Human-Toxicological Effect and Damage Factors of Carcinogenic and Noncarcinogenic Chemicals for Life Cycle Impact Assessment

Transcription

1 Integrated Environmental Assessment and Management Volume 1, Number 3 pp Ó 2005 SETAC 181 Human-Toxicological Effect and Damage Factors of Carcinogenic and Noncarcinogenic Chemicals for Life Cycle Impact Assessment Mark A.J. Huijbregts,*À Linda J.A. Rombouts,À Ad M.J. Ragas,À and Dik van de MeentÀ` ÀDepartment of Environmental Science, Institute for Wetland and Water Research, Faculty of Science, Radboud University Nijmegen, 6500GL, Nijmegen, The Netherlands `National Institute of Public Health and the Environment, Laboratory for Ecological Risk Assessment, 3720 BA, Bilthoven, The Netherlands (Received 1 October 2004; Accepted 2 February 2005) ABSTRACT Chemical fate, effect, and damage should be accounted for in the analysis of human health impacts by toxic chemicals in lifecycle assessment (LCA). The goal of this article is to present a new method to derive human damage and effect factors of toxic pollutants, starting from a lognormal dose response function. Human damage factors are expressed as disability-adjusted life years (DALYs). Human effect factors contain a disease-specific and a substance-specific component. The disease-specific component depends on the probability of disease occurrence and the distribution of sensitivities in the human population. The substance-specific component, equal to the inverse of the ED50, represents the toxic potency of a substance. The new method has been applied to calculate combined human damage and effect factors for 1,192 substances. The total range of 7 to 9 orders of magnitude between the substances is dominated by the range in toxic potencies. For the combined factors, the typical uncertainty, represented by the square root of the ratio of the 97.5th and 2.5th percentile, is a factor of 25 for carcinogenic effects and a factor of 125 for noncarcinogenic effects. The interspecies conversion factor, the (non)cancer effect conversion factor, and the average noncancer damage factor dominate the overall uncertainty. Original Research Keywords: Human toxicity Chemicals Effect factors Damage factors Life-cycle impact assessment INTRODUCTION Life-cycle assessment (LCA) deals with the environmental impact of product systems (Consoli et al. 1993). It considers the life cycle of a product from resource extraction to waste disposal. LCA starts from a definition of the so-called functional unit of product. In the inventory analysis, the releases and resource extractions that occur as a consequence of the production of 1 extra functional unit of product are obtained. In a subsequent impact assessment, the extra impact of these emissions and extractions are quantified. This step is called life-cycle impact assessment. Life-cycle impact assesment also includes the damage of toxic emissions to human health (Figure 1). For the analysis of human health impacts by toxic chemicals, life-cycle impact assessment comprises a fate analysis step, in which the marginal increase in human intake is calculated from the increase in release; an effect analysis step, in which the increase of effect per unit of dose increase is assessed; and a damage analysis step, in which the increase in damage per unit of effect increase is included. Fate calculations are usually done by means of a multimedia mass-balance model of the Mackay type combined with a human-exposure model (Hertwich et al. 2002). For the effect analysis, 2 es of methods can be identified. One of effect methods assumes a linear dose response relationship at low-dose levels (see Krewitt et al. [2002] for an overview). For instance, Huijbregts et al. (2000) assumed linearity in responses at human intake below the acceptable daily intake for noncarcinogenic substances and below the virtual safe dose for genotoxic and carcinogenic substances. Recently, a linear dose response method has been proposed by Pennington et al. * To whom correspondence may be addressed m.huijbregts@science.ru.nl (2002) and Crettaz et al. (2002), which assumes linearity below the effect dose affecting 10% of the individuals (ED10). The other of methods explicitly accounts for the nonlinearity in dose response curves by calculating the marginal change in human response because of a marginal dose-intake change in an ambient-background situation. In ecotoxicology, the main representative of this of methods is the potentially affected fraction of species, reflecting the fraction of all species that are exposed above the no-observed effect concentration (Klepper et al. 1998). For human health effects, however, effect factors accounting for the nonlinearity in dose response curves have not been derived. Human health damage factors are generally based on human health statistics on life years lost and disabled, referred to as disability-adjusted life years (DA- LYs), for various diseases such as skin cancer (Hofstetter 1998). DALYs have been reported for a number of cancer types (Hofstetter 1998; Frischknecht et al. 2000; Crettaz et al. 2002). However, DALY estimates for noncancer-effect types suitable for LCA are generally scarce, except for respiratory effects (Hofstetter 1998). The goal of this article is to present a new methodology to derive human damage and effect factors of toxic pollutants, starting from a lognormal dose response function. The new method has been applied to calculate combined human damage and effect factors for 1,192 substances. Uncertainties in the damage and effect factors are also reported. Finally, differences found in effect and damage factors are discussed in relation to the uncertainties involved. METHODOLOGY Calculation procedure Hofstetter (1998), the concept of DALY is a powerful concept to address human health damages in LCA. The overall

2 182 Integr Environ Assess Manag 1, 2005 M.A.J. Huijbregts et al. Figure 1. General structure of the life-cycle impact assessment framework for damage to human health (adapted from Jolliet et al. 2004). human population damage, expressed as DALY caused by a number of diseases (DALY ov ), can be estimated by Damage ¼ DALY ov ¼ N pop X DALY e R e ð1þ e where N pop is the total population number, DALY e is the DALY for disease type e, and R e is the probability of occurrence of disease type e in the human population. Taking DALY ov as a measure of overall human population damage, the human-toxicological characterization factor for substance x, emitted to compartment i, and taken in by exposure route r, CF x,i,r, can be obtained by differentiation of this equation (Huijbregts et al. 2002; Van de Meent and Huijbregts 2005). The derivative ]Damage/]Emission can be split into 3 factors: ]Damage/]Effect, describing the change in damage per type of effect added in the human population; ]Effect/]Intake, describing the amount of effect per unit of intake by the human population; and ]Intake/]Emission, describing the amount of population intake per unit of emission: CF x ¼ DF x EF x FF x ¼ ]Damage ]Ef f ect ]Ef f ect ]Intake ]Intake ]Emission ð2þ where DF is the damage factor, EF is the effect factor, and FF is the fate factor of pollutant x. The effect factor ]Effect/]Intake can be further split into 2 factors: ]Ef f ect ]Intake ¼ ]Ef f ect ]ToxicUnit ]ToxicUnit ð3þ ]Intake where ]Effect/]ToxicUnit is the mode of the action-specific part of the effect factor, describing the change in the occurrence of the human health effect e per added relative dose (toxic unit) of chemicals with the same mode of action; and ]ToxicUnit/]Intake is the substance-specific part of the effect factor, describing the change in toxic unit per unit of intake via inhalation and/or ingestion. The fate factor ]Intake/]Emission can be directly calculated with fate and exposure models. Detailed information on the application of fate models in LCA context can be found elsewhere (Hertwich et al. 2001; Hertwich et al. 2002; Huijbregts et al. 2005). Uncertainty factors In the damage and effect factor calculations, we accounted for parameter uncertainty and lack of information concerning the human disease caused by a pollutant. Parameter uncertainty is quantified by Monte Carlo simulation, propagating known parameter uncertainties into an uncertainty distribution of the output variable. A lognormal uncertainty distribution was chosen to represent uncertainty in parameter values because it avoids negative values, it captures a large range of values, and the uncertainty in many parameters follows a skewed distribution (Slob 1994). Parameteruncertainty estimates in input data were represented by an uncertainty factor k i, which is defined so that 95% of the values of a stochastic variable (X) are within a factor k i from the median M(X) (Slob 1994). To assign a damage factor and a mode of the action-specific part of the effect factor to a toxic pollutant, information on the human disease type caused by the pollutant should be available. Particularly for chemicals causing noncarcinogenic effects, however, this information is not available for humans. A nonparametric, bootstrapping procedure was applied to address this type of uncertainty and combined with the Monte Carlo simulation to quantify the resulting overall output uncertainty (Huijbregts et al. 2003). In the bootstrap simulation, a probability is assigned to each damage factor and mode of the action-specific part of the effect factor based on the incidence rates of the diseases connected to the damage and mode of the action-specific part of the effect factors. The uncertainty in the output variables is summarized by sffiffiffiffiffiffiffiffiffiffiffi 97:5p k o ¼ 2:5p where k o is the uncertainty factor of the output variable, such as DF and EF; and 97.5p and 2.5p are, respectively, the 97.5th and 2.5th percentile of the output distribution. Note that the k o is described in the same way as k i, except that k o does not involve the assumption of a lognormal uncertainty distribution. ]DAMAGE/]EFFECT The damage factor ]Damage/]Effect can be calculated by ]Damage ¼ ]DALY N pop X ðdaly e ]R e Þ ov e ¼ ]Ef f ect N pop ]R e;p N pop ]R e;p X ðdaly e X ]R e;p Þ e p ¼ ¼ DALY e ð5þ ]R e;p where R e,p is the probability of occurrence of disease type e in the human population caused by mode of action p. Damage factors were derived from the extensive burden of disease and health statistics provided by Murray and Lopez (1996a, 1996b) on a world level for Applying equal weightings for the importance of 1 y of life lost for all ages and no discounting for future damages, DALY e is the sum of years of life lost (YLL e ) and years of life disabled (YLD e ) caused by disease type e: DALY e ¼ YLL e þ YLD e ð6þ Table 1 shows the YLLs, YLDs, and DALYs for 49 noncommunicable diseases, representative for the world in Following Hofstetter (1998), we assumed that the YLL estimate of every disease has an uncertainty factor k i of 1.4, and the YLD estimate of every disease an uncertainty factor k i of 2. Uncertainty factors k o of the DALY estimates are shown ð4þ

3 Human Effect and Damage Factors for LCIA Integr Environ Assess Manag 1, Figure 2. The relative dose response relationship for r log of 0.26 and in Table 1. The average DALY, weighted by incidence cases, of carcinogenic effects is 11.5 y lost with an uncertainty factor k o of 2.8. The average noncancer DALY is 2.7 y lost with an uncertainty factor k o of Compared with Crettaz et al. (2002), the average DALY for carcinogenic substances is about a factor of 2 higher. This is because Crettaz et al. (2002) applied discounted and ageweighted YLLs in their calculation of DALYs for carcinogenic diseases, whereas we derived YLLs without age weighting and discounting. ]EFFECT/]TOXIC UNIT Starting from dose addition ( joint action or simple similar action) for chemicals with the same mode of action (Plackett and Hewlett 1952; Könemann and Pieters 1996; Wilkinson et al. 2000; Chen et al. 2001) and a lognormal human-relative dose response function (Hattis 1996,1997; Renwick and Lazarus 1998), the probability of occurrence of disease type e in the human population caused by mode of action p (R e,p ) can be defined as ZTU p R e;p ¼ pffiffiffiffiffiffiffiffiffi r log;e;p 2 p TU p ln10 e 1 2 logðtupþ log;e;p AdTU p Figure 2 shows the relative dose (TU p ) to cumulative probability of occurrence (R e,p ) relation for 2 different spreads (r log ) in human sensitivity. TU p represents the relative dose in effective toxic units of chemicals for mode of action p: TU p ¼ X ð8þ ED50 x x2p where I xep is the daily, human, dose intake of substance x with mode of action p (mg kg 1 d 1 ), and ED50 xep is the chronic dose of substance x with mode of action p affecting 50% of the human population (mg kg 1 d 1 ). The derivative ]R e,p /]TU p of Equation 7, representing the mode of action specific-part of the human-toxicological effect factor, can be calculated by ]Ef f ect ]ToxicUnit ¼ N pop ]R e;p N pop ]TU p where ]R e,p /]TU p 1 ¼ pffiffiffiffiffiffiffiffiffi r log;e;p 2 p 0 I x2p ð7þ 2 TU p ln10 2 logðtupþ r log;e;p ð9þ e 1 indicates the marginal change in the probability of occurrence of human health effect e by a marginal change in the ambient ToxicUnit of the mode of action p. Dose-addition holds for chemicals or stressors that cause the same critical effect, act on the same molecular target at the same target tissue, and act by the same biochemical mechanism of action or share a common (toxic) intermediate (Wilkinson et al. 2000; Chen et al. 2001). Because of the lack of specific information about the mode of action, we were only able to derive the spread in human sensitivity for the categories carcinogenic effects and noncarcinogenic effects. For carcinogenic effects, Hattis (1997) has estimated an r log of 0.59, based on the observed variability in metabolic activation, detoxification, DNA repair, systemic pharmacokinetics, and uptake variabilities. The uncertainty in r log for carcinogenic effects can be expressed with a factor k i of 1.7 (Hattis 1997). For noncarcinogenic effects, Slob and Pieters (1998) proposed a r log of 0.26 with an uncertainty factor k i of 2.5. The r log of 0.59 and 0.26 were applied for all carcinogenic and noncarcinogenic modes of action, respectively. The statistics for the probability of disease occurrence attributed to a specific mode of action can be converted to ambient effect units at log-scale by taking the inverse of the cumulative normal distribution. This conversion gives the opportunity to use the statistics for the probability of disease occurrence in the calculation of. Figures 3 and 4 show the dependence of ]R e,p /]TU p for carcinogenic and noncarcinogenic substances, respectively, toward the probability of occurrence for typical, small, and large r log estimates. The ]R e,p /]TU p values can range from virtually 0 for relatively low probability of occurrence up to 3 for relatively large probability of occurrence, implying that ]R e,p /]TU p strongly depends on the ambient probability of occurrence and the estimated r log estimates for humans. A restriction in the actual derivation of ]R e,p /]TU p is the lack of information on the mode of action-specific disease probability of occurrences (Seed et al. 1995; Könemann and Pieters 1996; Chen et al. 2001). As a tentative solution, we assumed that the probability of occurrence of disease type e is dominantly caused by only 1 mode of action. This implies that R e,p can be approximated by R e;p R e ð10þ Under this assumption, disease probability of occurrences can be directly used to calculate the mode of action-specific part of the effect factor. In this way, ]R e /]TU effectively becomes disease-specific instead of specific to the mode of action. To derive ]R e /]TU for the 49 disease types identified, we assumed for all carcinogenic effect types, a r log of 0.59 and for all noncarcinogenic effect types, a r log of Disease probability of occurrence statistics were provided by Murray and Lopez (1996b) on a world level for As congenital diseases always occur as a developmental effect, probability of occurrences at birth were used in the ]R e /]TU calculation. Based on Murray and Lopez (1996b), it was assumed that the probability of occurrence has an uncertainty factor k i of 1.5 for all disease types. Table 2 shows the disease-specific part of the effect factor for 49 noncommunicable diseases, representative for the world in The uncertainty factors k o of the ]R e /]TU estimates are also shown in Table 2. The average ]R e /]TU, weighted by incidence cases, of carcinogenic effects is 0.03 with an uncertainty factor k o of 4.2. The average noncancer ]R e /]TU is 0.16 with an uncertainty factor k o of 4.9.

4 184 Integr Environ Assess Manag 1, 2005 M.A.J. Huijbregts et al. Table 1. Yearly incidence cases (INC), years of life lost (YLL e ), years of life disabled (YLD e ), disability adjusted life years (DALY e ), and uncertainty factors for the DALY estimates (k DALY ) per disease type, based on world data in 1990 (Murray and Lopez 1996a, 1996b) Disease type INC YLL e YLD e DALY e k DALY Cancer Mouth and oropharynx cancer Oesophagus cancer Stomach cancer Colon and rectum cancer Liver cancer Pancreas cancer Trachea, bronchus and lung cancer Melanoma and other skin cancer Breast cancer Cervix uteri cancer Corpus uteri cancer Ovary cancer Prostate cancer Bladder cancer Lymphomas and multiple myeloma Leukemia Cancer average Neuropsychiatric conditions Bipolar disorder Schizophrenia Epilepsy Dementia Parkinson disease Multiple sclerosis Obsessive-compulsive disorder Panic disorder Sense-organ diseases Glaucoma Cataract Cardiovascular diseases Rheumatic heart disease Ischemic heart disease Cerebrovascular heart disease Inflammatory heart disease Respiratory diseases Chronic obstructive pulmonary disease Asthma Diabetes mellitus

5 Human Effect and Damage Factors for LCIA Integr Environ Assess Manag 1, Table 1. Continued Disease type INC YLL e YLD e DALY e k DALY Digestive diseases Peptic ulcer Liver cirrhosis Genitourinary diseases Nephritis and nephrosis Benign prostate hypertrophy Musculoskeletal diseases Rheumatoid arthritis Osteoarthritis Congenital anomalies a Abdominal wall defect Anencephaly Anorectal atresia Cleft lip Cleft palate Oesophageal atresia Renal agenesis Down syndrome Congenital heart anomalies Spina bifida Noncancer average a Probability of occurrence of congenital diseases at birth. ]TOXIC UNIT/]INTAKE Toxic potency The ]ToxicUnit/]Intake is equal to the inverse of the chronic dose affecting 50% of the human population of the substance added (ED50), also called toxic potency: ]ToxicUnit ]Intake ED50 ¼ ]TU N pop X p x;r ¼ ]I x2p;r ]I x2p;r ED50 x;r N pop ]I x2p;r ¼ 1 ED50 x2p;r ð11þ For 755 chemicals, the carcinogenic chronic dose affecting 50% of a laboratory species of the substance added (ED50) was obtained from the Carcinogenic Potency Database (CPDB), developed by Gold and Zeiger (1997). ED50s were reported as an average of all exposure routes considered (Gold and Zeiger 1997; Gold 2004). If a substance in the CPDB has a carcinogenic ED50 for more than 1 species, the order of preference is monkey, then dog, rat, hamster, and mouse. In case of noncarcinogenic effects, chronic ED50s were not readily available. Based on dose response data reported in the Integrated Risk Information System (IRIS) of the U.S. Environmental Protection Agency (USEPA 2004), we calculated ingestion ED50s for 12 substances and inhalation ED50s for 9 substances. The ED50 of the binary dose response data was determined by fitting a probit model to the data. The ED50 of the continuous dose response data was determined by fitting a linear dose response model to the data using USEPA s Benchmark Dose Software, version (USEPA 2000). Besides the probit and linear dose response model, many other types of dose response models are available. The sensitivity of the calculated ED50 for the selected dose response model type was investigated by fitting other dose response models to the binary and continuous data. The linear 2-stage model, the c-multihit model, the logit model, and the Weibull model were fitted to the binary data, whereas the polynomial model, power model, and Hill model were used for the continuous data (Huijbregts et al. 2004). It was found that all models produced ED50s within a factor of 2, indicating that the uncertainty resulting from the choice of the dose response model is relatively small compared with other types of uncertainty (Huijbregts et al. 2004). After deriving the ED50 for laboratory species, the lifetime ED50 for humans can be calculated by extrapolation from test species to humans and extrapolation from subacute or semichronic exposure to chronic exposure: ED50 x;r ¼ ED50 a;t;x;r X r LT N ð12þ CF a CF t where ED50 x,r is the lifetime dose (kg) of substance x via exposure route r affecting 50% of the human population, ED50 a,t,x,r is the effect dose (kg kg 1 d 1 or kg m 3 )of

6 186 Integr Environ Assess Manag 1, 2005 M.A.J. Huijbregts et al. Figure 3. Dependency of ]Effect/]ToxicUnit () for carcinogenic substances on the probability of occurrence of a disease for typical, low, and high r log estimates. Figure 4. Dependency of ]Effect/]ToxicUnit () for noncarcinogenic substances on the probability of occurrence of a disease for typical, low, and high r log estimates. substance x via exposure route r affecting 50% of the population test species a for exposure duration t, CF a is the conversion factor for interspecies differences, CF t is the conversion factor for differences in time of exposure, X r is the average body weight of humans (70 kg) or the average breath intake of humans (13 m 3 d 1 ), LT is the average lifetime of humans (75 y), and N the number of days per year (365 d y 1 ). Interspecies conversion factor In cases of oral exposure, the interspecies conversion factor can be calculated (Vermeire et al. 2001) by CF a ¼ BW 0:25 h ð13þ BW a where BW a is the average body weight (kg) of test species a. For exposure via inhalation, a default interspecies conversion factor of 1 was assumed (Vermeire et al. 2001), except for substances for which a substance-specific interspecies factor was available (USEPA 2004). Interspecies conversion factors are applied for both carcinogenic and noncarcinogenic substances. Typical body weights were taken from Vermeire et al. (2001) and Baird et al. (1996). The interspecies conversion factor has an uncertainty factor k i of 19 for both inhalatory and oral exposure (Vermeire et al. 2001). Time conversion factor For noncarcinogenic substances, a distinction is made in 3 periods of exposure: subacute, semichronic, and chronic. The conversion factor CF t for subacute-to-chronic exposure is 5, and for semichronic-to-chronic exposure, the CF t is 2, with an uncertainty factor k i of 12 (Vermeire et al. 2001). The conversion of the carcinogenic toxicity data to chronic conditions was already included in the CPDB, based on the standard lifespan of the experimental species (Gold and Zeiger 1997). Cancer effect conversion factor For substances lacking a carcinogenic ED50 in the CPDB, the carcinogenic, low-dose, slope factor q x was used to estimate the carcinogenic ED50: ED50 x ¼ CF q 1 q ð14þ x where CF q is the 1/q*-to-ED50 conversion factor. The carcinogenic, low-dose, slope factors were taken from IRIS (USEPA 2004), and ED50 values were derived from the CPDB (Gold 2004). The CF q was derived by taking the geometric average of the human-equivalent ED50-to-1/q* ratios of 64 substances (Figure 5). From the data set, a CF q of 0.8 was calculated with an uncertainty factor k i of 47. The CF q of 0.8 is in good agreement with the findings of Krewski et al. (1993) and Crettaz et al. (2002). For 21 substances, a carcinogenic ED50 was calculated using the low-dose, carcinogenic, slope factors q from IRIS (USEPA 2004). Noncancer-effect conversion factor For most of the substances, insufficient data were available to derive a noncarcinogenic ED50 with dose response models. In these cases, the ED50 has been estimated from the no-observed effect level (NOEL) by ED50 x;r ¼ CF N NOEL x;r ð15þ where CF N is the NOEL-to-ED50 conversion factor. The CF N is set equal to the geometric average of a data set of 21 human-equivalent ED50-to-NOEL ratios, derived from information in IRIS (USEPA 2004). Figure 6 shows the comparison between the noncarcinogenic ED50s and the NOELs. The CF N is set at 9 with an uncertainty factor k i of 11. In those cases when a lowest-observed effect level (LOEL) is reported instead of a NOEL, the LOELs are extrapolated to a NOEL by NOEL x;r ¼ CF L LOEL x;r ð16þ where CF L is the LOEL-to-NOEL conversion factor. Pieters et al. (1998) reported a CF L of 0.25 with an uncertainty factor k i of 4. NOELs and LOELs were derived from IRIS (USEPA 2004) and from the World Health Organization (Lu 1995; JMPR 2004). If the World Health Organization and the USEPA both reported a NOEL or LOEL value for the same substance, the World Health Organization information was preferred. NOELs and LOELs for oral exposures were obtained for 463 substances and for inhalatory exposure for 58 substances. Outcomes Table 3 gives a summary of the human-equivalent toxic potencies of the substances included causing, respectively, carcinogenic effects; oral, noncarcinogenic effects; and inhalatory, noncarcinogenic effects. The potency of toxic substances typically centers around 10 1 kg 1. For carcinogenic effects, the toxic potency ranges 8.8 orders of magnitude from kg 1 for 1,1,1,2-tetrafluoroethane to kg 1 for 2,3,7,8-tetrachlorodibenzo-p-

7 Human Effect and Damage Factors for LCIA Integr Environ Assess Manag 1, Figure 5. Comparison between the ED50 and 1/q* of 64 carcinogenic substances. dioxin (2,3,7,8-TCDD). For noncarcinogenic effects after oral exposure, the toxic potency ranges 7 orders of magnitude from kg 1 for chromium III to kg 1 for tetraethyllead. For noncarcinogenic effects after inhalatory exposure, the toxic potency ranges 8.5 orders of magnitude from kg 1 for 1,1-difluoro-1- chloroethane to kg 1 for beryllium. Depending on the number and type of conversion factors applied, uncertainty factors for toxic potencies range from 1 to 127. ]DAMAGE/]INTAKE Damage factors and effect factors can be combined by ]Damage ]Intake ¼ ]Damage ]Ef f ect ]Ef f ect ]ToxicUnit ]ToxicUnit ð17þ ]Intake To obtain combined damage factors and effect factors, information on the (critical) effect of a chemical is needed. For carcinogenic effects, this is not a significant problem because the type of cancers caused by a chemical is reported for the majority of the substances included in this article (Gold and Zeiger 1997). In cases in which the ED50 of a substance represents more than 1 cancer disease, the cancer with the highest DALY is taken as the damage factor. For carcinogenic substances lacking critical-effect information, the average cancer DALY of 11.5 y per incidence case and the average disease-specific part of the carcinogenic effect factor of 0.03 can be used as good alternatives because the uncertainty factors k o of, respectively, 2.8 and 4.2 are relatively low when compared with the uncertainty reported for the toxic potencies of the majority of the carcinogenic substances. For noncarcinogenic effects, however, the situation is more problematic. Standard toxicological-response variables in test species, such as decrease in body weight, are, in most cases, not specific for disease genesis in humans and, therefore, cannot be properly translated to real-life conditions (De Hollander et al. 1999; Owens 2002). Furthermore, DALYs are currently not available for all relevant noncarcinogenic health effects potentially caused by chemical exposure. As a tentative solution, we applied the average noncarcinogenic DALY of 2.7 y and the average diseasespecific part of the effect factor as 0.16 for all chemicals with noncarcinogenic effects. Particularly, the use of an average noncarcinogenic DALY causes substantial error in the combined damage and effect scores (k o of 13.0). Table 4 gives a summary of the combined, human, toxicological damage and effect factors in y per kg of intake Figure 6. Comparison between the ED50 and the NOEL of 21 noncarcinogenic substances. of the substances included causing, respectively, carcinogenic effects; oral, noncarcinogenic effects; and inhalatory, noncarcinogenic effects. The combined damage and effect factor of toxic substances typically centers around 10 2 y kg 1.For carcinogenic effects, the combined damage and effect factor ranges 9 orders of magnitude from y kg 1 for cinnamyl anthranilate to y kg 1 for 2,3,7,8- TCDD. For noncarcinogenic effects after oral exposure, the combined damage and effect factor ranges 7 orders of magnitude from y kg 1 for chromium III to y kg 1 for tetraethyllead. For noncarcinogenic effects after inhalatory exposure, the toxic potency ranges 8.5 orders of magnitude from y kg 1 for 1,1-difluoro-1-chloroethane to y kg 1 for beryllium. Depending on the number and type of conversion factors applied, uncertainty factors for combined damage and effect factors range from 16 to 277. Typical uncertainty factors for combined damage and effect factors are, respectively, 25 and 125 for carcinogenic and noncarcinogenic substances. DISCUSSION Damage factors We were able to calculate damage factors for 16 cancer diseases and 33 noncancer diseases. The main benefit of the inclusion of damage factors in the assessment of human impacts of toxic chemicals in LCA is the possibility for further meaningful aggregation with other impact categories affecting human health, such as ionizing radiation (Frischknecht et al. 2000). The concept of DALYs has proven to be a useful metric in the assessment of human health damages in LCA (Hofstetter 1998). Human damage assessment from toxic chemicals in LCA, however, is not without difficulties because the calculation of DALYs depends on a number of subjective assumptions. First, DALYs are presented for the world in Applying these damage factors in LCA case studies implies that it is assumed that human health damages occurring from life-cycle emissions can be represented by world averages. For LCA case studies focusing on region-specific human health impacts, however, these DALY estimates should be used with care. Taking another region in the world as a starting point for the DALY calculation can cause results to change. For instance, for established market economies in 1990, DALYs are up to a factor of 2 lower for cancer diseases

8 188 Integr Environ Assess Manag 1, 2005 M.A.J. Huijbregts et al. Table 2. Probability of occurrence (R), ]Effect/]ToxicUnit (]E/]TU), and corresponding uncertainty factors (k ]E/]TU ) per disease type, based on world data in 1990 (Murray and Lopez 1996a, 1996b) Disease type R ]E/]TU k ]E/]TU Cancer Mouth and oropharynx cancer Oesophagus cancer Stomach cancer Colon and rectum cancer Liver cancer Pancreas cancer Trachea, bronchus and lung cancer Melanoma and other skin cancer Breast cancer Cervix uteri cancer Corpus uteri cancer Ovary cancer Prostate cancer Bladder cancer Lymphomas and multiple myeloma Leukemia Cancer average Neuropsychiatric conditions Bipolar disorder Schizophrenia Epilepsy Dementia Parkison disease Multiple sclerosis Obsessive-compulsive disorder Panic disorder Sense-organ diseases Glaucoma Cateract Cardiovascular diseases Rheumatic heart disease Ischaemic heart disease Cerebrovascular heart disease Inflammatory heart disease Respiratory diseases Chronic obstructive pulmonary disease Asthma Diabetes mellitus

9 Human Effect and Damage Factors for LCIA Integr Environ Assess Manag 1, Table 2. Continued Disease type R ]E/]TU k ]E/]TU Digestive diseases Peptic ulcer Liver cirrhosis Genitourinary diseases Nephritis and nephrosis Benign prostate hypertrophy Musculoskeletal diseases Rheumatoid arthritis Osteoarthritis Congenital anomalies a Abdominal wall defect Anencephaly Anorectal atresia Cleft lip Cleft palate Oesophageal atresia Renal agenesis Down syndrome Congenital heart anomalies Spina bifida Noncancer average a Probability of occurrence of congenital diseases at birth. and up to a factor of 5 lower for noncancer diseases when compared with average world DALYs (results not shown, based on Murray and Lopez [1996a]). This can be explained by the fact that medical health care is much more advanced in the established market economies when compared with the world average. For the same reason, differences in medical health care in 1990 compared with the (far) future may result in differences in DALYs. This may be particularly important for emissions occurring now but having their impact in the future, such as emissions of carcinogenic substances. Second, the damage factors presented in this article are calculated without applying age-specific weighting and without discounting future health damages. These starting points, however, are a matter of debate (Hofstetter and Hammitt 2002). For instance, using nonuniform age weights and a future discount rate of 0.03, as proposed by Murray and Lopez (1996a), DALY estimates typically decrease by a factor of 2. Third, the use of YLDs includes subjective judgment of the weighting of health disabilities (Krewitt et al. 2002). For cancer diseases, DALYs and years of life lost differ by up to a factor of 1.2, indicating that the inclusion of years of life disabled does not have a large influence on the DALY outcomes. The situation is different for a number of noncancer diseases, such as for musculoskeletal, neuropsychiatric, and sense-organ diseases. For these disease types, the years of life disabled has a dominant contribution to the DALY estimates. As health-preference measurements tend to be rather stable across groups of individuals and regions of the world (Hofstetter and Hammitt 2002), it is expected that the influence of subjective judgment in years of life disabled estimates on the DALY outcomes will be small. As shown by Hofstetter (1998), subjective choices as mentioned above can be condensed to a relatively small number of scenarios, reflecting different archetypes from cultural theory. This requires further study, particularly toward the influence of future scenarios on the DALYs. Effect factors The current effect-factor calculations consist of a diseasespecific part and a substance-specific part. The diseasespecific component of the effect factor depends on probability of occurrence levels in a rather complex, nonlinear way. Figures 3 and 4 showed that the outcomes can range several orders of magnitude, depending on the differences in the sensitivity of the human population and the probability of disease occurrence. is calculated assuming a lognormal dose response relationship, although the choice of distribution function is, to some extent, arbitrary. Nevertheless, lognormally distributed variables have been reported in many scientific fields, including animal studies to drugs and measurements of human beings, such as body weight and blood pressure (Gaddum 1945; Slob 1994). Furthermore, the central limit theorem, stating that the product of a large

10 190 Integr Environ Assess Manag 1, 2005 M.A.J. Huijbregts et al. Table 3. Summary statistics of the human-equivalent toxic potency, equal to 1/ED50, for, respectively, carcinogenic; oral, noncarcinogenic; and inhalation, noncarcinogenic effects a group N Median 2.5-p 97.5-p Minimum Maximum Carcinogens Noncarcinogens, oral Noncarcinogens, inhalation a N = the number of chemicals included; 2.5-p = 2.5th percentile of the cumulative substance distribution; 97.5-p = 97.5th percentile of the cumulative substance distribution; minimum = lowest human-equivalent potency; maximum = highest human-equivalent potency. number of independent variables will be lognormally distributed, also indicates that the assumption of a lognormal distribution is reasonable. For modes of action with a recognized sensitive subpopulation in the population exposed, however, the assumption of an unimodal, lognormal dose response distribution will fail (Naumann et al. 2001; Dorne et al. 2003). In these cases, a bimodal dose response distribution would be more appropriate, which, in turn, can have great numerical consequences for the outcomes. A 2nd restriction in the actual derivation of is the lack of information on the probability of disease occurrence per mode of action. As a tentative solution, we assumed per disease only 1 mode of action is dominantly responsible for the probability of occurrence. This assumption may result in an overestimation of the (see Figures 3 and 4). For instance, an overestimation of the relevant probability of disease occurrence by a factor of 100 results in an overestimation of the by approximately a factor 20 to 60, depending on the spread in human sensitivity and the working point of the curve. The substance-specific part of the effect factor represents the toxic potency of a specific substance, which is equal to the inverse of the ED50. We were able to calculate for 1,192 substances the carcinogenic and/or noncarcinogenic toxic potencies with a total range of 7 to 8.8 orders of magnitude. Concerning the carcinogenity of a substance, it should be noted that not all substances with an carcinogenic ED50 are necessarily known carcinogenics to humans. The International Agency for Research on Cancer (), part of the World Health Organization, evaluated the carcinogenic risk of 844 substances (mixtures) to humans by assigning a carcinogenity to each substance ( 2004). The es reflects the strength of the evidence for carcinogenity derived from studies in humans and in experimental animals and from other relevant data. This information, reported in Appendix 1, can be readily used for including subjective elements in LCA studies. For instance, Hofstetter (1998) proposed the following scenarios to include carcinogenic substances in LCA, reflecting 3 cultural theory perspectives: The individualist scenario includes only substances that are proven carcinogenic to humans ( category 1); The hierarchist scenario includes the substances with sufficient evidence of carcinogenity ( categories 1, 2A, and 2B); The egalitarian scenario also includes the substances with insufficient evidence of carcinogenity ( categories 1, 2A, 2B, and 3). s not ified by the may also be added to the egalitarian scenario. Carcinogenic potencies are based on ED50s representing all exposure routes (Gold and Zeiger 1997). Note, however, that this assumption may not always be valid. For instance, cadmium, beryllium, and chromium VI compounds cause lung cancer only after inhalatory intake (USEPA 2004). Noncarcinogenic effect factors are based on ED50s representing oral and inhalatory exposure routes separately. For substances that lack relevant data on the exposure routes of interest, route-to-route extrapolation rules can be used, for instance, based on oral and inhalatory absorption data (Vermeire et al 1993). Route-to-route extrapolation will, however, introduce additional uncertainty into the effectfactor calculations because we lack adequate experimentvalidated rules for route-to-route extrapolation (Vermeire et al. 1999). Carcinogenic and noncarcinogenic toxic potencies were calculated by using a combination of conversion factors to account, for instance, for interspecies differences. For carcinogenic toxic potencies, a typical uncertainty factor of 19 was found, reflecting uncertainty in the interspecies conversion factor. For some substances, uncertainties up to a factor of 127 were reported, mainly because of uncertainty in the extrapolation from the carcinogenic-slope factor q* to the ED50. For noncarcinogenic toxic potencies, the typical uncertainty factor is 45, reflecting the combined uncertainty in the interspecies conversion factor and the NOEL-to-ED50 conversion factor. Apart from these uncertainties, uncertainty in the time conversion factor and the LOEL-to-NOEL Table 4. Summary statistics of the combined damage and effect factor (y kg 1 ) for, respectively, carcinogenic; oral, noncarcinogenic; and inhalation, noncarcinogenic effects a group N Median 2.5-p 97.5-p Minimum Maximum Carcinogens Noncarcinogens, oral Noncarcinogens, inhalation a N = the number of chemicals included; 2.5-p = 2.5th percentile of the cumulative substance distribution; 97.5-p = 97.5th percentile of the cumulative substance distribution; minimum = lowest human-equivalent potency; maximum = highest human-equivalent potency.

11 Human Effect and Damage Factors for LCIA Integr Environ Assess Manag 1, conversion factor cause uncertainties up to a factor of 115 for noncarcinogenic toxic potency estimates. How do the nonlinear, human toxicological effect factors relate to previous investigations concerning linear low-dose response methods? Combining our average estimate of the disease-specific part with the substance-specific part of the effect factor, we derive the following general equations to obtain carcinogenic and noncarcinogenic effect factors: EF carc;x ¼ ]R x ]TU x ¼ 0:03 0:024 q x ð18þ ]TU x ]I x ED50 x and EF non carc;x ¼ ]R x ]TU x ¼ 0:16 0:002 ð19þ ]TU x ]I x ED50 x NOEL x These effect-factor equations can be readily compared with the outcomes of the linear low-dose response method proposed by Crettaz et al. (2002) for carcinogenic substances and Pennington et al. (2002) for noncarcinogenic substances. Recalculation of the linear effect-factor equations for carcinogenic and noncarcinogenic effect factors given in Crettaz et al. (2002) and Pennington et al. (2002) result in EF carc;x ¼ 0:1 0:5 q ð20þ ED10 x and EF non carc;x ¼ 0:1 0:07 ð21þ ED10 x NOEL x The constants in the linear, carcinogenic and noncarcinogenic effect-factor method are, respectively, a factor of 21 and 35 higher when compared with the constants derived for the lognormal effect-factor method. This finding indicates the potential conservative nature of the linear effect-factor variant. CONCLUSION This article presented human damage factors, expressed as DALYs per case, for 49 noncommunicable diseases representative for the world in The average DALY for carcinogenic and noncarcinogenic effects is, respectively, 11.5 and 2.7 y. The article also implemented a new procedure to calculate human toxicological effect factors for LCA purposes, starting from a lognormal dose response function. The effect factor consists of a disease-specific and a substancespecific component. The disease-specific component depends on the probability of occurrence of a disease and the spread in sensitivity of the human population. The average diseasespecific part of the carcinogenic and noncarcinogenic effect factor is, respectively, 0.03 and The substance-specific component, equal to the inverse of the ED50, represents the toxic potency of a substance. Carcinogenic and noncarcinogenic toxic potencies were calculated for 1,192 substances with a total range of 7 to 8.8 orders of magnitude. For combined damage and effect factors, the total range is dominantly clarified by the range in toxic potencies. The typical uncertainty in the combined damage and effect factor is 25 for carcinogenic effects and 125 for noncarcinogenic effects. The interspecies-conversion factor and the (non)- cancer-effect conversion factor, needed in the calculation of toxic potency, and the average noncancer-damage factor are the largest contributions to the overall uncertainty. Acknowledgement This work, part of the ReCiPe project, was financed by the Ministry of Housing, Spatial Planning and Environment, under project KVI/ We thank Reinout Heijungs, Suzanne Effting, Mark Goedkoop, and Jaap Struijs for useful discussions. REFERENCES Baird SJS, Cohen J, Graham JD, Shlyakhter AI, Evans JS Noncancer risk assessment: A probabalistic alternative to current practice. Human and Ecological Risk Assessment 2: Chen JJ, Chen Y-J, Rice G, Teuschler L, Hamernik K, Protzel A, Kodell RL Using dose addition to estimate cumulative risks from exposures to multiple chemicals. Regul Toxicol Pharmacol 34: Consoli F, Allen D, Boustead I, Fava J, Franklin W, Jensen AA, De Oude N, Parrish R, Perriman R, Postlethwaite D, Quay B, Séguin J, Vigon B Guidelines for life-cycle assessment: A Code of Practice. Pensacola (FL), USA: Society of Environmental Toxicology and Chemistry (SETAC). 79 p. Crettaz P, Pennington D, Rhomberg L, Brand K, Jolliet O Assessing human health response in life cycle assessment using ED 10 s and DALYs: Part 1 Cancer effects. Risk Analysis 22: De Hollander AEM, Melse JM, Lebret E, Kramers PGN An aggregate public health indicator to represent the impact of multiple environmental exposures. Epidemiology 10: Dorne JLCM, Walton K, Renwick AG Polymorphic CYP2C19 and N- acetylation: Human variability in kinetics and pathway-related uncertainty factors. Food Chem Toxicol 41: Frischknecht R, Braunschweig A, Hofstetter P, Suter P Human health damages due to ionizing radiation in life cycle impact assessment. Environmental Impact Assessment Review 20: Gaddum JH Lognormal distributions. Nature 156: Gold LS Carcinogenic Potency Database (CPDB). potency.berkeley.edu. Accessed 15 May Gold LS, Zeiger E Handbook of carcinogenic potency and genotoxicity databases. Boca Raton (FL), USA: CRC. 754 p. Hattis D Human interindividual variability in susceptibility to toxic effects: From annoying detail to central determinant of risk. Toxicology 111:5 14. Hattis D Human variability in susceptibility: How big, how often, for what responses to what agents. Environ Toxicol Pharmacol 4: Hertwich EG, Jolliet O, Pennington D, Hauschild M, Schulze C, Krewitt W, Huijbregts M Fate and exposure assessment in the life-cycle impact assessment of toxic chemicals. In: HA Udo de Haes, editor, Life-cycle impact assessment: Striving towards best practice. Pensacola (FL), USA: Society of Environmental Toxicology and Chemistry (SETAC). Ch. 4. Hertwich EG, Mateles SF, Pease WS, McKone TE Human toxicity potentials for life cycle assessment and toxics release inventory risk screening. Environ Toxicol Chem 20: Hofstetter P Perspectives in life cycle impact assessment: A structured approach to combine models of the technosphere, ecosphere and valuesphere. Dordrecht, The Netherlands: Kluwer. 484 p. Hofstetter P, Hammitt JK Selecting human health metrics for environmental decision-support tools. Risk Analysis 22: Huijbregts MAJ, Gilijamse W, Ragas AMJ, Reijnders L Evaluating uncertainty in environmental life-cycle assessment. A case study comparing two insulation options for a Dutch one-family dwelling. Environ Sci Technol 37: Huijbregts MAJ, Rombouts LJA, Ragas AMJ Human-toxicological effect and damage factors for life cycle impact assessment of carcinogenic and noncarcinogenic chemicals. The Netherlands: Department of Environmental Studies, University of Nijmegen. Report 253. Huijbregts MAJ, Struijs J, Goedkoop M, Heijungs R, Hendriks AJ, Van de Meent D Human population intake fraction and environmental fate factors of toxic pollutants. Chemosphere (forthcoming). Huijbregts MAJ, Thissen U, Guinée JB, Jager T, Van de Meent D, Ragas AMJ, Wegener Sleeswijk A, Reijnders L Priority assessment of toxic substances in life cycle assessment, I: Calculation of toxicity potentials for 181 substances with the nested multi-media fate, exposure and effects model USES LCA. Chemosphere 41: Huijbregts MAJ, Van de Meent D, Goedkoop M, Spriensma R Ecotoxicological impacts in life cycle assessment. In: Posthuma L, Suter GW II, Traas TP, editors, Species sensitivity distributions in ecotoxicology. Boca Raton (FL), USA: Lewis. p