A quantitative risk assessment method based on population and exposure distributions using Australian air quality data

EnvironmentInternational,Vol. 25, No. 617, pp. 887-898, 1999 Copyright 01999 Elsevier Scimce Ltd Printed in the USA. All rightsresewed 0 160-4 120/99/S-see front matter

Pergamon

PI1SO160-4120(99)00064-l

A QUANTITATIVE RISK ASSESSMENT METHOD BASED ON POPULATION AND EXPOSURE DISTRIBUTIONS USING AUSTRALIAN AIR QUALITY DATA Tom Beer and Paolo F. Ricci* CSIRO Atmospheric Research, Aspendale, Australia University of San Francisco, San Francisco, CA, USA, and University of Queensland, Australia

EI 9903-I 43 M (Received 30 March 1999; accepted I2 July 1999)

This paper develops a practical probabilistic method for assessing aggregate population health risks from different types ofpopulation exposures. The method consists ofcalculating the product oftwo functions: a population-weighted distribution of concentrations and a concentration-response distribution. This operation yields the corresponding aggregated health-risk distribution function. The method can use alternative exposure-response distributions and populations-specific exposure patterns, depending on the context of the assessment. A deterministic sensitivity analysis is included in the methodological aspects of this research. The distributions of concentrations are generated by combining area-specific population densities with atmospheric concentrations for each of the areas where exposure to air pollutants occurs. The exposure-response functions are developed from the literature. The method is exemplified using alternative exposure probabilities to carbon monoxide, nitrogen dioxide, particulate matter (PM,,,), and exposure-response models developed specifically for these pollutants for assessing health risks, and applied to data from a number ofAustralian cities. The results, which hold when the functions are monotonic, show single maximum per pollutant, regardless of the choice of exposure and exposure-response distribution. Although those maxima are often below the Australian Air Pollution Standards, there are instances when this is not the case. 01999Elsevier Science Ltd

INTRODUCTION effects follows from accounting for the direct inhalation (mass of the original pollutant per volume of inhaled air by the individual) which can be related to the disease or mortality-related response, as described by Ricci (1985). The relation between exposure and response is generally described by a cumulative probability distribution of response, given exposure for a particular health endpoint. The endpoint is a single but critical adverse health outcome, out of possibly many adverse outcomes for the same, single air pollutant to which individuals can be exposed. The choice of that outcome, as

The connection between sources of pollution and their health effects is based on the transformation into a suitably averaged concentration (mass of pollutant per unit volume of air) of the mass of the pollutants emitted per unit time. The reason for this is that the concentration of atmospheric pollutants is usually a conservative surrogate for biologically effective dose (mg/kg-d). These conversions require using an appropriate time average, which means that the timeaveraged concentration can be equated with exposure. The link between the concentration and the health *Corresponding author. 887

888

Coherssen and Covello (1989) suggest, is a mixture of policy and science: generally, it is the most severe outcome that is chosen for regulatory risk analysis in part for the sake of economy. In this paper, the authors are concerned with developing a simple but accurate method for population-wide risk assessment and applying it to several major air pollutants. These are carbon monoxide (CO), ozone, sulfur dioxide, nitrogen dioxide, particulate matter (as PM,,), and lead, assessed individually, rather than as a mixture. Although this is a simplification, it follows the regulatory reasoning used in the U.S., the U.K., and Australia, among others. In this paper, for brevity, only a subset of the analyses is reported. This paper describes a method in which the distribution of alternative exposure measures to airborne pollutants and their exposure-response distributions are explicitly combined to form a function that fully characterizes these risks. Ricci and Wyzga( 1983) described the basis of health risk analysis in the context of air pollution and statistical estimation ofthe parameters of exposure-response models. The data are used in the statistical estimation of the coefficients of the exposure-response model, given a form of exposure pattern over time. More recently, for instance, in studying the relationship between daily mortality counts and exposure to particulate matter, sulfur dioxide, and other airborne pollutants, HE1 (1995) and Moolgavkar and Luebeck (1996) used a statistical risk model based on the Poisson regression. Ballester et al. (1996) have justified this form of regression model for rare, daily mortality counts related to a number of independent variables that account for those factors that contribute to mortality, but are uncorrelated with each other. Saldiva et al. (1995) and Ricci et al. (1996) used other risk models, such as the Auto Regressive, Integrated, Moving Average time series analysis as well as more complex formulations, directly to account for temporally changing exposure and response daily, monthly, and yearly data over long periods of time. This approach can use these methods as well as birthdeath stochastic processes and other statistical methods. However, the application that is discussed in this paper is based on a set of simple functional relationships for the exposure-response (the set consists of an exposure-response function for each pollutant), and combines them with the corresponding area-specific frequencies of exposures, to obtain the distribution of the health risks. Not only are the adverse health effects complex to model, and differing in severity, but the

T. Beer and P.F. Ricci

importance attached to particular health endpoints requires discussions. No similar approach has been found in the literature, as reviewed by Gratt (1996). A SUMMARY

OF HEALTH EFFECTS

Given the enormous literature, and reviews of that literature, only some of its salient aspects are described to orient the reader. This is done through a limited discussion of the types of exposure patterns and the adverse response associated with the major air pollutants in this work. In a later section, the specific exposureresponse functions used are described and the rationale for those choices is given. Lippmann et al. (1983) found that ozone (0,) has been found to reduce children’s FVC and FEV, at concentrations ranging from 0.024 to 0.24 ppm (40470 mg/m3) hourly average, in summer. Ozone with sulfates and the hydrogen ion, as Lippmann (1989) reported, increase the number of admissions of asthmatics in hospitals at concentrations from 0.001 to 0.21 ppm (2 to 400 mg/m’) hourly average. PM,, particle air pollution, with sulfates and ozone, also increased hospital admissions ofasthmatics, cough and other respiratory symptoms, as well as mortality, as reported by Dockery et al. (1993) and Pope et al. (1992), among others. Other studies are those by Schwartz (1994a), Schwartz et al. (1993), Schwartz and Dockery (1992), and Scarlett et al. (1995), who developed the statistical association of particulate matter and sulfur dioxide with selected health endpoints. Thus, they found that 10 mg/m’ of PM,, increases respiratory mortality rates by 3.4% and cardiovascular diseases by 1.O%, and that PM,,, up to about 30 mg/m3 is associated with 37% increase in cardiovascular mortality. A number of epidemiological studies that have focused on the effect of PM,, and SO, on human health were conducted in a wide range of geographic areas with diverse climatic conditions. The groups at risk were often children and elderly populations with preexisting respiratory illness. The length of study period ranged from weeks to 10 y and even longer. Most analyses were conducted with time-series data by using different statistical analysis models such as logistic regression, Poisson regression, multiple regression, and statistical time-series analyses. A number of PM,, studies, in different locations, were conducted by Schwartz (1994b; 1995), Saldiva et al. (1994), Docker-y et al. (1993), Pope and Kanner (1993), Pope et al. (1992),

Assessing aggregate population health risks

Moolgavkar et al. ( 1995), and Ricci et al. (1996). These authors often reported similar significant associations between long-term air pollution exposure and respiratory illnesses or symptoms, hospital emergency room visits and admissions, and mortality from chronic obstructive pulmonary diseases (COPD), cardiovascular diseases (CV), and lung and respiratory cancers. However, unlike the PM,,, findings, the overall results of SO, studies were not as consistent in finding a biological gradient as did the PM,, studies. However, Imay et al. (1986) reported that short-term exposure to SO, concentrations of 0.25 ppm (650 mg/m3) produced short-term results in reversible bronchial spasms in asthmatics. Mean levels of NO, for periods less than 24 h, ranging from 0.1 to 0.6 ppm (200 to 1000 mg/m’) were reported by Euler et al. (1988) and Harrington and Krupnick (1995) and others, to affect asthmatics, patients with COPD, as well as the general public and children. Short-term exposure to nitrogen dioxide at 0.3 ppm (500 ms/m3) for a 4-h exposure period, caused lung function decreases in those exposed. Some of the studies accounted for co-exposure to SO,; nevertheless, there is debate about those results. The general adverse health effect from exposure to CO, as reported by Aronow et al. (1977) and others, is that this gas increases carboxy-hemoglobin and thus reduces the amount of oxygen available in the blood. In angina patients, for instance, exposure to CO decreased the time to the onset of angina by approximately between 1 and 22%, for COHb between 2 and 6%. CO concentrations from about 7 ppm (8 mg/m3) to 20 ppm (23 mg/m’) were reported to cause between 5 to 8 excess hospital admissions per day. The U.S. EPA (1986) has causally associated lead, another important air pollutant in many countries, with neurological, IQ-related changes and hypertension. For example, some results indicate that blood containing 500 ug/L to 700 ug/L total lead results in a 5 point decrease in IQ in children. Dentin lead levels higher than 20 ppm (mg/kg)resulted in disabilities (Odds Ratio = 5.8) relative to less than 10 ppm (mg/kg) levels. The literature suggests that single pollutant analysis is generally complicated by the fact that ambient air contains a mixture of inorganic and organic pollutants, as well as other agents such as pollens and so on. A few specific aspects of the potential effect in the context of any choice of exposure-response functions include the following:

889

1. Ambient and indoor air temperature, other omitted pollutants, and factors such as diet and other risk factors - associated with the health endpoints of relevance to the adverse health effect of the air pollutant of interest-can confound its actual effect. As to temperature, for instance, some associations have been reported in winter by Lippmann and Ito (1995) and Roemer et al. (1993) in cities where exposure to particles peaks in the warm months during summer and also in cities where it peaks in the cold months. 2. If some pollutants included in the multivariate exposure-response models are correlated, it may be impossible to determine which subset of specific pollutants is responsible for the causal association of air ‘pollutants with the adverse health effects being studied, as discussed by Ricci and Wyzga (1983). For instance, the association of SO, with health effects can be reduced after introducing particulate matter in the exposure-response model. The result is that the contribution to the adverse health effect by a specific component of air pollution perhaps cannot be singled out. 3. The behavior of individuals at risk during the day and the night, and their exposure to air pollutants differs significantly to the exposure recorded by instruments monitoring the ambient air at fixed monitoring stations. The monitoring data take no account of the area in which people carry out their daily tasks or the time they spend inside buildings, as suggested by Ricci and Ospital(1996). Individual exposure will also vary because of time spent outside, physical activities, personal living pattern, and occupation. 4. Hospital emergency room visits and admissions used in attempting to develop a causal model of response from ambient air pollution can be affected by “admission bias”, as suggested by Sunyer et al. (1993) and Walters et al. (1994). Different hospitals in different countries (and within regions or cities of a country as well) can have different criteria for admissions thus affecting the response. Finally, the diagnostic criteria used in a period of time may change, thus affecting the result obtained by studies that use long periods of time. Added causal specificity about the form of the relation between exposure and response was contributed by the Health Effects Institute (HE1 1995). It re-studied the form of the exposure-response functions for mortality associated with particulate matter and SOZ. An example of the results of time series of data for Philadelphia, PA, USA, is that the relationship between relative total mortality and short-term exposure to SO,

T. Beer and P.F. Ricci

Table 1. Summary of the relationships between daily mortality and bxposureto particles and SO*.

City and time series of daily data

Dependent variable [range; ]

Model

Philadelphia, PA, USA (1973 to 1980)

Total daily Log-linear (Poisson) mortality [26 to 92; <48.2>]

St. Louis, MO, USA (9/85 to 8186)

Log-linear Total daily mortality (Poisson) [31 to 81; <55.9>]

Birmingham, AL, USA (8/85 to 12/88). Eastern Tennessee, USA (9/85 to 8/86)

Total daily mortality mR to 33;<17.1>] Total daily . mortality [5 to 29, <15.5>]

Santa Clara County, CA, USA (1980 to 1986)

Utah Valley, UT, USA (4/85 to 12/89)

Independent variables exposure metric [range and ]

Estimated coefficient (s.e.), [t-value]

Controls or available data [range; ]*

Mean of same day and previous day TSP [0 to 137; <21.0>(pg/m3] SOz[22t0338,<77.2>ppb*] Prior day PM,, [1 to 97; <27.6> @g/m’].

0.5, (0.16), [2.8] 1.21,90.34), [2.9]

Temp. [4 to 89, <548> (F”)]; Dew pt. [ 1 to 76; <42.8> (F”)]; Season; Date; and Year.

1.5, (0.71), [2.1]

Log-linear (Poisson)

Mean of 3 prior day PM,, [NR to 163; <47.9>]

1.5 (0.4) [2.5]

PM, 5 [ 1 to 75, ]; SO4 [0 to 38, <8.0>]; H+.[Oto 87.8, <9.8>]; SO, [0 to 47.0, <8.9>]; NO* [5.7 to 5 1.4, <20>]; 0, [-1.6 to 63.7, <22.5>]; Temp. [8 to 91; <57.8>]; Dew pt. [-7 to 75, <45>]; Temp. WR to 88, <62.6>]; Dew pt. [NR to 75, <50.8>].

Log-linear (Poisson)

Prior day PM,,, [4 to 67, <30.1>]

1.6 (1.4) [l.O]

Total daily mortality [NR, ]

Linear regression

Concurrent day Coefficient of Haze (COH) * 1000 [59.7 to 79.6, <65.9>]

6.7* 1000 (3.0* 1000) and p-val = 0.03

Total daily mortality [0 to 12, <2.7>]

Log-linear (Poisson)

5-d lagged average PM,, [ 11.2 to 296.7; <47.2>]

1.6 (0.4) [2.5]

PM, 5 [4 to 758, <21>]; SO4 [l to 27, <8.7>]; H’ [0 to 289.8, <36.1>]; SO, [-1.3 to 29.2, <5.2>]; NO, [3.8 to 33.5, <12.6>]; O3 [-0.08 to 49.1, <23.0>]; Temp. [ 11 to 86; <59.8>]; Dew pt. r-1 to 71, <47.9>]. Year indicator variable; Temp.; Rel. Humidity; 3d order polynomial for each year; Interaction between winter day and years. Temp. [NR to 8, <62.6>]; Dew point WR to 75, <50.8>].

*The estimated coefficients reported in this table may not reflect these controls; **ppb = nL/L; NR means not reported in the HEI (1995) reference; s.e. stands for standard error of the estimate.

and total suspended particulate matter (TSP) is nonlinear and non-monotonic. However, for SO,, the exposure-mortality relationship for this chemical (measured and reported in parts per billion, ppb or nL/L) is linear from 0 to approximately 30 ppb (nL/L), for total mortality, mortality in the less than 65 year old, and mortality in those older than 65 years (HE1 1995). TSP, on the other hand, does not show this relationship at exposures to concentrations ranging from 0 to approximately 60 mg/m3 of TSP. The overall effect of particulate matter (measured either as TSP or PM,,), and SO, for three American cities is summarized in Table 1 (HE1 1995). The response is total mortality excluding “external causes of death (e.g., accidents)”

and each estimated coeffkient is multiplied by 1000 to yield the approximate percentage change in total daily mortality per lo-unit change in daily particulate or sulfur dioxide exposure (HE1 1995). The physical units for the hydrogen ion (H’) are nmol/m3. Those for SO, are mg/m’. All other physical units are either in mg/m’ for particles and in ppb for the gases. METHOD

This section contains the description of the method of analysis in which exposure is estimated from the actual monitoring of air quality parameters as well as different choices of exposure-response functions.

Assessing aggregate population health risks

Australia was used as a case study because Beer and Walsh (1997) and Walsh and Beer (1998) collated the exposure data used in the method that was developed. The components of this method are: 1. Exposure assessment. It is formulated to portray either repetitious exposure or persons-affected distributions for several air pollutants. 2. Exposure-response assessment. It is formulated to describe the relation between the exposure assessment and the probability of adverse health response, given the exposure patterns developed in the exposure assessment. 2. Risk assessment. It is formulated as the multiplication of the functions generated in the exposure assessment with the exposure-response function, for each air pollutant and for each type of exposure. The method described is simple and reflects the probabilistic nature of exposure and its relation to the equally probabilistic exposure-response functions. Specifically, the method can deal with threshold and nothreshold adverse health effects and is independent of the type of the adverse health effect. It can be applied to carcinogens, toxicants, mutagens, and other causal agents that can lead to increased health risk. Thus far, however, univariate exposure-response models were used and limited the analyses to toxic health endpoints. Cancer, mutagenic, or teratogenic endpoints were not used, although the method can readily account for stochastic dose-response models such as the multistage family ofmodels and other linear or non-linear models. Furthermore, extending the method to include multivariate models - such as several of those discussed in the literature reviewed in the first sections of this paper - follows directly; this extension is not discussed.

Exposure assessment Population-at-risk data is developed from available statistical summaries for the regions, air basins, or other suitable areas to be studied. Although this is not a specific requirement, it is used in the Australian example. Thus, population data from the Australian Bureau of the Census for 1990 has been converted to a population density value for squares of 1 km by 1 km, for several Australian regions for which these density calculations make sense. Calculations by Walsh and Beer (1998) proceed by working through the occurrences of all concentrations from zero up to the annual maximum, and by working through all grid squares

891

within the boundaries ofthe relevant city. The concentration data measured around Australia was collated by the EPA of Victoria and interpolated onto a grid, as described subsequently in this paper. Three aspects of exposure were evaluated: 1) the cumulative frequency distribution of each pollutant; 2) an estimate of persons affected; and 3) an estimate of repetitious exposure assessment that combines the number of persons affected with the duration of their exposure. Two of these measures (repetitious population exposure and population affected) are exposure metrics. 1. Averagefrequency of exceedance. This measure concerns the spatial approach developed for the Repetitious Population Exposure, and then is used to calculate the average of the exceedances in each 1 km by 1 km cell to establish how often an environmentally tolerable value is surpassed, within a specific cell. 2. Populations affected (or at risk). This measure describes those elements in the population that are potentially affected by one or more exceedances of the pertinent chemical per year or other unit of time. 3. Repetitious population exposure. This measure is based on the frequency of exceedances per (suitable) time period. Interpolation methods and spatial data fitting methods are used to generate spatially continuous (including fractional) values of the exceedances for a particular chemical. The isopleths of exceedances are plotted on population density maps to obtain population-weighted exceedence values. An aspect of the method, so far as exposure assessment is concerned, is that the distribution of exposure (measured by concentrations) is either purely exponential: f(c) = { 1 - exp[-c/Q>]}, (1) where < . > denotes the expected value, for instance, the annual average concentration, and c is concentration, or is a functional form, such as tanh, that involves exponentiais. The use of a known statistical distribution permits an analyst to relate particular exposure averaging times (e.g., 1 h, 4 h, and so on) as a multiple of the annual average concentration. Beer and Walsh (1997) produced Australia-wide results for lead, sulfur dioxide, carbon monoxide, particulate matter, nitrogen dioxide, and ozone. The estimates of persons affected, known as high end exposure estimates @IEEE), were poorly fitted by the exponential function, suggesting instead:

892

T. Beer and P.F. Ricci

y

=

4 [l

- tanh (k {c - c,})]

(2)

The parameter of this function is k, such that the expected value of c is l/k. For the persons-affected measure, the function selected is the hyperbolic tangent. In the domain (1, -I), this function has two parameters: a non-central parameter cL and the shape parameter k. The exposure-response function chosen for the measure persons-exposed is exponential. This was determined by applying these functions to a number of Australian cities and determining that the statistical measure of fit was satisfactory for all cases considered. The extrapolating function for the exposure likely to be found in areas where the data are sparse is also exponential. Thus, whether the repetitious exposure or the high-levels exposures are being assessed, the general shape of the distribution functions is the same. What changes is its parameterized form of the distribution: each area will have a different estimated parameter value. The computer software developed by Beer and Durre (199 1) uses a combination of interpolation and boundary conditions because some Australian cities have poor monitoring coverage in their outer regions and the direct extrapolation of inner city monitoring data would over-estimate the exposures. Ricci and Beer (1997) and Beer and Ricci (1998) used these boundary values which were set at 8 points; the corners and the mid-sides of the rectangular region of study. These are the results used in this paper. Kriging, a form of estimation for spatial analysis, was also used but the results were not significantly different from spatial interpolation. More comprehensive spatial modeling can be done to verify that the spatial interpolations are reasonable through correlations between local pollutant levels and nearby population density. In the Beer and Walsh (1997) study, for instance, the population density was averaged over a 10 km by 10 km region around each monitoring station. This correlation was then applied over the entire airshed to predict the frequency of exceedance at any point. This approach may not work for ozone, which is often negatively correlated with local population, because of atmospheric phenomena such as scavenging. However, for the other pollutants where the correlations are reasonable (e.g., CO), this method is satisfactory.

Consider, for instance, that for an area with a constant number of sources of pollution, the number of persons exposed will decrease, as the concentrations increase, and that the probability of response increases as the concentrations increase. Thus, the health risk at any specific concentration is calculated by the product of the probability of being exposed times the probability of being affected, given exposure. The cumulative distribution of exposure takes the value 1.0, at zero concentration, and tends to zero, as the exposure increases. The exposure measure is either personevents per year or persons-affected per event. Multiplying the population and the number of events for each square in the grid, and then summing them over all grids, for a particular area or air basin, calculates the person-events per year. The persons-affected measure is found by calculating the number of event per year per grid, with values greater than one set to be one event/y, values less than one were set to equal zero. The multiplication and summation over all grids then followed, as done for the persons-exposed measure. Exposure-response functions for nitrogen dioxide, carbon monoxide, and particulate matter A linear exposure-response function (a cumulative distribution function) is part of the portfolio of functions likely to be used in the risk assessment. This function is bounded between 0 and 1, with the zero value being associated with either a toxicological measure of response, such as the No Observed Adverse Effect Levels (NOAEL) perhaps suitably divided by a safety factor, or an epidemiological measure such as the relative risk or the odds ratio. This exposure-response function therefore admits a threshold. Pharmaco-kinetic models can be used to determine biologically effective dose (in units of mg/kg*d), from exposure. In this work, exposure-response, rather than dose-response functions, were used in part to keep the example coherent with respect to the exposure measure. Nevertheless, this method can use dose-response models, when those are available. Although the literature review suggests a number of possible exposure-response models, in the examples, the simplest ones were used, namely the linear. This is in part due to the fact that, in Australia, there are insufficient data on air pollutants in Australian cities to justify the use of more complicated models and to apply them preferentially to functions found in the literature. For this study, an adequate first approximation a linear relation between the response (r) and

Assessing aggregate population health risks

893

the dose, c were taken. The pollutant-specific aspects of the exposure-response models used in this paper are discussed next.

Nitrogen dioxide. Small changes in pulmonary function in asthmatic individuals, as the USEPA (1995) has concluded, have occurred at concentrations between 0.2 and ,0.5 ppm @L/L) but not at much higher concentrations, even up to 4 ppm (uL/L). The USEPA (1995) notes that, in healthy individuals, there is no evidence of lung function decrements or change in airway responsiveness at concentrations below 1.Oppm @L/L) NO,. Despite the uncertainties that surround responses at low concentrations, adverse health effects, such as pulmonary edema or death, have been reported following accidental short-term exposure of humans to 150 ppm @L/L) of NO, or greater, as stated by the USEPA (1995). Animal studies provide evidence of emphysema caused by long-term exposures to greater than 8 ppm @L/L) of NOz. The assumed concentration-response relationship used in this analysis is that of a linear function with no threshold at low concentration. For l-h exposure, it is assumed that all subjects exhibit acute health effects at 150 ppm (FL/L) so that Y(C) = c/150 for c < 150 ppm (uL/L)

(3a)

r(c) = I for c 2 150 ppm @L/L)

(3b)

For 24 h exposure,

The fetus and newborn infant appear to be particularly susceptible to even minor increases in COHb levels above 20 g/L, from EPA Victoria (1976) and USEPA (1992). In normal persons, COHb levels of around 200 g/L are associated with headaches, and loss of perceptive function, especially if the exposure is prolonged over extended periods such as 8 h or more, following the EPA Victoria (1976) and USEPA (1992). On these results, these concentration-response functions were used: r(c3=c’/lS-

l/9

(5)

(6) r(cJ= lforc’>20

(7)

where c ‘is the COHb level in percent COHb. The relationship between CO and COHb at continuous exposure is given by Bascom et al. (1996) as: c ‘= 0.16~

(8)

where c ‘is the COHb level in %, and c is the CO concentration in ppm. This relationship does not allow for non-steady state COHb at various durations of exposure. Results using a non-linear model are given in USEPA (1992). The assumed relations between COHb and CO are depicted in Table 2.

these two functions were used:

r(c) = c/8 for c < 8 ppm @L/L)

Table 2. Relations between COHb (g/L) and CO concentration (ppm or pL/L).

(4a)

r(c) = I for c 2 8 ppm @L/L)

Carbon monoxide. The health effects of carbon monoxide are reported as percentage of blood carboxyhemoglobin (COHb) levels. Natural levels can reach 10 g/L, or even 15 g/L, in certain anemic individuals; healthy smokers in urban environments have up to 20 g/L COHb content, as suggested by EPA Victoria (1979) and USEPA (1992). No discernible physiological or psychometric effects have been demonstrated at these levels. Between 25 g/L and 50 g/L COHb content, there is evidence of an increasing incidence of angina pectoris in those with coronary artery disease. Above 30 g/L COHb, there is also increasing psychometric dysfunction. Elevated COHb is recognized as a significant cause of lower birth weight.

Moderate activity COHb level

1-h exposure

20

24

200

240

8-h exposure 12.5 125

Particulate matter. Deck et al. (1996) conducted

a particulate matter risk assessment for Philadelphia and Los Angeles. The concentration-response functions used in the analysis were empirically estimated relationships between average ambient concentrations of the pollutant of interest (PM,, or PM,,,) and the health endpoints of interest reported by epidemiological studies. Deck et al. ( 1996) emphasize that reporting the results as “a 10 ug/m’ increase in PM,, results in a 1% increase in daily mortality” assumes that the concentration-response function is exponential

T. Beer and P.F. Ricci

894

r(c) = 40) exp(74

(9)

The expository human health review earlier in this paper notes that there are a large number of studies that have found significant associations between particulate matter and daily mortality from respiratory and cardiovascular causes. The strength of these findings is the consistency of the results, despite the stated concern that the relationship lacks biological plausibility. We have followed Deck et al. (1996) and used a relative risk of 1.025 (a 2.5% increase) for mortality as a result of a 50 pg/m3 increase in PM,,, using a l-d averaging time - and a relative risk of 1.068 for long-term (> 2 d) averaging times. The relative risk (RR) for a concentration change is:

(10) So that y = 4.9 x 1O-’was used. The value of r(o) was determined on the basis of the baseline health effects incidence rates reported by Deck et al. (1996). Annual mortality rates for Philadelphia (1280 per 100 000 general population per year, 1280/10-‘), Los Angeles (667 x 10m5),and the U.S. National Average (830 x 10”) were used. For this study, it would be more appropriate to use the Australian annual mortality rate of 714 deaths per 100000 people. We have thus used mortality as our endpoint and assumed: r(c) = 714 x 10” exp((4.9

x 10a)c)

(11)

with no threshold or upper bound for c -the PM,, concentration in pg/m3.

In every instance that was developed, there is a concentration at which the health risk has a maximum value. The results for nitrogen dioxide, ozone, and particles, using this method to a sample of Australian exposure data are as follows. The examples are limited for brevity because the calculations and results for other pollutants can be found in Ricci and Beer (1997). As an example of the results obtained, the human health risks related to the present situation with respect to NO, can reach 2 x 10” for the high-end exposure estimate associated with 24 h exposure to NO,. Human health risks related to the present situation with respect to 0, can reach high values of 10” for the I-IEEE, especially for DFEV > 10%. This means that there is about a 15% annual chance of an urban resident being subject to at least one ozone episode that results in a DFEV > 10%. When we calculate the human health risks related to mortality associated with PM,,, we obtain a maximum of about 7 x 10”. Under normal conditions, health risk values greater than low6(the “one in a million excess lifetime risk”) are a possible cause for policy concern. Although the “one-in-a-million” increased risk is limited to the context of cancer risk assessment, it is a probabilistic number and therefore can be used with any exposure-response model that is a distribution function. Because the high value determined is due to the exponential concentration-response relationship, it is critical that, in any policy application of this method, the most accurate exposure-response function is used. The repetitious exposure assessment yields probabilities, pR, that vary as exp(-kc). The concentrationresponse relationship for particles was taken as being exponential such that r(c) = r(o) exp (yc). The product of the two yields the human health risk function: PH

RESULTS FROM APPLYING ASSESSMENT

THE METHOD:

=

Pa&)

RISK

The evaluation of the health risk combines the exposure assessment with that of the concentration-response curves, by multiplying them. The result is a function that expresses the actual number of people likely to be affected, as shown in Fig. 1. Tables 3 and 4 display the summary of these maximum values as the maximum risks and the concentrations at which the risks actually occur. The health endpoints indicated by DFEV refer to a decreased forced expiratory volume, whereas FEP stands for free erythrocyte porphyrin.

=

r(o) exp(1~- klc)

(12)

In this context, the argument that small probability numbers are not meaningful and are equivalent to zero is incorrect. In addition, attention is drawn to the fact that the peak of the distribution, the value, c,, at which maximal health effects may be expected to occur as a result of the time series of air quality variables that arose as a result of the present regulatory regime. Such peak will exist no matter how small the value of the probability.

Assessing aggregate population health risks

895

robabilily

Concenlralion-response function Exposure probabilily

Maximum heallh risk

1.5

2

3

Concentraticit5

3.5

4

4.5

Fig. 1. Health risk assessment methodology. Table 3. Maximum values of health risk for repetitive exposure.

Exposure time Pollutant

03

Concentration at maximum

(h)

Health endpoint

Maximum health risk

1

DFEV > 10%

6.19 x 1O-6

0.079

(ppm or WL)

0,

1

DFEV > 20%

5.75 x 10-7

0.098

0,

8

DFEV > 10%

3.16 x 1O-3

0.052

0,

8

DFEV > 20%

1.08 x lOA

0.061

NO,

1

Acute

2.24 x lo-’

0.009

NO,

24

Acute

3.55 x lOA

0.008

SO,

1

4.33 x lOA

0.267

so*

24

co co PM,, Pb

Bronchospasm Mortality

1.54 x 10”

1

Heart disease

1.85 x lo-”

25.1

8

Heart disease

6.18 x 10-7

13.9

Mortality FEP

3.46 x 1O-3 1.81 x lo-”

19 Wm3) 6.3 (w/m’)

24 (monthly)

The primary question to be answered relates to the size of the affected population and to the magnitude of the peak concentration. The question becomes: Is it acceptable for a population c$, to be affected to some degree, measured by a probability number, and to suffer certain symptoms? If the answer to this question is yes -because the value of 4, is sufficiently small then no changes are required. However, if the answer to the question is no, then the decision maker has established a coherent case for more stringent air quality guidelines or standards.

0.0029

The second question is: Is the level c, above or below current regulatory guidelines? If c, is above current regulatory guidelines, then maximal health effects occur only during exceedances, and it may then be argued that there is no need to alter the guideline values. The need is to diminish the number of exceedances. But if such exceedances are already rare events, then c, will be below present regulatory guideline (or standard) values and consideration will need to be given to reducing these guideline values.

T. Beer and P.F. Ricci

896

Table 4. Maximum values of health risk for population affected.

Health endpoint

Maximum health risk

NQz NQz SO* SO, co co

1 1 8 8 1 24 1 24 1 8

DFEV > 10% DFEV > 20% DFEV > 10% DFEV >20% Acute Acute Bronchospasm Mortality Heart disease Heart disease

1.29 x 10-Z 1.44 x 10-3 1.35 x 10-I 8.92 x 10-* 2.09 x lOA 2.28 x lo-’ 1.45 x lo4 1.24 x 10“ 9.52 x 10-r 4.28 x 10d

0.082 0.098 0.051 0.044 0.043 0102 0.258 0.012 25.3 13.2

PM,, Pb

24 (monthly)

Mortality FEP

1.79 x 10-Z 2.4 x 1O-26

45 @g/m’) 6.2 (ug/m3)

Polhmmt 0, 03 03 03

Table 5. Australian Ambient Air Quality Standards, July 1998.

Pollutant Ozone Nitrogen dioxide Sulfur dioxide Sulfur dioxide Sulfur dioxide Carbon monoxide Particles as PM,, Lead

Averaging time lh 4h Ih LY lh Id JY 8h Id IV

Concentration (ppm or uL/L) 0.10 0.08 0.125 0.03 0.20 0.08 0.02 9 50 Wm3) 0.5 (w/m’)

The variability of the estimates on the final results can be described using Monte Carlo simulations. These simulations calculate the variability about the estimated coefficients (statistical parameters) of the functions used in this methodology. Alternatively, the researcher can use, when appropriate, the large sample (asymptotic) statistical properties of the estimators used in determining the parameters of the functions. To put the concentrations depicted in Tables 3 and 4 into perspective, Table 5 depicts the 1998 Australian National Environment Protection Measures (Standards) for Ambient Air Quality as formulated by the National Environment Protection Council (NEPC 1998). Sensitivity

Concentration at maximum (ppm or uLW

Exposure time (h)

analysis

The purpose of simple deterministic sensitivity is to study the effect that changes in the coefficients of the

functions used in assessing risks have on the risk results. Two aspects of deterministic sensitivity analysis used to study the effect that changes in the coefficients of the risk function have on the risk results are discussed. The choice of the NOAEL determines the concentration of maximum health risk; ifthere is no-threshold value for the exposure-response function, then the risk will be high. This is true for particulate matter (24-h PM,,) for which the linear exposure-response had no threshold. In this case, the maximum health risk of mortality for the population exposed is greaterthan 1W3, and for the population exposed it is approximately 2.0 x lo-*. The value of k and c,_ can be changed to reflect the characteristics of each area being studied. Using the l-h NO, values, for which there is uncertainty about the value of the NOAEL, the population exposure assessment yields k = 0.12 ppb (nL/L) and l/k = 9 ppb (nL/L). If the NOAEL is changed to 200 ppb, the maximum health risk occurs at 0.21 ppm (nL/L), but the maximum health risk is now 2.0 x 10-15. CONCLUSIONS

On the basis of this study, the method that was developed and the results obtained were: 1. Show all of the possible values for exposure, based on currently available information and scientific judgments. The method developed in this paper is practical and transparent; it can be used to investigate changes in risk by alternative choices of exposure distributions and exposure-response models.

Assessing aggregate population health risks

2. Allow the decision maker to understand the implications of choices of risk values such as the “excess deaths per million individuals” by making explicit the full set of distributions inherent to that choice, rather then relying on single numbers such as the expected value and variance. 3. Are independent of the choice of exposure-response and dose-response model. Thus, as new public health information becomes available suggesting a more plausible model than that used, the method developed can directly incorporate it within its structure. The method is practical, transparent, and corresponds with the probabilistic aspects of health risk assessment. It can be used to investigate changes in risk by alternative choices of exposure distributions and exposureresponse models. An extension to the method developed and exemplified in this paper would include multivariate exposure-response functions and multiple dependencies in the structure of the exposure-response models through Bayesian networks. Finally, probabilistic sensitivity analysis, after simple deterministic sensitivity analysis should be part of the full implementation of the method that is described in this paper. REFERENCES Aronow, W.S.; Ferlinz, J.; Glauser, F. Effect of carbon monoxide on exercise performance in chronic obstructive pulmonary disease. Am. J. Med. 63: 904-908; 1977. Ballester, F.; Corella, D.; Perez-Hoyos, S.; Hervas, A. Air pollution and mortality in Valencia, Spain: A study using the APHEA methodology. J. Epidemiol. Community Health 50: 527-533; 1996. Bascom, R.; Bromberg, P.A.; Costa, D.L. Health effects of outdoor pollution. Am. J. Respir. Crit. Care Med. 153: 477-498; 1996. Beer, T.; Durre, M. Mesoscale extension of station observations to field locations. In: Cheney, N.P.; Gills, A.M., eds. Proc. conference on bushtire modelling and tire danger rating systems. 199 1: 91-100. Available from: Commonwealth Scientific Research Organization, Canberra, Australia. Beer, T.; Walsh, S. Ambient Air Quality National Environment Protection Measure (NEPM) - Exposure assessment. Report SB/1/297F3C. Victoria, Australia: CSIRO Division of Atmospheric Research; 1997. Beer, T.; Ricci, P.F. A method for population-based non-cancer health risk assessment. Proc. 14* international clean air and environment conference. 1998: 12 I- 124. Available from: Clean Air Society of Australia and New Zealand, Melbourne, Australia. Coherssen, J.J.; Covello, V.T. Risk analysis: A guide to principles and methods for analyzing health and environmental risks. United States Council on Environmental Quality. Washington, D.C.: National Academy Press; 1989. Deck, L.; Post, E.; Wiener, M.; Dunningham, K. A particulate matter risk assessment for Philadelphia and Los Angeles. Report to USEPA. Cambridge, MA: Abt Associates; 1996.

897

Dockery, D.W.; Pope, A. III; Xu, X.; Spengler, J.D.; Ware, J.H.; Fay, M.E.; Ferris, B.G.; Speizer, F.E. An association between air pollution and mortality in six U.S. cities. N. Engl. J. Med. 329: 1753-1759; 1993. EPA (Environmental Protection Authority) Victoria. Draft state environment protection policy for the air environment of Victoria and explanatory document. Melbourne, Australia: Environment Protection Authority of Victoria; 1979. Euler, G.L.; Abbey, D.E.; Hodgkin, J.E.; Magie, A.R. Chronic obstructive pulmonary disease symptom effects of long-term cumulative exposure to ambient levels of total oxidants and nitrogen dioxide in California Seventh-day Adventist residents. Arch. Environ. Health 43: 279-285; 1988. Gratt, L.B. Air toxic risk assessment and management: Public health risk from normal operations. New York, NY: Van Nostrand Reinhold; 1996. Harrington, W.; Krupnick, A.J. Short-term nitrogen dioxide exposure and acute respiratory disease in children. J. Air Pollut. Control Assoc. 35: 1061-1067; 1995. HE1 (Health Effects Institute). Particulate air pollution and daily mortality: Replication and validation of selected studies (Phase I report of the particle epidemiology evaluation project). Cambridge, MA: HEI; 1995. Imay, M.; Yoshida, K.; Kitabatake, M. Mortality from asthma and chronic bronchitis associated with changes in sulfur oxides pollution. Arch. Environ. Health 4 1: 29-35; 1986. Lippmann, M.; Lioy, P.J.; Leikauf, G.; Greene, K.B.; Baxter, D.; Morandi, M.; Pasternak, B.S.; Fife, D.; Speizer, F.E. The effects of ozone on the pulmonary fucntion of children. Adv. Mod. Environ. Tox. 5: 423-446; 1983. Lippmann, M. Health effects of ozone: A critical review. J. Air Waste Manage. Assoc. 39: 672-7120; 1989. Lippmann, M.; Ito, K. Separating the effects of temperature and season on daily mortality from those of air pollution in London: 1965-1972. Inh. Tox. 7: 85-97; 1995. Moolgavkar, S.H.; Luebeck, E.G.; Hall, T.A.; Anderson, H.R. Air pollution and daily mortality in Philadelphia. Epidemiology 6: 476-484; 1995. Moolgavkar, S.H.; Luebeck, E.G. A critical review of the evidence on particulate air pollution and mortality. Epidemiology 5: 420428; 1996. NECP (National Environment Protection Council).Australian National Environment Protection Measures (Standards) for Ambient Air Quality. Adelaide, S.A. Australia: NEPC; 1998. Pope, C.A. III; Schwartz, J.; Ransom, M.R. Daily mortality and PM,,, pollution in Utah Valley. Arch. Env. Health 47: 211-217; 1992. Pope, C.A. III; Kanner, R.E. Acute effect of PM,, pollution on pulmonary function of smokers with mild to moderate chronic obstructive pulmonary disease. Am. Rev. Resp. Diseases 147: 1336-1440; 1993. Ricci, P.F.; Wyzga, R. An overview of cross-sectional studies of mortality and air pollution and related statistical issues. Environ. Int. 9: 177-194; 1983. Ricci, P.F. Principles of risk assessment. Englewood-Cliffs,-NJ: Prentice-Hall; 1985. Ricci, P.F.; Catalano, J.; Kelsh, M. Time series (1963-1991) of mortality and ambient air pollution in California: An assessment with annual data. Inh. Tox. 8: 95-106; 1996. Ricci, P.F.; Ospital, J. Cancer, cardiovascular& chronic obstructive pulmonary mortality and ambient air pollution: A study of 11 California counties, 1963 to 1992, using monthly data. 2nd

898

Colloquium on Particulate Air Pollution & Health, Park City, UT. l-3 May 1996. Available from: UC Irvine, Irvine, CA. Ricci, P.F.; Beer, T. Quantitiative risk assessment of six air pollutants for the National Environmental Protection Council (NEPC). Unpublished Report. New South Wales, Australia: University of Wollongong; 1997. Roemer, W.; Hoek, G.; Brunekreef, B. Effect of ambient winter pollution on respiratory health of children with chronic respiratory symptoms. Am. Rev. Resp. Disease 147: 118-124; 1993. Saldiva, P.H.N.; Lichtenfels, A.J.F.C.; Paiva, P.S.O.; Barone, I.A.; Martins, M.A.; Massad, E.; Pereira, J.C.R.; Xavier, V.P.; Singer, J.M.; Bohm, G.M. Association between air pollution and mortality due to respiratory diseases in children in Sao Paula, Brazil: A preliminary report. Environ. Res. 65: 218-225; 1994. Saldiva, P.H.N.; Pope, C.A. III; Schwartz, J.; Dockery, D.W.; Lichtenfels, A.J.; Salge, J.M.; Barone, I.; Bohm, G.M. Air pollution and mortality in elderly people: A time-series study in Sao Paulo, Brazil. Arch. Environ. Health 50: 159-163; 1995. Scarlett, J.F.; Griffrths, J.M.; Strachan, D.P.; Anderson, H.R. Effect of ambient levels of smoke and sulphur dioxide on the health of a national sample of 23 year old subjects in 1981. Thorax 50: 764-768; 1995. Schwartz, J.; Dockery, D.W. Increased mortality in Philadelphia associated with daily air pollution concentrations. Am. Rev. Resp. Diseases 145: 600-604; 1992. Schwartz, J.; Slater, D.; Larson, T.V.; Pierson, W.E.; Koenig, J.Q. Particulate air pollution and hospital emergency room visits for asthma in Seattle. Am. Rev. Resp. Diseases 147: 826-831; 1993. Schwartz, J. Air pollution and hospital admission for the elderly in Birmingham, Alabama. Am. J. Epidemiol. 139: 589-598; 1994a.

T. Beer and P.F. Ricci

Schwartz, J. PM,,, ozone, and hospital admissions for the elderly in Minneapolis-St. Paul, Minnesota. Arch. Environ. Health49: 366374; 1994b. Schwartz, J. Short term fluctuations in air pollution and hospital admissions of the elderly for respiratory disease. Thorax 50: 53 l538; 1995. Sunyer, J.; Saez, M.; Murillo, C.; Castellsague, J.; Martinez, F.; Anto, J.M. Air pollution and emergency room admissions for chronic obstructive pulmonary diseases: A 5-year study. Am. J. Epidemiol. 137: 701-705; 1993. USEPA (United States Environmental Protection Agency). Air quality criteria for lead. Report EPA-600/8-83/028aF. Research Triangle Park, NC: Offtce of Research and Development; 1986. USEPA (United States Environmental Protection Agency). Review of the national ambient air quality standards for carbon monoxide: Assessment of scientific and technical information. Report EPA-452/R-92-004. Staff Paper. Research Triangle Park, NC: Office of Air Quality Planning and Standards; 1992. USEPA (United States Environmental Protection Agency). Review of the national ambient air quality standards for nitrogen dioxide - Assessment of scientific and technical information. Report EPA-452/R-95-005. Staff Paper. Research Triangle Park, NC: Office of Air Quality Planning and Standards; 1995. Walsh, S.; Beer, T. A new methodology for the assessment of exposure to ambient air pollution in Australia. Proc. 14* international clean air and environment conference. 1998: 237-242. Available from: Clean Air Society ofAustralia and New Zealand, Melbourne, Australia. Walters, S.; Griffiths, R.K.; Ayres, J.G. Temporal association between hospital admissions for asthma in Birmingham and ambient levels of sulphur dioxide and smoke. Thorax 49: 133140; 1994.