Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 129
B Coverage Does Matter: The Value of Health Forgone by the Uninsured Elizabeth Richardson Vigdor In the United States, 16.5 percent of the nonelderly population lacks health insurance (Fronstin, 2002). This translates into approximately 41 million people who are exposed to the potential risks of being uninsured. This problem has not attenuated over time. In fact, the uninsured proportion of the population has been on a generally upward trend for more than 10 years (Fronstin, 2002).1 There has long been concern in the policy world about the cost to society of this phenomenon. Health insurance is a key component of access to timely and effective medical care. Without the latter, individuals may end up consuming unnecessarily costly medical care at a later date, and productivity may be adversely affected if individuals are unable to work. One factor that is often overlooked, however, is the cost of forgone health experienced by the uninsured individuals. This paper examines the loss in health capital—the imputed dollar value of health that individuals will have over their remaining lifetimes—that accrues to society from this lack of health insurance. To measure this, I apply a variation of the methodology previously developed by Cutler and Richardson (1997) to measure “health capital,” following Grossman (1972). This measure combines several different dimensions of health to estimate the present value of the stock of present and future quality-adjusted life years. This analysis measures health capital empirically, using data on the length of life, the 1 At the time this analysis was conducted, the latest Current Population Survey estimates of insurance coverage were for calendar year 2000. Henceforth, this paper refers to those data, as presented in Fronstin (2001b).
OCR for page 130
prevalence of various health conditions, and the quality of life conditional on having those conditions. I estimate the average health capital of the insured and the uninsured first assuming perfect health, then incorporating morbidity. I calculate a lower-bound estimate of morbidity-adjusted health capital assuming no difference in morbidity between the two groups, and an upper-bound estimate using observed cross-sectional differences in morbidity. I then estimate the amount of health capital lost by not insuring the uninsured. I examine this under two scenarios. One assumes that everyone who is uninsured today will remain so until age 65, and another assumes that the uninsured face the average probability of being uninsured in a given year. Using benchmark assumptions of a 3 percent discount rate and a $160,000 value of a life year, I estimate that the value of future forgone health to an uninsured 45-year-old is between $7,800 and $83,000 using the years of life (YOL) approach, and between $6,000 and $102,000 using the quality-adjusted life year (QALY) approach. The value of future forgone health to an uninsured infant is between $4,600 and $50,000 with the years of life approach, and between $3,800 and $98,000 when morbidity is incorporated. These numbers add up to an extremely large social cost. Reasonable, conservative estimates of the total cost to society of forgone health are between $250 billion and $3.3 trillion, depending on the assumptions about lifetime insurance status. If health insurance were extended to the currently uninsured population, the average gain in healthier years of life would be between $1,600 and $4,400 per additional year of coverage provided. WHOSE HEALTH LOSS ARE WE MEASURING? This analysis addresses the question of how much health is lost due to a lack of universal insurance coverage. Another way to frame that question is to ask how much health would be gained if we were to suddenly provide coverage to the uninsured. In order to do this, there are several steps that one must take. First, one must identify the population of interest. Second, one must determine the precise intervention to be undertaken. Many individuals transition in and out of insurance, remaining uninsured for only a short spell. Others remain uninsured for a long time. Clearly, the impact of granting universal coverage will have a very different impact on the two groups. Similarly, there is a difference between insuring someone for the rest of their life and giving them insurance for a short period of time, such as a year. Indeed, there are an infinite number of permutations for such an intervention when one considers timing, degree of coverage, cost-sharing arrangements, and so on. Next, we need to determine the differences in the underlying components of health capital that arise from being uninsured. Finally, we must calculate health capital before and after the intervention, and sum the difference over the relevant population.
OCR for page 131
TABLE B.1. Percentage of Population Uninsured by Age Category, 2000 Male Female Under 18 11.6 11.6 18–24 30.1 24.4 25–34 23.9 18.9 35–44 16.9 14.3 45–54 12.0 12.1 55–64 12.0 15.2 SOURCE: Fronstin, 2001b. Identifying the Population The population comprises the uninsured from ages 0 to 64 in 2000 as estimated by the March Current Population Survey (CPS) (Fronstin, 2001b). The March CPS health insurance questions refer back to the previous year. If people are answering the questions correctly, the Fronstin estimates represent people who were uninsured for the whole year.2 The percentage of the population without health insurance by age category is presented in Table B.1. Men are more likely to be uninsured than women until about age 45, at which point women have a higher likelihood of lacking insurance. For both men and women, one is most likely to be uninsured between ages 18 and 24. Nearly 30 percent of men and a quarter of women in this age group did not have health insurance in 2000. I use the estimates from Fronstin to determine the size and age distribution of the population of interest, multiplying the proportion of individuals without insurance at a given age by the sex- and year of age-specific population for 2000 (CPS, 2001). I assume that the age-group probability of being uninsured applies to the midpoint of the age range, and extrapolate linearly for individual years within categories.3 The comparison group of interest is individuals with private health insurance coverage. Determining the Intervention This analysis assumes that the intervention is to provide lifetime health insurance to each uninsured person at a given age. That is, the individual will be covered from his or her current age until age 65, at which time Medicare coverage 2 See Fronstin (2001a, pp. 25–28) for a discussion of the questions and their limitations. 3 An alternative specification would be to assume that the percentage remains constant for individual years within age-group categories. Using this specification does not substantially alter the results; the extrapolation method leads to slightly more conservative estimates.
OCR for page 132
begins. Because not all insurance plans are alike, the actual impact on health is likely to vary with the design of the particular plan. Without the detailed data to construct health capital measures that vary by plan feature, the effective comparison is that uninsured individuals have the “average” private health plan. In order to determine accurately the impact of the intervention of providing health insurance, one needs to know the counterfactual. This is problematic because of great heterogeneity in the duration and frequency of spells without insurance. Some people are uninsured for only a short period of time, and others lack coverage more often than not (IOM, 2001a; Short, 2001). Not surprisingly, the impact on health is greater for those who are uninsured for longer periods of time (Ayanian et al., 2000; Kasper et al., 2000). Although there is a fairly large literature that examines duration and frequency of periods without coverage,4 there is not enough information to map out expected patterns of insurance over a lifetime and incorporate the differential effect of duration of uninsurance on health outcomes.5 Given the practical need to make some generalizing assumptions, I will measure the difference in health between the insured and uninsured under two scenarios. In the first, the counterfactual is that the individual would have otherwise remained uninsured until age 65, at which point Medicare coverage would begin. In the second, I assume that being uninsured in time t is independent of the probability of being uninsured in period t + 1. Thus the counterfactual for the individual receiving the intervention is that his expected health in the year occurring at age a is pr(uninsureda) *Hunins + (1 – pr(uninsureda)) *Hins, where pr(uninsured) is the proportion uninsured at age a. This assumes that the overall rate of insurance at a particular age will remain constant over time. Reality probably lies somewhere in between these two scenarios. Few individuals will actually remain uninsured until age 65 in the absence of the intervention. Many will eventually obtain private or public insurance coverage and realize health benefits from doing so. Therefore this approach will overestimate the impact of providing lifetime insurance, assuming that there are health gains to be made that in fact have already occurred as people transition back into the insured state. On the other hand, insurance status is not a random draw every year. The latter scenario captures the fact that the probability of being without health insurance is not constant across the lifespan. It does not, however, account for any difference in the probability of being uninsured in the future conditional on being uninsured now. It also ignores any residual effects on health of being uninsured in the previous period. If the effect of insurance status were instantaneous and non-persistent, this method would accurately capture the average difference in health capital by insurance status. However, if being uninsured now increases the prob- 4 See Institute of Medicine (2001a) and Short (2001) for summaries of this literature. 5 Even if we did have this information, such a mapping would be a monumental undertaking.
OCR for page 133
ability of being uninsured in the future and the length of uninsured spell has an adverse impact on health outcomes, this scenario will lead to an underestimate of the benefit from providing health insurance. Furthermore, if rates of uninsurance continue on an upward trend, this scenario will provide a conservative estimate of the gains from providing insurance.6 I compute health capital under three sets of assumptions, which will be described. I calculate each set of estimates for the two alternative insurance scenarios. Because one approach will overestimate the health gains from insurance and the other will underestimate these gains, we can use the results from each scenario to bracket the actual effect we would expect to see under each set of assumptions. HEALTH CAPITAL: A FRAMEWORK FOR MEASURING HEALTH We are interested in measuring the gains in health from insuring the uninsured. But “health” has many dimensions, and is a difficult concept to define and measure. One component of health is generally straightforward: Is the person alive? Health has other physical and mental attributes as well. Is the person in pain? Does the person need help caring for herself, or is she unable to work or otherwise function normally? Is the person happy and well adjusted? We need to combine all of these elements into a single measure of health. One approach combines these elements into a measure of quality-adjusted life. In previous work, we employed such a methodology to measure changes in the health of the U.S. population over time (Cutler and Richardson, 1997). Consider H(a) to be an individual’s health at age a. Because health has no natural units, we can scale H however we want. Suppose we scale H from 0 to 1, where 0 is death and 1 is perfect health. Any diseases or impairments the person has will reduce his quality of life so that it falls somewhere in the 0 to 1 range. This definition of H is frequently referred to as a health-related quality of life (HRQL) weight (Gold et al., 1996a). A year in a health state with a particular HRQL weight is referred to as a “quality-adjusted life year,” or QALY (Zeckhauser and Shepard, 1976). Because a dead person has H = 0, expected health is simply Pr[alive at a]*Q(a) where Q(a) is the average quality of life among those who are alive at that age. We then defined health capital as the utility7 resulting from the stock of current and future quality-adjusted life years: 6 Conversely, if rates of uninsurance go down, the gains from insurance estimated under this approach will be biased upward. Given historical patterns, however, it seems unlikely that rates of uninsurance will decline dramatically in the future. 7 Utility is an economic term meaning the satisfaction or benefit that people receive from consuming goods and services.
OCR for page 134
where r is the discount rate and V is the marginal rate of substitution between health and other goods and services. Because health is measured in life years, V is the value of an additional year in perfect health. The bracketed term in the above equation is simply the discounted value of the expected number of QALYs that a person has remaining at age a. Health capital can be thought of as analogous to human capital, a concept frequently used in labor economics. Human capital is the present value of the income one can expect to receive over the course of one’s life as a function of educational attainment. Having a better education allows a person to earn more in the future. Similarly, having more health makes a person happier (and possibly more productive). Health capital is the present value of the utility resulting from a person’s health. For this paper, we are interested in finding the difference between the health capital of an insured person and an uninsured person at age a, and summing these differences over the current population of the uninsured. Note that in theory we want to isolate the difference in average health capital that is caused by lack of insurance, and not differences that arise as a result of different distributions of underlying characteristics, such as gender or income. In other words, we want to hold constant individual and environmental factors, and change only the variables that are affected by insurance coverage. The change in health capital can be decomposed into two terms: the change in the present value of the number of quality-adjusted life years times the dollar value of those life years. We are making two different assumptions about future insurance status without the intervention. In the first case, we are assuming that an uninsured person would have remained in that state until he or she reaches age 65. For a person at age a, the difference in health capital is: In the second case, we are assuming that the uninsured person has a probability less than 1 of being uninsured at any given year in the future. In that case the difference in health coverage from the intervention is:8 8 In our previous work (Cutler and Richardson, 1997), we raised an important concern that can arise when determining changes in the average health capital for a population. If the health of the marginal survivor is worse than average, it will appear that health capital is declining when in fact it is
OCR for page 135
The Discount Rate Which discount rate to use is a longstanding issue in economics. The appro-priate discount rate is the one that trades off utility across different years. Although market interest rates are often very high, these are discount rates for dollars rather than utility. Interest rates will be higher than the discount rate for utility for a number of reasons, including taxes and risk. I use a range of discount rates that are in line with the range considered appropriate in the health literature (Lipscomb et al., 1996): 0 percent, 3 percent, and 6 percent. The benchmark assumption is a discount rate of 3 percent. The Value of a Life Year In addition to measuring years of life, we need to value them in dollars. The value of a human life or a quality-adjusted life year is a subject of much debate.9 Although there is an extensive literature on the value of life, remarkably little work has been done to date on the value of a life year. This is true despite the fact that QALYs are frequently used as an outcome measure in cost-effectiveness analyses of medical technologies and health care programs. One reason QALYs have become so widespread in the literature is almost certainly the appeal of being able to compare options without putting a dollar value on health. The problem with this, of course, is that it does not fully solve the problem of how to allocate resources efficiently (Johannesson and Jonsson, 1991). What is frequently done in evaluation studies is to compare the cost per QALY to commonly used medical technologies and draw a conclusion relative to accepted practice (Mason et al., 1993). One widely used benchmark (Tolley et al., 1994) set forth by Kaplan and Bush (1982) is that a policy is cost-effective if the cost per QALY is less than $34,000, updated to 1999 dollars, which is not always done going up—someone is now alive who formerly was not, so the actual health of every person in the population is the same or better. To address this, we calculated average health capital over the population that is potentially alive, rather than actually alive. This is not an issue here because the focus changes to the health capital of people who are already alive, and we are examining the issue cross-sectionally. For simplicity, I will use the average health capital of those who are actually alive in each group. 9 See Viscusi (1993) for a review.
OCR for page 136
when used as a benchmark (Hirth et al., 2000), and questionable if it is greater than $238,000 (1999 dollars). The area in between was deemed controversial, but acceptable by the standards of currently employed technologies. Another commonly used benchmark of acceptability is $40,000 (Bala and Zarkin, 2000). Of course, using a benchmark value for determining what is “cost-effective” is implicitly assigning a value to a life year, albeit a range of acceptable values. However, these and other benchmarks are quite arbitrary, and there is no evidence that they represent societal willingness to pay for a life year (Hirth et al., 2000; Johannesson and Meltzer, 1998). From a welfare economics perspective, the theoretically correct way to determine whether a policy should be adopted is to compare the total benefit that would accrue to society from the policy with the total cost to society of implementing the policy (Mishan, 1972). If an individual will receive a total benefit of X from the consumption of a good, it follows that he will be willing to pay a maximum of X to obtain that good. Therefore to value the social benefit from health outcomes such as QALYs, one would need to sum each individual’s willingness to pay for the QALYs gained by them and by others (i.e., any external effects that someone else’s improved health might have on a particular individual). With market-produced goods, the consumer’s maximum willingness to pay is the area under the demand curve for the units consumed. It is difficult enough to construct a demand curve when a market exists, but for most health outcomes there is no direct market. Thus the challenge is to determine the maximum willingness to pay for these health outcomes. In this exercise, we need to determine willingness to pay for an individual year of life. One way to determine societal willingness to pay for a life year is to impute it from the value of a life literature. The bulk of this literature uses one of three approaches to measuring the value of life: the human capital method, revealed preference, and contingent valuation. The human capital approach (e.g., Rice and Cooper, 1967) uses the labor market to estimate the value of life. A life is worth the sum of expected future earnings. Although it is relatively easy to measure, principal limitations of this approach are as follows: indirect costs will be zero for individuals not participating in the labor market, such as retirees, homemakers, or children; it fails to consider any disutility from illness over and above forgone earnings; and it does not allow for altruism (Mishan, 1971). Furthermore, people value leisure time as well as time spent in the labor force. Keeler (2001) shows that if the average worker places the same value on leisure time and time spent in the labor force, the value of his life is 5 to 10 times the value of future earnings.10 Thus the human capital method does not capture an individual’s full willingness to pay and is likely to underestimate social benefits. 10 Keeler (2001) also points out that correcting for the value of leisure time brings human capital estimates of the value of a life close to estimates obtained from the revealed preference and contingent valuation methods.
OCR for page 137
An alternative approach is to measure willingness to pay for reduction in health risks through revealed preference. By observing individuals’ behavior in existing markets, we can evaluate willingness to pay indirectly for features such as automobile safety or lower risk of mortality on the job. A great deal of work has been done in examining compensating wage differentials to estimate willingness to pay for lower risk of fatal and nonfatal injury in the labor market (see Viscusi, 1993, for a review). Consumer market studies have also examined implied willingness to pay for reduced morbidity and mortality risk through the purchase of products such as smoke detectors (Dardis, 1980). Because the consumer market studies are limited, much of what is known about the value of a statistical life comes from the labor market literature. However, this is problematic in that it represents a subset of society—primarily workers in blue-collar jobs. How much these results can be generalized to the overall population is unknown. These studies also assume unlimited job mobility, which may not be the case for workers in low-paying jobs (Lanoie et al., 1995). Furthermore, both labor and consumer market studies suffer from possible omitted variable bias due to the inherent difficulty in identifying other job or product amenities that may be highly correlated with lower risk jobs or products (Viscusi, 1993; Gerking et al., 1988). For example, someone who buys a smoke detector is not simply purchasing a reduction in the risk of death, but also a decreased chance of injury, property damage, and psychological harm. All of these elements have value, making it difficult to isolate the actual value of a life. Finally, many health interventions that need to be valued are not traded even in an implicit market (Viscusi, 1993). The approach that is most consistent with the theoretical foundations of costbenefit analysis is to measure willingness to pay through contingent valuation (CV) (Diener et al., 1998; O’Brien and Gafni, 1996). CV is a survey-based methodology for eliciting consumers’ willingness to pay for benefits from a particular policy, usually expressed as a small change in risk. The advantage of this approach is that it directly elicits total willingness to pay for a benefit, which is precisely the theoretically desirable measure. CV studies can be conducted from different perspectives, which determine how the results are interpreted. A societal perspective asks all individuals affected by a policy about their willingness to pay for that policy. This gives the total societal benefit of the policy. On the other hand, asking an individual how much she is willing to pay for a reduction in personal risk provides that individual’s valuation of her life, but theoretically will not capture any impact of this change on others. Developed originally in environmental economics, CV is now a widely accepted tool for assessing the benefits of environmental programs (Mitchell and Carson, 1989). In the health field, CV studies have been conducted to measure a wide range of benefits, including blood donation (Lee et al., 1998; Eastaugh, 1991), arthritis treatment (Thompson et al., 1984; Blumenschein and Johannesson, 1998; Thompson, 1986), in-vitro fertilization (Neumann and Johannesson, 1994), and hypertension therapy (Johannesson and Jonsson, 1991).
OCR for page 138
Although CV provides a direct and theoretically appealing way to measure willingness to pay for a risk reduction, it can be difficult to implement in practice. It suffers from all the standard limitations of survey questions, such as anchoring, framing, and interviewer bias. In addition there may be concern about the validity of people’s responses to questions that have no real implications for their lives. There is also the issue of whose willingness to pay should be considered in an analysis. For example, do we survey only those directly affected or a representative sample of the whole population, because some people may have an altruistic interest in a policy? These issues are real, but much can be done to address them. For example, O’Brien and Gafni (1996) have set forth a valuable conceptual framework for contingent valuation studies in the health field. Among their recommendations are that (1) the entire population (or a representative sample) be surveyed; (2) an ex-ante insurance framework be utilized; and (3) the question be framed as a tax referendum or some other type of compulsory payment scheme. One criticism of both the human capital and contingent valuation methods is that they value individuals with low income or wealth less than those with high income or wealth. Indeed, how one “should” value a life year within a lifetime or across individuals is a controversial and extremely subjective subject. A good illustration of the inherent challenges is the administration of the Victim’s Compensation Fund established for victims of the September 11 attacks. The grand master of the fund, Kenneth Feinberg, was directed by Congress to consider the economic loss to a victim’s family in making awards. Recent articles in the popular press have documented his controversial attempt to balance the lost economic potential of each victim with some degree of equity—for example, by capping awards (Kolbert, 2002; Belkin, 2002). Aside from an individual’s valuation of her own life, there are other reasons why the value of a year of perfect health might vary across people. Society has invested more in some people than in others, and some contribute more back to society than others. Thus, one might vary the value of a life year with the amount that one contributes to society or that society has invested in a person. This is essentially the approach taken in the literature on disability-adjusted life years. In that methodology, it is assumed that society values young adults more than children or older adults, but there is no variation by other factors such as income or the number of dependents an individual is supporting.11 The type of weights to use, let alone the values to employ, are questions about the social welfare function. Because there is not a standard social welfare function, the choice of weights in this context does not have clear theoretical rationale. In this paper, as in previous work, I do not vary the value of a life across people, age, or time. This assumes that society values a healthy year of life the 11 Murray and Lopez (1996). This is based on the empirical observation that many people express a preference for saving the lives or life years of middle-aged people more than the very young or very old. The exclusion of other criteria in valuing people is made on a priori grounds.
OCR for page 139
same for everyone, at any point in their life. Another way to think of this is that it defines the social value of a life year to be the average of all individual values.12 If as a society we value equity, this is a perfectly reasonable assumption. In previous work, we used $100,000 as the value of a life year (Cutler and Richardson, 1997). This value came from a brief synthesis of the literature done by Tolley et al. (1994), which concluded that a range of $70,000 to $175,000 per life year is reasonable. A subsequent study by Hirth and colleagues (2000), however, reviewed the value-of-life literature more thoroughly and with the express purpose of determining the value of a QALY. Hirth and colleagues identified 42 studies that used one of the three valuation methods described earlier and were appropriate for inclusion in their analysis. To estimate the value of a QALY from this literature, they first converted all the values to 1997 dollars and determined the average remaining life expectancy of the sample population. If age was not reported in the study, they assumed an average age of 40, with a sensitivity analysis ranging from 35 to 45. They then applied age-specific HRQL weights from the literature and assumed a 3 percent discount rate (with sensitivity analysis using 0, 5, and 7 percent). Not surprisingly, Hirth and colleagues found that the median value of a life year varied tremendously. Not only were there large differences based on the methodology used, but within methods values also varied greatly. They found that the median value per life year was approximately: $25,000 in the human capital studies; $93,000 in the revealed preference for safety studies; $161,000 in the contingent valuation studies; and $428,000 in the revealed preference for job safety studies. The authors presented their sensitivity analysis in terms of the percentage change to the life year value of the relevant studies; for example, changing the assumption about the average age from 40 to 35 lowered the value per life year in those studies by 7 percent. I recalculated the median value per life year for each methodology, incorporating the high and low ends of the range of estimates they tested for the discount rate and the average age of the population (when it was assumed). These numbers, along with their benchmark median values, are presented in Table B.2. Under the assumption of no discounting and an average study age of 35, the value per life year ranges from $14,000 for the human capital approach to $256,000 for the revealed-preference-for-job-risk method. Under the looser assumptions of 7 percent discounting and an average study age of 45, the 12 One additional problem with the value of a life literature is that market-based values of a life are marginal rather than average. Similarly, if people value their own life years unequally, the average value of a life year will depend on the number of years of life remaining. These are issues that need to be explored further in future research.
OCR for page 159
TABLE B.8 Relationship Between Chronic Conditions and Self-Reported Health Coefficient Standard Error QALY Disease (Conditions included in health capital) Heart disease -0.459 (0.024) 0.93 Stroke -0.577 (0.043) 0.91 Asthma -0.070 (0.034) 0.99 Bronchitis -0.129 (0.031) 0.98 Emphysema -0.439 (0.042) 0.93 Cancer -0.279 (0.022) 0.96 Diabetes -0.630 (0.022) 0.91 Kidney disorder -0.500 (0.041) 0.92 Liver disease -0.536 (0.047) 0.92 Deaf -0.205 (0.067) 0.97 Bad hearing -0.113 (0.015) 0.98 Blind -0.412 (0.066) 0.94 Bad vision -0.172 (0.019) 0.97 Joint pain -0.242 (0.014) 0.96 Head/back/neck pain -0.255 (0.012) 0.96 Interactions (Change in QALY weight) Heart disease*joint pain 0.084 (0.029) 0.01 Joint pain*other pain -0.051 (0.019) -0.01 Heart disease*other pain 0.105 (0.029) 0.02 Poor hearing*poor vision -0.097 (0.032) -0.01 Poor hearing*diabetes 0.092 (0.038) 0.01 Heart disease*stroke 0.186 (0.060) 0.03 Lung disease*diabetes 0.146 (0.044) 0.02 Poor hearing*cancer 0.143 (0.040) 0.02 Diabetes*stroke 0.203 (0.070) 0.03 Cancer*uninsured -0.293 (0.086) -0.04 Diabetes*stroke* uninsured -1.254 (0.495) -0.19 Control conditions Hypertension -0.321 (0.011) 0.95 Other circulatory -0.612 (0.076) 0.91 Other lung disease -0.185 (0.036) 0.97 Other endocrine -0.678 (0.094) 0.90 Nervous system -0.748 (0.047) 0.89 Ulcer -0.190 (0.016) 0.97 Digestive disorders -0.863 (0.081) 0.87 Skin conditions -0.492 (0.221) 0.93 Blood conditions -0.771 (0.194) 0.88
OCR for page 160
Coefficient Standard Error QALY Amputee –0.259 (0.138) 0.96 Injuries –0.496 (0.061) 0.93 Other musculoskeletal –0.565 (0.048) 0.92 Genitourinary disorders –0.709 (0.092) 0.89 Mental health disorders –0.821 (0.038) 0.88 Uninsured (change in QALY weight) –0.075 (0.027) –0.01 Age –0.019 (0.002) Age2 0.00011 (0.00002) Male 0.309 (0.077) Male*Age –0.011 (0.003) Male*Age2 0.00009 (0.00003) White 0.179 (0.020) Black –0.072 (0.023) Hispanic –0.133 (0.015) High school graduate 0.301 (0.013) Some college 0.545 (0.015) College graduate 0.739 (0.020) Income $5–10,000 –0.060 (0.034) Income $10–15,000 –0.007 (0.034) Income $15–20,000 0.059 (0.034) Income $20–25,000 0.078 (0.034) Income $25–35,000 0.144 (0.032) Income $35–45,000 0.194 (0.033) Income $45–55,000 0.259 (0.033) Income >$55,000 0.281 (0.031) Married 0.032 (0.010) Lives in urban area 0.088 (0.012) Lives in urban area*uninsured –0.053 (0.031) Northeast region –0.038 (0.014) Midwest region –0.053 (0.013) South region –0.064 (0.013) N 84,738 Max predicted h 0.985 Min predicted h –5.67 NOTE: Model is an ordered probit of self-reported health, with categories of excellent (5), very good, good, fair, and poor (1).
OCR for page 161
TABLE B.9 Health Capital by Insurance Status, Sex, and Age: Lower-Bound QALY Approach Health Capital Levels and Differences (thousands of dollars) Benefit per Year of Insurance (dollars) (1) (2) (3) Insured Until 65 Uninsured Until 65 Average pr(Ins) Until 65 Difference (1)–(2) Difference (1)–(3) Uninsured Until 65 Average pr(Ins) Until 65 Men 0 4,308 4,265 4,301 42 7 1,490 1,198 18 3,707 3,654 3,698 53 9 2,121 1,536 25 3,446 3,389 3,437 57 9 2,491 1,812 35 3,023 2,959 3,014 64 9 3,304 2,403 45 2,535 2,469 2,527 67 8 4,584 3,119 55 1,996 1,941 1,989 55 7 6,582 3,565 65 1,434 1,434 1,434 Women 0 4,356 4,329 4,352 27 4 941 754 18 3,842 3,811 3,837 31 5 1,208 920 25 3,605 3,570 3,600 34 5 1,481 1,164 35 3,215 3,175 3,209 40 6 2,047 1,624 45 2,757 2,715 2,751 43 6 2,868 2,131 55 2,229 2,193 2,223 36 6 4,198 2,594 65 1,660 1,660 1,660 NOTE: Calculations assume a value of a life year of $160,000 and a real discount rate of 3 percent. Benefit per year of insurance is calculated by dividing the gain in health capital by the discounted years of insurance coverage provided (see Tables B.5A, B.5B).
OCR for page 162
FIGURE B.1 HRQL weights by insurance status, age 0–64. picking up the impact of omitted factors that are correlated with insurance status and health. This could be omitted chronic conditions, acute conditions, or personal characteristics. This negative impact of being uninsured is included in the overall HRQL weight for the uninsured state. As in previous work, I also include age factors for men and women. Figure B.1 shows the overall expected HRQL weight for the insured state and the uninsured state from ages 1 to 64. The insured have a slightly higher HRQL weight in any given year, with a difference of about 0.01 to 0.02. The gap between them narrows throughout childhood and then gradually starts to widen a bit in the early twenties. The QALY Approach: Lower-Bound Estimates Table B.9 shows health capital at certain ages by sex and insurance category. In these estimates, the HRQL weights and disease prevalence are assumed to be
OCR for page 163
the same for the uninsured and the insured. These estimates differ from those in Table B.4 only in that they are adjusted for morbidity. If morbidity were evenly distributed by age, this would only affect the level of health capital and not the relative difference in health capital between the uninsured and the insured. Because the disease burden is not constant over the lifespan, however, the difference is likely to change as well. Because we are incorporating morbidity, the levels of health capital should be lower at every age for every group. Table B.9 shows that they are lower, but still very high. Health capital for an insured infant is about $4.3 million, slightly lower than it was using the years of life method. There are much larger decreases in health capital as one ages, however, compared with the Table B.4 estimates. At age 65, health capital is $1.4 million for men, and $1.7 million for women, quite a bit lower than the YOL method. Again, this is because the burden of disease increases with age. At birth the decrease in health capital once morbidity has been incorporated gets relatively little weight once we discount, whereas the decline in QALYs is more imminent at older ages. Incorporating morbidity under this conservative approach actually decreases the differences in health capital between the insured and the uninsured. For a newborn, the forgone health capital from lacking insurance ranges from $4,000 to $42,000. As before, the difference is greatest in late middle age, reaching $6,000 to $67,000 at age 45. The left-hand columns of Table B.5B show the impact on the health capital differences by insurance of using a range of interest rates. The pattern is exactly like that observed under the YOL approach. With no discounting, the magnitude of the difference declines with age, while the difference increases slightly with age when a higher discount rate is used. The health capital difference between insured and uninsured 45-year-olds is between $3,000 and $36,000, using a 6 percent discount rate, and between $13,000 and $134,000 with a discount rate of zero. The last two columns of Table B.9 show the gain in health capital per year of insurance coverage provided. Because the increases in health capital from insurance are lower with this approach than with the years of life method, the gain per year of insurance will be correspondingly lower as well. A 25-year-old male realizes an improvement in health capital of between $1,800 and $2,500 per incremental year of coverage. A 25-year-old female experiences gains of between $1,200 and $1,500 for each additional year of health insurance provided. A large difference in the health gains for men and women remains. The improvements in health capital are still driven by mortality because morbidity does not vary by insurance status. Health Capital Estimates: Upper-Bound QALY Approach Table B.10 shows the analogous results to Table B.9 when we allow the HRQL weight to differ by insurance status. As expected, the differences in health
OCR for page 164
TABLE B.10 Health Capital by Insurance Status, Sex, and Age: Upper-Bound QALY Approach Health Capital Levels and Differences (thousands of dollars) Benefit per Year of Insurance (dollars) (1) (2) (3) Insured Until 65 Uninsured Until 65 Average pr(Ins) Until 65 Difference (1)–(2) Difference (1)–(3) Uninsured Until 65 Average pr(Ins) Until 65 Men 0 4,308 4,210 4,292 98 16 3,435 2,842 18 3,707 3,602 3,688 105 19 4,239 3,288 25 3,446 3,337 3,428 109 18 4,753 3,646 35 3,023 2,912 3,007 111 15 5,758 4,286 45 2,535 2,433 2,523 102 13 7,012 4,766 55 1,996 1,921 1,987 75 9 8,879 4,806 65 1,434 1,434 1,434 Women 0 4,356 4,276 4,344 79 12 2,759 2,273 18 3,842 3,763 3,828 78 13 3,090 2,482 25 3,605 3,526 3,593 79 12 3,400 2,705 35 3,215 3,135 3,204 81 12 4,099 3,191 45 2,757 2,684 2,747 74 11 4,958 3,610 55 2,229 2,176 2,220 52 8 6,134 3,780 65 1,660 1,660 1,660 NOTE: Calculations assume a value of a life year of $160,000 and a real discount rate of 3 percent. Benefit per year of insurance is calculated by dividing the gain in health capital by the discounted years of insurance coverage provided (see Tables B.5A, B.5B).
OCR for page 165
capital between the insured and the uninsured increase dramatically; they are generally 2 to 3 times larger once morbidity is allowed to vary. For an uninsured newborn, the reduction in health capital relative to an insured newborn ranges from $12,000 to $98,000. At age 35, the difference is between $12,000 and $111,000. The distribution of the difference over the lifespan changes as well. The difference in health capital is now greater at the younger ages and generally declines with age. This reflects the differences in the relative risk of disease by age as well as the fact that the decreased quality of life from being uninsured is felt more heavily at the younger ages. The last four columns of Table B.5B show that, without discounting, the value of health capital is between $40,000 and $354,000 greater for an insured newborn; the difference shrinks to $5,000 to $43,000 with a 6 percent discount rate. Looking at the last two columns of Table B.10, we see some big differences in the health capital benefits per year of insurance under this approach compared with the previous two approaches. First, the gains are much larger, reflecting the improvements in morbidity from obtaining insurance coverage. Now a 25-year-old male has a gain in health capital of between $3,600 and $4,800 per year of coverage, while a 25-year-old female has an increase in health of between $2,700 and $3,400 per year of insurance. Now that morbidity gains are incorporated, the change in women’s health capital is much closer to the change in the health capital of men. To summarize, our benchmark estimates of the reduction in health capital from being uninsured range from $27,000 to $98,000 for a newborn who will remain uninsured until age 65 and from $4,000 to $16,000 for an uninsured newborn who faces the average probability of being uninsured each year. These results are quite sensitive to the discount rate. The overall range of estimates we get by varying the discount rate from 0 to 6 percent widens to $8,000 to $354,000 for a newborn who will remain uninsured until age 65, and to a range of $1,000 to $55,000 for an uninsured newborn who faces the average probability of being uninsured each year. This is obviously an extremely wide range of estimates. Figure B.2 shows the pattern of health capital across the lifespan using the different approaches and scenarios. This simply presents more clearly the patterns that were visible in the tables. If we normalize the gains in health capital from insurance by the number of years of insurance coverage provided, a newborn who will remain uninsured until age 65 sees an increase in health of between $940 and $3,400 per year. An uninsured newborn facing the average probability of being uninsured realizes a gain in health capital of $750 to $2,800 per year of coverage. For a 25-year-old, the corresponding increases per year of insurance range from $1,500 to $4,800 and from $1,200 to $3,600, respectively.
OCR for page 166
FIGURE B.2 Difference in health capital by insurance scenario and age. CONCLUSION: TOTAL COST OF HEALTH FORGONE BY THE UNINSURED The previous section presented a range of estimates for the difference in health capital between the insured and the uninsured at various ages. If we suddenly gave a lifetime of health insurance to the 40 million people who are currently uninsured, how much health would we gain? One key question is how much of the difference in health capital or other measures of health status would disappear and how much would persist. In constructing the health capital measures described earlier, I have tried to focus only on the difference that arises purely from a lack of insurance. The two measures of health capital using the QALY approach should provide an upper and lower bound to this impact. If that is the case, then to estimate the total value of the health that society would gain from insuring the uninsured, we simply multiply the average gain in health capital at age a by the number of uninsured at that age and sum over all ages. Table B.11 presents the results of that exercise. The total value of forgone health
OCR for page 167
TABLE B.11 Total Cost of Health Forgone by Lack of Insurance Assume Otherwise Remain Uninsured Assume Faces Average pr(Unins) 3% discount rate Total cost ($ millions) Health capital, YOL approach 2,087,157 306,097 Health capital, lower-bound QALY approach 1,698,366 250,049 Health capital, upper-bound QALY approach 3,266,083 498,595 Cost per year of insurance ($) Health capital, YOL approach 2,783 2,014 Health capital, lower-bound QALY approach 2,265 1,645 Health capital, upper-bound QALY approach 4,356 3,280 0% discount rate Total cost ($ millions) Health capital, YOL approach 6,269,248 913,831 Health capital, lower-bound QALY approach 5,053,450 738,543 Health capital, upper-bound QALY approach 7,771,782 1,157,662 6% discount rate Total cost ($ millions) Health capital, YOL approach 895,041 141,256 Health capital, lower-bound QALY approach 735,379 109,174 Health capital, upper-bound QALY approach 1,777,432 277,677 NOTE: Calculated by summing the average difference in health capital by age over all uninsured people under age 65 in the United States in 2000. Assumes value of a life year of $160,000. is extremely large. Using the benchmark discount rate of 3 percent, the conservative estimate of the value of this health is between $250 billion and $499 billion. This assumes that each uninsured individual faces the average probability of being uninsured each year in the future. If they otherwise would have remained uninsured until age 65, the value of this health would be between $1.7 and $3.3 trillion. In reality, the value probably falls somewhere in between. This corresponds to an average increase in health capital per year of insurance coverage provided of between $1,600 and $4,400. Using a range of discount rates from 0 to 6 percent, the conservative estimate of the total value of health is between $109 billion and $1.2 trillion. The highest upper-bound estimate is nearly $7.8 trillion. Even using the most conservative measures, the estimates of lost health capital are substantial. There is no doubt that this forgone health imposes tremendous disutility on our society. Furthermore, it is important to remember that measure of health capital only captures the value of an individual’s health to that individual. It
OCR for page 168
does not include any possible spillover effects to others in society from that one person being in better health. There are numerous limitations to a study such as this one. Obviously, great uncertainty is inherent in all the numbers presented. Many implicit and explicit assumptions could shift the numbers in either direction. For example, the results are sensitive to the choice of discount rate, the value of a life year, the magnitude of any mortality reduction, and estimates of the HRQL weight. Also, we cannot be certain how much of the difference in observed health capital can truly be attributed to lack of insurance coverage. There may be omitted variable bias from unobservable differences between the insured and uninsured that are correlated with health outcomes. Even if there were a causal relationship, there is no guarantee that providing coverage will restore health outcomes to a preuninsured state. A related point is that this analysis also assumes that expanding coverage to a large number of people will not have any macroeconomic effects that would influence health outcomes. In reality, a large-scale expansion might have a measurable impact on access to care by insured individuals, incomes of health-sector workers, the productivity of society in general, and any number of other things, all of which might in turn affect population health. Another limitation is the assumption that differences in mortality and morbidity by insurance status are eliminated once someone turns 65 and becomes eligible for Medicare. In fact, individuals who lack insurance suffer morbidity consequences that cannot be rectified immediately. Therefore, providing insurance coverage prior to Medicare is likely to lead to additional gains after age 65 that are not measured in this analysis. This assumption is true for the under-65 population as well. Permanent decreases to health capital as a result of previous spells of uninsurance are not taken into account. Conditional on reaching a certain age, any adverse events prior to that age have no bearing on future health. This is an unrealistic but necessary assumption, given the data limitations. Consequently, the levels of health capital for individuals with a history of being uninsured are overestimated, although it is unclear a priori what effect this will have on changes in health capital. I also make several assumptions about the uninsured population. I extrapolate rates of uninsurance within age categories to obtain uninsurance rates by year of age. The specification in which each individual faces the average probability of being uninsured assumes that rates of uninsurance will not change, when in fact they appear to be on an upward trend. Wherever possible, all these assumptions were made conservatively, that is, to understate the loss in health capital to the uninsured. The estimates of health capital depend heavily on the value assigned to a life year. In this analysis, $160,000 per life year is the value used, which is derived from the literature on stated willingness to pay for a statistical life. This value falls in the middle of the range of estimates considered appropriate in the literature and used by government agencies. However, one feature of this measure of health capital is that it is quite straightforward to revise the value of a life year. One need
OCR for page 169
only divide the health capital estimate of interest by $160,000 and multiply it by a different value for a single life year. Even by conservative estimates, the lost health capital due to lack of health insurance is substantial. Further research is necessary to increase the precision of each component of this analysis. This will better enable us to examine changes in population health from public policies, improvements in medical technology, or any number of health-related interventions.
Representative terms from entire chapter: