August 27, 1970
Page 30198
THE HEALTH COSTS OF AIR POLLUTION
Mr. MUSKIE. Mr. President, last week the Subcommittee on Air and Water Pollution completed action on major amendments to the Clean Air Act. The bill sent to the full committee is a strong piece of legislation. It will, when enacted, have a major impact on the way we live. The subcommittee did not act in haste. It held some 20 days of hearings and 12 executive sessions over the course of the summer.
One of the bill's innovations is the requirement that levels of air quality necessary to protect public health be achieved within a period of time set by the statute. This must be done regardless of cost but costs work both ways.
Air pollution costs the American people billions of dollars in health losses, property damage, increased accidents, and in the deterioration of the quality of life. A paper by Lester Lave and Eugene P. Seskin entitled "Air Pollution and Human Health" appeared recently in the publication of the American Association for the Advancement of Sciences. In that paper the authors calculate the health costs of air pollution to the American people.
Mr. President, the costs presented were made without regard to increased accidents, property damage, psychological effects, esthetic values, or effects on vegetation or wildlife. The paper concludes that 50 percent reduction in air pollution levels in this country would produce a health savings to the American people of some $2.1 billion annually.
The concluding paragraphs from this paper summarize what I consider to be important and relevant information as the Senate prepares to consider air pollution legislation. It demonstrates the need to act immediately to provide protection of the public health:
We estimate the total annual cost that would be saved by a 50 percent reduction in air pollution levels in major urban areas, in terms of decreased morbidity and mortality, to be $2080 million. A more relevant indication of the cost would be the estimate that 4.5 percent of all economic costs associated with morbidity and mortality would be saved by a 50 percent reduction in air pollution in major urban areas. This percentage estimate is a robust figure; it is not sensitive to the exact figures chosen far calculating the economic cost of ill health.
Mr. President, I ask unanimous consent that the paper by Messrs: Lave and Seskin be printed in the RECORD.
There being no objection, the article was ordered to be printed in the RECORD, as. follows:
AIR POLLUTION AND HUMAN HEALTH
(By Lester B. Lave and Eugene P. Seskin)
The quantitative effect, with an estimate of the dollar benefit of pollution abatement, is considered.
Air pollution is a problem of growing importance; public interest seems to have risen faster than the level of pollution in recent years. Presidential messages and news stories have reflected the opinion of scientists and civic leaders that pollution must be abated. This concern has manifested itself in tightened local ordinances (and, more importantly, in increased enforcement of existing ordinances), in federal legislation, and in extensive research to find ways of controlling the emission of pollutants from automobiles and smokestacks. Pollutants are natural constituents of the air. Even without man and his technology, plants, animals, and natural activity would cause some pollution. For example, animals vent carbon dioxide, volcanic action produces sulfur oxides, and wind movement insures that there will be suspended particulates; there is no possibility of removing all pollution from the air. Instead, the problem is one of balancing the need of polluters to vent residuals against the damage suffered by society as a result of the increased pollution (1). To find an optimum level, we must know the marginal costs and marginal benefits associated with abatement. This article is focused on measuring one aspect of the benefit of pollution abatement.
Polluted air affects the health of human beings and of all animals and plants (2). It soils and deteriorates property, impairs various production processes (for example, the widespread use of "clean rooms" is an attempt to reduce contamination from the air), raises the rate of automobile and airline accidents (3), and generally makes living things less comfortable and less happy.
Some of these effects are quite definite and measurable, but most are ill-defined and difficult to measure, even conceptually. Thus, scientists still disagree on the quantitative effect of pollution on animals, plants, and materials. Some estimates of the cost of the soiling and deterioration of property have been made, but the estimates are only a step beyond guesses (4). We conjecture that the major benefit of pollution abatement will be found in a general increase in human happiness or improvement in the "quality of life," rather than in one of the specific, more easily measurable categories. Nonetheless, the "hard" costs are real and at least theoretically measurable.
In this article we report an investigation of the effect of air pollution on human health; we characterize the problem of isolating health effects; we derive quantitative estimates of the effect of air pollution on various diseases and point out reasons for viewing some earlier estimates with caution; we discuss the economic costs of ill health; and we estimate the costs of effects attributed to air pollution.
THE EFFECT OF AIR POLLUTION ON HUMAN HEALTH
In no area of the world is the mean annual level of air pollution high enough to cause continuous acute health problems. Emitted pollutants are diluted in the atmosphere and swept away by winds, except during an inversion; then, for a period that varies from a few hours to a week or more, pollutants are trapped and the dilution process is impeded. When an inversion persists for a week or more, pollution increases substantially, and there is an accompanying increase in the death rate.
Much time has been spent in investigating short-term episodes of air pollution (5). We are more concerned with the long-term effects of growing up in, and living in, a polluted atmosphere. Few scientists would be surprised to find that air pollution is associated with respiratory diseases of many sorts, including lung cancer and emphysema. A number of studies have established a qualitative link between air pollution and ill health.
A qualitative link, however, is of little use. To estimate the benefit of pollution abatement, we must know how the incidence of a disease varies with the level of pollution. The number of studies that allow one to infer a quantitative association is much smaller.
Quantifying the relation.
Our objective is to determine the amount of morbidity and mortality for specific diseases that can be ascribed to air pollution. The state of one's health depends on factors (both present and past) such as inherited characteristics (that cause a predisposition to certain diseases), personal habits such as smoking and exercise, general physical condition, diet (including the amount of pollutants ingested with food), living conditions, urban and occupational air pollution, and water pollution (6, 7). Health is a complex matter, and it is exceedingly difficult to sort out the contributions of the various factors. In trying to determine the contribution of any single factor one must be careful neither to include spurious effects nor to conclude on the basis of a single insignificant correlation that there is no association. Laboratory experimentation is of little help in the sorting process (8).
The model implicit in the studies we have examined is a simple linear equation wherein the mortality or morbidity rate is a linear function of the measured level of pollution and, possibly, of an additional socioeconomic variable. In only a few cases do the investigators go beyond calculating a simple or partial correlation.
A number of criticisms can be leveled at this simple model. No account is taken of possibly important factors such as occupational exposure to air pollution and personal habits. These and other factors influencing health must be uncorrelated with the level of pollution, if the estimated effect of pollution is to be an unbiased estimate. In addition, the linear form of the function is not very plausible, except insofar as one considers it a. linear approximation over a small range.
Both because of the rather crude nature of the studies and because of the statistical estimation, there is a range of uncertainty concerning the quantitative effect of pollution on human health. This range is reflected in the estimate of the benefit of pollution abatement, discussed below.
Epidemiological studies.
Epidemiological data are the kind of health data best adapted to the estimation of air pollution effects. These data are in the form of mortality (or morbidity) rates for a particular group, generally defined geographically (9). For example, an analyst may try to account for variations in the mortality rate among the various census tracts in a city. While these vital statistics are tabulated by the government and so are easily available, there are problems with the accuracy of the classification of the cause of death (since few diagnoses are verified by autopsy and not all physicians take equal care in finding the cause of death). Other problems stem from unmeasured variables such as smoking habits, occupations, occupational exposure to air pollution, and genetic health factors. Whenever a variable is unmeasured, the analyst is implicitly assuming either that it is constant across groups or else that it varies randomly with respect to the level of air pollution. Since there are many unmeasured variables, one should not be surprised to discover that some studies fail to find a significant relationship or that others find a spurious one. For the same reason, one should not expect the quantitative effect to be identical across various groups; even when the relationship in each group is statistically significant.
Sample surveys are a means of gathering a more complete set of data. For example, a retrospective analysis might begin with a sample of people who died from a particular disease. Through questionnaires and interviews, the smoking habits and residence patterns of the deceased can be established. The analysis would then consist of an attempt to find the factors implicated in the death of these individuals. Two types of problems arising from such a study are the proper measurement of variables such as exposure to air pollution (there are many pollutants and many patterns of lifetime exposure) and the possible contributions of variables which still are unobserved, such as occupational hazards, socioeconomic characteristics, and personal habits.
Whatever the source of data, the investigators must rest their cases by concluding that the associations which they find are so strong that it is extremely unlikely that omitted variables could have given rise to the observed correlations; they cannot account for all possible variables.
Episodic relationships.
Another method of investigating the effects of pollution involves an attempt to relate daily or weekly morality (or morbidity) rates to indices of air pollution during the interval in question (10). The conclusions of these studies are of limited interest, for two reasons. First, someone who is killed by an increase in air pollution is likely to be gravely ill. Air pollution is a rather subtle irritant, and it is unlikely that a healthy 25-year-old will succumb to a rise in pollution levels. Our interest should be focused on the initial cause of illness rather than on the factor that is the immediate determinant of death. Thus, morbidity data are more useful than mortality data.
Second (and more important for the morbidity studies), there are many factors that affect the daily morbidity rate or daily rate of employee absences. Absence rates tend to be high on Mondays and Fridays for reasons that have nothing to do with air pollution or illness. One would expect little change in these absence rates if air pollution were reduced. Other factors, such as absence around holidays, give rise to spurious variation; this can be handled by ignoring the periods in question or by gathering enough data so that this spurious variation is averaged away.
Some of these factors (such as high absence rates on Fridays and seasonal absence rates) may be correlated with variations in air pollution and no amount of data or of averaging will separate the effects. We have chosen to disregard the results of these episodic studies, with a few exceptions, cited below.
It is difficult to isolate the pollutants that have the most important effects on health on the basis of the studies we survey here. Measurement techniques have been crude, and there has been a tendency to base concentration figures on a single measurement for a large area. A more important problem is the fact that in most of these studies only a single pollutant was reported.
Discovering which pollutants are most harmful is an important area, where further exploration is necessary. We have tried, nevertheless, to differentiate among pollutants in the survey that follows (11). The problem is complicated, since pollution has increased over time, and since lifetime exposure might bear little relation to currently measured levels. These problems are discussed elsewhere (12).
A REVIEW OF THE LITERATURE
We will proceed with a detailed review of studies made in an attempt to find an association between mortality or morbidity and air pollution indices.
Air pollution and bronchitis.
Studies link morbidity and mortality from bronchitis to air pollution in England (13), the United States (14, 15), Japan (16), and other countries (17). Mortality rates by country boroughs in England and Wales have been correlated with pollution (as measured by the sulfation rate, total concentration of solids in the air, a deposit index, and the density of suspended particulates) and with socioeconomic variables (such as population density and social class). The smoking habits of the individuals studied have also been investigated. The conclusion of these studies is that air pollution accounts for a doubling of the bronchitis mortality rate for urban, as compared to rural, areas.
We took data reported by Stocks (18, 19) and by Ashley (20) and performed a multiple regression analysis, as shown in Table I. We fit the following equation to the data [EQUATION OMITTED] where MR is the mortality rate for a particular disease in country borough i, P1 is a measure of air pollution in that borough, S4 is a measure of socioeconomic status in borough i, and e4 is an error term with a mean of zero. (We also fit other functional forms, as discussed below.) Under general assumptions, the estimated coefficients (ao, al, as) will be best linear, unbiased estimates (21). Only if we want to perform significance tests must we make an assumption about the distribution of the error term (for example, the assumption that it is distributed normally).
The first regression in Table 1 relates the bronchitis mortality rate for men to a deposit index (see Table 1, footnote t), and the population density in each of 53 country boroughs. Thirty-nine percent of the variation in the mortality rate (across boroughs) is "explained" by the regression. It is estimated that a unit increase in the deposit index (1 gram per 100 square meters per month) leads to an increase of 0.18 percent in the bronchitis mortality rate (with population per acre held constant). An increase of 0.1 person per acre in the population density is estimated to lead to an increase of 0.02 percent in the mortality rate (with air pollution held constant). As indicated in Table 1 by the t statistics (the values in parentheses below the estimated coefficients), the air pollution variable is extremely important, whereas the socioeconomic variable contributes nothing to the explanatory power of the regression.
The first ten regressions in Table 1 are an attempt to explain the bronchitis death rate. Four different data sets are used, along with three measures of pollution and two socioeconomic variables. The coefficient of determination, R2 (the proportion of the variation in the mortality rate explained by the regression), ranges from 3 to8. Air pollution is a significant explanatory variable in all cases. In only three cases is the socioeconomic variable significant.
The implication of the first regression is that a 10 percent decrease in the deposit rate (38 g 100 m-8 month -1) would lead to a 7 percent decrease in the bronchitis death rate. Another way of illustrating the effect of air pollution on health is to note that, if all the boroughs were to improve the quality of their air to that enjoyed by the borough having the best air of all those in the sample (a standard deposit rate for all boroughs of 96 g 100 m-2 month -1), the average mortality rate (for this sample) would fall from 129 to 77. Thus, cleaning the air to the level of cleanliness enjoyed by the area with the best air would mean a 40 percent drop in the bronchitis death rate among males. In the fifth regression the pollution index is a smoke index (Table 1, footnote §), and a different set of areas is considered. This is a more successful regression in terms of the percentage of variation explained. As before, the air pollution coefficient is extremely significant, and the implication is that cleaning the air to the level of cleanliness currently enjoyed by the area with the best air (15 mg/100 m3) would lower the average bronchitis mortality rate from 106 to 30, a drop of 70 percent. Results of the other regression analyses based on bronchitis mortality data have similar implications. Note that the effect is almost the same for males and females.
This indicates reliability and suggests that the effect is independent of occupational exposure. Winkelstein et al (14) collected data on 21 areas in and around Buffalo, New York. A cross tabulation of census tracts by income level and pollution level shows that the mortality rate for asthma, bronchitis, and emphysema (in white males 50 to 69 years old) increases by more than 100 percent as pollution rises from level 1 to level 4 (see 22).
These studies indicate a strong relationship between bronchitis mortality and a number of indices of air pollution. We conclude that bronchitis mortality could be reduced by from 25 to 50 percent depending on the particular location and deposit index, by reducing pollution to the lowest level currently prevailing in these regions. For example, if the air in all of Buffalo were made as clean as the air in those parts of the area that have the best air, a reduction of approximately 50 percent in bronchitis mortality would probably result.
Air pollution and lung cancer.
The rate of death from lung cancer has been correlated with several indices of pollution and socioeconomic variables in studies that provided controls for smoking habits and other factors.
For English nonsmokers, Stocks and Campbell (23, 24) found a tenfold difference between the death rates for rural and urban areas. Daly (25), in comparing death rates in urban and rural areas of England and Wales, found the urban rate twice as high. Evidence for other parts of Europe also shows an association between lung cancer and air pollution (26).
Regressions 11 through 16 (Table 1) show our reworking of the data for lung cancer mortality for England and Wales (there is no control for smoking). Regressions 11 through 14 imply that, if the quality of air of all boroughs were improved to that of the borough with the best air, the rate of death from lung cancer would fall by between 11 and 44 percent. Regressions 15 and 16 show a relationship between air pollution and lung cancer which is either insignificant or inverse. The only contrary results come from Ashley's data. In the absence of more complete evidence, we must remain curious about these results. Use of such small samples and inadequate controls is certain to lead to some contrary results, but they are disconcerting when they appear.
In a study of 187,783 white American males (50 to 69 years old), Hammond and Horn (27) reported that the age-standardized rate of death due to lung cancer was 34 (per 100,000) in rural areas as compared to 56 in cities of population over 50,000. When standardized with respect both to smoking habits and to age, the rate was 39 in rural areas and 52 in cities of over 50,000.
Haenszel et al. (28) analyzed 2191 lung cancer deaths among white American males that had occurred in 46 states, and data for a control group consisting of males who died from other causes. They found the crude rate of death from lung cancer to have been 1.56 times as high in the urban areas of their study as in the rural areas in 1958 and 1.82 times as high in the period 1948-49 (in subjects 35 years and older, with adjustments made for age). When adjustments are made for both age and smoking history, the ratio is 1.43. Also the ratio increased with duration of residence in the urban or rural area, from 1.08 for residence of less than 1 year to 2.00 for lifetime residence. Haenszel and Taeuber (29) report similar results for white American females. In a number of additional studies the association between air pollution and lung cancer is examined (30).
Buell and Dunn (31) review the evidence on lung cancer and air pollution; a summary of their findings is given in Table 2. For smokers, death rates (adjusted for age and smoking) ranged from 25 to 123 percent higher in urban areas than in rural areas. For nonsmokers, all differences exceed 120 percent. "The etiological roles for lung cancer of urban living and cigarette smoking
seem each to be complete," they say, "in that the urban factor is evident when viewing nonsmokers exclusively, and the smoking factor is evident when viewing rural dwellers exclusively." They argue that differences in the quality of diagnosis could not account for the observed differences for urban and rural areas.
Nonrespiratory-tract cancers and air pollution.
Our reworking of data from England on rates of death from non-respiratory-tract cancer is presented in Table 1 (regressions 17 through 26). In the regressions, stomach cancer is significantly related to a deposit index and a smoke index. The effects are nearly identical for males and females. Intestinal cancer appears to be only marginally related to indices of either deposit or smoke. For 26 areas in northern England and Wales, there appears to be little relationship between non-respiratory-tract cancers and a smoke index. The single exception in the four regressions occurs for males when the socioeconomic variable is social class; here the smoke index explains a significant amount of the variation in the cancer mortality rate. (Apparently population density and smoke index are so highly related in these 26 areas that neither has significant power to explain such variation).
Winkelstein and Kantor (32) investigated rates of mortality from stomach cancer in Buffalo, New York, and the immediate environs. Their measure of pollution is an index of suspended particulates averaged over a 2-year period. They found the rate of mortality due to stomach cancer to be more than twice as great in areas of high pollution as in areas of low pollution (33).
Hagstrom et al. (34) tabulated rates of death from cancer among white middle class residents of Nashville, Tennessee, between 1949 and 1960, using four measures of air pollution. They found the cancer mortality rate to be 25 percent higher in polluted areas than in areas of relatively clean air (35). They also found significant mortality-rate increases associated with individual categories of cancer, such as stomach cancer, cancer of the esophagus, and cancer of the bladder. The individual mortality rates are more closely related to air pollution after the data are broken down by sex and race.
Levin et al. (36) report, for all types of cancer, these relationships: The age-adjusted cancer incidence rates for urban males was 24 percent higher than that for rural males in New York State (exclusive of New York City) (1949-51), 36 percent higher in Connecticut (1947-51), and 40 percent higher in Iowa (1950) ; the incidence rate for urban females was 14 percent higher than that for rural females in New York State, 28 percent higher in Connecticut, and 34 percent higher in Iowa. For both males and females, the incidence rate for each of 16 categories of cancer was higher in urban than in rural areas.
Cardiovascular disease and air pollution.
Enterline et al. (37) found that mortality from heart disease is higher in central-city counties than in suburban counties, and, in turn, higher in suburban counties than in non-metropolitan counties. Ziedberg et al. (38) found that both morbidity and mortality rates for heart disease are associated with air pollution levels in Nashville. The morbidity rate was about twice as high in areas of polluted air as in areas of clean air. The mortality rate was less closely associated; it was 10 to 20 percent higher in areas of polluted air than in areas of clear air (39).
Friedman (40) correlated the rate of mortality from coronary heart disease in white males aged 45 to 64 with the proportion of this group living in urban areas. The simple correlation for 33 states is 79. When cigarette consumption is held constant, the partial correlation is 67.
On the basis of these studies we conclude that a substantial abatement of air pollution would lead to 10 to 15 percent reduction in the mortality and morbidity rates for heart disease. We caution the reader that the evidence relating cardiovascular disease to air pollution is less comprehensive than that linking bronchitis and lung cancer to air pollution.
Total respiratory disease (41).
Daly (25) found significant correlations between air pollution and death rates for all respiratory diseases (and for non-respiratory diseases as well) in England. Douglas and Waller (42) found significant relationships between air pollution and respiratory disease in 3866 British school children. Fairbairn and Reid (43) found significant correlations between air pollution and morbidity rates (for bronchitis, pneumonia, pulmonary tuberculosis, and lung cancer) in England. Regressions 27 through 3.0 in Table 1 show pneumonia mortality to be related only marginally to a smoke index.
Zildberg et al. (44) questioned 9313 Nashville residents about recent illnesses. Among males aged 55 and older from white middle-class families, the numbers of illnesses per respondent during the past year were 1.92, 1.15, and 1.26 for areas of high, moderate, and low pollution, respectively. There are a number of other comparisons, based on other measures of air pollution and on data for females and nonwhites (some of these are given in 45). However, we should add a word of caution: although the sample size in this study was large enough and controls for many socioeconomic variables were included, many important factors were ignored – for example, smoking habits and length of residence. Nonetheless, the finding is extremely strong and seems unlikely to be an artifact of unmeasured variables.
Hammond (46) studied over 50,000 men to find the relationships between emphysema, age, occupational exposure to pollution, urban exposure, and smoking; His results indicated that the effect of air pollution is significant and that heavy smokers have a much higher morbidity rate in cities than in rural areas; the effect becomes more marked as age increases.
Ishikawa et al. (47) estimated the incidence of emphysema in Winnipeg (Canada) and St. Louis. They examined the lungs of 300 corpses in each city (the samples were comparable). Findings for each age group (over 25 years old) indicated that the incidence and severity of emphysema is higher in St. Louis, the city with the more polluted air. (In the 45-year-old group 5 percent of
those in Winnipeg and 46 percent of those in St. Louis Showed evidence of emphysema.)
A number of studies have been made in England on homogeneous occupational 3.27 groups, such as postmen. The results are relatively pure in that all members of the 3.80 sample have comparable incomes, working conditions, and social status. Holland and Reid (48) found that the rates of occurrence of severe respiratory symptoms were 25 to 50 percent higher for London postmen than for small-town postmen (Sample Size, 770). Reid (49) found that in the postmen of his study, absences due to bronchitis rose from an index number of 100 for the area of least air pollution, to 120 for an area of moderate pollution, to 250 and 283 for the areas of highest pollution. Corresponding figures for absences due to other respiratory illness were 100, 100, 150, and 151, respectively, and for absences due to infectious and parasitic diseases, 100, 115, 130, and 140. Cornwall and Raffle (50) made a similar study of bus drivers in London. They found that 20 to 35 percent of absences due to sickness of any kind could be ascribed to air pollution (they used a fog index as a measure of pollution). Fairbairn and (Reid (43) tabulated absences due to sickness of pollution). Fairbairn and Reid (43) tabulated absences due to sickness for postmen, for males working indoors, and for females working indoors. They found that the age-standardized morbidity rate for bronchitis and pneumonia in the postmen of their study rose from 40 man-years per 100 man-years for the area of lowest air pollution (of the four areas Studied) to 122 for the area of highest air pollution. Corresponding figures for morbidity from colds were 75 and 171 man-years, and for morbidity from influenza, 131 and 184 man-years. For males working indoors, the low and high morbidity rates were as follows: bronchitis and pneumonia, 32 and 39; colds, 53 and 64; influenza, 88 and 102.
Doban (fifty-one) studied absences (of more than 7 days) of female employees in eight Radio Corporation of America plants. He found a correlation of 96 between atmospheric concentrations of SO, and absences due to respiratory disease in the five cities for which complete data were available. During Asian flu epidemics there was a 200 percent increase in illness in cities with polluted air and only a 20 percent increase in those with relatively unpolluted air.
Infant mortality and total mortality rates.
Sprague and Hagstrom (fifty-two) compared air-pollution date for Nashville with fetal and infant mortality rates for Nashville as given in census tracts (for 1955 through 1960). Controls for socioeconomic factors were not included. For infant death rates (ages 28 days to 11 months), the highest correlation was with atmospheric concentrations of SO (in milligrams per 100 square centimeters per day) and was70. For the neonatal death rates (ages 1 day to 27 days), the highest correlation was with dustfall and was 49. For infants dying during their first day whose death certificate includes mention of immaturity, the highest correlation was with dustfall and was 45. The correlation of the fetal death rate with dustfall was 58.
In a study just being completed (fifty-three), we have collected data for 114 Standard Metropolitan Statistical Areas in the United States and have attempted to relate total death rates and infant mortality rates to air pollution and other factors. Socioeconomic data, death rates, and air pollution data were taken from U.S. government publications (fifty-four). Regression 1 (Table 3) shows how the total death rate in 1960 varies with air pollution levels and with socioeconomic factors. As the (biweekly) minimum level of suspended particulates increases, the death rate rises significantly. Moreover, the death rate increases with (i) the density of population of the area, (ii) the proportion of nonwhites, (iii) the proportion of people over age 65, and (iv) the proportion of poor families. Eighty percent of the variation in the death rate across these 114 statistical areas is explained by the regression.
Regression 3 shows how the 1960 infant death rate (age less than 1 year) varies. A smaller proportion (55 percent) of the variation in the death rate is explained by the regression, although the minimum air pollution level, the percentage of nonwhites, and the proportion of poor families continue to be significant explanatory variables. Regression 5 is an attempt to explain variation in the neonatal death rate. The results are quite similar to those of regression 3. The fetal death rate is examined in regression 7. Here the minimum air-pollution level, population density, the percentage of nonwhites, and the percentage of poor families are all significant explanatory variables.
Regressions 2, 4, 6, and 8 are an attempt to relate these death rates to the atmospheric concentrations of sulfates for the 114 statistical areas of the Study. Regression 2 shows that the total death rate is significantly related to the minimum level of sulfate pollution, to population density, and to the percentage of people over age 65; 81 percent of the variation is explained.
Regressions 4, 6, and 8 show that the minimum atmospheric concentration of sulfates is a significant explanatory variable in three categories of infant death rates.
One might put these results in perspective by noting estimates on how small decreases in the air-pollution level affect the various death rates. A 10 percent decrease in the minimum concentration of measured particulates would decrease the total death rate by 0.5 percent, the infant death rate by 0.7 percent, the neonatal death rate by 0.6 percent, and the fetal death rate by 0.9 percent. Note that a 10 percent decrease in the percentage of poor families would decrease the total death rate by 0.2 percent and the fetal death rate by 2 percent. A 10 percent decrease in the minimum concentration of sulfates would decrease the total death rate by 0.4 percent, the infant mortality rate by 0.3 percent, the neonatal death rate by 0.4 percent, and the fetal death rate by 0.5 percent.
Each of the relations in Tables 1 and 3 was estimated in alternative ways, including transformation into logarithms, a general quadratic, and a "piecewise" linear form as documented elsewhere (12). The implications about the roles of air pollution and of the socioeconomic variables were unchanged by use of the different functional forms. Another result to be stressed is that, in Table 1, comparable regressions for males and females show almost precisely the same effects for air pollution. This suggests that occupational exposure does not affect these results; the result lends credence to the estimates. A result that we document elsewhere (53) is that it is the minimum level of air pollution that is important, not the occasional peaks. People dealing with this problem should worry about abating air pollution at all times, instead of confining their concern to increased pollution during inversions.
SOME CAVEATS
In preceding sections we have described a number of studies which quantify the relationship between air pollution and both morbidity and mortality. Is the evidence conclusive? Is it possible for a reasonable man still to object that there is no evidence of a substantial quantitative association? We believe that there is conclusive evidence of such association (55).
In the studies discussed, a number of countries are considered, and differences in morbidity and mortality rates among different geographical areas, among people within an occupational group, and among children are examined. Various methods are used; ranging from individual medical examinations and interviews to questionnaires and tabulations of existing data. While individual studies may be attacked on the grounds that none manages to provide controls for all causes of ill health, the number of studies and the variety of approaches are persuasive. It is difficult to imagine how factors such as general habits, inherited characteristics, and lifetime exercise patterns could be taken into account.
To discredit the results, a critic would have to argue that the relationships found by the investigators are spurious because the level of air pollution is correlated with a third factor, which is the "real" cause of ill health. For example, many studies do not take into account smoking habits, occupational exposure, and the general pace of life. Perhaps city dwellers smoke more, get less exercise, tend to be more overweight, and generally live a more strained, tense life than rural dwellers. If so, morbidity and mortality rates would be higher for city dwellers, yet air pollution would be irrelevant. This explanation cannot account for the relationships found.
Apparently there is little systemic relationship between relevant "third" factors and the level of air pollution. An English study (19) in which smoking habits are examined reveals little evidence of differences by residence. There is evidence in the United States that smoking is more prevalent among lower socioeconomic groups (56) but income or other socioeconomic variables would account for this effect and still leave the pollution coefficient unbiased. More importantly, the correlations between air pollution and mortality are better when one is comparing areas within a city (where more factors are held constant) than when one is comparing rural and urban areas (57). It is especially hard to believe that the apparent relation between air pollution and ill health is spurious when significant effects are found in studies comparing individuals within strictly defined occupational groups, such as postmen or bus drivers (where incomes and working conditions are comparable and unmeasured habits are likely to be similar).
When there are uncontrolled factors, some studies may show inconclusive or even negative results; only by collecting samples large enough to "average away" spurious effects can dependable results be guaranteed. In the main, each of the studies cited above was based on a substantial sample. It is the body of studies as a whole that we find persuasive.
An examination of contrary results.
Uncontrolled factors, together with small samples, are certain to lead to some results contrary to the weight of evidence and to our expectations. For example, in some studies (58) no attempt is made to control even for income or social status. From the evidence of studies which did provide such controls, we know that failure to control for income leads to biased results, and so we place little credence in either the positive or negative findings of studies lacking these controls.
Sampling error can be extremely important. For example, Zeidberg et al. (59) find mixed results in cross-tabulating respiratory disease mortality with level of air pollution and with income class. In general the relationships are in the expected direction, but they are often insignificant.
Insignificant results might occur often, if the samples are small, even if air pollution is extremely significant, since sampling errors dominate the explanatory variables.
Another study in which sampling error is important is reported by Ferris and Whittenberger (7). They compared individuals in Berlin, New Hampshire, with residents of Chilliwack (British Columbia), Canada, and – not surprisingly in view of the small samples – failed to find significant differences in the occurrence of respiratory disease. Prindle et al. (60) compared two Pennsylvania towns in the same fashion. These two studies are admirable in that individuals were subjected to careful medical examinations. However, only a few hundred individuals were studied, and this means that sampling errors tend to obscure the effects of air pollution.
Moreover, there were no controls for other factors, such as smoking. Also, one must be careful to control for a host of other variables if the sample is small. For example, the ethnic origins of the population and their general habits and occupations are known to affect mortality rate. It is exceedingly difficult to control for these factors; use of carefully constructed large samples seems the best answer. Finally, air pollution is measured currently, and it is generally assumed that relative levels have been constant over time and that people have lived at their present addresses for a long period. It is hardly surprising that statistical significance is not always obtained when such assumptions are necessary.
Since investigators are more reluctant to publish negative results than positive ones, and since it is more difficult to get negative results published, it is probable that we are unaware of other studies that fail to find a strong association between air pollution and ill health. We are somewhat reluctant to come to strong conclusions without knowledge of such negative results. However, there seems to be no reasonable alternative to evaluating the evidence at hand and allowing for uncertainty. Thus, we conclude that an objective observer would have to agree that there is an important association between air pollution and various morbidity and mortality rates.
THE ECONOMIC COSTS OF DISEASE
Having found a quantitative association between air pollution and both morbidity and mortality, the next question is that of translating the increased sickness and death into dollar units. The relevant question is,
How much is society willing to spend to improve health (to lower the incidence of disease)? In other words, how much is it worth to society to relieve painful symptoms, increase the level of comfort of suffers, prevent disability, and prolong life? It has become common practice to estimate what society is willing to pay by totaling the amount that is spent on medical care and the value of earnings "forgone" as a result of the disability or death (61). This cost seems a vast underestimate for the United States in the late 1960's. Society seems willing to spend substantial sums to prolong life or relieve pain. For example, someone with kidney failure can be kept alive by renal dialysis at a cost of $15,000 to $25,000 per year; this sum is substantially in excess of forgone earnings, but today many kidney patients receive this treatment. Another example is leukemia in children; enormous sums are spent to prolong life for a few months, with no economic benefit to society. If ways could be found to keep patients with chronic bronchitis alive and active longer, it seems likely that people would be willing to spend sums substantially greater than the foregone earnings of those helped. So far as preventing disease is concerned, society is willing to spend consider able sums for public health programs such as chest x-rays, inoculation, fluoridation, pure water, and garbage disposal and for private health care programs such as annual ''physical checkups.”
While we believe that the value of earnings forgone as a result of morbidity and mortality provides a gross underestimate of the amount society is willing to pay to lessen pain and premature death caused by disease, we have no other way of deriving numerical estimates of the dollar value of air pollution abatement. Thus, we proceed with a conventional benefit calculation, using these forgone earnings despite our reservations.
Direct and indirect costs.
Our figures for the cost of disease are based on Estimating the Cost of Illness, by Dorothy P. Rice (61). Unfortunately, Rice calculated disease costs in quite aggregate terms, and so the category "diseases of the respiratory system" must be broken down. It seems reasonable to assume that both direct and indirect costs would be proportional to the period of hospitalization (total patient-days in hospitals) by disease category (62).
Rice defines a category of direct disease costs as including expenditures for hospital and nursing home care and for services of physicians, dentists, and members of other health professions.
"Other direct costs" (which would add about 50 percent to those just enumerated) consist of a variety of personal and nonpersonal expenditures (such as drugs, eyeglasses, and appliances), school health services, industrial in-plant health services, medical activities in federal units other than hospitals, medical research, construction of medical facilities, government public health activities, administrative expenditures of voluntary health agencies, and the net cost of insurance.
Since Rice does not allocate "other direct costs" among diseases, we omit it from our cost estimates. However, we conjecture that respiratory diseases represent a substantial portion of this category. Thus, our direct cost estimate is likely to be a substantial underestimate of "true" direct costs (probably more than 50 percent too low).
Estimating indirect cost is an attempt to measure the losses to the nation's economy caused by illness, disability, and premature death. We would argue that such a calculation gives a lower bound for the amount people would be willing to pay to lower the morbidity and mortality rates. These costs are calculated in terms of the earnings forgone by those who are sick, disabled, or prematurely dead (63).
THE HEALTH COST OF AIR POLLUTION
The studies cited earlier in this article show a close association between air pollution and ill health. The evidence is extremely good for some diseases (such as bronchitis and lung cancer) and only suggestive for others (such as cardiovascular disease and non-respiratory-tract cancers.)
Not all factors have been taken into account, but we argue that an unbiased observer would have to concede the association. More effort can and should be spent on refining the estimates.
However, the point of this exercise is to estimate the health cost of air pollution. We believe that the evidence is sufficiently complete to allow us to infer, roughly, the quantitative associations. We do so with caution, and proceed to translate the effects into dollars. We have attempted to choose our point estimates from the conservative end of the range.
We interpret the studies cited as indicating that mortality from bronchitis would be reduced by about 50 percent if air pollution were lowered to levels currently prevailing in urban areas with relatively clean air. We therefore make the assumption that there would be a 25 to 50 percent reduction in morbidity and mortality, due to bronchitis if air pollution in the major urban areas were abated by about 50 percent. Since the cost of bronchitis (in terms of forgone income and current medical expenditures) is $930 million per, year, we conclude that from $250 million to $500 million per year would be saved by a 50 percent abatement of air pollution in the major urban areas.
Approximately 25 percent of mortality from lung cancer can be saved by a 50 percent reduction in air pollution, according to the studies cited above. This amounts to an annual cost of about $33 million.
The studies document a strong relationship between all respiratory disease and air pollution. It seems likely that 25 percent of all morbidity and mortality due to respiratory disease could be saved by a 50 percent abatement in air pollution levels. Since the annual cost of respiratory disease is $4887 million, the amount saved by a 50 percent reduction in air pollution in major urban areas would be $1222 million.
There is evidence that over 20 percent of cardiovascular morbidity and about 20 percent of cardiovascular mortality could be saved if air pollution were reduced by 50 percent. We have chosen to put this saving at only 10 percent – that is, $468 million per year.
Finally, there is a good deal of evidence connecting all mortality from cancer with air pollution. It is difficult to arrive at a single figure, but we have estimated that 15 percent of the cost of cancer would be saved by a 50 percent reduction in air pollution – a total of $390 million per year.
Not all of these cost estimates are equally certain. The connection between bronchitis or lung cancer and air pollution is much better documented than the connection between all cancers or all cardiovascular disease and air pollution. The reader may aggregate the costs as he chooses.
We estimate the total amount annual cost that would be saved by a 50 percent reduction in air pollution levels in major urban areas, in terms of decreased morbidity and mortality, to be $2080 million. A more relevant indication of the cost would be the estimate that 4.5 percent of all economic costs associated with morbidity and mortality would be saved by a 50 percent reduction in air pollution in major urban areas (64). This percentage estimate is a robust figure; it is not sensitive to the exact figures chosen for calculating the economic cost of ill health.
A final point is that these dollar figures are surely underestimates of the relevant costs. The relevant measure is what people would be willing to pay to reduce morbidity and mortality (for example, to reduce lung cancer by 25 percent). It seems evident that the value used for forgone earnings is a gross underestimate of the actual amount. An additional argument is that many health effects have not been considered in arriving at these costs. For example, relatively low levels of carbon monoxide can affect the central nervous system sufficiently to reduce work efficiency and increase the accident rate (65). Psychological and esthetic effects are likely to be important, and additional costs associated with the effect of air pollution on vegetation, cleanliness, and the deterioration of materials have not been included in these estimates (66).