts of uninsured populations.
MECHANISMS AND METHODS 37 BOX 2.3 Socioeconomic Status SES affects health through a variety of mechanisms. While some of these are directly related to limited financial resources, others are related to health less di- rectly, through belief, behaviors, and environmental factors. Socioeconomic disad- vantage can have a continuing or cumulative effect throughout an individualâs life. For example, poverty in childhood may affect the health status of an adult (Lynch et al., 1994,1997; Krieger et al., 2001). The relationship between SES and health is bidirectional. In addition to SES influencing health, poor health may cause declines in SES related to loss of employment or income (Adler et al., 1994). This makes the interpretation of cross-sectional studies examining the relationship be- tween SES and health particularly challenging. Education and income are critical. Low educational attainment, poverty, and economic hardship have each been associated with higher rates of chronic illness, poor self-reported health status, disability, and lower life expectancy (Haan et al., 1987; Pincus et al., 1987; Marmot et al., 1991; Winkleby et al., 1992; Guralnik et al., 1993; Elo and Preston, 1996; Ostrove et al., 2000; van Rossum et al., 2000). Even among those with access to care, individuals with less education may be less able to communicate with their provider, understand risk factors or symptoms of disease, schedule an appointment, or manage their own health condition (Kunst and Mackenbach, 1994; Behera et al., 2000). Health behaviors such as diet, physical activity, smoking, and alcohol or illicit drug use are also strongly associated with both health outcomes and SES (Otten et al., 1990; Ford et al., 1991; Winkleby et al., 1992; Dixon et al., 2001). A strong inverse association between number of years of education and smoking, choles- terol level, blood pressure, and body mass index exists for the U.S. population (Winkleby et al., 1992). Other potential pathways by which SES may affect health include the following: â¢ living in neighborhoods with more crowding (with more potential for expo- sure to communicable diseases), higher rates of violence, and more pollution (Samet et al., 2000); â¢ greater exposure to stressful life events with associated risk of illness (Co- hen et al., 1991; Ruberman et al, 1991); and â¢ less ability to take time off from work to see the doctor or to arrange for child care or transportation for a visit. National Surveys and Databases Many of the studies reviewed draw on a number of publicly sponsored surveys with national or state-level probability samples, epidemiological databases, and disease registries. Box 2.4 identifies these surveys and data sources. The size of these sample surveys, the comprehensiveness of reporting systems, and the collec- tion of comparable data periodically over several years and even decades contrib- ute to the quality of the information available from these sources.
38 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE BOX 2.4 Surveys and Databases Surveys The National Health Interview Survey (NHIS) is a continuous nationwide household interview survey conducted by the National Center for Health Statistics (NCHS). It is designed to allow the development of national estimates of health status and health services utilization of the U.S. civilian noninstitutionalized popu- lation. The NHIS has been conducted since 1957, with the core content of the survey being updated approximately every 10â15 years. Interviewers obtain infor- mation on personal and demographic characteristics, illnesses, injuries, impair- ments, chronic conditions, utilization of health services, health insurance, and oth- er topics from about 43,000 households comprising approximately 106,000 individuals. The National Health and Nutrition Examination Survey (NHANES) conduct- ed by the NCHS is designed to assess the health and nutritional status of adults and children in the United States through interviews and direct physical examina- tions. Approximately 5,000 people are examined each year. NHANES was con- ducted periodically from 1960 to 1994 with seven national surveys completed dur- ing that time. The most recently published data are for the six-year survey conducted between 1988 and 1994, referred to as NHANES III. Since 1999 sur- veys have been conducted on an annual basis. NHANES collects information on chronic disease prevalence and conditions, risk factors, diet and nutritional status, immunization status, and specific clinical measures such as blood pressure, blood cholesterol, blood sugar, mental health, oral health, and physical functioning. The Behavioral Risk Factor Surveillance System (BRFSS) is an ongoing telephone-based surveillance system designed by the Centers for Disease Control and Prevention (CDC) and conducted separately by each state. Approximately 150,000 adults are interviewed each year. BRFSS monitors modifiable risk behav- iors and other factors contributing to the leading causes of morbidity and mortality in the U.S. population. A standardized questionnaire is used to collect information on self-reported health habits and risk factors that contribute to the development of chronic diseases. States participate in the selection of questions that will be asked uniformly across the states. The primary data included in this survey are alcohol and tobacco use, use of preventive health services, HIV/AIDS, health status, limi- tations of activity, and health care access and utilization. The Medical Expenditure Panel Survey (MEPS), conducted by the Agency for Healthcare Research and Quality (AHRQ) for the first time in 1996, affords a comprehensive view of national health care utilization, household health expendi- tures, and health insurance coverage. MEPS consists of four linked, integrated surveys: 1. The Household ComponentâA sample of families and individuals across the nation, drawn from a subsample of households that participated in the prior yearâs NHIS. Approximately 13,000 families and 36,000 persons are represented in five interviews conducted over 30 months. Data are collected on health status, health insurance coverage, health care use and expenditures, and sources of pay- ment for health services. 2. Nursing Home ComponentâA national sample of nursing homes and resi-
MECHANISMS AND METHODS 39 dents. Data are collected on the characteristics of facilities and services offered, expenditures and sources of payment by resident, and resident characteristics. 3. Medical Provider ComponentâA national sample of hospitals, physicians, and home health care providers. Data are gathered to supplement information from the household component and can be used to estimate the expenses of peo- ple enrolled in health maintenance organizations and other types of managed care. 4. Insurance ComponentâInformation from the Household Component and from a sample of businesses and government agencies that offer insurance cover- age. For the Household Component, data are collected on the health insurance that respondents have had or that has been offered to them. The additional sample consists of data gathered from health plans and health insurance sponsors. This survey is used to develop national, regional, and state-level estimates of the amount, type, and cost of employment-based health insurance. The National Medical Expenditure Survey (NMES) was conducted in 1987 and is a predecessor to MEPS. NMES was a three-part national survey of the U.S. civilian noninstitutionalized population, including surveys of households, medical providers, health plans, and employers. Approximately 13,500 households were surveyed about their demographic background, medical use, costs, and payments. The National Medical Care Expenditure Survey (NMCES), conducted in 1977, assessed the cost and extent of health insurance in the United States. NM- CES included about 14,000 households. It is now known as the National Medical Expenditure Survey-1. The Health and Retirement Study (HRS), sponsored by the National Institute on Aging and conducted by the University of Michigan, gathers information on the various characteristics of those near or in retirement. It is a longitudinal national panel study that at baseline (1992) sampled 7,600 households with at least one member between the ages of 51 and 61, with follow-ups every 2 years, planned for 12 years. The data collected include economic, demographic, and health informa- tion, generally related to retirement issues, for more than 22,000 Americans over the age of 50. The Current Population Survey (CPS) is a nationwide sample survey of about 50,000 households representing 116,000 persons conducted monthly for the Bu- reau of Labor Statistics by the U.S. Bureau of the Census. It is the primary source of information on the employment characteristics of the U.S. civilian noninstitution- alized population. Since 1980, the CPS has included questions about health insur- ance coverage along with other demographic characteristics. Both the February and the March supplements to the CPS contain information related to health insur- ance. The HIV Cost and Services Utilization Study (HCSUS) is a major research effort to collect information on a nationally representative sample of people under care for HIV infection. The HCSUS is funded by AHRQ and conducted by RAND. The original survey (1994â2000) collected data on more than 3,700 HIV-positive people in hospitals, clinics, and private practices in 28 urban and 24 clusters of rural counties. The HCSUS design includes a baseline in-person interview, two follow-up interviews scheduled for 6 months and 12 months after the baseline, and data from patientsâ medical, pharmaceutical, and billing records. The HCSUS is composed of a core study and seven supplemental studies. The core study examines cost, use of health services, quality of care, access to and unmet needs for care, quality of life, social support, knowledge of HIV, clinical
40 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE BOX 2-4 Continued outcomes, mental health, and the relationship of these variables to provider type and patient characteristics. Databases The Healthcare Cost and Utilization Project (HCUP) is a database main- tained by AHRQ. HCUP is a partnership among states, the private sector, and the federal government to develop a standardized, multistate health data system on the use of health care services in small areas, the states, and the nation. Currently, 22 states contribute data to the project. HCUP includes a wide array of uniformly formatted administrative databases containing detailed information about hospital inpatients, ambulatory surgeries, and emergency services. The Surveillance, Epidemiology, and End Results Program (SEER) is a comprehensive source of population-based information on cancer incidence and survival data coordinated by the National Cancer Institute. Data are obtained from 11 population-based cancer registries and 3 supplemental registries covering ap- proximately 14 percent of the U.S. population. Information on more than 2.5 million cancer cases is included in the SEER database, and about 160,000 new cases are added each year. Incidence rates are estimated using statistics from the U.S. Bu- reau of the Census. The SEER registries routinely collect data on patient demo- graphics, primary tumor site, morphology, stage at diagnosis, first course of treat- ment, and follow-up for survival. The National Registry of Myocardial Infarction (NRMI) provides information on the current treatment rates at hospitals across the country for patients dis- charged following myocardial infarction. It is sponsored by Genentech, Inc. More than 1,500 hospitals of all types and from all regions are involved in the registry, and information is collected for more than 5 percent of the total hospitalized acute myocardial infarction (AMI) patients in the United States. NRMI allows hospitals to compare their own treatment rates to this national data set. Since 1990, NRMI has collected data on more than 1.7 million AMI patients. The United States Renal Data System (USRDS) is a national data system that collects, analyzes, and distributes information about the incidence, prevalence, treatment, morbidity, and mortality associated with end-stage renal disease (ESRD). The USRDS is funded directly by the National Institute of Diabetes and Digestive and Kidney Diseases in conjunction with the Centers for Medicare and Medicaid Services. Data are obtained on ESRD patients whose treatment is fund- ed by Medicare, which covers about 93 percent of all treated ESRD patients. Cur- rently, data are available for 581,000 patients treated between 1977 and 1995. METHODS OF THE SYSTEMATIC LITERATURE REVIEW AND SYNTHESIS The Committeeâs literature review updates and broadens the scope of a number of extensive reviews of research measuring the effects of health insurance status on health-related outcomes. Notable prior contributions include Does Health Insurance Make a Difference, a background paper prepared by the Congressional
MECHANISMS AND METHODS 41 Office of Technology Assessment (OTA, 1992); Falling Through the Safety Net: Insurance Status and Access to Health Care (Weissman and Epstein, 1994); âMonitor- ing the Consequences of Uninsurance: A Review of the Methodologiesâ (Brown et al., 1998); and No Health Insurance? Itâs Enough to Make You Sick (American College of PhysiciansâAmerican Society of Internal Medicine, 1999). To these earlier reviews, the Committee contributes the development and application of explicit criteria used to identify and select studies for inclusion and to assess their methodological strength. This section describes the selection crite- ria, the evaluation of research quality, and how the individual study findings are presented in Chapter 3. Identification of Studies and Inclusion Criteria The systematic literature review includes clinical and health services research and population surveys that are structured to examine the independent effect of health insurance on some health-related outcome. It includes studies with the following dependent variables or outcomes: â¢ General health status (self-reported or medically evaluated) â¢ Disease-specific clinical indicators (e.g., blood pressure) and stage of dis- ease at diagnosis or treatment â¢ Mortality (e.g., in-hospital; longer-term survival rates) â¢ Functional status, limitations, disability â¢ Use of services for specific conditions that are associated with improved health outcomes (e.g., periodic dilated eye exams for diabetics) â¢ Screening and other secondary preventive services â¢ Use of appropriate procedures (e.g., diagnostic and treatment services after acute myocardial infarction) â¢ Adverse events due to medical mismanagement â¢ Hospital admissions for preventable conditions The Committee excluded those studies that measured only basic access to care (e.g., number of physician visits per year, presence of a regular source of care, difficulty reported in obtaining care when needed) because the relationship be- tween health insurance and access is well established. This literature is discussed earlier in this chapter and in the Committeeâs first report, Coverage Matters (IOM, 2001a). All studies selected for systematic review include uninsured subjects. Some excluded studies, for example, were limited to comparisons of health-related outcomes among types of insurance coverage (e.g., fee-for-service or indemnity coverage versus health maintenance organization or managed care) and included no information on uninsured patients. The following three categories are the most common classification of insurance status groups for the purposes of the Committeeâs analysis:
42 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE 1. privately insured (employment-based or individually purchased coverage); 2. publicly insured (Medicaid, State Childrenâs Health Insurance Program, or Medicare for disabled or end-stage renal disease beneficiaries); and 3. uninsured. Studies that combine publicly insured and uninsured persons within a single category are included in the literature review because they may offer some insight into the factors that affect health-related outcomes, even though they do not yield results specific to uninsured adults and thus are of limited value in measuring the effects of uninsurance as such. Likewise, studies of health services utilization that report combined results for publicly and privately insured adults are included. Reporting findings for a single category of publicly and privately insured adults also presents a problem for interpretation because these insured groups differ in their health status, with publicly insured adults tending to have worse-than- average health status and privately insured adults better-than-average health status. This difference is rarely adequately controlled for analytically. The literature review focuses on outcomes for U.S. adults between 18 and 65. This review excludes perinatal and pediatric studies (as noted earlier, these will be reviewed in the next report of the Committee) and studies that are limited to the population over age 65, virtually all of whom have at least medical and hospital insurance, primarily through the Medicare program. To the extent that the re- search is discussed by disease category (e.g., diabetes care and outcomes), some studies may include children. These studies may be considered in the next Com- mittee report as well. Almost all of the studies reviewed share a conventional and imprecise defini- tion of health insurance as meaning general medical and hospital coverage. A few of the studies that examine preventive or mental health services assess health- related outcomes as a function of a specific insurance benefit package. These exceptions are noted in the discussion in Chapter 3. Studies that address dental services have been excluded from the review because coverage for dental services is minimal or missing from most basic health insurance plans for adults (KPMG, 1998). Furthermore, studies of dental care and outcomes tend to identify coverage for dental services specifically. Likewise, studies that exclusively consider institu- tional, long-term, or custodial care as a function of insurance coverage were excluded because these services are not usually included in health insurance ben- efit packages. Studies of rehabilitation services were included. The Committee applied its selection criteria to studies identified in PubMed searches conducted between March and June 2001 and updated monthly thereaf- ter through November 2001.11 Studies cited in the published literature surveys 11The electronic search included publication dates back to 1965; however, most citations were for studies published after 1985. The search terms included âinsurance status,â âinsurance, longitudinal,â âinsurance, cohort,â âuninsured, longitudinal,â âuninsured, cohort,â âpayer status,â âpayer source,â âmedically indigent,â and âuncompensated care.â
MECHANISMS AND METHODS 43 noted earlier in this chapter were also included if they met the criteria set out above. In addition, unpublished studies of which the Committee became aware through experts were included if they met the criteria listed above. A total of 131 primary research studies were reviewed and rated.12 This primary research bibliog- raphy is included in Appendix B. Evaluation of Research Quality The goal of the Committeeâs literature review is to evaluate the nature and quality of the evidence in the aggregate for particular kinds of health-related out- comes, rather than to judge the results of specific research studies. Thus, individual studies were evaluated in light of the information that they could contribute to the body of evidence on health insurance effects. To carry out that evaluation, two reviewers rated each study in the research bibliography (see Appendix B). Box 2- 5 presents the methodological review criteria. One reviewer was a member of the Subcommittee on Health Outcomes for the Uninsured and the other was an Institute of Medicine staff member or consultant with training complementary to that of the first reviewer. Studies that were judged to be of poor quality by both reviewers were not used in formulating the Committeeâs findings. If studies were judged to be of fair or good methodological quality by one reviewer and poor by the other, they were submitted to a third reviewer (a Subcommittee or Commit- tee member). Quantitative results are presented in Chapters 3 and 4 only for studies that received a fair or good evaluation from two reviewers. Presentation of Committee Findings Chapter 3 presents the Committeeâs findings based on the literature review, with the evaluation of studies organized into categories that reflect either specific diseases or type of service such as preventive and screening services or hospital care. The categories reflect practical considerations, including ease of presentation and summary of results. Findings regarding specific health conditions are more easily understood within the context of similar clinical research and can be related to larger populations at risk. Service-based categories permit synthesis of findings across studies with consistent outcome measures. This may be especially useful because health insurance coverage rules are often structured by service categories (e.g., preventive and screening services may be excluded or covered without any cost sharing to promote their use) but may also specify exclusions based on condition (e.g., mental illness). 12Separate articles using the same sample and analytic design, as noted in Appendix B, are counted as a single study here.
44 CARE WITHOUT COVERAGE: TOO LITTLE, TOO LATE BOX 2.5 Study Methodology Review Criteria The methodological quality of each study was rated on a three-point scale: Good = 3, Fair = 2, or Poor = 1. These ratings were based on the following considerations: â¢ The validity and relevance of the outcome measure (e.g., whether the study provides information on a result that is of interest). For example, if a process-of- care or preventive service measure was the focus of the study (the dependent variable), does an evidence base link it to a health outcome? If a health outcome is the dependent variable, is reporting of that outcome unbiased and reliable? â¢ Whether the health insurance status categories are well defined and con- sistently reported. For example, is health insurance status directly measured? â¢ The adequacy of adjustments to minimize biases introduced by covariates of health insurance status. At a minimum, studies should adjust or control for age, sex, and raceâethnicity. It should be noted, however, that by controlling for social and demographic covariates (e.g., raceâethnicity), cumulative effects or interac- tions between such factors and health insurance status may be missed in the analysis. Ideally, variables measuring income or educational attainment, health status, and site of care (for a process or quality-of-care outcome measure) should be included. In ranking a studyâs quality, reviewers determined which of these likely covariates were most important, given the specific outcome being measured. â¢ The study design, including consideration of the sample size, response rate, amount of missing data, representativeness of the study population, and consider- ation given to potential selection biases. In developing its summary findings, the Committee considered the volume of evidence within a given category in terms of the number and representativeness of studies and the degree of consistency among studies. The Committee gave greater weight to longitudinal than to cross-sectional studies and, among cross-sectional studies, placed more importance on those that measured and appropriately consid- ered key covariates. Regardless of the Committeeâs evaluation of methodological quality, however, all studies with results inconsistent with the Committeeâs sum- mary findings are discussed in Chapter 3 and included in Appendix B.