Tobacco Documents Online

12 The Costs of Poor Health Habits likely than never smokers to visit a doctor. We included former smok- ers because their heavier use of medical care may be caused by the effects of their previous smoking. The picture was different for hospi- tal care. Smokers and former smokers were hospitalized much more: current smokers had 38 percent, and former smokers 13 percent, more hospitalizations. Smokers also had lower taxes on earnings. Some of these external costs were offset by the reduction in life expectancy, which lowered retirement and disability costs for these individuals. For Heavy Drinking Considering only the same cost components used in the smoking and exercise analyses, we found that the lifetime external cost for a heavy drinker is only $4,600. This value is driven largely by the differ- ences in medical care, sick leave, and taxes on earnings. Most of the costs for medical care and sick Icave result from differences for former drinkers. just as we regarded former cigarette smokers as "ever" smokers in estimating the costs of smoking, we regarded former drinkers as though they were heavy or problem drinkers. Current drinking has little effect on outpatient care. In fact, "ab- stainers" had 13 to 17 percent more outpatient episodes than light drinkers. But former drinkers had 22 percent more outpatient epi- sodes than light drinkers. Former drinkers also had 110 percent more hospital admissions than light drinkers, and significantly more sick leave. Among current drinker, we found no significant difference for light and heavy drinkers on any of these counts. While drinking does not have the drastic effect on life expectancy that smoking has, heavy drinkers tend to retire earlier and thus have even greater reductions in taxes on earnings than smokers do. What accounts for the difference between the $4,600 in external costs for these components and the $42,OW estimate we presented above? By far, the largest external costs of drinking are imposed by loss of innocent lives, property damage, and their concomitant effects on public systems and programs. We estimate that innocent lives lost in alcohol-related traffic accidents alone cost society $24,000 per heavv drinker. Loss of property, strain on the criminal justice system, and social programs add another 814,0of1 per heavy drinker. These costs account for a large part of the difference between the external costs of drinking and those of smoking and lack of exercise. I An Overview 13 For Lack of Exercise The primary external cost factors for sedentary people are higher use of medical care, work loss, and taxes on earnings. We found that lack of exercise had relatively little effect on outpatient medical costs. Moderate exercisers had 12 percent fewer outpatient episodes, but heavy exercisers had only 8 percent fewer, than light exercisers. The story for inpatient care was different, but only for strenuous exercis- ers and diagnoses related to poor health habits. These individuals had about 30 percent lower use rates than people who did not exer- cise. If work loss is any indication of gener:d health, however, exercis- ers clearly benefit (and have lower external costs): moderate exercis- ers had 18 percent, and strenuous exercisers 32 percent, less work loss than did light exercisers. In examining the costs of a sedentary life-style, we contrasted mod- erate and heavy exercisers only with those inactive people who had no constraints on their role or physical activity. Our purpose was to avoid attributing to exercise the adverse effects of health problems that simultaneously raise costs and limit the individual's ability to exercise. f-tow Reasonable Are Our Estimates? We believe they are reasonable, even conservative. The study does have some limitations, which are discussed later and summarized in Chapter 7. When we tested our estimates in analyses that used differ ent assumptions and different data, they fell about midway in the range of estimates generated by these assumptions. The tests support our belief that the estimates presented above are approximately correct. Table 1-3 summarizes the results of our sensitivity analyses, in which we used different data or made dif ferent assumptions than we did in the base-case analyses. For each habit we conducted a separate cost analysis (1) using data from the NHIS fnr all age groups; (2) comparing costs of people who had the habits with costs of people who actually did not (rather than our constructed group of, for exam- ple, nonsmoking smokers); and (3) limiting medical costs to care only for diagnoses possibly related to poor health habits.13 The second type of analysis should give an upper bound on costs because it S6806S480Z

2 The Costs of Poor Health Habits An Overview 3 that ultimately impairs health and shortens life. Indeed, health pro- motion has been a major element of the Surgeon General's agenda for two and a half decades. Belief in the efficacy of good health habits prompted Joseph Cali- fana, at the time Secretary of I lealth and Human Services, to say in the Iey9 Surgeon General's report: "A wealth of scientific research reveals that the key to whether a person will he healthy or sick, live a long life or die prematurely, can be found in several simple personal habits .-. One study found that people who practiced seven of these simple habits lived, on the average, eleven years longer than those who practiced none of them." 1'hese seven habits are not smoking; limiting consumption uf akohe+l; keeping weight within normal lim- its; reducing intake of fats, salt, and sugar; exercising regularly; hav- ing periodic medical checkups that screen for high blood pressure and certain cancers; and observing speed laws and using seat belts. Social concern thus far has focused on how health habits affect the health and well-being of individuals, their families, and the social fabric. T here is another compelling concern, however, that is not often considered-the economic costs that people who have these habits impose on otherc. 'lhe-,e "external costs," their magnitude, and what can be done to lower them are the subjects of this book. 7he habits we focus on are smoking, drinking, and lack of exercise. Why Do External Costs Matter? People with poor health habib+ can impose costs on others in various ways, not all of them financial. But the financial costs of health care are among the most obvious and significant-and the rise in those costs has been a critical public concern for the last two decades. In 1950, spending on health care accounted for 4.4 percent of gross national product (GNP). By L98N, the percentage was 11. l; and pro- jections are for still higher costs in the future. It is true that there is little empirical evidence that people with bad health habits use more health care, or that their increased use of such care is the prime mover behind escalating health expenditures. Indeed, health care costs have risen while smoking has fallen sharply. Nevertheless, poor health habits, including smoking, heavy drink- ing, and lack of exercise, are considered to be among the primary causes of illness and death in the United States. It seems plausible that a healthier nation (or corporation or union) might have to spend less on health care. The increase in medical costs has certainly provided an impetus for government, public and private health insurers, and the employ- ers who pay the premiums, to support programs that encourage bet- ter health habits. According to the Office of'I'echnoingy Assessment (1985), cigarette smoking may account annually for 5.3 million person-years of life lost, $22 billion of medical care costs, and $43 billion in lost productivity. Alcohol abuse may account annually for 22,4H) traffic deaths, 15,400 other deaths, $11.9 billion of medical care costs, and $20.6 billion in lost productivity (U.S. Department of "rransportation, 1986; Luce and Schweitver, 1978). To date there has been no similar research on the costs of sedentary living, but a rea- sonable estimate is that they are about one-tenth of the costs of smoking.'' people with these unhealthy habits, and their families, certainly bear some of the costs directly. 'Phey lose wages, pay a portion of their medical costs, and suffer from disability and premature death. These are what we define as internal costs. 'I'he costs we emphasize in this book are "external costs": that is, what smokers cost nonsmok- ers, what heavy drinkers cost abstainer. or moderate drinkers, and what voluntarily inactive people cost those who exerdse regularly. As we shall see, the existence of extern;d costs is a major reason for government concern about health habit,. Moreover, the magnitude of these external costs can be used to gauge the appropriate level of excise taxes on cigarettes and alcnhol. Collectively financed programs are a leading source of external cost. Such programs may cover some or:dl of medical care costs, sick leave, life insurance, nursing home care, and retirement pensions. Typically, the premiums or taxes for these benefits do not vary ac- cording to a person's health habits (this situation is especially likely in employer group health and retirement plans and public programs). That is, if John Doe and Jane Roe both have health insurance coverage through their employer, their premiums will be the same even though John is, say, a heavy drinker and Jane has only an occasional drink. If John uses more medical care than Jane-because of drinking-related health problems-Jane is in effect subsidizing his heavy drinking through the insurance program. The same is true for other collectively financed costs. At the same time, if his drinking makes John work less and retire 04806S680Z

32 The Costs of Poor I lealth Habits For heavy drinking we also include the costs of accidents and criminal activity. Assumptions Made in Lifetime Cost Analysis Cross-Sertion Used as Cohort Although the cost analyses follow a hypothetical cohort over time, the information on what happens at each age is based on recent (cross-sectional) experience for that age, and not on projections of what life will be like, say, in the year 2050 when those 20 years old in 1990 turn 80. Thus, we did not estimate the effects of secular and biomedical trends on smoking and its health effects. Instead, we used current estimates for parameters such as retirement, medical and nursing home costs, education, and life insurance arrangements. This simplifying assumption is commonly made (for instance, by the Na- tional Center for Health Statistics in computing life expectancy) be- cause the alternative is too difficult and conjectural. Costs fnr Others in Smoker's Household Costs imposed on other family members are difficult lo classify as either internal or external to the smoker. It is not clear whether these costs are taken into account by the smoker in decisions about where and how often to smoke. lf they are, then they should be treated as internal. Although our base-case estimates classify such costs as internal, we show the effect of treating certain costs bome by other family members as extemal.10 Bounding t)tr Ef/ects of Smokiqq Poor general and mental health, physical limitations, and chronic disease significantly increase medical and other health costs. Smok- ing also affects these costs directly. lt may affect them indirectly as well, through its effects on those other risk factors. Because we do not know how much of the difference between smokers and nonsmokers should be attributed to smoking and how much to other risk factors, we tried to hnund the true effects uf smoking. Lower hound. If smoking has little effect on intermediate risk factors such as high blood pressure and weight, then we should control for any differences between smokers and "nonsmoking smokers" in Conceptual Framework 33 such factors. The result is a lower bound on the effeds of smoking because we assume that smoking does not affect these other risk factors. We computed a lower bound in two different ways. In the first method, for mortality, we used the llealth Risk Appraisal (HRA) program (described later in this chapter) to calculate survival rates for nonsmoking smokers, without altering the values for other factors (such as blood pressure and weight) thal smoking might affed. That is, we assumed that nonsmoking smokers would have the same ac- tual values of blood pressure, weight, and the like that the smokers did. For covered medical and work-luss costs, we made a similar assumption, including other habits and health measures in the re- grc:ssion. In the second method, we estimated effects on medical costs only for those diseases probably related to smoking and other poor health habits." Upper lrotmd. If smoking has strong effects on intermediate risk factors, then the hypothetical nonsmoking smokers may exhibit the characteristics of actual never smokers For example, they may have lower blood pressure. In our cost estimation we obtained an upper bound on the effects of smoking by simply comparing actual smokers with actual nonsmokers, without controlling far othe r habits and risk factors. In the I IRA model we used nonsmokers' mortality as Ihe estimate of smokers' mortality if they had never smoked. The bound is an upper bound because people with one poor habit tend to have others, but the comparison attributes all the health and mortality costs to smoking. These two extreme assumptions should bound the true effects of smoking. Underreporting Various studies comparing self-reported consumption with national sales have shown that people tend to underreport bad health habits, smoking among them (Warner, 1978). Our study nsed self-reported data on the amount of smoking to estimate external costs per pack of cigarettes. Consequently, we multiplied reported packs per day in the data by 1.5 to correct for the difference between reported and aciual consumption. People also underreport cnnsumption nf alcohol. Pernanen f1974) noted that self-reported consumptinn nf akohulic beverages accounts for only 40 to 60 percent of alcohol salos. (lur comparison of NHIS SZg0s900Z

6 The Costs of Poor Health Habits TABLE I-L Cumponents of exterasl costs Collectively frnutced costs Medicatcsre Sick leave Group life insurance Nursing twme cam Retirement pension Insurance to cover direct costs Direct costs Motor vehicle accidents (lost lives, propeny damage, other) Criminal justice F'ues Taxes on eamings istered l0 22,148 people (20 percent of the NIIIS sample) by the Na- tional Center for Health Stati,lics. The Current Population Survey, the Centers for Disease Control, and other sources provided supple- mentary information. Our analysis differed from previous studies in several ways that should improve on their estimates. (I) lb estimate custs of medical care, most previous work has relied on imputed differences in uve by people with and without poor health habits. These imputed differences are often judged from differ- ences in mortality for the two groups. Being sick, using medical ser- vices, and dying are correlated, but not perfectly. For example, if John Doe is a smoker and has a quick and deadly heart attack, he will have little medical expense from heart disease. Prorating medical expenses by death rates would give the opposite impression. In con- trast, we based our estimates on observed differences in the actual use of medical care among peopfe with varying health habits. . (2) For people with a specific bad habit, previous work has tended to attribute all the differential costs of medical care, sick leave, and the like to the adverse effects of that habit. "ihere are two problems with this attribution. First, a host of other factors can affect general health and use of medical care. I'eople who have bad health habits may differ in those other factors. (For example, smokers tend to be less well educated.) In addition, bad health habits are often them- selves positively correlated--for example, people who smoke are likely to drink as well. When studies attribute all the differences to An Overview 7 one habit, they probably overestimate its costs. In other words, when they examine each habit in isolation, they effectively double count. To overcome these problems, we controlled for the other characteris- tics of smokers, heavy drinkers, and sedentary people, including their other bad health habits. We could do this because our two main data sources contained details on all three habits. (3) Previous studies have focused primarily on medical costs and sick leave to estimate the costs of habits. They have failed to consider the potential effects that bad health habits have on early retirement and disability, which can be considerable.° Our analysis examined all collectively financed costs. (4) Studies that look at use of all medical .services may incorrectly estimate costs. It seems unlikely, for example, that smoking explains why a smoker is treated for food poisoning and a nonsmoker is not.to To allow for this problem, we tested the sensitivity of our results to alternative assumptions about what categories of treatment are re- lated to each habit. We analyzed differenres in use for all care (exclud- ing maternity and welltare), as well as use for diagnoses that have been linked to the habits. (5) Much of the previous research looks only at current costs. It is true that some costs are immediate, for example, costs of smoking- caused fires and of drunk-driving accidents. For these costs we di- vided the estimated national annual costs by the annual packs of cigarettes or excess ounces of liquor rnnsumed. Smoking, heavy drinking, and lack of exercise also have long"term effects that may result in savings as well as costs. For example, both smokers and nonsmokers are covered by collectively financed programs. Smoking causes or aggravates a number of health problems, and it reduces life expectancy. The practical effect is that although current and former smokers may need more medical care while they are alive, they will collect less in pension benefits than nonsmokers will. If we consider only current costs, former and current smokers will seem to be subsi- dized by nonsmokers. If we factor in the long-term costs, the subsidy drops because smokers in effect subsidize the pensions and nursing home care of people who have never smoked. They do so by paying premiums and taxes to finance pensions and nursing home care, but collecting fewer benefits. (6) If we consider long-term effects, we have to discount future costs, such as pensions, in order to make comparable costs that occur at different times. A dollar paid into a pension plan today, for in- stance, is worth more than a dollar re,rived fifteen years frerm now Z4806S480Z

34 The Costs uf Poor 1 lealth Habits 1953 estimates with alcohol sales (USD1itI.S, 1983a) also yielded a 40 percent figure. Therefore, we multiplied reported alcohol consump- tiun by 2.5. We assumed that underreporting of both smoking and drinking was proportional to consumption, because we had no information that underreporting varies at different consumption levels. We also corrected pension and transfer income figures, because there is evidence that they too are underreported. Respondents to the Current Population Survey underreport pension and transfer in- come by 21 percent (U.S. Department of Commerce, 1986, pp. 165- 170), but negligibly misreport other income. Inflafimr Inflation can cause problems in combining cost estimates from differ- ent years. All components of costs +hould be expressed in dollars of a common year, so that thep can be added together. All estimates in the rest of this book are given in 1986 dollars. When an estimate for a particular cost component was available only for a different year, we assumed costs grew at the rate nf inflation and simply multiplied the estimates by the ratio of the 19t36 CI'1 to the CI'I in the year the costs were e.stimated. All our estimates of costs and taxes are stated in 1986 dollars, as they were in Keeler et al. (1989) and Manning et al. (1989). To convert to lune 199o dollars, multiply by 1.185. 'Phe Components of the Model Mortality The death rates for our abridged life table came from applying esti- mates of the relative risk of smoking to the 1980 National Center for Ilealth Statistics (USDHHS, 1984b) abridged life table of the U.S. population. Our two goals were to make the computed mortality for our sample match the national data for each five-year age group, and to correctly estimate the mortality ratio between smokers (both current and former) and nonsmokers. The relative risk of dying fnr smokers and nonsmokers was derived from the 1984 Centers for tNs- case Control version of the I IRA program. Robbins and Hall (1970) developed the HRA as a health promotion technique for use in a doctor's office as part of a physical examination. Originally designed for manual odculation, the program has been Conceptual Framework 35 updated several times by the Centers I'or Disease Control and is now available in numerous computerized versions. In the clinical setting, the patient usually completes a questionnaire on personal characteris- tics, family history, behaviors, and certain physiologic measure- ments. The individual's risk of dying in the next ten years is calcu- lated, from this inforniation in conjunctinn with national mortality slatistics and data from epidemiologic studies. The next step is esti- mating how much the individual could reduce that risk by modifying his or her behavior. The results are summarized and presented to the patient, accompanied by a list of suf;gested life-style changes that could improve the chances of a longer life. We have adapted this procedure for use in calculating the ten-year probability of dying for each individual in our two data sources (de- scribed in Chapter 3). We modified the input and output sections of the 1984 Centers for Disease Control version of HRA to accommodate our need to process data on thousands of people rather than a small group. Data from the two sources were fed into the HRA program, after translation into the format required by the program. 1'able B-1, in Appendix B, lists the 33 major variables incorporated in the IIRA calculations and summarizes the program's response to a missing value for each variable." The most important components of the risk calculation were the mortality probabilities for each of the leading causes of death by sex, race, and five-year age group. These probabilities were based on mor- tality data for the United States for 1975, 1976, and 1977, obtained from the National Center for Iiealth Statistics. The risk of dying was adjusted up or down from the averaW by applying "risk multipliers" formulated for the person's characteristics, health- related behavior, and physiologic measurements. The magnitude of the risk multiplier was based on information from major prospective epicfemiofogic studies, actuarial experience, and expert opinion regarding the rela- tionship between the risk factor and mortality rate due to a specific cause. The risk multipliers were combined to form a composite risk multi- plier for each major cause of death. (See Appendix B for an example of the method used.) The average probabilily for that cause of death was multiplied by the risk multiplier io produce a risk projection for each individual. 't'he sum of the cause-specific risk projections to- gether with the risk for "other cau.ae,;" yielded the risk of death for all causes. Numerous researchers have critici7eai the methods employed by 9Z8069690Z

56 The Costs of Poor Health Habits Each measure is based on the responses to the HIE medical history questionnaire." All data on health status were collected at the begin- ning of the HIE study and are summarized below. General health {rerceptions. The General Health Index is a continuous score (0-100, 100 being best) based on questionnaire items about perceptions of past, present, and future health. The items also ad- dress resistance to illness and self-concern about health. Twenty-two questions were asked nf individuals 14 and over, and seven of younger children. The index refers to health in general and does not focus on a particular component. The scale is a subjective asses,ment of personal health status. The reliability and validity of the index have been extensively studied and documented (Ware, 1976; Davies and Ware, 1981; Eisen et al., 1980). For example, the impact of hypertension, everything else being equal, is equivalent to 5 index points (Brook et al., 1983). The death rate in the study was 25 in 1,000 for those with index values lower than 63, 6 in 1,000 for those with index values from 63 to 76, and I in 1,0(1D for those with index values from 77 to 100. 1'hy.sical or role lirnitations. This measure is scored dichotomously (= I if limited, = 0 otherwise) to indicate one or more limitations in four categories: self-care (eating, bathing, dressing); mobility (con- fined, or able to use public or private transportation); physical activity (walking, bending, lifting, stooping, climbing stairs, running); and usual role activities (work, home, school)- Data for this measure came from twelve questionnaire items for adults and five items for chil- dren. The reliability and validity of these measures have been studied and documented by Stewart et al. (1977, 1978, 1981a,b), and Eisen et al. (1980). Chronic diseases and complaints. This measure is a simple count of diseases or health problems (from a possible twenty-six) for individu- als aged 14 or older (Mannin'g et al., 1981). The list of conditions includes kidney disease and urinary tract infections, eye problems, bronchitis, hay fever, gum problems, joint problems, diabetes, acne, anemia, heart problems, stomach problems, varicose veins, hemor- rhoids, hearing problems, high blood pressure, hyperthyroidism, and so forth. Mental health status. The Mental Health Inventory for adults is a continuous score (0-100) based on thirty-eight questionnaire items. They measure both psychological distress and psychological well- being, as reflected in anxiety, depression, behavioral and emotional Data and Statistical Methods 57 control, general positive affect, and interpersonal ties. The reliability and validity of this measure have been studied and documented by Veit and Ware (1983), Ware et al. (1979, 1980b), and Williams et al. (1981). A similar construct has been developed for children aged 5 to 13, based on twelve questionnaire items (Eisen et al., 1980). OTHER fOVARIATfiS Our analysis also included covariates for age, sex, race, family in- come, family size, education, and the use of seat belts. With the exception of family income, all these variables were measured before the study or at enrollment." We used measures of education and seat-belt use to reduce any bias that might result from differenres in health attitudes between people who do and do not have poor health habits. These attitudes may affect work loss and use of medical services-independent of smoking, drinking, or lack of exercise. For example, cautious individ- uals may have fewer illnesses becausr they take better care of them- selves. They may also be less likely to smoke. Cbnsequently, we might find a negative association between smoking and use of medi- cal care that reflects caution rather than the effects of smoking. Simi- larly, individuals with more future orientation are more likely to go to college and to get preventive care, and are less likely to smoke. Adding measures of seat-belt use and education should reduce the bias from possible differences in health attitudes among those with different habits. Data from the National Health Interview Survey (NIiIS) Throughout our analyses we used data from the NItIS for the elderly (those aged 60 and older). We also compared the nonelderly NHIS responses on habits with the HrE data. 'the former are from a 1983 supplemental questionnaire on habita that was administered to 20 percent of the NHIS sample (N = 22,418 persons) aged 18 or older. Outcome Meusures We used the responses to questions atxrut the twelve-month physi- cian visit and hospitalization rates as our medical use measures. The two-week work-loss response providrs uur meacure of work loss. 1£8065680Z

84 The Costs of Poor Health Habits Low birthweight is one of the strongest predictors for use of neonatal intensive care units (NICUs). An Office of Technology Assessment report (1987) on neonatal intensive care indicates that between 150,1)00 and 200,000 infants are treated annually in NICUs, and 50 to 80 percent are low-birthweighl babies. The average cost per baby of NICU use is $12,1)DO to $39,00(1. If we use the midpoint of each range, the estimated total cost for neonatal care of low-birthweight babies is $2.9 billion. According to the NIiIS, 35 percent of all women between 20 and 29 years were current smokerv in 1983. If as many as one-third of pregnant women persist in smoking, then smoking may be responsi- ble for as many as one-quartet of all NICU costs for low-birthweight babies. In 1986 Americans purchased 652 billion cigarettes (Statistical Abstract of the United Slates, 19NA, p. 719). Our estimates of the direct dollar costs of smoking, then, are probably too low by up to 2 cents per pack. In addition, the Surgeon Gcneral estimates that 2,500 fetal deaths occur because of smoking during pregnancy. If we were to value these infant deaths at the same $1.66 million that we use for adult deaths, and assume that the smoking mothers ignore the risks to their babies, then the external cost not considered by the smoker is 14 cents per pack. Our estimates for the external cost of fires take into account the likelihood that an innocent bystander will he killed in a smoking- related fire. In 1984 there were 1,6(10 smoking-related deaths from fires in the United States (John Hull, National Fire Protection Associ- ation, and John Ottoson, U.S. Fire Administration, personal commu- nications). Based on a willingness-to-pay for a human life of $1.66 million in 1986 dollars," and the total volume of smoking from the NHIS 1983 survey (corrected for underreporting), we estimate the value of the lost lives in smoking-related fires to be 9 cents per pack of cigarettes. Because virtually all of these deaths are smokers or members of their families, these costs can be considered internal, not external. Summary The major determinants of the external costs of cigarette smoking are medical costs, pensions, and taxes on earnings. Although the results from our two data sets are similar, the estimated magnitude of the The External Costs of Smoking 85 external costs of smoking are quite sensitive to methods, especially to the discount rate used. Because smoking increases these costs among the young and middle-aged but decreases the need for sup- port of the aged, the discount rate is a critical determinant of the net costs of smoking. Without discounting, smoking appears to save nonsmokers money because it reduces the period of aged depen- dency. At reasonable (real) discount rates, smoking appears to cost nonsmokers 15 or more cents per pack. Including all of the other costs discussed, the estimated external costs of smoking range from 31 to 52 cents per pack. 458069680Z

I The External Costs of Smoking Our results provide still more evidence that smoking is a pernicious habit for smokers and for the rest ot society as well. To recapitulate, the I lealth Risk Appraisal model applied to our data shows that it reduces the life expectancy of z 20-year-old by about 4.3 years, or 7 minutes per cigarette. Cnrrent smokers are more likely than former smokers or people who have never smoked to (I) be hospitalized for any reason and (2) visit their doctors for conditions related to smok- ing. Nonsmokers subsidize this higher demand for medical services. Because nonsmokers live longer, however, smokers effectively subsi- dize their pensions. tn fact, if the costs of smoking are not dis- counted, smoking has negative external costs. With a 5 percent discount rate, smoking has nei lifetime external costs of $1,(NAI per smoker. The external cnst per pack nf cigarettes (at a 5 percent dis- counl rate) is 15 cents. This chapter describes the analyses that led to these and other findings on the effects of cmok.ing. II shows the prevalence of smok- ing in the two filE and NI IIS s.imples and smokinR s association with both the use of medical servi:rs and work loss. It presents results adjusted only for age and sex, as well as results controlled for a fuller set of factors that may atfect health nutconies: age, sex, education, family income and size, and other health habits. Finally, it presents our estimates of the external costs uf smoking, that is, the costs im- posed by smokers on others through health insurance, sick leave, retirement, and othercnllectivrly financed programs. Prevalence of Smoking As noted earlier, to establish the prevalence and incidence of smok- ing, we classified the samples into four groups: never smokers, for- The External Costs of Smoking 63 mer cigarette smokers, current cigarette smokers, and pipe or cigar smokers. We then subdivided the groups byage, sex, race, residence, and years of education. HealtH Insurance Fxperinrent Data CIGARETTE SMOKING Pafterns. Table 4-1 shows this breakdown for the fIIE sample.' At the beginning of the HIE, 42 percrnl of people 20 through 59 years of age smoked cigarettes and an additional 17 percent were former smokers. The prevalence of cigarene smoking was highest for per- sons with less than a high school education (52 percent) and men in their thirties (49 percent). It was lowest for individuals with postgrad- uate education (25 percent) and rwomen in their fifties (33 percent). The relationship between smoking and education was particularly striking: the more education, the less likely people were to smoke. The highest prevalence of never smokers for any subgroup was among those with postgraduate education (55 percent). People in nonmetropolitan areas (Franklin County, Massachusetts, and Georgetown County, South Carolina) were also much less likely to smoke than city dwellers were. There was no appreciable racial differ- ence in prevalence of smoking. More blacks reported mever smoking and more nonblacks claimed to have quit. Duralinm Table 4-2 shows the duration of cigarette smoking for current and former smokers. Among fanner smokers, 11 percent re- ported less than two years and about 14 percent reported more than twenty years of smoking. Not surprisingly, duration was higher among current smokers: fewer than 5 percent had smoked less than two years and about 25 percent had smoked longer than twenty years. The two groups had similar percentages of hravy smokers. Because former smokers had not .moked as long or quite as much as current smokers, their pack-year exposure was lowrr.' Sfabifity qf status. Cigarette smoking status in the f1IE was very stable, as Table 4-3 shows. From en rollment to the end of the experi- ment, 94 percent of never smokers, 75 percent of former smokers, and 82 percent of current smokers maintained their status. Overall, 86 percent of persons 20 through 59 years of age did not change their smoking status during the three to five years uf the experiment, and 6 percent changed from smoker at enrollment to former smoker at exit. OV8065 L80Z

22 The Costs of Poor I lealth Habits _ mind-beyond the exigencies of their respective budgets and a ten- dency of states that produce these items to tax their products at very low rates (see Appendix F). At the turn of the centurv, excise taxes were the federal govern- ment's principal source of revenue. In 1902, 36 percent of federal tax revenue came from alcohol taxes and another 10 percent from tobacco taxes.P4 With the advent of income tax, the emphasis on progressive ta,ation, and the increase in social insurance payroll taxes, excise taxes have become much less important (Clark, 1984). Tobacco and alcohol laxes in 1984 each generated about 1 percent of overall tax revenues. Taxes on both substances have rffeclively declined, despite the nrnuntirrg evidence that their cnnsennption causes adverse health ef- fects and imposes considerable external costs. Between 1951 and Irr85, the Consumer I'rice Index quadrupled but the nominal rates (if cigarette and alcohol taxes changed much less. From 1951 to 1983, the federal excise tax on cigarettes stood at 8 cents a pack. In 1983, it was raised to 16 cents, and cvc have seen that as a result of OBI2A 1990 federal taxes have increased another 8 cents per pack. In 1951, the tax represented 42 percent of fhe purchase price; by 1982, only I I percent. It increased to 18 percent in 1983, when the rate doubled, and has fallen off somewhat since then. 'I'he story for alcohol taxes is similar. Between 1951 and 1985, the federal excise tax remained at $10. +0 per proof gallon of spirits but increased to $12.50 in 1985.=` Since 1951, federal excise taxes nn wine and beer have remained constant at 17 cents for a gallon (if wine with an alcohol content of 14 percent or lecs, and 29 cents for a gallon of beer (Distilled Spirits Council of thr United States, 1985). Given these declines in real taxes, it seems evident that taxation has not been applied in anv consistent way to the problem of limiting consumption, much less to the issue of external costs. 14eczuse nomi- nal external cnsts rise with the raie of inflation, we recommend that so-called sin taxes (on tohacco and alcohol) be indexed by inflation to prevent future erosion. When we look across states, the impression of inconsistency is even more pronounced- If taxes did reflect concern about either health or external costs, we would have to conclude that some states worry a lot more than others about the health habits (if their citizens and the consequent external costs. In 1987, state excise taxes ranged from a low of 2 cents per pack in the tobacco-producing state of North Carolina to a high of 38 cents in Minnesota. (Table F-2 in Appendix OZg059V80Z _ - An Overview 23 F shows taxes per pack by state.) Looking at state taxes on alcohol requires breaking them down by kind of beverage.'" In 1985, tax rates on distilled spirits ranged from $1.50 per gallon in Maryland to $6.50 in Florida. Taxes on wine varied from I cent in the wine-producing state of California to $2.25 in Florida. For beer, the tax variation was from 4 cents a gallon in Arkansas to 777 cents in South Carolina. If it were decided to increase cigarette and alcohol taxes to discour- age consumption and/ur reflect external costs, the tax increases should be made at the federal, not the state, level. Minimal variation among states is desirable to prevent bootlegging across state lines and, in the case of alcohol, driving to anuther stale to drink and then driving home. If taxes were raised, how much would be passed on to consumers, and how would they respond?'I'heory and history suggest that pro- ducers will pass all or most of the incre.+se along to consumers. The exact degree of consumer response is uncertain, but the preponder- ance of the literature suggests that con::umers, even Ihose addicted, will reduce their frequency of smoking:md drinking." What Afxatf Inck of f;xercfse? The difOculty with exercise is that there is no obvious way to make people who do not exercise pay their way. Lack of exercise differs from smoking and drinking as sins of omission differ from sins (if commission. From the standpoint of taxes, society can tax people for "wrongful consumption'"-per unit of the substance consumed. It is hard to imagine how they can be taxed for not doing the "right" thing. They could, however, be"rewarded" for doing it. Thus, the way to lower the external costs of sedentary lifestyles is by encouraging and rewarding exercise. Options here are educational efforts (including advertising), benefits to people who exercise, and subsidies to facilities and programs that promote exercise. These might include weliness programs in public and private institutions and corporations, public parks and other facilities that charge no fees or have fees below cost, or rebates on life and group insurance premiums for those who demonstrate that they are fit. Southern Cali- fornia Edison provides an example of the last alternative. The com- pany gives a rebate to cover a portion of employee health insurance premiums to those with good health-as measured by nonsmoking, low body mass index, low cholesterol, and low blood pressure (Mark Kailin, personal communication). w3r

60 The Costs of Pour Health Habits effects of poor health habits on the use of medical services (outpatient visits, outpatient episodes of treatment, and inpatient admissions) and on work loss. We used direct age and sex adjustment to provide simple contrasts for these outcomes, purged of the known association between age, sex, and habits. We augmented these results with estimates based on a negative binomial regression model. We chose the negative binomial tech- nique because of three characteri,tics of the distribution of medical expenses and work loss. First, a Iarge proportion of the participants used no medical services or had no work loss during the year. Sec- ond, the distribution of expenses among users and work loss among workers is strongly skewed. Third, the distribution of medical use is quite different for individuals with only outpatient use than it is for individuals with inpatient use. Accordingly, we examined inpatient and outpatient use separately. Because of these characteristics, techniques like ANOVA (includ- ing direct age and sex adjustmi-nt) and the analysis of covariance (ANOCOVA) yield imprecise though consistent estimates of the ef- fects of health habits on the use of medical services and work loss, even for a sample as large as the NHIS 1983 habits supplement. A model that exploits the characteristics of the distributions of medical expense and work loss vields more precise and robust estimates. We used a model based on the negative binomial distribution to estimate how admissions, outpatient episodes of treatment, and work loss respond to poor health habits. The negative binomial is an appealing distribution because it ran yield a large proportion of zeros and a skewed distribution of positive outcomes. It is also attractive because of its ability to adjust the estimates for different time frames for different individuals--that is, its convolution properties with re- spect to time observed. We have counts on episodes of treatment, admissions, and work loss that cover varying periods of time-from one day to five years. The technique can effectively annualize all of our estimates, while controlling for age, sex, and other confounding variables. See Appendix D for a formal description of the statistical methods. The negative binomial reKression model is more appealing than a Poisson regression because it allows for unmeasured characteristics generating overdispersion, that is, a variance greater than the mean. (Indeed a Poisson regression can be a special case of the negative binomial model.) Kg069WOZ Data and Statistical Methods 61 Correlation in the Responses Although we have several thousand observations, we do not have the information we would get from the same number of independent observations, because of substantial positive correlations in the error terms among family members and uver time among observations on the same person. These correlations exist in all of our outcome measures. Failure to account for them in the analysis yields inefficient estimates of the coefficients and statistically inconsistent estimates of the standard errors. As a result, the inference statistics (/, F, and X') calculated in the usual way (without odjusting for these correlations) can be too large. In the results presented in the rent of the text, we have used a nonparametric approach to correct the inference statistics for this pos- itive intrafamily correlation. The comrtion is similar tn that (or the random effects least-squares model, or equivalently the intracluster correlation model (Searle, 1971). The model is described in Rogers (1983) and Brook et al. (1984), based nn prior work by lluber (1967) on the variance of a robust regression.