93s fO POLICY RESEARCH WORKING PAPER 1301 ..How-does oneassess the.. Estimating the Health health benefits oF iir polltion Effects of Air Pollutants ... nr6lDoesresponse functions applied to data an . , , l~~~~~~~~~~~-aicaria reveal t:h-iiai air qai A Method with an Application ia reatair.quaIe imp'rav'rnents mAi'm-ll redu to Jakarta illziin prernue death dn : l learning disabi!ides in Bart Ostro children. Lead and respirable partides are thenmost important problems. The World Bank Policy Fesearch Departmrnt Public Economiics Division May 1994 POLICY RE3SEARCH WORKING PAPERt 1301 Summary findings To develop efficient strategies for pollution control, it is indoor air pollutants. For example, if annual essential to assess both the costs of control and the concentrations of particulate matter were reduced to the benefits that may result. These benefits will often include midpoint of the World Health Organization guideline improvements in public health, including reductions in (and former U.S ambient standard), the estimates both morbidity and premature mortality. indicate a reduction per year of 1,400 prematurc deaths Until recently, there has been little guidance about (with a range of 900 to 1,900), 49,uOO emergency room how to calculate the benefits of air pollution controls visits, 600,000 asthma attacks, 7.6 million restricted- and how to use those estimates to assign priorities to activity days (including work loss), 124,000 cases of differenr air pollution control strategies. Ostro describes bronchitis in children, and 37 million minor respiratory a method for quantifying the benefits of reduced ambient symptoms. concentrations of pollutants (such as ozone and In the case of Jakarta, the methodology suggests that parriculare matter) typically found in urban areas reducing exposure to lead and nitrogen dioxide should worldwide. He -en applies the method to data on also be a high priority. Jakara, Indonesia, an arca characterized by little wind, An important consequence of ambient lead pollution is hibih population density (8 million people), congested a reduction in learning abilities for children, measured as roads, and ambient air pollution. I.Q. loss. Apart from that, reducing the proportion of The magnitude of the benefits of pollution control respirable particles can reduce the amount of illness and depend on the level of air pollution, the expected effects premature mortality. on health of the pollutants (dose-rcsponsc), the size of Clearly, air pollution represents a significant public L¶ie population affected, and the economic value of these health hazard to residents of Jakarta and other cities effects. consistently exposed to high levels of air pollution, such The results for Jakarta suggest that significant benefits as Bangkok, Mexico City, and Santiago, Chile. reslt from reducing exposure to both outdoor andc Thispaper-a product of the PublicEconomicsDivision, Policy Research Department-is part of a larger cffortin thedepartmnent to analyze the economics of pollution control in developingcountrics. Thestudywasbeingfunded by the Banks Rcsearch Support Budget under the rcsearch projccmtPoilution and the Choicc of Economic Policy Instruments in Developing Countrics" (RPO 676- 48). Copics of the paper are availablc free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Carlina Jones, room NIO-063, extension 37699 (63 pagcs). May 1994. The Poliy Researc Wrriung Paper Series disseminates the firdings of work in pgess so encouage the exJ*Chge of ideas about devlopment issues. An obi ctivtofse series is to get the finding out quickly. evn if thepresentaions are ssthanf(idypoisbe7he papers cary the names of the authors and should be used and cited accordingly- The inbigs, inteprezs ions, and cAnclusions are the asrbiors'om and sbould not be attributed to the World Bank. its Executive Board of Direcao o any ofits member mouries Produced by the Policy Research Disscmination Centcr Estimating the Health Effects of Air Pollutants A Method with an Application to Jakarta by Bart Ostro, PhJD.* *Dr. Ostro is Chief, Air Pollution Epidemiology Unit. Office of Environmental Health Hazard Assessment California Environmental Protection Agency. This study was undertaken in his capacity as an independent consultant and has not been reviewed or endorsed by the California Environmental Protection Agency. The author expresses his appreciation of Gunnar Eskeland for his oversight and technical assistance in this project This research was undertaken as part of PRDPE'B Research Project 676-48 in association with Bank sector work (Studies 11871 IND and 12083 MID) and with additional fimding from LATEN. Table of Contents INTRODUCTION ....................................... . 1 H. METHODOLOGY AND BASELINE ASSUMFnIONS FOR ESTIMATING -EALTH EFFECTS ................ 3 A. Methodology .......................................... 3 B. Baseline Assumpons ..... 6 m. REVIEW OF HEALTH EFFECTS AND PROVISION OF DOSE-RESPONSE FUNCTIONS ..... 8 A. Selection Criteria .... 8 B. Development of Dose-R oeEstimates ....................... 10 1. Particulate Matter ..10 - Premature Mortality ...................... .......... 1 - Respiratory Hospita Admissions .......................... 16 - Emergency Room Visits .--------------- 17 - Restricted Activity Days .17 - Lower Respiratory Illness in Children.. 18 - Asthma Attacks ...... ............................... 20 - Respiratory Symptoms .---------------.-..-...-.-21 - Chronic Bronchitis .... ............................... 22 - Other (Non-quantified) Health Effects .22 2. Sulfur Dioxide . ------------------ 23 - Premature Mortality .... 24 - Respiratory Symptoms ................................. 26 3. Ozone .-------------------28 - Respiratory Hospital Admissions ........................... 29 - Restrictions in Activity ................................. 29 - Respiratory Symptoms ............... 31 -EyeIrritation .............. 31 - Asthma Exacerbation .............. 32 U - 4. Lead ............................................. 33 - Effeet of Lead on Blood Pressure in Adults ................... 34 - Hypertenion .................. 35 - Pemature Mortality .................................. 37 -Effcts of Blood Lead on Neurodevelopment in Children ............. 38 5. Nitrogen Diadde . ..................................... 40 6. Carbon MonOide .............................. 41 7. Carbon Dioide .42 IV. APPLUCATION TO JAKARTA, INDONESIA ....................... 43 A. Badcroud Information .................. 43 B. Fstimated Health Effects.47 V. IMPROVING THE ESTIMATES OF AIR POLLUTION DAMAGES UNCERTAINTIES AND FUTRE NEEDS ------------------ .. As48 REFERENCES -54 I. INTRODUCTION Until recently, there has been litde guidance about the calculation of the social costs of air pollution and about using these costs to evaluate altemative air pollution control strategies. With limited resources, rational decision-makdng requires some quantification of the potential benefits of controlling air pollution. These benefits are dependent on the expected health effects of the pollutant. the magnitude of the effect in response to air pollution (dose-response), the economic valuation of the adverse effect, and the existence of subpopulations particularly sensitive to air pollution. Information about health and economic effects of air pollution needs to be categorized for pollutants commonly discharged by mobile and stationary sources. This report descnrbes a method for determining quantitative estimates of the benefits of reducing ambient concentraions of five pollutants: particulate matter, sulfur dioxide, nitrogen dioxide, ozone and lead. This methodology is then applied to Jakara. Indonesia. A brief review of the effects of carbon monoxide and carbon dioxide is also provided. Once the benefits (both quantified and unquantified) of control are calculated, they can be incorporated in decisiorLs about prioritizing contol sategies. For cost-benefit analysis of air poilution control, a common denomination for various health effects would be used- It could be based on willingness to pay, mzdicai treatment costs and the value of lost productive days and years. Such valuation is beyond the scope of this paper, however. It should be acknowledged, however, that large uncertainties about the existence and magnitude of the health effects of air pollution continue to exist. Therefore, the analysis provided below should be viewed as an attempt to present, in the judgement of this author, the most likely and well-documented health impacts for which quantitative information exists. This assessment will probably change over time as new clinical, epidemiologic and economic research is completed. In the past, the U.S. Environmental Protection Agency has estimated the health and welfare effects of air pollution in its Regulatory Impact Analysis for national ambient air quality standards, as required by the Presidemial Execuive Order #12291 issued in 19812' Additional information and methodological improvements were incorporated in the subsequent analysis of economic benefits of air -2 - quality programs in selected U.S. locations.33 Recently, broad estimates of the health benefits of controlling ozone and particulate matter were provided for both the U.S. and for the ambitious control plans being considered in Southem Califomia. 4.6 The analysis reported here uses a similar approach to estixizt. health effects of criteria pollutants in Jakarta, with two improvemens: the most recent set of research fmdings are utilized and a full range of health endpoints are included. Dose-response funcions that relate various health outcomes to air pollution are taken from the available peer revI,ewcd literature. Estimates of selected health effects of air pollution are generated by applying these functions to ambient levels either observed from monitoring stations located throughout the city or estimated from available dispersion models. Using results from both time-series and cross-sectional epidemiologic analyses from the United States, Canada, and Britain, effects are estimated for such health outcomes as premature mortality, hospital visits and admissions, emergency room visits, restrictions in activity, acute respiratory symptoms, acute bronchitis in children, asthma attacks, IQ loss, and blood pressure changes. At this time,.however, because of uncertainties about the coverage and representativeness of the existing monitors in the city, and about the applicability of health studies undertaken in the U.S. to the developing world, the results should be viewed as providing only general estimates of the impacts of air pollution. Following this introduction, there are four sections in this report. Section II describes the methodology for estimating the health effects associated with changes in air pollution. The section also details the data and baseline assumptions that are necessary fo: such estimations. Section 11 provides a brief review of the literature that quantitatively links changes in air pollutants with adverse health outcomes. From this review, dose-response functions, along with associated confidence intervals, are developed. Also, suspected health effects for which quantitative estimates cannot be provided, are -3 - indicated. Section IV applies the methodology and provides estimates of health effects of air pollution for Jakarta. Section V discusses the results and indicates the research and data on developing countries needed to enhance the accuracy of these estimates. II. METHODOLOGY AND BASEUNE ASSUMPTIONS FOR ESlTMATING HEALTH EFFECTS A. Meth¶ dolo:y The estimation of the health and economic effects of air pollution involves the use of methodology similar to that used by the U.S. Enviromnental Protection Agency (EPA) in their Regulatory Impact Analysis for a new national air quality standard for particulate matter.' Estimation techniques are also derived from the analyses of economic benefits of air quality control programs in selected U.S. locations.4 To estimate the economic value associated with changes in air pollution, four factors must be determined: the dose-response relationships, the susceptible populations, the relevant change in air pollution, and an economic valuation of the health endpoints. In this paper. health effects for a range of health outcomes are provided, while valuation of these is not performed. The first step is to develop estimates of the effects of air pollution on various health outcomes. Dose-response functions chat relate health impacts to ambient levels of air pollution are taken from the published epidemiologic literature. This step involves calculating the partial derivative (or slope, b) of the dose-response fimction, to provide an estimate of the change in the prevalence of a given health effect associated with a change in outdoor air quality (A). Sufficient infonration is provided in this report to understand the sources of the selected dose-response functions, but a more complete review of the literature can be obtained in the EPA scientific review of the health effects of criteria pollutants? -4 - The next step involves multiplying this slope by the relevant population that is believed to be exposed and susceptible to the air pollutant effect under consideration (POP@. For certain pollution-related health effects this may include the entire exposed population; for other effects there may be particularly sensitive subgroups such as children or asthmatics. A third step in the calculation of health effects of air pollution involves the change in air quality (dA) under consideration. The actual change is dejsendent on both the policy issue under consideration and the available data. For example, it may be relevant to consider the change from current air pollution levels to some ambient air quality standard, either a local one, the EPA standard, or the WHO air quality guideline. A second change that might be relevant for consideration is a given percent reduction, such as 10 percent. A third method of determining the relevant change in air pollution is to assume that air quality changes in some simple proportion to the change in emissions, as in a simple linear rollback model. In that case, a 10 reduction in the total tonnage of particulate emissions, for example, is assumed to reduce ambient particulate air pollution and health effects by 10 percent. Fnally, the ambient changes associated with a given change in a stationary or area-wide pollution source can be calculated through use of computer models, if the necessary data are available. In this report, we examine a change from existing levels to several alternative ambient standards. including: (1) proposed Indonesian standards, (2) EPA ambient air quality standards, (3) WHO guidelines, and (4) California state ambient air quality standards. For the case of lead, we also calculate the benefits associated with a 90 percent reduction of ambient lead, assumed to be accomplished through a ban on leaded gasoline. The relevant standards, expressed in ternm of the annual average concentration in micrograms per cubic meter (j4glirr), are as followvs: -5 - Ambient Air Quality Standards for Annual Averages (micrograms/M3) Proposed Pollutant Indonesian EPA WHO California Total Suspended Particles (TSP) 90 75* 60-90 55# Lead 0.5 N/A 0.5-1.0 N/A Nitrogen Dioxide 100 100 N/A N/A Sulfur Dioxide 60 80 50 N/A Ozone 200 240 150 180 *In 1986, the EPA standard was replaced by a PM1O standard. The California standard has been converted from the current PMl0 standard to a TS? equivalent for the purpose of this analysis. N/A signifies standards with averaging time other than annual average. With this infonnation, the estimated health impact can be represented as follows: dHi = b * POP; * dA where: dH; = change in population risk of health effect i bi = slope from dose-response curve POP; = population at risk of health effect i dA = change in air pollution under consideration To complete benefit estimation for health effects, one would calculate the economic valuaLion of this effect (V.), as well. The valuation could be developed from estimates of the willingness to pay (WTP) for reducing risk, in order to atach values to the expected changes in prmature mortality, or a modified cost of illness (CO1). approach, to value changes in morbidity. Thus, the change in total social -6 - value (dT) of the hea;th effects due to the change in air pollution under consideration is the surnuation of all effects and can be represented by: dT = YV1dH, Unfortunately, there is still a great deal of uncertainty and controversy about much of the research on which these estimates are based. Recognizing this uncertainty, upper and lower bound estimnaes are provided to indicate the ranges within which the actual health effects are likely to fall. In addition, the categories presented in this paper are not all-inclusive, since quantitative evidence is not available for every health effect suspected of being associated with air pollution. Also, air pollution has been associated with non-health effects, including materials damage, soiling, vegetation lr..ses and visibility degradation. These types of omissions suggest that the results of this analysis are likely to underestimate the hea1th effects, and will certainly underestimate the total effects, resulting from air pollution. B. Baseline Assumptions An important question in all of the health effects estimates is whether a threshold level exists, below which health effects no longer occur, or whether the slope of the dose-response function diminishes significantly at lower concentrations. There is a presumption by some that a threshold exists at the EPA air quality standard, or at the WHO ambient guidelines for criteria pollutants. Most of the studies reported here have estimated linear or log-linear functions suggesting a continuum of effects down to the lowest levels, and have not specifically identified a threshold level. When efforts have been made to identify a threshold, little conclusive evidence has been found that one exists. In fact, many recent epidemiologic studies report an association between air pollution and health at ambient concentrations at or below the current federal standard. The former Administrator of the EPA has stated, 'in a heterogeneous population it is unlikely that, for any pollutant, there will be a single scientifically defensible threshold applicable to all people. Instead, there will be a series of thresholds for different -7 - sensitive populations and a threshold of zero for some people".' Therefore, for this report, we calculate the effects of bringing pollution down to alternative standards, without taking a position on what would happen at even lower pollution levels. A basic assumption of Lhe -model is that the association between air pollution and health estimated in the cited studies can be applied to estimate the health impact in Jakarta. These studies show that when the readings at fixed site monitors change, there is a change in the observed incidence of many health effects. Although the monitors do not measure actual exposures, they do provide a general measure of air quality which is obviously related to ultimate exposure. The use of these results implicitly assumes a similar distribution of baseline factors - health status (e.g., incidence of chronic disease), chemical composition of pollutants, occupational exposures, seasonality, time spent out of doors, general activity - and that results from other studies can be applied to the study area. As described in greater detail in Section V, since the baseline health status in developing countries tends to be poorer than that experienced in the western, developed world, this assumption will likely lead to an underestimate of the more severe health outcomes. Another source of underestimation will be present since the population is assigned a pollution concentration based on their residential location. Effects of air pollution will be higher if the person commutes to the central business district and the subsequent higher exposures are incorporated into the analysis. Therefore, the quantitative assessment of health effects presenred below are likely to be underestimates. To the extent that the original studies were primarily time-series studies relating daily changes in air pollution to the daily incidence of a health effect, the likelihood of confounding from other factors is minimized. For example, if a study was conducted over a 3-month period, and daily emergency room visits were associated with PM10, it is extremely unlikely that a change in smoking habits, occupational exposure, diet, exercise and activity patterns, indoor exposure, etc. would change on a daily basis and be correlated with daily particulate matter enough to drive the observed association. Our use of epidemiologic studies also assumes that the spatial relationship between pollution monitors and population that exists in the original studies is generally similar in Jakarta. Thus, with these assumptions, the relationship between the levels of air pollution and subsequent health effects in the cited studies can be extrapolated to estimate the health impact in Jakarta. m. REVIEW OF HEALTH EFFIECTS AND PROVISION OF DOSE-RESPONSE FUNCTIONS A. Selection Critena For this report, dose-response fimctions have been identified and adapted from published epidemiologic and economics literature. These functions allow the estimatiou of the change in health effects that would be expected to occur with changes in ambient pollution levels. For each health effect, a range is presented within which the estimated effect is likely to fall. The central estimate is typicafly selected from the middle of the range reported in a given study, or is based on the most recent study using the most reliable estination methods available. When several different studies are available for a given health effect, the range reflects the variation in results observed across the studies. When only one study is available, the range is based on the statistical confidence that can be placed on the reported results. The reported epidemiologic investigations involve two principal study designs: statistical inference based on time-series and cross-sectional dat sets. Time-series analysis examines changes in a health outcome within a specific area as air pollution levels fluctuate over time. A cross-sectional analysis compares the rate or prevalence of given health outcomes across several locations for a given point in time. The time-series studies have the distinct advantage of reducing or elimiating the problems associated with confounding or omitted variables, a common concem in the cross-sectional studies. Since the population characteristics are basically constant over the study period, Ebe only factors that may vary with daily mortality are environmental and meteorologic conditions. In general, researchers are able to -9 - more easily elicit tie effects of air poUution and weadLer on health in the time-series studies. Therefore, this review focuses primarily on time-series studies. The use and extrapolation of results from time-series analysis, however, is predicated on its applicability to other areas and for ocher time periods. Several specific criteria bad to be met for a paper to be included in this rview. First, a propar study design and methodology were required. Therefore, there was a focus on time-series regression analyses relating daily health effects to air pollution in a single city or metropolitan area. Second, studies that minimized confounding and omitted variables were included. For example, research that compared two cities or regions and characterized diem as 'high" and "low" pollution area were not included because of potential confounding by other factors in the respective areas Third, concern for the effects of seasonality and weather had to be demonstrated. This could be accomplished by either staifying and analyzing the data by season, by pre-filtering to reduce pattens in the data, by examining the idependent effects of temperature and humidity, andlor by correcting the model for possible autocorrelation. A fourth criterion for inclusion was that the study had to include a reasonably complete analysis of the data. Such analysis would include an carefil exploration of the primary hypothesis as well as an examination of the robusmess and sensitivity of the results to alternative functional forms. specifications, and influental data points. When studies reported the results of these alternative analyses, die findings judged as most representuive of the overall findings were those that were summarized in this paper. Fifth, for inclusion in this review, the study had to provide an air pollution measure that could be converted into a common miric. For example, studies that used weekly or mondtly average concentrations or that involved measurements in poorly characterized metropolitan areas (e.g., one monitor representing a large region) were not included in this review- In addition, studies that used measures of partculate matter that could not be converted into total suspended particulates (TSP or particles of all sizes) or particulate matter below 10 microns in diameter (PM10) were not used. Sixth, the study had to involve relevant levels of - 10- air pollution. The air pollution levels in Jakarta are well within the range of those observed in the epidemiologic studies used in this report. B. Development of Dose-Response Estimates 1. Particulate Matter Epidemiologic studies provide dose-response relationships between concentraons of ambient particulate matter and several adverse health outcomes including: mortality, respiratory hospital admissions, emergency room visits, restricted activity days for adults, lower respiratory illness for children. asthma atacks, and chronic disease. Among these sudies, statstically significant relationships have been found using several alternative measures of particulate matter, including TSP, fine particles (particles less than 2.5 microns in diameter), British smoke. coefficient of haze (COH) and sulfaes. Frw have involved measurement of PMIO. the metric used by the EPA in the national ambient air quality standard. The studies have been conducted in several different cities and seasons, thereby inoorporating a wide range of ciimates, chemical compositions of particulate matter, and populations- For comparison of results and the calculation of final dose-response functions, alternative measures of particulate matt were converted into PMIO. Ideally, this would be accomplished by comparing co-located monitors at each study site. Unfortnately, for many of the measures, these data are not available and we are forced to use broad estimates of the relationships between alternative measures of particulate matter. The results of our analysis of consistency, however, indicates that the findings are generally robust to these assumptions. To convert from TSP to PMIO, we relied on the EPA estimate7 that PMIO is between 0.5 and 0.6 of TSP, and use the mean of 0.55. Using the reported averages from 100 cities in 1980, we assumed that sulfates constitute approximately 0.14 of TSP.9 Therefore, the ratio of sulfate to PM10 is 0.25. The "British Smoke" (BS) measurenent is based on the amount of light reflected through a filter paper stained by ambient air flowing through the paper. Since monitors for BS do not admit particles greater than 4.5 microns in diameter, they are indicators of - 11 - concentrations of fairly srnall particles, but they do not measure particle mass like TSP and PM10. However, available data for co-located BS and TSP monitors in London indicate an average ratio of BSITSP of 0.55, the same as the ratio of PM1O to TSP." Based on this and additional analysis by the California Air Resources Board." it is assumed that PMIO is roughly equivalent to British smoke. Premature Mortality. Since the effects of particulate air pollution on mortality is the source of such large potential benefits, the evidence for an effect and its potential magnitude will be reviewed in detail The most relevant studies are reviewed below. London. Among the earliest empirical estimate of mortality outcomes associared with particulate mater is the analysis of data from London for the winter of 1958-59, where a stistically significant relationship was found between daily deaths and daily levels (24-hour average) of British smoke. London data for 14 winters, 1958-59 through 1971-72 have been analyzed by Mazurmdar et aL,'3 Ostro 1 and Schwar and Marcus,'S'6 and reviewed by the U.S. EPA.'7 En the earlier winters, the levels of British smoke were extremely high; the mean for the first seven winters was 270 psglm3. However, the mean for the last seven winters was 80 gg/m3, and in three of the last four years the mean concentrations were below 70 gg/m3. The concentrations of British smoke in London in the last years are more comparable to those commondy found in the U.S- Although these analyses involve several different statistical methds, the following general conclusions can be drawn: (1) there is a strong relationship between particulate concentatons and daily mortality in London, which holds both for the entire data set and for individual years (the larer years exhibited almost an order of magnitude decrease in air poliution concenrations); (2) there is no indication of a "no effects level' (i.e-, a threshold) at the lower concentrations of air pollution experienced in London; (3) the association between air pollution and mortality cannot be 'explained away" by meteorologic factors or by serial correlation in the data; and (4) regardless of the model specified, the quantitative implications of the studies are very simfilar. For this review, quantitative estmates of the - 12 - London data are taken from Schwartz and Marcus" which involves the most complete examination of the effects of temperature and humidity, auEocorrelation, and functional form. Their results over the 14 irears suggest the following for all-cause mortality: Daily mortality = 2.31 * (daily average BS)" in London The standard error of the estimated regression coefficient was 0.160. During the period of study, there was an average of 280 deaths per day and a mean concentration of BS of 174 gglm? in London. After taldng the derivative of the above and substituting in the mean of daily deaths, the effects of PM10 can be expressed as: % change in mortality = (0.4125 PM1Of) * change in PM1O Thus, at the mean level of pollution, a 10 yg/rn3 increase in PM10 is associated with a 0.31 percent increase in mortality. Using plus or minus one standard error from the estmate, a confidence interval of 0.29 to 0.33 percent is obtained. This change is similar to that predicted from the linear models described above. Onrio, Canada. Plagiannakos and Parker"' used pooled cross-sectional and timne-series data for nine counties in Southern Ontario, Canada, for the period 1976 to 1982. Their model attempted to explain mortality as a function of several socioeconomic factors (education, population over age 65, alcohol consumption), time, meteorology, and air pollution including TSP, sulfates, and sulfur dioxide. There was no correction for autocorrelation. Since mean ambient concentrations were not provided by the authors, grap1hical displays were used to estimate pollution levels. The mean for TSP appeared to be approxirnately 70 g//mI3 while the mean for sulfate was approximately 12 pLgm3. A statistically significant association was found between all-cause mortality and sulfates, and between respiratory-related mortlity and both TSP and sulfates. For all-cause mortality, the model with the highest association between air pollution and mortality, is represented by the following quantitative relationship: - 13 - log (annual mortaliLy in Ontario) = 0.047 log (sulfate) The standard error of the estimated coefficient is 0.0235. To estimate the change in mortality per pg/n3 oE PM10, we take the total derivative of (3) and obtain: % change in mortality = 0.047 (change in sulfatelmean sulfate) Substituting the mean concentration of sulfates and the ratio of sulfate to PM1O: % change in mortality = 0.098 * change in PMlO This indicates that a change in PMIO of 10 /ggm3 corresponds to a 0.98 percent change in all- cause mortality. Applying plus and minus one standard error, a 10 Agrm3 change in PM10 generates an effect ranging from 0.49 to 1.47 percent. Steubeville and Philadelphia. Two recent time-series studies"'9 used similar methods to examine the association between dafly mortality and TSP Both studies controlled for the effects of weather and seasonality, and in both a statistically significant relationship was found between TSP and mortality. After converting to PM1O equivalence, the Steubenville study implies that a 10 jglIm3 change in PM10 corresponds to a 0.64. percent increase in mortality over baseline. The confidence interval, based on plus or minus one standard error, is 0.44 to 0.94 percent. The Philadelphia study implies that a 10 iLglm3 in PMIO is associated with a 1.20 peret increase in mortaity, with a one standard error confidence interval of 0.96 to 1.44 percent. Santa CZara County. A recent time-series analysis examined the relationship between coefficient oE haze (COH) and mortality for the metropolitan area surrounding San Jose, Califomia3' Daily mortaliy and suspended particles measured as COH were compared between 1980 and 1986. A statistically significant association was found between COH and both all-cause mortality and respiratory- related mortality, after controlling for temperature and humidity. The models were also tested for the influence of year. season, day, and weather, with little change in the overall results. - 14- The general model for all-cause mortality, chosen by the author as most representative, is indicated the following: Daily mortality in Santa Clara = 0.0084 COH The standard error of the esEimated coefficient mas 0.0029. To obtain the effect of a 10 ,gqmh change in PM1O on the percent change in mortality, several adjustments must be made. First, as discussed by dhe author, only readings from the central city monitor were used in the study. This monitor averaged about one-third higher concentrations than the metropolitan-wide average, a metric typically used in the other studies. The author also found that the COH to TSP ratio was at least one. Therefore, we assume tha PMIO is approximately .50*COH. Finally, the study population averaged about 20 deaths per day. Therefore, we adjust the coefficient by (413)(115)(1120) to obtain ?% change in morality = 0.112 1 change in PM10 This implies that a 10 ggm3 change in PMIO results in a 1 12% change in mortliy. Applying dte standard error, we obtain a range of effect of 0.73 to 1.51 percent associated with a 10 lzgIm' change in PMIO. Ozkayzak and Thuwsron. Additional evidence is provided by a study using cross-sectional data Ozkaynak and Thurstonr examined roughly 100 metropolitan areas in the United States using the 1980 vital statistics. This study controlled for socioeconomic characteristics and conducted additional sensitivity analysis to detemine the impact of certain cities and alternative model specifications. The authors found sutistically significant relationships between mortaliLy ra;es and alternative measures of ambient particulate matter including sulfates and fine particulates. Specifically, the study reports tah existing sulfate concentrations (mean of 11.1 gImg) correspond to a 4 to 9 percent increase in mortality. Assuming a ratio of sulfates to PM10 of 0.25 as above. this suggests that a 10 gg/m3 change in PM1O corresponds to a 0.92 to 2.06 percent change in all-cause mortality, with a mean of 1.49 percent. - 15 - Summary of mortaity effct. This review suggests that the recent studies linking partilate matter to mortality generate remarkably consistent results-' (see Table 1). The mean effect of a 10 pg/imn change in PM1O implied by these studies varies between 0.31 and 1.49 percent, with a mean of 0.96 percent. Several more recent studies support this magnitude of effect" and indicate that a 10 pgrn change in PMIO relates to a 1 percent change in mortality. In our range of studies, the upper confidence level varies between 0.33 and 2.06 percent with a mean of 1.30 percent, with a mean lower bound of 0.62. Although similar studies have not been undertaken ir Indonesia, there is one set of data available to test for the existence of an effect Annual mortality and TSP in Bardhig have been reported for 1983 through 1989?2 Regressions run on these dat suggest that a 10 pg/rn change in TSP is ssociated with a 0.695 percent change in mortality, if a crude mortality rate of 0.01 is assumed. This corrsponds to 1.26 percent change in mortality associated with a 10 pglm3 in PM10 If the crude mortality rate is 0.007, a 10 ,uglm change in PM1O is associated with a 0.99 percent change in mortality. Regardless of the assumed crude mortaity rate, the estimated air pollution effects are fairly similar to those denrved from the studies summarizd above. Therefore, we assume the following associton: Central percent change in mortity = 0.096 * change in PM1O Table I: Summary of Mortality Studies Indicating Percent Change in All-Cause Mortality Association with 10 pg/rn Change in PMIO Central High Estimate Estmate London, UK .31 .33 Ontario, Canada .98 1.47 Steubenville, Ohio .64 .94 Philadelphia, PA 12 1.44 Santa Clara County, CA 1.12 1.51 US Metropolitan Area 1.49 2.06 * A high (low) estimate is obtained by increasing (decreasing) the coefficient by one estimated standard deviation. The crude mortality rate in the U.S. is approximately 0.007. - 16 - The upper and lower percent changes in mortality have coefficients of 0.130 and 0.062, respectively. The central estimate for the number of cases of premature mortality, can be expressed as: Change ir. mortality = 0.096 * change in PM1O * (1/100) crude mortality rate * exped population The crude mortality rate in the U.S. is approximately 0.007, while in Indonesia the rate is 0.0095. However, since Jakarta is expected to have a lower mortality rate than the rest of Indonesia, we assume a rae of 0.007. Therefore, the range of changes in mortality (per person) is: Upper change in mortality = 9.10 * 10 * change in PM10 Central change in mortality= 672 *10 * change in PMIO Lower change in mortality =4.47* 1( * change in PM10 ResuiaMor Hospital Admissions. As described abdve, Plagiannakos and Parker'8 used pooled cross-sectional and time-series data for nime counties in their study for the period 1976 to 1982 Soluthern Ontario, Canada. A sttistcaliy significant relationship was found betwen the incidence of hospital admissions due- to respiratory diseases qRHA) and ambient sumiate levels- Additional evidence for an effect of partimlates on hospital admissions is provided by a study by Pope.' In this study a sttistcally significant assocition was found between monthly RHA, including admissions for pneumonia, asthma and bronchitis, and monthly average PM1O in two valleys in Utah studied between 1985 and 1989. Ozone concentrations were close to baseline during the winter seasons when both PM1O and RHA were elevated so the effect appears to be mosdy related to particles. After analyzing, the results suggest the following fiuctions: Upper change in RHA per 100,000= 1.56 * change in PM10 Central change in RHA per [00,000 1.20 * change in PM1O Lower change in RHA per 100,000 = 0.657 * change in PM1O - 17 - Emergency Room Visits. Samet ct al.A analyzed the relationship between emerger,cy room visils (ERV) and air pollution levels in Steubenville, Ohio, an industrial town in the midwestern United States. Daily ERV (mean 94.3) at the primary hospital in the area were matched with daily levels of total suspended particulates (TSP), sulfur dioxide levels, and nitrogen dioxide levels for March, April, October, and November of 1974 through 1977. Daily ERV were regressed on maximum temperature and each of the pollutants in separate runs. The particulates and sulfur dioxide coefficients were statistically significant in separate regression, but these measures were highly correlated. We have selected the estimated regression coefficient for TSP as the best estimate and have used plus or minus one standard deviation from the coefficient to generate high and low estimates. The results obtained by Szmet et al. indicate the following relationship: Change in daily ERV = .011 * change in daily TSP in Steubenvilie Since the appmximate population in Steubenville during this period was 31,000, and PM10 is 0.55 of TSP, we annualize this equation and obtain an estimate of- Centra change in ERV per 100,000 = 23.54 * annmal change in PM10 The upper and lower coefficients are 34.25 and 12.83, respectively- Restricted Activity Days. Restricted activity days (RAD) include days spent in bed, days missed fron work, and other days when activities are significantly restricted due to illness. Ostro examined the relationship between adult RAD in a two week period and fine particles (FP. diameter less than 2.5 microns) in the same two week period for 49 metropolitan areas in the United States. The RAD data were from the Health Interview Survey conducted annually by the National Center for Health Statistics. The FP data were estimated from visual range data available for airports in each area. Since fine particles - 18- have a more significant impact on visual range than do large suspended particles, a direct relationship can be estimated between visual range and FP. Separate regression estimates were obtained for 6 years, 1976-1981.3 A statistically significant relationship between FP and RAD was found in each year and supported earlier findings relating RAD to TSP. We selected the approximate average of the six coefficients in calculating a central esdmate, and derived the upper and lower estimates from the range in the coefficients over the six years. The form of the estinated relationship was such that the coefficient for FP gives the percentage change in RAD associated with a unit change in FP. Specifically, the results indicate: Change in RAD per adult per year = b * annual RAD * change in FP where the high. central, and low estimate of b are 0.0076, 0.0048, and 0.0034 respectively. To covert this function for our use, we use the following information from the original study: the annual average RAD per adult was about 19 days and sulfates were 40 percent of FP. therefore, we analyze the results and convert from FP to sulftes to PM10, to obtain the following relationship: Upper change in RAD per person per year = 0.0903 * change in PM1O Central change in RAD per person per year = 0.0575 * change in PM10 Lower change in RAD per person per year = 0.0404 'change in PM1O These estimates are applied to all adults. Subsequent work by OstroP" focused on currendy working males and obtained generally sinilar results. Lower Respiratory Illness in Children. Estimates of lower respiratory illness in children are based on an analysis by Dockery et al.3 of children in six cities in the United States. The study related TSP, PM15, PM2.5 and sulfate levels to the presence of chronic cough, bronchitis, and a composite - 19 - index of respiratory illrsss (prevalence of cough, bronchitis, or respiratory illness) as measured during health examinations of samples of children in each city. A logistic regression analysis was used to estimate the relationship between the probability of an illness being present and the average of the 24-hour mean concentrations during the year preceding the health examination. Due to the likely overlap of the health endpoint measures, only the results for bronchitis is used, noting that this could include chronic cough or a more general respiratory illness. The results are applied to the population age 17 and below (17.07 percent of the total population). The basic findings that are used suggest the following: Log odds of bronchitis = log (Bk(l-B)) = 0.02368 * PM15 where B = the baseline probability of bronchitis The change in the probability of bronchitis due to a change in PM10 can be calculated since, after taking the partial derivative of the above, the following holds: dB = b p (1-pJ) * change in PM15 where: dB = change in probability of bronchitis, b = estimated regression coefficient, and po = baseline probability of bronchitis. To determine the effect of a change in PMIO, we assume that it is 0.9 of PM15 and use the baseline probability of bronchitis of 6.47 percent. The central estimate uses the estimated regression coefficient reported by Dockery et al. (0.02368) and the upper and lower ranges are plus or minus one standard! error from this coefficient (0.03543 and 0.01197). Therefore. the central estimate for the effect of a unit change in PMIO equals (0.02368)(1/0.9)(0.0647)(0.9353) = 0.00169. incorporating the above data, the following relationship for changes in the annual risk of bronchitis in children are determined as: -20 - Upper change in bronchitis 0.00238 * change in PMlO Central change in bronchitis = 0.00169 * change in PM10 Lower change in bronchitis 0.0008 * change in PM10 Asthma Attacks. Several studies have related air pollution to increases in exacerbation of asthma. For example, in a study of asthmatics in Los Angeles, Whittemore and Korn34 reported a relationship between exacerbations of asthma and daily concentrations of TSP and ozone using logistic regression analysis. Also, Ostro et al. recently reported an association between several different air pollutants, including sulfates, and inreases in asthma atacks among adults residing in Denver. Additional evidence for an effect of particulate matter on asthmatic children is provided by Pope et al.3? This study examines the effects of air pollution on a clinic-based sample and from a school-based (and relatively untreated) sample. Associations were reported between particulate matter, measured as PM10, and both respiratory symptoms and the use of medication. T'he Ostro et al. study took place during the winter months when asthma attacks were influenced by respiraotyr infections. Thus, this study is used to help derive the upper estimate. Specifically, the central estimate of this study is averaged with the upper estimate of Whittenore and Kom study to generate the upper bound. The central and lower bound estimates are denved from Whittemore and Kom's central regression estimate and minus one standard error. For our calculations, their reported asthma attack incidence of 0.26 was halved to better represent the general population of asthmatics, since many asthmatics with low attack prevalence were dropped from the analysis. The regression model includes both TSP and ozone so the total effects on asthma are apportioned to these two pollutants in accordance with their findings. As described in greater detail in Section IV, based on available Indonesian data, we have assumed that 8.25 percent of the population of Jakarta are astlunatic versus appwximately 5% in the U.S.'. The -21 - estimates for increases in the annual probability of an asthma attack (per asthmatic), based on annual changes in particles, are: Upper change in asthma attacks 0.273 * change in PMlO Central change in asthma attacks = 0.0326 * change in PM1O Lower change in asthma attacks = 0.0163 * change in PMIO Respirator Svmptorns. Respiratory symptoms are an additional measure of acute effects of air pollution. Results of Krupnick et al.3' can be used to determine the effects of particulate matter. This study examined the daily occurrence of upper and lower respiratory symptom among a panel of adults in Southern California. A Markov process model was developed to deternine the effects of air pollution on health which incorporated the probability of illness on the prior day and controlled for autocorrelation. Among the pollutants examined independently, coefficient of haze (COH) was found to be statistically associated with the probability of reporting a symptom (b=0.0126, s.e. = .0032). Data from the study suggest a ratio of COH(units/100 ft) and TSP of 0.116. The marginal effect of COH was calculated by incorporating the stationary probabilities as described in the paper. ThereCore, using the results of regressions when COH was the sole pollutant included as an explanatory variable, the following ranges were determined: Upper change in symptom days per year per person = 0.274 * change annual PM1O Central change in symptom days per year per person = 0.183 * change armual PM10 Lower change in symptom days per year per person = 0.091 "'change annual PM1O -22 - Chronic Bronchitis. Recent epidemiologic studies have related long-term exposure of air pollution to a higher prevalence of chronic respiratory disease or significant decrements in lung function. For example, Detels et al.39 found that residents living in the Los Angeles air basin who were exp.sed over a long period of time to relatively high levels of particulates and oxides of sulfur and nitrogen had significanly lower lung function than a cohort less exposed. Abbey et al.A' conducted a study on 6,600 Seventh Day Adventists, nonsmokers who had lived for at least 11 years in California. In this study, participants above age 25 (n=3914) were matched with 10 years of exposure to ambient pollutants based on their monthly residential location. New cases of chronic bronchitis between 1977 and 1987 were recorded. A logistic model was estimated that included adjustment for sex, past and passive smoldng and occupational exposure. A statistically significant association was reported between long-term exposure to TSP and chronic bronchitis. Using the mean incidence rate of bronchitis during the 10-year period, the functions were linearized and converted to PM10 equivalence, and annual number of cases of bronchitis can be estimated. The functions are as follows: Upper change in chronic bronchitis = 9.18 x 105 * change in annual PM10 Central change in chronic bronchitis 6.12 x 10r5* change in anmual PM1O Lower change in chronic bronchitis = 3.06 x 10YS * change in annual PMLO Other (Non-uantified) Health Effects. There is limited evidence linking long term exposure to TSP to increases in cancer in women.' Since TSP includes seveal materials known or believed to be carcinogenic, reductions in particulate matter will likely reduce cancer cases, as well- There also may be other acute and cromnic effects for which no empincal information is curreny available. For example, the mortality effects calculated in this study only relate to acute exposures; longer term exposures to particles may increase the likelihood of prematre mortality. Finally, no estimates are -23 - provided for changes in lung function that are likely to occur after exposure to certain forms and levels of particulate matter. Table 2 sunmarizes the morbidity outcomes associated with particulates in this review. Table 2: Morbidity Effects of 10 pg/r3 Change in PM1O Morbidity Central High' RHA/100,000 12.0 15.6 ERV/100,000 235.4 342.5 RAD/person 0.575 0-903 LRI/child/per asthmatic 0.0169 0.0238 Ashma attacksfper asthmatic* 0.326 2-73 Respiratory Symptoms/person 1.83 2.74 Chronic Bronchitisll00,000 61.2 91.8 *Applies to the 8.25 percent of the Indonesian population that is assumed asthmatic. #A high (low) estimte is obtained by increasing (decreasing) the coefficient by one estimated standard derivation. 2. Sulfur Dioxide Effects of sulfur dioxide (SO2 on the respiratory system have been observed after either short- term (less than one hour average) or longer term (24-hour average or longer) exposures. Several recent epidemiologic studies indicate that changes in 24-hour average exposure to S02 my affect lung funion, the incidence of respiratory symptoIs and diseases, and risks of mortality. These studies have been conducted in different geographic locations and climates, and with different populations and covarying pollutants. Although many of these investigations also indicate that particulate matter or ozone was associated with these adverse health outcomes, several studies appear to show an effect of SO2 alone. Furthemnore, in some of the publications reporting an effect of both S02 and parciculates, they are highly correlated, but in others, the correlation of the daily levels is only weak to moderate. Thus, it is possible -24 - to infer an effect of SOQ or a sulfur species highly related to SO. Below, a brief review of several relevant studies and available dose-response relaonships are provided. Premature Mortaliy. Epidemiologic studies undertakcen in several locations indicate that S02, acting alone or as a surrogate for other sulfur-related species, is associated with increased risk of mortality- These includes studies in France,4 England,0. Poland," and Athensu@ Unfortunately, most of the available studies do not provide dose-response functions. Our estimate is derived from latzais et al.Y5 A brief summary of the studies indicating an effec of S1 on mortality follows. It is also important to note that after oxidation in the atmosphere, some of the SO, will tun into sulfate. As reported in the earlier section of particulate mater, changes in sulfates are associated with both mortality and morbidity. Therefore, some of the benefits of reducing SO relate to the reductions in particulate matter described above, and one should be aware of potential double counting of health benefits from reduced polBution. Derriennic et alYe analyzrzed daily mortality for individuals over 65 years of age in Marseilles and Lyons, France between 1974 and 1976. Daily averges of SO and suspended particulates were .03 Ppm and 106 pg/m3, respectively, but monthly SO2 averages were above .07 ppm during certain times of the year. These two pollutants were moderately correlated (r = .46). Seasonal influences were prefiltered from the data, which were also corrected for autocorrelation. The authors noted that temperature, which was inversely correlated with SO, was correlated with respiratory-related mortality in Lyons and with cardiovascular-related mortality in Marseilles. The results, based on multiple regression and controlling for temperature, indicated a statisticay significant association between S02 and respiratory deaths in both cities, and between SO, and circulatory deaths in Marseiles. The authors argue that fte SO effecs are independent of the impt of te m , since the regression coefficient relating SO, to respiratory mortality is similar for the two cities, but in only one is temperature correlated with mortality. Similarly, the association of SQ with circulatory deaths in Marseilles (but not Lyons) may be explained by the high -25 - correlation in that city between SO% and temperature, and tempeature and mortality. No association between particulates and morality was detected in either city. Chinn et al.'3 investigated the association between mortlity of people aged 45 tO 74 and air pollution in London, England and Wales. Mean SO2 and smoke levels for 1971 are not provided explicitly, but visual inpectio of the relevant graph in the tcxt suggests means of .06 ppm and 80 pg/mt. respectively. Two age groups for bothmn and women were analyzed: those aged 45 to 64 and 65 to 74- In addition to total mortality, several specific causes of death were considered. including hypertensive disease, influenza, and chronic brnchitis. There was little correlation between either SCQ or smoke and morUtliy, prompting the authors to suggest this was a negative study. However, one particularly significant finding was a correlation between S02 and morality from chronic bronchitis among men above 65 (r= .22) and women betwe 45 and 65 (r = .26). Kyrzanowski and Wojtniak" examined the association between individual-specific daily mortality and air pollution over a ten-year period for residents living in Cracow, Poland versus dtose living just outside the city. The results indicated a significa sttical relationship between air pofluidon, measured as particulate matter and SO2, and all-cause mortality for men. Hatzakis et al.6 explored the relationship between daily morality and air polution in Athens. Greece during 1973-1982. Mean daily levels of 50 and British smoke were 85 pg/n3 and 63 g/rn3, respeivey, with an average of 28.48 deaths per day. The poflutmnts were firly strongly correlated (r = .55). Mortality was adjusted for seasonal pa over time by calculatig an observed minus predicted measure. Regression analysis was used to control for temp humidity. holidays, and annual. seasonal, monthly and weeldy trends. S02 and excess all-cause mormlity were correlated when all other independet variables were taken ito accounlt Particulates, however, were not related to mortality. The high estimate of the mortality efect is taken from the crude (i.e. no covariates included) regressionresults of this study. The linear coefficient relating S% to daily deats was 0.0346. Thus a - 26- 10 jggm3 (or 0.004 ppm) change in SO, is associated with a daily increase of 0.346 deaths or a 0.346/28.48 = 1.21 percent increase. For the lower range of the effect, we use the regression equations that includes adjustment for seasonality, year, interactions betweer. year and season, day of smdy, several meteorologic faitors, and dummy variables for months. These adjustments result in a lowering of the coefficient to 0-0058 (standard error = 0.0029, p = 0.046), suggesting a 10 pglm3 change in SO0 is associated with a 0.2 percent change in morclicy. The central estimate uses ihe model that adjusts for seasonality and year and implies that a 10 pglm3 change is associated with a 0.48 percent change in mortality. The ranges are as follows: Upper percet change in mormal-ty = 0 121 * change in SO Central percent change in mortality = 0.048 * change in SO% Lower percent change in mortality = 0.020 change in SO, RVespirator Snmptoms. Recent studies chat provide evidence of an effect of SO, on symptoms include Charpin et al..4' Bates and Sizto,4'', Ponkal4' Dodge et al.," and Schwartz et alY5l2 Dose- response normation can be generated from the later two studies. Schwartz et al.51 relate daily levels of SO, to respiratory symptoms among a sample of approximately 280 children in Watertown, Massachusetts who were part of the Harvard Six-Cities Study. A daily diary completed by parents recorded several acute symptoms of their children including upper respiratory illness and cough. The corlmion among pollutants was not reported. A logistic regrssion was used to examine the relationship of pollution to these symptoms. Sulfur dioxide had a statstically significant association with cough. The impacts of other pollutants were unclear from this primarily methodological article. Nevertheless, the results suggest the following: logit (cough) = 0.0130 SO2 (ppm) - 27 - The standard error of the estimated regression coefficient was 0.0059 and the mean incidence rate was one percent. Taking the derivative, converting into pglm3 (1 ppm = 2,860 Ag/lm3). annalizing and substituting, we obtain the following functions for children: Upper change in the probability of cough per 1,000 kids per year 0.0262 * change in SOz Central change in the probability of cough per 1,000 kids per year = 0.0181 * change in SCQ Lower change in the probability of cough per 1,000 kids per year = 0.010 tchange in SQt Schwartz et al2- examined the effects of air pollution among a population begining nursing school in Los Angeles in the early 1970s. Daily diaries were completed and provided information on incidence of symptoms including cough, phlegm, and chest discomfort. PoUutant under investigation included oxidants, sulfiir dioxide, nitrogen dioxide and carbon monoxide. In models corrected for autocorrelation, a signifi-cant association was found between SQ and chest discomfort. Daiy concentrations of Sa averaged approximately 0.09 ppmr Specifically. the results indicated: logit (chest discomfort) = 1.88 * S% (ppm) The standard error of the estimated regression coefficient was 0.094. Taking the derivative, annualizing, converting into pglm' and substituting the mean rate of chest discomfort of 0.04, the following fimcdons are obtained: Upper change in probability of chest discomfort per year = 0.015 * change in SQ2 Central change in the probability of chest discomfort per year = 0.010 * change in SQ2 Lower change in the probability of chest discomfort per year =0.005* change in SQ -28 - Table 3 summarizes the health effects that have been quantfied for SO.. Table 3: Effects of 10 pglm3 Change in SO1 Concentrations Cental High' Sulphur Dioxide Estimate Estimnate Mortality (percent change) 0.48 1.21 Respiratury Syinptomss 1,000 child/year 0.18 .26 Chest Discomfortfadultlyear 0.10 .15 #A high (low) esumate is obtained by increaing (decreasing) the coefficient by one estimated derivation- 3. Ozone Ozone is the primary component of photochemical smog. As such, it has been assodated with several adverse respiratory outcomes including incrmeased upper and lower respiratory symptoms, eye irritation (oxidants), restrictions in activity, and exacerbation of asthma. Most of the evidence of the effects of ozone is derived from clinical studies in which subjects are exposed to a known amount of ozone in a controlled setting. For example, healthy individuals may be exposed to moderate levels of ozone in a chamber while engaging in moderate exercise. Unfortunately, these studies usually focus on changes in lung fimction and less so on increases in symptoms. Also, it is difficult to develop dose- response functions from some of these studies or extrapolate from their findings to the fiee-living population. However, severl epidemiologic studies are available and provide a basis for dose-response estimates. - 29 - Respiratoxy Hospital Admissions. Current evidence indicates that ozone may be associated with hospital admissions related to respiratory disease (RHA)35- This possibility is supported by findings from panel studies of asthmatics indicating that exacerbations occur in response to ozone?34-' Clearly, some of these exacerbations may result in either emergency room visits or hospital admissions. Unfortunately, because of the high covariation between ozone and other pollutants in the summer when most of the studies have been undertaken, it is difficult to determine the effect on RHA attnrbutable to ozone alone. However, by using information from several studies, it is possible to begin to apportion the effects of the different pollutants. Thurston et al.53 found a significant association between RHA and both ozone and sulfates in New York City (the Bronx) and Buffalo in the summer 1988- In this analysis, correccions for autocorrelation and day-of-week effects were made. At the mean, the effect of ozone was approximately twice the effect observed for sulfates. Burnett et al.4 also reported a staistcally significant association between hospital admissions and both ozone and sulfates in Ontario, Canada for the years 1983 through 1988. Their findings suggest that the ozone effect was approximately 3 times tbat of sulfates, based on a regression equation that included both pollutants. Therefore, it is reasonable to apportion the effects of RHA based on the relative coefficients in Thurston et al. The average of the coefficients for the two cities in that paper is 2L .3 RHA per day/ million/ppm ozone. which becomes the central estimate, with a standard error of 10.9. Thus, after annualizing, the functions for RIHA become: Upper change in RHA per person 0.012 * change in daily 1-hour max ozone (ppm) Central change in RHA per person = 010077 * change in daily 1-hour max ozone (ppm) Lower change in RHA per person -0.0038 * change in daily 1-hour max ozone (ppm) Restrictions in Activitv. Portney and Mullahy1' used the 1979 Health Interview Survey conducted by the National Center for Health Statistics. to examine the relationship between ozone and -30 - the occurrence of minor restictions in activity (MRAD). These involve restrictions in activity that do not result in either work loss or bed disability Individual-level health data for 14,000 adults living throughout the United States were combined with data on air pollution and meteorology. The health outcome was based on a two-week recall period. The general regression model included socioeconomic and demographic factors, chronic healdt status, and city-wide variables. A statistically significant association was found between the number of MRAD in a two-week period and ozone concentrtions during a similar period. For the most general results, using a poisson model, the coefficient on ozone represents the percent change in MRAD per person per two weeks fir a one part per million (ppm) change in ozone. The estimated beta coefficient was 6.883, wiEh a standard error of 3.4 (p C .05) Therefore, the centmal estimate becomes: % change in MRAD per two week period = 6.883 * change in two week average of 1-hour daily maximum ozone Since the mean MRAD per two week period is roughly 0.19, the above equation becomes: $ change in MRAD per year = (6.883)(.19)(26) change in annual average of I-hour naximum ozone (ppm) This implies that reducing the average daily maximun ozone concentration by 0.01 ppm for one year would reduce the number of MRAD per person by 0-34. The high and low ranges are developed by using plus and minus one standard error of the estimate. The annual increase in the number of cases per person for a change in the annual average of the 1-hour daily maximum of ozone (ppm) is: - 31 - Upper MRAD per person per year =51.0 * change in l-hr max ozone (ppm) Central MRAD per person per year 34.0 * change in 1-hr max ozone (ppm) Lower MRAD per person per year 17.0 * change in I-hr max ozone (ppm) Respiratory Svmptoms. Results of Krupnick et al_38 can be used to estimate the effects of ozone on respiratory symptoms. As noted above, this study examined the daily occurrence of upper and lower respiratory symptoms among a panel of adults in Southern California. A Markov process model, which incorporated the probability of illness on the prior day and controlled for aumocorrelation, was developed to determine the effects of air pollution on health- Many regression models included both ozone and particulate matter. Therefore, to prevent double counting, the effects of these pollutants were apportioned according to the regression results when the pollutants were examined together. The marginal effect of ozone was calculated by incorporating the stationary probabilities as described in the paper. The high estimate is obtained from the specification that includes all other pollutants, the central estimate includes only some of the pollutants and the low estimate is one standard error below the central estimate. Therefore, the following ranges were determined for the number of symptom days related to a change in the annual average of 1-hour daily maxinum ozone: Upper change in symptom days per year per person 96.6 * change in 1-hour max ozone (ppm) Central change in symptom days per year per person = 54.75 * change in 1-hour max ozone (ppm) Lower change in symptom days per year per person = 28 11 * change in 1-hour max ozone (ppm) Eve Irritation. Schwarez et al.57 provide empirical estimates relating oxidants to eye irritation using the data from a population beginning nursing school in Los Angeles, as described above. Using -32 - logistic models corrected for autocorrelation, a significant association was found between oxidants and daily incidence of eye irritation. Specifically, the results indicated: logit (eye irritation) = 0.0202 * ozone (pphm) The standard error of the estimated regression coefEfcient was 0.0018. Taking the derivative, analyzing, and substituting the mean incidence rate of eye discomfort of 3.75 percent, the following functions are obtained for adults, in terms of cases per year per annual change in average 1-hour maximum ozone (ppm): Upper change in eye irritations = 29.9 * change in 1-hour max ozone (ppm) Central change eye imtations = 26.6 * change in 1-hour max ozone (ppm) Lower change in eye irritations = 23A * change in 1-hour max ozone (ppm) Asthma Exacerbation. Whittemore and Korrt? studied the acute effect of oxidants (including ozone) on the increased probability of a daily asthma attack. The data were taken from 16 panels of adult asthmatics located in six Los Angeles communities. The median one-hour maximum oxidant level ranged from 0.03 to 0.15 ppm across the communities studied. with single day peaks of 0.40 ppm. The study used a statistically powerful approach to estimate both individual-level and group effects. Oxidants were found to be statistically related to exacerbation of asthma, after controlling for asthma status on the previous day, temperature. hunidity, and day of study. Using a logistic model. they obtained a regression coefficient of 1.66 and a standard error of 0.72. We use a baseline daily probability of asthnma attack of 0.13 as discussed above. In a similar study using daily data on asthmatics in Houston. Stock et al-55 report an association between ozone and the likelihood of an asthma attack. The regression specification included several pollutants, pollen, humiditv, temperature and whether an effect occurred on the previous day. The results from the regression model that included particulate measurements was used to generate the upper bound. - 33 - The results were not used for the central estimate since the sample size was so small (n=41). Therefore, the functions for the effect of ozone in terms of ppm 1-hour maximum are: Upper change in asthma attacks per year 189.8 * change in ozone (ppm) Central change in astima attacks per year = 68.44 * change in ozone (ppm) Lower change in asthma attacks per year -38.69 & change in ozone (ppm) Table 4 summarizes the outcomes that have been quantified for ozone. Table 4: Effects of 1 ppm Change in Ozone Central Estimate* High Estimate Hospital Admissionslpersons 0.0077 0.012 Minor Resictions in Activitylperson 34.0 51.0 Respiratory Symptoms/person 54.75 96.6 Eye Irritation/adult 26.6 29.9 Astma Exacerbation/asthmatic 68.44 189.8 4The coefficients apply to the annual average of the daily 1-hour maximum ozone. 4. Lead Exposure to ambient lead occurs primarily from leaded fuel in automobiles and from stationary sources including primary and secondary smelters and battery recycling plants. Once absorbed, lead is distributed throughout the body and is only slowly removed. Lead has been reported to cause many different health effects- Based on current knowledge, clinical effects that may occur at the lowest blood lead concentations include neurodevelopmental effects in children, and hypertnion and related - 34 - cardiovascular conditions in adults. These two effects provide the basis for our estimates of the impact of lead. A thorough review of health outcomes associated with lead exposure is provided by U.S. EPA54 D and ATSD.6Y Effects of Lead on Blood Pressure in Adults. The association between lead and increased blood pressure was first observed in animals. This effect has been shown across a range of doses and in several species," and has been examined in occupational and population-based epidemiologic studies. The population-based studies will be briefly reviewed here. Several investigators66 have used NHANES II. data, published by the National Center for Health Statistics, to investigate the relationship between blood lead level and blood pressure. NHANES IL is a large, individual-level database that includes information on a variety of potentially confounding factors. Therefore, these studies avoided common study design problems (e.g., healthy worker effect, workplace exposures to other toxic agents, selection bias, and problems of control group selection). Using these data, Harlan er al.< demonstrated statistically significant linear associations (p