Document NeM7wvvXemy9mgz7Qwp8ekr0R

Ofiza September 25, 1990 TO: Vinylidene Chloride Panel SUBJECTS: (1) EPA's Calculation of Cancer Risk from Exposure to Indoor Air Containing Vinylidene Chloride (2) ATSDR and Data Needs Report (3) European Study on Mechanisms of Toxicity Bill Hayes (Dow) asked that I distribute the enclosed study to the Vinylidene Chloride Panel. There are actually two papers enclosed; however, they are slightly different papers of the same study. The study author is Lance Wallace of EPA and the study concerns the cancer risks associated with exposure to various indoor air pollutants, one of which is vinylidege chloride. Wallace has calculated an excess cancer risk of 320 x 10 from exposure to vinylidene chloride. Please review the Wallace study. The Vinylidene Chloride Panel may want to respond to the study and to Wallace in some fashion. Wallace himself notes several study limitations. I have enclosed a document recently published by the Office of Management and Budget (0MB) which outlines general weaknesses of the risk assessment process. This document may be helpful in identifying risk assessment criticisms which can be used to fault the Wallace study. CMA is beginning to use this 0MB document in its advocacy efforts as a "guide" to respond to a variety of environmental and toxicological issues. I will be talking with Dr. Cebulus of ATSDR in the next day or so about what the status of vinylidene chloride is in terms of ATSDR developing a Data Needs report. If ATSDR has identified vinylidene chloride as a "high priority" chemical for which additional data will have to be developed, then the Panel may want to respond to ATSDR. Bill Hayes also indicated that he would be contacting the Europeans planning to conduct a study on the mechanisms of vinylidene chloride toxicity. He will find out more about the study for the Panel. One issue is whether or not the Panel wants to participate financially in this European study. If you have any questions, please call me at 202/887-1189. Sincerely, cc: Enclosures (2) . _____ Busch Manager, Vinylidene Chloride Panel Vex 3l>/ - CMA ASH SL 064711 90-159.9 CANCER RISKS FROM 35 ORGANIC CHEMICALS IN AIR AND DRINKING WATER: RESULTS FROM EPA'S TEAM STUDIES OF VOCS AND PESTICIDES Lance A. Wallace U.S. Environmental Protection Agency Warrentown, Virginia Air & Waste Management association Sines 1907 For Presentation at the 83rd Annual Meeting & Exhibition Pittsburgh, Pennsylvania June 24-29,1990 SL 064712 90*159.9 INTRODUCTION EPA's TEAM Studies have provided data on the exposures to organic chemicals of about 1000 persons representing about 1,000,000 residents of ten U. S. cities. About 35 of these chemicals cause cancer in animals and may cause cancer in man. In this paper, we calculate the upper-bound lifetime risk associated with airborne and waterborne exposures to each chemical. Although the absolute magnitudes of these upper-bound risks are very uncertain, the relative rankings of these chemicals may be useful in highlighting the more important on s for exposure reduction priorities. METHODS The calculation of cancer risk requires two factors: carcinogenic potencies of chemicals and mean exposures of people. Chemical potencies are taken from EPA sources1 *z. Exposure data for 12 volatile organic compounds (VOCs) in personal air and drinking water are taken from TEAM Studies carried out in eight U.S. cities between 1980 and 19873*4. Exposure data for 23 carcinogenic pesticides in air and drinking water were collected in two cities between 1986 and 1988s. Outdoor air measurements were made for all chemicals in the back yards of the residents; therefore an estimate can be made of the relative contribution of outdoor and indoor air to total airborne exposure to all the target VOCs and pesticides. In a previous study of cancer risks of six prevalent VOCs6, the TEAM cities were divided into "metropolitan" and "non metropolitan" categories. The mean exposures calculated for ach city were averaged to provide a risk associated with each of th two categories. Assuming that the TEAM cities represented typical Values, U.S. Census figures were employed to calculate a risk for the U.S. population. The results from that study have been reproduced here, with two additions: calculated risks fr m exposures to benzene during smoking and chloroform during showering. (Both of these exposures could not be measured using the personal monitors employed in the TEAM Studies; however, they could be estimated using breath measurements for smokers7 and models for exposure during showers6}. An additional six VOCs have been added. Two of these (styrene and 1,1,1-trichloroethane) are prevalent and their concentrations .are well characterized, but their carcinogenicity is in doubt. A third chemical (methylene chloride) is probably prevalent but very few measurements of personal exposure have been made due to its high volatility. The remaining three chemicals (vinylidene chloride, 1,2-dichloroethane, and 1,2dibromoethane) are well-established animal carcinogens but are 2 06*n3 Si* 90-X59.9 much less prevalent. They have been measured in only a few percent of the personal and outdoor air samples collected in the TEAM Studies. Thus the risk estimates for these six VOCs are more uncertain than the estimates for the original six VOCs. The pesticide exposures are the unweighted mean of the seasonally-averaged values for each city. Since Jacksonville, FL was chosen as a high-use area and Springfield, MA as a low-use area, the average of the two may represent a closer approach to actual mean exposures than either one separately. Risks are calculated as a simple multiple of the exposur and the potency. If the potency is given in units of [mg/kgday]'1, the exposure is translated to a daily dose by assuming 20 m3 inspired air per day and a body weight of 70 kg. RESULTS The mean measured exposures, carcinogenic potencies and calculated upper-bound cancer risks are displayed for 12 VOCs in Table I and for 23 pesticides in Table II. The percent of total airborne exposure due to indoor sources is also provided. Six VOCs and four pesticides exceeded the de minimus or negligible risk level of 10** by a factor of 10 or more. The six VOCs were benzene, vinvlidene chloride, y-dichlorobenzene, chloroform, methylene chloride, and carbon tetrachloride. The four pesticides were' chlordane, heptachlor, aldrin and dieldrin. All four have been banned by EFA. Despite the bans, exposures remain high, due perhaps to their long life in the soil and their movement into homes after being injected in the soil as termiticides. carbon tetrachloride has also been banned from consumer products, but its long life in the atmosphere has led t a global background that is sufficiently high to result in a nonnegligible risk. Three additional VOCs and four additional pesticides are at or above the 10** risk level, but by less than an order of magnitude. The three VOCs are tetrachloroethylene, trichloroethylene, and ethylene dichloride. The four pesticides (or degradation products) are dichlorvos, o-BHC. 7-BHC (lindane), and heptachlor epoxide. Two additional VOCs and four additional pesticides had upper-bound risks between 10'* and 10'7: styrene, 1,1,1trichloroethane, hexachlorobenzene, propoxur, DDT, and DDE. Nine chemicals (all pesticides) were below the 10*7 risk level for airborne exposure. 3 SL 90-159.9 Finally, one VOC and one pesticide were seldom or never detected. However, detection limits were high for ethylene dibromide and for pentachlorophenol, leaving open the possibility that these chemicals could represent non-negligible lifetime risks of cancer. Additional exposure studies with lower detection limits for these chemicals are necessary before a more trustworthy estimate of their risks is possible. Indoor sources accounted for the great majority (80-100%) of the_total airborne risk associated with most of these chemicals. Carbon tetrachloride is the only one of the target chemicals for which outdoor sources account for a majority of the airborne risk, indicating the effectiveness of the ban on its use ` ' consumer products. DISCUSSION Uncertainty of Estimates Great uncertainty accompanies most risk estimates. Th major uncertainties~lnvolved in potency calculations are well known: the extrapolation from animals to man and from high dose to low dose. These uncertainties are such that a given chemical may not cause human cancer at allthe actual cancer risk may be exactly zero. Even if the risk is not zero, the estimates could easily be wrong by factors of 10, 100, or more, depending on th shape of the dose-response curve, the possible existence of a threshold due to DHA repair or other mechanisms, and many other factors. Considerably less uncertainty is associated with some.of the .exposure estimates.- The overall mean VOC exposure in eight cities was usually with! a factor of 3 othhe extremes for an individual city, whether the city was rural^ suburban, urban, or heavily industrialized. The reason for this predictability appears to be the relative importance of consumer products, personal activities, and building materials to human exposure; such factors do not vary greatly across the country. Of c urse, personal activities can result in very high exposures for sh rt periods, but we are concerned here with long-term exposures. Somewhat sore variation was noted for pesticides, with differences of a factor of 10 in exposure noted for a number f pesticides in Jacksonville and Springfield. For the more prevalent VOCs and pesticides, relatively little error is associated with the estimates of the relative contribution of indoor and outdoor sources. The reason is that the same instrumentation was used to measure both indoor and outdoor air--even if the instruments were biased, the relative proportions would remain nearly unchanged. For the more rarely 4 SL 06471$ i 90-159.9 found chemicals (e.g., vinylidene chloride, ethylene dichloride), the indoor/outdoor ratio is less certain. Only one of these chemicals is considered a human carcinogen: benzene. Therefore the risk estimate associated with benzene is on more solid ground than any of the others. Benzene is also the only one of these chemicals with human epidemiological studies showing a possible influence of environmental levels of exposure on cancer risk: two studies show that children of smokers die of leukemia at two or more times the rate of children of nonsmokers9*10. The higher mortality rate is consistent with the measured elevated levels of benzene in the breath of smokers (suggesting exposure of the fetus in the womb of the pregnant smoker). Elevated levels of benzene in th air of homes have also been documented by the TEAM Study and by a study in West Germany"; however, the increase (on the order f 50% in both studies) does not seem enough to explain the increase in the mortality rate unless children are more susceptible to benzene-induced leukemia at some point in the first 8-9 years of life. /* The case-^of. vinylidene chloride requires further comment. - This chemical is highly volatile and therefore "breaks thr ugh" t/ ; the Tenax monitor after only a portion of the monitoring period. , The concentration is calculated on the basis of the "breakthrough volume" rather than the actual sampling volume, and depending on the pattern of exposure during the monitoring period, may be either an over- or under-estimate of the actual concentration. Also, because the sampling volume of 20 liters is effectively / reduced to the breakthrough volume of only a few liters, the N sensitivity is reduced by the same factor. The limits of detection for vinylidene chloride ranged from 3 to 14 pg/m3, about an order of magnitude worse than for most of the other target VOCs. out of 1085 personal air samples collected from 355 New Jersey residents over three different seasons, only 77 (7%) had measurable concentrations of vinylidene chloride. (An ther 107 (10%) showed trace concentrations.) ' The population risk for such rarely detected chemicals can be calculated, but the interpretation of the risk presents -- difficulties. For example, the single highest measured exposure to VDC was 120,000 4g/zr, which was incurred by a cabinetmaker. (The second highest value of 14,000 Mg/m3 was also measured for this same person in a different season.) Taken together, these two values agcounted for more than 80% of the total calculated exposure (and therefore the risk) for the population. If we include these values in our calculation of risk, then vinylidene chloride exposures average 150 jig/m3, and the upper-bound risk is 7.5 X 10*3--greater than the risks from radon and passive smoking combined! If we drop these two values, the population exposure 5 SL 064716 90-159.9 decreases to 28 jig/m3, and the risk to 1,4 X 10'3--still very large. However, four other exposures exceeded 1000 /ig/m3. If these values are also dropped from the risk calculation, the average exposure decreases to .5 jtg/m3, and the associated upper-bound risk to 3.2 X 10'4. Thus the population average of 150 ug/n3 for all 355 persons is actually composed of an average of 6.5 fig/m3 for about 350 persons, and an average of about 30,000 jig/m3 for about 5 persons. This corresponds to a difference in risk of the two groups of a factor of 5000! Because of the tremendous effect of a few measurements, the calculated upper-bound risk for vinylidene chloride must be considered tentative. Only additional data on personal exposures (collected by methods with a sensitivity of 1 pg/m3 or less) will provide the informat, n necessary for an adequate risk assessment. Exposures Through Other Routes Exposures through routes other than air and water may be possible for some of the chemicals. All of the above chemicals were measured in drinking water, and found to present less than 1% of the risk due to airborne exposures with a single exception: chloroform. All of the VOCs were also measured in food and beverages; again chloroform was the only VOC found in significant amounts in food. Many of the pesticides have been measured in food by the FDA for years; however, exposures in food account for only a small proportion of total exposure to the four pesticides of highest risk through airborne routes. Food exposures outweigh air exposures for some of the other pesticides (e.g, Captan). House dust may provide an important reservoir for any or all of the pesticides, and possibly also for the least volatile of the VOCs (p-dichlorobenzene, tetrachloroethylene). Ingestion of the dust by toddlers could be an important additional source of risk. Comparison with Other Environmental Risks Several organic chemicals of interest were not monitored in the TEAM Studies. Among these are formaldehyde and 1,3-buta diene. Risk estimates for these chemicals may be compared to the risks calculated above. The carcinogenicity of formaldehyde is controversial, due to the unusual metabolic pathway associated with its carcinogenicity in rodents. Estimates for the cancer risk of formaldehyde12 range over extreme limits, from zero to 10*3. Employing an intermediate potency factor (unit risk of 1.3 X 10*5 (jjg/m3)"1) and measured values of 40 ng/m3 for normal (non-mobile home) housing 6 SL 064717 90-159.9 stock results in a risk of about 5 X 10`4. Recent animal studies of 1,3-butadiene have resulted in revising its potency upward by nearly three orders of magnitude. Although no personal exposure data are available for this chemical, a recent study has measured the level in sidestream smoke at about 400 ng per cigarette13. This is approximately the level of benzene in sidestream smoke (330 ng per cigarette); therefore, if 1,3-butadiene is not too reactive, we can calculate that it will be elevated by about 4 fig/m3 in smoking homes and by 13 pg/m3 in workplaces allowing smoking. Using the revised unit risk value of 2.8 X 10'4, and assuming 38 million homes with smokers averaging three residents each, 75 million workers in workplaces allowing smoking, and 26 million non-workers exposed to cigarette smoke, we arrive at an upper-bound risk associated with 1,3-butadiene of 6 X 10`4. Thus the individual risks for formaldehyde and 1,3-butadiene are greater than the airborne risk of any of the other 35 VOCs and pesticides considered in this report. However, the great uncertainty in the carcinogenic potency of formaldehyde, including the uncertainty as to whether it is a human carcinogen at all, and the lack of exposure data for 1,3-butadiene make the risk estimates for these two chemicals particularly speculative. The combined upper-bound risk of about 10*3 associated with these 37 predominantly indoor organic chemicals appears to be similar to the risks associated with the most severe environmental hazards (radon and passive smoking). For example, the risk associated with nonsmokers' exposure to radon has been estimated to be about 10*3 and that with passive smoking has been estimated14 at 2 X 10`3. It should be noted, however, that the risk estimates for radon and passive smoking are based on human epidemiology studies, and are therefore on firmer ground than all of the risk estimates for the organic chemicals with the exception of benzene. The risk estimates for these organic chemicals are considerably higher than the risks associated with some EPA regulations (NESHAFS chemicals and Superfund clean-up criteria). Similar conclusions regarding the importance of indoor air pollution compared to other environmental hazards have been reached by EPA Headquarters15 and by three Regions16; both of these reports rank indoor_air_pollution as among the top two or three environmental threats to public health. Two other risk estimates for personal exposure to VOCs hav been published12,17. McCann arrived at similar risk estimates for most of the chemicals; Tancrede estimated 5-10 times higher risks, due partly to using a different method for calculating 7 SL 064718 potencies from animal data and partly to considering explicitly several additional sources of uncertainty. No previous risk estimates for most of these pesticides have been possible, du to the lack of exposure information. Actions to avoid these risks may be taken by individuals. Since the sources of the risks are often personal activities (smoking, using air fresheners), these activities can be halted or modified. (For example, smokers could establish a room in the home with separate ventilation.) Exposures from chloroform could be reduced by drinking bottled water or using an activated carbon filter on the water supply. Exposures from petroleum-based products could be reduced by discarding or storing used paint cans and sprays in a detached garage or tool shed. Dry-clean d clothes could be hunc outdoors for a day-- ne study indicates that 20-30$ of tetrac..loroeth> me residues on the clothes will outgas during the first.day. The reason for the large number of pesticides observed in indoor air in the latest TEAM Study is not well understood. Termiticides, like radon gas, may be entering the basement du t soil gas movement; it may be that the same techniques to contr 1 radon (sealing the foundation, providing separate ducting at the entrance points) may also control termiticide entry. Other pesticides, particularly the long-lived chlorinated hydrocarb ns such as DDT, may be entering the home by being tracked in on people's shoes. If so, removing shoes before entering the home, and reducing or eliminating the use of carpets or rugs (which collect large amounts of dust containing pesticides and metals as well), should reduce pesticide exposures. SUMMARY AND CONCLUSIONS Measured personal exposures to 12 VOCs and 23 pesticides in EPA's TEAM Studies have been used to arrive at upper-bound lifetime cancer risk estimates. Seven VOCs and seven pesticides have upper-bound risks ranging from 10** to 10'4. The combined upper-bound risk of about 10`3 from these organic indoor air pollutants is nearly comparable to the estimates of risk from radon and environmental tobacco smoke. (However, the latter two estimates are based on human epidemiology studies, and are therefore subject to far less uncertainty.) These upper-b und risks are much greater than the health risks associated with most other environmental problems. Despite the recognized large uncertainty in these risk estimates, these findings provide additional support for the conclusion of two recent comparative rankings of environmental risk by EPA: that indoor air polluticn is one of the greatest threats to public health of all environmental problems. 8 SL 064719 4 90-159.9 REFERENCES 1. Integrated Risk Information System (IRIS), US EPA, Washington, DC. 2. Engler, R., memorandum to Office of Pesticide Programs Division Directors, USEPA, October 27, 1989. 3. Wallace, Lance A., Pellizzari, E.D., Hartwell, T.D., Sparacino, C., Whitmore, R., Sheldon, L., Zelon, H. and Perritt, R., "The TEAM Study: personal exposures to toxic substances in air, drinking water, and breath of 400 residents of New Jersey, North Carolina, and North Dakota" Environmental Research 43: 290-307 (1987). 4. Wallace, L. A., Pellizzari, E.D., Hartwell, T.D., Whitmore, R., Zelon, H., Perritt, R. and Sheldon, L. "The California TEAM Study: breath concentrations and personal exposures to 26 volatile compounds in air and drinking water of 188 residents of Los Angeles, Antioch, and Pittsburg, CA." Atmos. Environ. 22: 2141-2163 (1988). 5. Immerman, F.W. and Schaum, J.L. Final Report of the Nonoccuoational Pesticide Exposure Study (N0PES1. US Environmental Protection Agency, Research Triangle Park, (1990). NC 6. Wallace, Lance A. "Cancer risks from organic chemicals in th home", in Environmental Risk Assessment: Is Analysis Useful?. APCA Specialty Conference Proceedings, Air Pollution Control Assoc., Pittsburg, PA. SP-55 (1986). 7. Wallace, Lance A., Pellizzari, E., Hartwell, T., Perritt, K., and Ziegenfus, R., "Exposures to benzene and other volatile organic compounds from active and passive smoking," Archives of Environmental Health 42: 272-279 (1987). 8. McKone, T.E. "Human exposure to VOCs in household tap water: The indoor inhalation pathway", Env. Sci. Tech. 21: 1194-1201 (1987) . 9. Sandler, D.P., Everson, R.B., Wilcox, A.J., and Browder, J.P. "Cancer risk in adulthood from early life exposure to parents' smoking". Am. .J. Public Health, 75:467 (1985). 10. Stjernfeldt, M., Berglund, K., Lindsten, J., and Ludvigsson, J. "Maternal smoking during pregnancy and risk of childhood cancer", Lancet, June 14, 1986:1350-52 (1986). 9 SL 064720 90-159.9 11. Krause, C., Mailahn, W., Nagel, R., Schulz, C., Seifert B., Ullrich D. "Occurrence of volatile organic compounds in the air of 500 Homes in the Federal Republic of Germany", in Proceedings of The 4th International Conference on Indoor Air .Quality and Climate. Vol. l, pp. 102-106. Institute for Soil, water, and Air Hygiene. Berlin (Nest) (1987). 12. McCann, J. Horn, L., Girman, J. and Nero, A.V. Potential risks from exposure to organic, compounds J.n J.ndoor .air- Lawrence Berkeley Laboratory, report f LBL-22473. Berkeley, CA (1986). 13. Lofroth, G. Burton, B., Forehand, L., Hammond, S.K., Sella, R., Zweidinger, R., and Levtas, J., "Characterization of environmental tobacco smoke," Env. Sci. and Tech,. : (1989). 14. Repace, J., "A quantitative estinw * of nonsmokers' lung cancer risk from passive smoking," Env. Int. 11:3-22 (1985). 15. usepa, Unfinished Business:_A Comparative Assessmentst Environmental Problems. Washington DC (1987). 16. USEPA, Comparing Risks and Setting Environmental Prioritits: pvsryjgy-QX-ThrtB R^gianaL Praltsta, office of Policy, Planning, and Evaluation, Washington, DC (1989). 17. Tancrede, M, Wilson, R. Zeise, L. and Crouch, E.A.C., "The carcinogenic risk of organic vapors indoors: a theoretical survey," Atmos. Env. 21: 2187-2205 (1987). 10 SL 064721 90-159.9 Table I. Upper-Bound Lifetimei Cancer Risks of 12 V0CS Measured in the TEAM[ Studies (1980-87) Indoor Air chemical EXPO^re". Potency. (Mg/mJ) (Mg/1)'1 (X 10*6) Risk (X10'4) Contribution' (percent) Benzene Air Smokers 15 8 12 0e 60 90 8 720e vinylidene chloride 6.5d chloroform Air 3 Showers (Inhalation) 2 Water 30* Food & Beverages 30* 50 23 23 2.3* 2.3* 320 70 50 70 70 86 80 100 -- p-Dichlorobenzene 1,2-Dibromoethane Methylene chloride Carbon tetrachloride Tetrachloroethylene Trichloroethylene 22 0.05 6f 1 15 7 4 510 4 15 0.6 1.3 90 25 24 15 9 9 98 40 60 20 80 80 y Styrene Air Smokers 1 0.3" 0.3 70 6 0.3 2 1,2-Dichloroethane 0.5 1,1,1-Trichlorethane 30 7 0.003 4 0.1 60 70 a Arithmetic means based on 24-hour average exposures of 750 persons in six urban areas measured in the TEAM Studies b Based on backyard measurements in 175 homes in six urban ar as c The risk estimates for benzene are based on human epidemiology and are therefore mean as opposed to upper-bound estimates d Six measurements exceeding 1000 ^g/m3 dropped from the calculation; inclusion of the measurements leads to an average exposure of 150 Mg/m3. These figures are in M9/L or ppb rather than M9/m3 f Based on only eight 24-hour indoor measurements in 1987. 8 Source: US EPA (1983) Review and Evaluation of Evidence for Cancer Associated with Air Pollution, EPA 450/5-83-006.11 11 SL 064722 90-159.9 Table II. Upper-bound Lifetime Cancer Risks from Airborne Exposures to 23 Pesticides Measured in the N0PE5 TEAM Study Pesticide Sxpgsar?" Potency RlsX (ng/m*) (kg-d/mg) (X10'*) Indoor Air Contribution (percent) Banned Termltlcides Heptachlor Chlordane Aldrin Dieldrin Heptachlor Epoxide DDE DDT 71 198 13 3 0.4 2.2 0.7 4.5 1.3 17 16 9.1 0.34 0.34 90 (19)e 70 (15) 60 (13) 14 (3) 1 (0.2) 0.2 (0.04) 0.1 (0.02) 90 94 100 92 80 100 88 Other Pesticides Dichlorvos 7-BHC (Lindane) a-BHC Propoxur Hexachlorobenzene Dicofol p-Phenylphenol 2,4-D Atrazine cis-Permethrin trans-Permethrin Chlorothalonil Folpet Captan DDD Pentachlorophenol 33 6.6 0.5 100 0.3 2.6 58 0.6 0.05 0.4 0.1 0.7 0.5 0.1 <4 <730 0.29 1.3 6.3 0.0079 1.67 0.34 0.0016 0.019 0.22 0.022 0.022 0.011 0.0035 0.0023 0.34 0.013 2.7 2.5 1 0.2 0.1 0.05 0.02 0.003 0.003 0.003 0.001 0.002 0.0005 0.00007 <0.4 <3 100 100 100 98 90 100 99 77 100 100 100 28 50 100 --- -- a Arithmetic mean of population-weighted and seasonally-weighted average personal exposures measured for 173 persons in Jacksonville, FL and 85 persons in Springfield/Chicopee, MA. b Percent of total airborne exposure only, based on outdoor measurements at each home in the two cities. e All risks calculated assuming 70-year lifetime exposure at the measured levels. For banned pesticides, whose environmental concentrations should decrease over time, an alternative calculation of risk (in parentheses) assuming a 10-year half-life in soil is provided. 12 SL 064723 Comparison of Risks from Outdoor and Indoor Exposure to Toxic Chemicals Lance A. Wallace Atmospheric Research and Exposure Assessment Laboratory U.S. Environmental Protection Agency Bldg 166, Bicher Road Vint Hill Farms station Warrenton, VA 22186-5129 For presentation at EPA-ILSI Joint Symposium on Methodology for Assessing Health Risks from Complex Mixtures in Indoor Air Arlington, VA April 17-19, 1990 SL 064724 INTRODUCTION For the last decade, EPA's TEAM Studies have been providing data on personal exposures (including indoor and outdoor concentrations) to organic chemicals for more than 1000 persons representing more than 1,000,000 residents of ten U. S. cities. About 35 of these chemicals cause cancer in animals and may cause cancer in man. In this paper, we calculate the upper-bound lifetime risk associated with airborne exposures to each chemical. We also try to apportion the risk between indoor and outdoor sources. Although the absolute magnitudes of these upper-bound risks are very uncertain, the relative rankings of the chemicals and their sources may be useful in focusing our attention on efficient ways to reduce exposure. METHODS The calculation of ca, risk requir two factors: carcinogenic potencies of : icals and me exposures of people. chemical potencies .re taken from EPA sources1*2. Exposure measurements (incluumg some overnight indoor air measurements) for 12 volatile organic compounds (VOCs) are taken from TEAM Studies carried out in eight U.S. cities between 1980 and 19875,4. Personal exposures and indoor air concentrations for 23 carcinogenic pesticides were measured in two cities between 1986 and 1988s. Outdoor air measurements were made for all chemicals in the back yards of the residents; therefore an estimate can be made of the relative contribution of outdoor and indoor air to total airborne exposure to all the target VOCa and pesticides. In a previous study of cancer risks of six prevalent VOCs*, the TEAM cities were divided into "metropolitan" and "non metropolitan" categories. The mean, exposures calculated for each city were averaged to provide a risk associ ad with each of the two categories. Assuming that the^TEAM cit is represented typical values, U.S. Census figure^ were employed to calculate a risk for the U.S. population. The results from that study have been reproduced here, with two additions: calculated risks fr m exposures to benzene during smoking and chloroform during showering* (Both of these exposures could not be measured using the personal monitors employed in the TEAM Studies; however, they could be estimated using breath measurements for smokers7 and models for exposure during showers8). An additional six VOCs have been added. Two of these (styrene and 1,1,1-trichloroethane) are prevalent and their personal exposures and outdoor air concentrations are well characterized, but their carcinogenicity is in doubt. A third chemical (methylene chloride) is probably prevalent but very few measurements of personal exposure have been made due to its high volatility. The remaining three chemicals (vinylidene chloride, 2 SU 06^7 25 1,2-dichloroethane, and 1,2-dibromo thane) are well-established animal carcinogens but are much less prevalent. They have been measured in only a few percent of the personal and outdoor air samples collected in the TEAM Studies. Thus the risk estimates for these six VOCs are more uncertain than the estimates for the original six VOCs. The pesticide exposures are the unweighted means of the seasonally-averaged values for each city, since Jacksonville, FL was chosen as a high-use area and Springfield, HA as a low-use area, the average of the two may represent a closer approach to actual mean exposures than either one separately. Risks are calculated as a simple multiple of the exposur and the potency. If the potency is given in units of [mg/kgday]'1, the exposure is translated to a daily dose by assuming 20 m3 inspired air per day and a body weight of 70 kg. RESULTS The mean measured exposures, outdoor air concentrations, carcinogenic potencies and calculated upper-bound cancer risks are displayed for 12 VOCs in Table I and for 23 pesticides in Table II. Seven VOCs and four pesticides exceeded the de minimus or negligible risk level of 10`4 by a factor of 10 or more. The seven VOCs were benzene, vinylidene chloride, p-dichlorobenzene, chloroform, ethylene dibromide, methylene chloride, and carbon tetrachloride. The four pesticides were chlordane, heptachlor, aldrin and dieldrin. All four have been banned by EPA. Despite the bans, exposures remain high, due perhaps to their long life in the soil and their movement into homes after being injected in the soil as termiticides. Carbon tetrachloride has also been banned from consumer products, but its long life in the atmosphere has led to a global background that is sufficiently high to result in a nonnegligible risk. Three additional VOCs and four additional pesticides are at or above the 10*4 risk level, but by less than an order of magnitude. The three VOCs are tetrachloroethylene, trichloroethylene, and ethylene dichloride. The four pesticides (or degradation products) are dichlorvos, a-BHC. 7-BHC (lindane), and heptachlor epoxide. Two additional VOCs and four additional pesticides had upper-bound risks between 10'4 and 10'7: styrene, 1,1,1trichloroethane, hexachlorobenzene, propoxur, DDT, and DDE. Nine chemicals (all pesticides) were below the 10'7 risk level for airborne exposure. 3 SL 064726 Finally, one pesticide--pentachlorophenol (PCP)--was never detected. However, the detection limit was very high for PCP, leaving open the possibility that this pesticide could represent non-negligible lifetime risks of cancer. Additional exposure studies with lower detection limits for pentachlorophenol, and also for some of the less prevalent VOCs such as ethylene dibromide and vinylidene chloride, are necessary before a more trustworthy estimate of their risks is possible. Indoor sources accounted for the great majority (80-100%) of the total airborne risk associated with most of these chemicals. Carbon tetrachloride is the only one of the target chemicals for which outdoor sources account for a majority of the airborne risk, indicating the effectiveness of the ban on its use in > consumer products. DISCUSSION Upper-bound vs. "best-guess" estimates of potency The potencies employed in the risk calculations above are, with one exception (benzene), "upper-bound" potencies calculated from animal experiments. This raises the question of what the "best-guess" potency might be. Unfortunately, the EFA has chosen not to calculate "best" estimates of potency, arguing that such estimates are inherently more unstable than the upper-bound estimates, without arguing this point, it may still be instructive to calculate "best" estimates of potency. For example, are such "best" estimates lower by a factor of 3, 10, or 100? Could the "best" estimate sometimes be zero? To answer such questions, we may examine the approach developed by a gr up at Harvard University9. In this approach, a median potency is calculated from the animal studies, together with an estimate of the natural logarithm of the geometric standard deviation (ox) of that animal potency. A median potency in humans is then calculated, based on an assumed uncertainty connected with extrapolation from animal to man (oy - 1.5), and an uncertainty (if necessary) associated with converting from oral dose to inhalation dose (st * 1.6). Assuming log-normal distributions for all the sources of uncertainty, estimates of mean and upperbound potencies can be calculated using standard relationships of log-normal distributions: Mean Median X exp (o2/2) where a2 * ox2 + oy2 + at2 When these calculations are carried out, using values for the median potencies and their uncertainties as calculated by the Harvard group, we find that 95% upper-bound potencies are typically about 7 times mean potencies. This finding is in4 4 SL 064727 general agreement with the results of a study10 that compared upper-bound potency estimates from animal studies with observed potencies from human epidemiology of about 20 chemicals that are both animal and human carcinogens. That study found that, in general, the upper-bound estimates from the animal data were about an order of magnitude higher than the "best" estimates from the human data. Therefore, the upper-bound risk estimates in Tables I and II (except for benzene) may be divided by a factor of 7 or 10 to provide "best" estimates of risk based on mean potencies and mean exposures. This results in benzene emerging as the single chemical with the highest risk of all 35 considered. For some of these chemicals, it is also possible to argue that the "best" estimate of risk is 0. For example, if a chemical causes cancer in animals but not in man, its carcinogenic risk in humans is zero by definition. Now suppose that a chemical is "more likely than not" to be a noncarcinogen-what is the "best" estimate of its carcinogenic risk? One way to answer this question is to say that there is a better than 50% probability that it is not a carcinogen, and therefore the "best" (median) estimate of risk is zero. Another approach might be to assign a probability that it is a carcinogen, calculate the risk on the basis of the animal studies, and then "dilute" that risk by multiplying by the assigned probability. Thus, if the animal studies are ambiguous, and we assign only a 10% probability that the chemical is a human carcinogen, the risk calculation would be multiplied by 0.10 to arrive at a "best" estimate of risk. In the case of the chemicals considered here, most are classified as "B2" (probable human carcinogens), but some are classified as "C" (possible human carcinogens). (Still others, such as tetrachloroethylene and p-dichlorobenzene, wobble back and forth between the two classifications.) If the "probables" were assigned a likelihood factor greater than 50%, while the "possibles" were assigned a factor less than 50%, one of the above two approaches could be employed to further adjust the risks calculated in Tables I and II. Assigning risk to population subgroups The risk calculations above have generally been made on the basis of the entire population studied in the TEAM Studies, or on extrapolating those results to the U.S. population. One exception has been the risks to smokers of benzene and styrene, which apply only to the 50 million active smokers in the U.S. However, if it were possible to identify sources of exposure, and characterize population subgroups on the basis of their exposure to those sources, it would be possible to refine our estimates of risk. In particular, we would find that risks are higher among the exposed subgroups, and lower (perhaps zero) among the less exposed or unexposed subgroups. 5 SL 064728 As an exampl of the above points, we nay consider the pesticide dichlorvos. This pesticide was found in about a third of Jacksonville hones and only 2% of Springfield hones. Outdoor concentrations were negligible in both regions. If the personal exposures cane nostly fron use of a consumer product containing dichlorvos, it seems reasonable to calculate a risk based on exposures to the users only. Thus the calculated risk to Jacksonville users would be three tines the risk averaged over the entire population, and the calculated risk to Springfield users would be 50 tines the risk averaged over the population. Since the calculated risk for the Jacksonville population was 5 X 10'6, the risk to users would be 15 X 10**; the calculated risk for the Springfield population was 3 X 10'7, leading to a risk to users of 15 X io*6--identical to the risk to Jacksonville users. This calculation has not changed the total population risk in either case--it has simply apportioned the risk across the populations. Thus we have gained a sharper definition of the risk. It is also interesting to note that the risks calculated in this way show that in both areas, upper-bound risks to users exceed the one in a million level by a factor of 15; the previous calculation indicated that the upper-bound risks averaged across the population were very close to this dividing line. This approach does not appreciably change the risks calculated for the more prevalent VOCs and pesticides, but has the potential for order-of-magnitude increases in calculated risks for the less prevalent chemicals. For example, applying this approach to other pesticides, we find that a-BHC, which was found in only 27% of Jacksonville homes and only 2% of Springfield homes, has a calculated average risk in these homes of 8 X 10'6 and 20 X 10**, respectively, compared to the value of only 1 X 10*6 averaged across the population. Other pesticides whose lifetime upper-bound risks in exposed homes approached or exceeded one in a million included 2,4-D, DDE, hexachlorobenzene, and dicofol. Uncertainty of Estimates Great uncertainty accompanies most risk estimates. The major uncertainties involved in potency calculations are well known: the extrapolation from animals to man and from high d se to low dose. These uncertainties are such that a given chemical may not cause human cancer at all--the actual cancer risk may be exactly zero. Even if the risk is not zero, the estimates could easily be wrong by factors of 10, 100, or more, depending on the shape of the dose-response curve, the possible existence of a threshold due to DNA repair or other mechanisms, and many other factors. Considerably less uncertainty is associated with some f the exposure estimates. The overall mean VOC exposure in eight cities was usually within a factor of 3 of the extremes for an 6 SL 064729 individual city, wh ther the city was rural, suburban, urban, or heavily industrialized. The reason for this predictability appears to be the relative importance of consumer products, personal activities, and building materials to human exposure; such factors do not vary greatly across the country. Of course, personal activities can result in very high exposures for short periods, but we are concerned here with long-term exposures. Somewhat more variation was noted for pesticides, with differences of a factor of 10 in exposure noted for a number of pesticides in Jacksonville and Springfield. For the nine prevalent VOCs, about 2000 measurements have been made of 12-hour average personal exposure, and more than 500 have been made of outdoor concentrations. Thus, for these more prevalent VOCs and pesticides, relatively little error is associated with the estimates of the relative contribution of indoor and outdoor sources. The reason is that the same instrumentation was used to measure both indoor and outdoor air-- even if the instruments were biased, the relative proportions would remain nearly unchanged. However, two other VOCs were measurable in only a small percentage of samples; vinylidene chloride (7*) and ethylene dibromide (2%). (The 12th VOC, ethylene dichloride, was measured in about 20% of the samples, but with most measurements hovering close to the detection limit.) For these more rarely found chemicals, the indoor/outdoor ratios are less certain, as are the risk estimates. Some chemicals were prevalent but do not have sufficient animal studies to establish their carcinogenicity. Among these are toluene and xylenes, not found to be carcinogenic in two-year rat and mouse studies conducted by the National Toxicology Program (NTP), but found to be carcinogenic in natural-lifetime rat and mouse studies carried out in Italy (Maltoni, personal communication). Limonene (used in lemon-scented products and also as a food additive) was the single VOC at the highest average concentrations in people's homes; a recent 2-year NTP study found clear evidence of carcinogenicity in one sex-speci s combination, but no evidence in the other three sex-species . combinations. Two pesticides,that were prevalent at relatively j high concentrations indoors were chlorpyrifos (Pursban) and J diazinon. Health studies of these pesticides do not completely "S rule out their possible carcinogenicity. Because of the / prevalence and high concentrations of these VOCs and pesticides, further health studies are indicated. i Several of the chemicals with the highest associated risks will be discussed separately. Benzene 7 SL 064730 Only one of these chemicals is considered a human carcinog n: benzen . Therefore the risk stimate associated with benzene is on more solid ground than any of the others. Benzene is also the only one of these chemicals with human epidemiological studies showing a possible influence of environmental levels of exposure on cancer risk: two studies show that children of smokers die of leukemia at two or mote times the rate of children of nonsmokers11'12. The higher mortality rate is consistent with the measured elevated levels of benzene in the breath of smokers (suggesting exposure of the fetus in the womb of the pregnant smoker). Elevated levels of benzene in the air of homes have also been documented by the TEAM Study and by a study in West Germany13; however, the increase (on the order of 50% in both studies) does not seem enough to explain the incr ase in the mortality rate unless children are more susceptible to benzene-induced leukemia at some point in the first 8-9 years of life. Major sources of exposure to benzene appear to be activ and passive smoking, driving and other personal activities associated with automobiles, use of attached garages for parking cars, storing gasoline and kerosene, and the use of certain consum r products (marking pens, paints, glues, rubber products). The major outdoor source is auto exhaust; emissions from stationary sources account for only a few percent of nationwide exposures. Vinylldene chloride This chemical is highly volatile and therefore "breaks through" the Tenax monitor after only a portion of the monitoring period. The concentration is calculated on the basis of the "breakthrough volume" rather than the actual sampling volume, and depending on the pattern of exposure during the monitoring period, may be either an over- or under-estimate of the actual concentration. Also, because the sampling volume of 20 liters is effectively reduced to the breakthrough volume of only a few liters, the sensitivity is reduced by the same factor. The limits of detection for vinylidene chloride ranged from 3 to 14 nq/m2, about an order of magnitude worse than for most of the other target VOCs. Out of 1085 personal air samples collected from 355 New Jersey residents over three different seasons, only 77 (7%) had measurable concentrations of vinylidene chloride. (Another 107 (10%) showed trace concentrations.) The population risk for such rarely detected chemicals can be calculated, but the interpretation of the risk presents difficulties. .For example, the single highest measured exposure to VDC was 120,000 7/m3, which was incurred by a cabinetmaker (The second higher alue of 14,000 jig/m3 was also measured fc this same person different season.) Taken together, the two value accoun for more than 80% of the total calculat exposure .and thei re the risk) for the population. If we 8 SL 064731 include these values in our calculation f risk, then vinylidene chloride exposures average ISO fiq/v?, and the upper-bound risk is 7.5 X io'3--greater than the risks from radon and passive smoking combined! If we drop these two values, the population exposure decreases to 28 ng/m*, and the risk to 1.4 X 10'3--still very large. However, four other exposures exceeded 1000 pq/mz. If these values are also dropped from the risk calculation, the average exposure decreases to 6.5 nq/ml, and the associated upper-bound risk to 3.2 X 10`4. Thus the population average of 150 nq/m* for all 355 persons is actually composed of an average of 6.5 nq/vi3 for about 350 persons, and an average of about 30,000 Mg/n3 for about 5 persons. This corresponds to a difference in risk of the two groups of a factor of 5000! Because of the tremendous effect of a few measurements, the calculated upper-bound risk for vinylidene chloride must be considered tentative. Only additional data on personal exposures (c llected by methods with a sensitivity of 1 nq/ml or less) will provide the information necessary for an adequate risk assessment. p-Dichlorobenzene This chemical has two major uses: as a moth repellent, it is a registered pesticide; as an air "freshener", it is an additiv ( ften unlabeled) to consumer products. From the TEAM Study results, it appears that about a third of homes employ p-DCB. Therefore the average risk to users is about 3 times the risk shown in the table. The risk to non-users should be no more than that associated with average outdoor concentrations, or about 1% of the risk to users. Chloroform This chemical is unique among the target VOCs in having many routes of exposure: air, food, water, beverages. Indoor and outdoor air levels and levels in drinking water have been documented in all the TEAM Studies. The pilot TEAM Study of 1980-82 also documented chloroform levels of 15-56 ppb (/*g/L) in milk, butter, cheese, and ice cream; and 9-178 ppb in soft drinks14. A recent Japanese study (Y. Sato: personal communication) of seven housewives indicated that they were exposed to chloroform through all routes, but that the diet provided more exposure (10.7 jig/day) than the air (2.0) and water routes (2.4 jjg/day). Methylene chloride This chemical is too volatile to be collected on Tenax; therefore few personal exposure measurements have been made. Several indoor and outdoor measurements were made in the 1987 TEAM Study in Los Angeles using evacuated canisters--these are 9 06^732 SL the values on which the risk estimate has been based. However, the existing data is s sparse that the estimate must be considered very speculative. Short-term exposures at the high ppm level from using paint strippers have been documented15. Such an exposure for one day would equal the lifetime exposure to ambient concentrations of methylene chloride. Therefore the population risk from this chemical might better be calculated from data on the number of people who use paint strippers and the amount of time they use them. Ethylene dibromide (1,2-dibromoethane) This chemical is widely used as * fungicide, particularly on grain, and therefore the main risk is thought to be through food. However, the potency is so high that-cven the very low airborne exposures measured in the TEAM Study? produce a non-negligible risk. Only 15 of 21 personal air samples (2.4%) exceeded th quantifiable limit of 0.05 M9/m3. Another $1 samples (12.2%) showed trace amounts. If we assume a value of 0 for the 545 nondetected samples, and the lowest possible value of 0.05 for the 61 trace samples, the average exposure is 0.014 Mg/m3* Assuming the maximum values for the trace samples (0.24 pq/w) and th non-detects (0.05 Mg/m3) results in an average exposure of 0.087 Mg/m3. A value between these two extremes is 0.05 pq/vr, resulting in a risk estimate of 25 X 10`*. Only 3 of 282 outdoor air samples were measurable (and only 5 were at trace levels), and the range of average values using the same assumptions as above is 0.003 to 0.06 Mg/m3. The maximum personal exposure was only 0.97 Mg/m3, so that the risk calculations are not extensively skewed by a few samples as they were in the case f vinylidene chloride. Chlordane and Heptachlor These pesticides were recently withdrawn (April 1988), after wide use as termiticides (85% of the market). They were applied primarily as a liquid poured or injected into soil around building foundations. Therefore their appearance in the NOPES study as airborne vapors may indicate widespread intrusion of soil gas into the home through cracks or drains in the basement or ground floor. Aldrin and Dieldrin These pesticides were withdrawn from use in the U.S. in the early 1980's. They were used mainly as termiticides (about 10% of the market). Their appearance in the NOPES study is further 10 SL 064733 indication of a long half-life in soil coupl d with some mechanism allowing intrusion into the h_me. Dichlorvos (DDVP) This insecticide was widely used on "pest strips" before such a use was banned. Measurements during different seasons in the two cities ranged from 98-99% not detectable in Springfield, and from 65-89% not detectable in Jacksonville. As discussed above, although the risk averaged over the entire population is close to the 10'6 level of risk, when averaged over the smaller population of users, the risk climbs to about 15 X 10*6 in both cities. Exposures Through Other Routes All of the above chemicals (both VOCs and pesticides) were measured in drinking water, and found to present less than 1% of the risk due to airborne exposures with the single exception of. chloroform. All of the chlorinated VOCs were also measured in food and beverages; again chloroform was the only VOC found in significant amounts in food. Exposures through routes other than air and water have been documented for some of the pesticides. Many of the pesticides have been measured in food by the FDA for years; however, exposures in food account for only a small proportion of total exposure to the four pesticides of highest risk through airborne routes. Food exposures outweigh air exposures for some of the ther pesticides (e.g, Captan). House dust may provide an important reservoir for any or all of the pesticides, and possibly also for the least volatile of the VOCs (p-dichlorobenzene, tetrachloroethylene). DDT was found in house dust in five of eight homes in the NOPES study. Since this chemical has been banned for nearly two decades, its appearance in house dust is troubling. Ingestion of the dust by toddlers could be an important additional source of risk. The DDT may be tracked in on people's shoes from outdoor soil. Comparison with Other Environmental Risks Several organic chemicals of interest were not monitored in the TEAM Studies. Among these are formaldehyde and 1,3-butadiene. Risk estimates for these chemicals may be compared to the risks calculated above. * The carcinogenicity of formaldehyde is controversial, due to the unusual metabolic pathway associated with its carcinogenicity in rodents. Estimates for the cancer risk of formaldehyde1* range over extreme limits, from zero to 10`3. Employing an intermediate potency factor (unit risk of 1.3 X 10`5 (/ig/m3)'1) and11 11 SL 064734 measured values of 40 ng/v? for normal (non-mobil home) housing stock results in a risk of about 5 X 10`4. (Average outdoor concentrations of formaldehyde have been about 4 Mg/m3 corresponding to a risk of about one tenth of this level.) Recent animal studies of 1,3-butadiene have resulted in revising its potency upward by nearly three orders of magnitude. Although no personal exposure data are available for this chemical, a recent study has measured the level in sidestream smoke at about 400 MS per cigarette17. This is approximately th level of benzene in sidestream smoke (330 m9 per cigarette); therefore, if 1,3-butadiene is not too reactive, we can calculat that it will be elevated by about 4 Mg/m? in smoking homes and by 13 Mg/m3 in workplaces allowing smoking. Using the revised unit risk value of 2.8 X 10'4, and assuming 38 million homes with smokers averaging three residents each, 75 million workers in workplaces allowing smoking, and 26 million non-workers exposed to cigarette smoke, we arrive at an upper-bound risk associated with exposure to 1,3-butadiene in environmental tobaccos smoke of 6 X 10`4. No data exist on personal exposures or indoor concentrations of l,3-butadiene. (Outdoor concentrations of this chemical have been estimated to lie within a range of 0.3-1.6 Mg/m3, corresponding to a risk of about 1-4 X 10*4.) Thus the individual risks for formaldehyde and 1,3-butadiene are greater than the airborne risk of any of the other 35 VOCs and pesticides considered in this report. However, the great uncertainty in the carcinogenic potency of formaldehyde , including the uncertainty as to whether it is a human carcinogen at all, and the lack of exposure data for 1,3-butadiene make the risk estimates for these two chemicals particularly speculative. The combined upper-bound risk of about 10*3 associated with these 37 predominantly indoor organic chemicals appears to be similar to the risks associated w h the most severe environmental hazards (radon and r. ssive smoking). For example, the risk associated with nonsmokei 1 exposure to radon has been estimated to be about 10*3 and that with passive smoking has b en estimated18 at 2 X 10'3. It should be noted, however, that the risk estimates for radon and passive smoking are based on human epidemiology studies, and are therefore on firmer ground than all of the risk estimates for the organic chemicals with the exception of benzene. The risk estimates for these organic chemicals are considerably higher than the risks associated with some EPA regulations (NESHAPS chemicals and Superfund clean-up criteria). Similar conclusions regarding the importance of indoor air pollution compared to other environmental hazards have been reached by EPA Headquarters19 and by three Regions20; both of these reports rank indoor air pollution as among the top two or three environmental threats to public health. 12 SL 064735 J Two other risk estimates for personal exposure to VOCs have been published9,16. McCann arrived at similar risk estimates for most of the chemicals; Tancrede estimated 5-10 times higher risks, due partly to using a different method for calculating potencies from animal data and partly to considering explicitly several additional sources of uncertainty. No previous risk estimates for most of these pesticides have been possible, due to the lack of exposure information. Actions to avoid these risks may be taken by individuals. Since the sources of the risks are often personal activities (smoking, using air fresheners), these activities can be halted or modified. (For example, smokers could establish a room in th home with separate ventilation.) Exposures from chloroform could be reduced by drinking bottled water or using an activated carb n filter on the water supply. Exposures from petroleum-based products could be reduced by discarding or storing used paint cans and sprays in a detached garage or tool shed. Dry-clean d clothes could be hung outdoors for a day--one study indicates that 20-30% of tetrachloroethylene residues on the clothes.will outgas during the first day. The reason for the large number of pesticides observed in indoor air in the latest TEAM Study is not well understood. Termiticides, like radon gas, may be entering the basement due to soil gas movement; it may be that the same techniques to control radon (sealing the foundation, providing separate ducting at th entrance points) may also control termiticide entry. Other pesticides, particularly the long-lived chlorinated hydrocarbons such as DDT, may be entering the home by being tracked in on people's shoes, if so, removing shoes before entering the home, and reducing or eliminating the use of carpets or rugs (which collect large amounts of dust containing pesticides and metals as well), should reduce pesticide exposures. SUMMARY AND CONCLUSIONS Measured personal exposures to 12 VOCs and 23 pesticides in EPA's TEAM Studies have been used to arrive at upper-bound lifetime cancer risk estimates. Seven VOCs and seven pesticides have upper-bound risks ranging from 10`6 to 10'4. The combined upper-bound risk of about 10'3 from these organic indoor air pollutants is nearly comparable to the estimates of risk from radon and environmental tobacco smoke. (However, the latter two estimates are based on human epidemiology studies, and are therefore subject to far less uncertainty.) These upper-bound risks are much greater than the health risks associated with most other environmental problems. Several chemicals for which we have inadequate information, either on exposure or potency, to calculate risk were identified; 13 SL 064736 J vinylidene chloride, methylene chloride, 1,3-butadiene, formald hyde, ethylene dibromide, chlorpyrifos, and diazinon. Despite the recognized large uncertainty in these risk estimates, these findings provide additional support for the conclusion of two recent comparative rankings of environmental risk by EPA: that indoor air pollution is one of the greatest threats to public health of all environmental problems. 14 SL 064737 REFERENCES 1. Integrated Risk Information System (IRIS), US ERA, Washington, DC. 2. Engler, R., memorandum to Office of Pesticide Programs Division Directors, USEPA, October 27, 1989. 3. Wallace, Lance A., Pellizzari, E.D., Hartwell, T.D., Sparacino, C., Whitmore, R., Sheldon, L., Zelon, H. and Perritt, R., "The TEAM Study: personal exposures to toxic substances in air, drinking water, and breath of 400 residents of New Jersey, North Carolina, and North Dakota" Environmental Research 43: 290-307 (1987). 4. Wallace, L. A., Pellizzari, E.D., Hartwell, T.D., Whitmore, R., Zelon, H., Perritt, R. and Sheldon, L. "The California TEAM Study: breath concentrations and personal exposures to 26 volatile compounds in air and drinking water of 188 residents of Los Angeles, Antioch, and Pittsburg, CA." Atmos. Environ^ 22: 2141-2163 (1988). 5. Immerman, F.W. and Schaum, J.L. Final Report of the Nonoccupational.Pesticide Exposure Study (NOPES). US Environmental Protection Agency, Research Triangle Park, (1990). NC 6. Wallace, Lance A. "Cancer risks from organic chemicals in the home", in Environmental Risk Assessment: Is Analysis Useful?. APCA Specialty Conference Proceedings, Air Pollution Control Assoc., Pittsburg, PA. SP-55 (1986). 7. Wallace, Lance A., Pellizzari, E., Hartwell, T., Perritt, K., and Ziegenfus, R., "Exposures to benzene and other volatile organic compounds from active and passive smoking," Archives of Environmental Health 42: 272-279 (1987). 8. McKone, T.E. "Human exposure to VOCs in household tap water: The indoor inhalation pathway", Env. Sci. Tech. 21: 1194-1201 (1987) . 9. Tancrede, M, Wilson, R. Zeise, L. and Crouch, E.A.C., "The carcinogenic risk of organic vapors indoors: a theoretical survey," Atmos. Env. 21: 2187-2205 (1987). ^ 10. Allen, B., Crump, K. and Shipp, A., "Correlation between Carcinogenic Potency of Chemicals in Animals and Humans," Risk Anal. 8:531-544 (1988). 11. Sandler, D.P., Everson, R.B., Wilcox, A.J., and Browder, J.P. "Cancer risk in adulthood from early life exposure to parents' smoking". Am. J. Public Health. 75:467 (1985). 15 Si 12. stjemfeldt, M., Berglund, K., Lindsten, J., and Ludvigsson, J. "Maternal smoking during pregnancy and risk of childhood cancer", t.apoet - June 14, 1986:1350-52 (1986). 13. Krause, C., Mailahn, W., Nagel, R., Schulz, C., Seifert B., Ullrich D. "Occurrence of volatile organic compounds in the air of 500 Homes in the Federal Republic of Germany", in Proceedings of The 4th International Conference on Indoor Air Quality and Climate, vol. 1, pp. 102-106. Institute for Soil, Water, and Air Hygiene. Berlin (West) (1987). 14. Entz, R.C., Thomas, K. and Diachenko, G. "Residues of Volatile Halocarbons in Foods Using Headspace Gas Chromatography." J. Agrlc. Food Chem. 30:846-849 (1982). 15. Gi in, J. R. and Hodgson, A.T. "Exposure to methylene chlori from controlled use of a paint remover in a residence" paper presented at 80th annual meeting of the Air Pollution Control Association, New York, NY June 21-26, 1987. 16. McCann, J. Horn, L., Girman, J. and Nero, A.V. Potential risks from exposure to organic compounds in indoor air. Lawrence Berkeley Laboratory, report # LBL-22473. Berkeley, CA (1986). 17. Lofroth, G. Burton, B., Forehand, L., Hammond, S.K., Seila, R., Zweidinger, R., and Lewtas, J., "Characterization of environmental tobacco smoke," Env. Sci. and Tech. : (1989). 18. Repace, J., "A quantitative estimate of nonsmokers' lung cancer risk from passive smoking," Env. Int. 11:3-22 (1985). 19. usepa, Unfinished Business: A Comparative Assessment.of Environmental Problems. Washington DC (1987).20 20. USEPA, Comparing Risks and Setting Environmental Prioritiesi Overview of Three Regional Projects. Office of Policy, Planning, and Evaluation, Washington, DC (1989). 16 SL 064739 4 Table I. Upper-Bound Lifetime Cancer Risks of 12 VOCs Measured in the TEAM Studies (1980-87 Outdoor Air Chemical Exposure* Potency Risk (Mg/m*) (Mg/m*)'1 Concentration1 (Mg/m*) (X 10`6) (X10'6) Benzene Air 15 8 12 0C 6 Smokers 90 8 720c Vinylidene chloride 6.5d Chloroform Air 3 Showers (Inhalation) 2 Water 30* Food & Beverages 30* 50 23 23 2.3* 2.3* ^ 320 70 50 70 70 <1 0.6 -- -- p-Dichlorobenzene 1,2-Dibromoethane Methylene chloride Carbon tetrachloride Tetrachloroethylene Trichloroethylene 22 0.05 6f 1 15 7 4 510 4 15 0.6 1.3 90 25 24 15 9 9 0.6 0.03 2f 0.6 3 1 Styrene Air Smokers 1 0.3" 0.3 0.3 6 0.3 2 ---- 1,2-Dichloroethane 0.5 1,1,1-Trichlorethane 30 7 0.003 4 0.1 0.2 7 * Arithmetic means based on 24-hour average exposures of 750 persons in six urban areas measured in the TEAM Studies b Based on backyard measurements in 175 homes in six urban areas e The risk estimates for benzene are based on human epidemiology and are therefore mean as opposed to upper-bound estimates d Six measurements exceeding 1000 ^g/m3 dropped from the calculation; inclusion of the measurements leads to an average exposure of 150 M9/3* These figures are in Mg/L or ppb rather than Mg/3 f Based on only, eight 24-hour measurements in 1987. 9 Source: US EPA (1983) Review and Evaluation of Evidence for Cancer Associated with Air Pollution, EPA 450/5-83-006. 17 Q6A740 SL Table II. Upper-bound Lifetime Cancer Risks from Airborne Exposur s 23 Pesticides Measured in the NOPES TEAM Study Pesticide Exposure* Eotensv Risk (ng/mJ) (kg-d/mg) (X10-6) Outdoor Air Concentration1 (ng/m*) Banned Termiticides Heptachlor Chlordane Aldrin Dieldrin Heptachlor Epoxide DDE DDT 71 198 13 3 0.4 2.2 0.7 4.5 1.3 17 16 9.1 0.34 0.34 90 (19) 70 (15) 60 (13) 14 (3) 1 (0.2) 0.2 (0.04) 0.1 (0.02) 7 14 0.1 0.2 0.1 ND* ND Other Pesticides Dichlorvos 7-BHC (Lindane) a-BHC Propoxur Hexachlorobenzene Dicofol fi-Phenylphenol 2,4-D Atrazine cis-Permethrin trans-Permethrin Chlorothalonil Folpet Captan DDD Pentachlorophenol 33 6.6 0.5 100 0.3 2.6 58 0.6 0.05 0.4 0.1 0.7 0.5 0.1 <4 <730 0.29 2.7 1.3 2.5 6.3 1 0.0079 0.2 1.67 0.1 0.34 0.05 0.0016 0.02 0.019 0.003 0.22 0.003 0.022 0.003 0.022 0.001 0.011 0.002 0.0035 0.0005 0.0023 0.00007 0.34 ^0.4 0.013 -**3 v. ND 0.4 ND 2.5 0.1 ND 0.6 0.1 ND ND ND 0.5 0.2 ND ND ND * Arithmetic mean of population-weighted and seasonally-weighted average personal exposures measured for 173 persons in Jacksonville, FL and 85 persons in Springfield/Chicopee, MA. b Based on outdoor measurements at each home in the two cities. c All risks calculated assuming 70-year lifetime exposure at the measured levels. For banned pesticides, whose environmental concentrations should decrease over time, an alternative calculation of risk (in parentheses) assuming a 10-year half-life in soil is provided. d Not Detected 18 SL 064741 REGULATORY PROGRAM OF THE UNITED STATES GOVERNMENT APRIL 1,1990 - MARCH 31,1991 SL 064742 ov xw 13 Supporten of this system argue that it woulc * Devalopir t record of actual expenditures while provide an incentive for better estimate* ofthe costs of iwitiimigy the recordkeeping burden on the legislative proposals and a basis for an explicit private sector; of the costs and tradeoffs of such proposals, pigh cost ceilings would focus attention on the expected benefits of the program, and alternative approaches; cost ceilings that were too low would prevent agencies from issuing implementing regula tions. Such an approach would, needless to say, give agencies an incentive to choose regulatory approaches Identifying an appropriate "baseline,* reeogniang that some costs would be incurred even in the absence of Federal regulation; and Estimating the costs of forgoing certain products whore Federal regulation prohibits production or distribution. } that would produce the greatest benefits at the lowest Each of these raises difficult issues in designing an effective regulatory budget process. For example, the coats of banning a product are not directly measurable and can only be estimated using complex statistical ISSUES AND AREAS FOR FURTHER STUDY mode's. However, measuring only the reel compli- ac 'its for oversight purposes creates a bias toward While the fiscal budget process provides a continu be g substances and products instead of control- ous record of actual expenditures, there is no compara hi am. ble record of the cost of meeting regulatory require a first step in determining the feasibility of the ments* Members of Congress and the past two regulatory budget concept, OMB has begun systemati Administrations have considered developing an ac cally to collect the costs of ah significant published counting framework to record direct regulatory expen regulatory actions. Analysis ofthese data should aid in ditures, but more work needs to be done to solve the the development of ways to overcome the problems of practical accounting problems inherent in measuring regulatory budgeting, uncover unforeseen' problems in the private expenditures that Federal regulations developing cost estimates, and more fully refine a mandate. These include: workable regulatory budgeting process. Current Regulatory Issues in Risk Assessment and Risk Management Many Federal agency regulatory decisions are intended to reduce risks to human life and health. Government regulations control which agricultural chemicals may be used to reduce insect damage, increase farm yields, and improve the quality of food products. Other rules govern hazards in the Nation's workplaces and emissions from its factories. There are regulations directing the way in which automobiles must be manufactured, commercial aircraft main tained, and trains operated. Hardly any widespread human activity that entails risk is free of some degree of social control, often achieved through government regulation. Regulatory decisions involving risk require agencies to address questions such as, "How safe is `safe*?* and *How clean is 'clean'?" When government agencies promulgate regulations intended to reduce a riak or mitigate a hazard, they are engaging in what has become known as risk management. These policy choices inevitably involve consideration of both the risks entailed by the underlying activity and the social consequences of regulatory intervention. Thus, the first challenge of risk management is to set priorities to determine which risks are worth reducing and which are not Far government to carry out its risk-management responsibilities, there must be an extensive invest ment in the careful assessment and quantification of risks. The term risk (uatttmtni means the applica tion of credible scientific principles and statistical methods to develop estimates of the likely effects of natural phenomena and human activities. The need to keep risk assessment and risk manage ment separate has long been the objective of responsi ble public officials. In 1983, the National Academy of Sciences (NAS) studied the process of managing risk * Research*TM, usinf different method*, aatumptions, end time period*, have formed incomplete cellmate* by adding up the coat individual regulation*. The** otimatet accordingly chow eontidcrabi* venation tor currant annual ooatc ranging from 960 billion to SI billion a year--4 to 15 percent of current Federal outlays. 1 i SL 064743 R^GULATOHY PROGRAM OF THE UNITED STATES GOVERNMENT in the Federal Govanua mt pnd offend the following (commendations, among-uibers: Recommendation 1: Ragulatory ageobae ahould take (tap* Id mtabliab and maintain a dear eocoeptual distinction botwoec aaeoaemont of risk* aad the oonmdarutina of riak management altarnativw; that is, the adontific finding* and policy judgment* embodied fat riak aiaaaantanta ahould b* mplidtly dietinguiohod from tha political, -- and i consideration* that teflusnee tha daaigB and Adw of initiatory aVatagiaa.u Jbeommendadon 2: Bafon as agency deride* vhethm a subatanm ahould or ahould not ba rafulatad aa a health bound, a datailad and onmprahanaive written risk assess ment ahould ba prepared and made publicly available. This written aaaetasimt ahould dandy distinguish be tween tha scientific baai* and tha pokey bada tar tha agincy** mneluaion*. Hie belief that risk assessment and riak manage ment should be kept separate enjoys widespread support among profeaiional risk-assessment practi tioners and risk-management officials. ^Others have emphasized the importance of ensuring that policy biases do not distort the analysis of alternative risk-management choices.*4 Hie NAS principles have also have been endorsed by a number of Federal agencies, including the Office of Science and Technol ogy Policy (OSTP), the Environmental Protection Agency (EPA), and the Department of Health and Human Services (HHS).* Unfortunately, risk-assessment practices continue to rely on conservative models and assumptions that effectively intermingle important policy judgments within the scientific assessment of risk. Policymakers must mike decisions based on risk assessments in which scientific findings cannot be readily differenti ated from embedded policy judgments. This policy environment makes it difficult to discern serious hazards from trivial ones, and distorts the ordering of the Government's regulatory priorities, hi some cases, the distortion of priorities may actually increase health and safety risks. Ibis section explores some of the continuing difficul ties that plague the practice of risk assessment, and describes briefly their policy implications. It can be summarised in three observations: The continued reliance on conservative (wont-case) assumptions distorts risk assessment, yielding esti mates that may overstate likely risks by several orders cf magnitude. Many risk assessments are based on animal bioassays utilizing sensitive rodent species dosed at extremely high levels. Conservative statisti cal models are used to predict low-dose human health risks, based on the assumption that human biological response mimics that observed to laboratory animals. Wont-case assumptions oanearning actual human exposure are commonly used instead of empirical data, Anther exaggerating predicted risk levels. Conservative biases embedded in risk assessment impart a substantial "margin of safety*. The choice of an appropriate margin of safety should remain the province of responsible risk-management officials, and should not be preempted through biased risk assess ments. Estimates of risk often fail to acknowledge the presence of considerable uncertainty, nor do they present the extant to which conservative assumptions overstate likely risks. Analyses of risk-management alternatives routinely ignore these uncertainties and treat the resulting upper-bound estimates as reliable guides to the likely consequences of regulatory action. Decisionmakers and the general public often incor rectly infer a leva! of scientific precision and accuracy in the risk-assessment process that does not exist. Conservatism in risk assessment distorts the regula tory priorities of the Federal Government, directing societal resources to reduce what are often trivial carcinogenic risks while failing to address more substantial threats to lift and health. Distortions are probably most severe in the area of cancer-risk assessment, because many conservative models and assumptions ware developed specifically far estunst- 1 j < " National Academy of Sdtncee, Rial Aseesement in the federal Government: Managing the Premet, Washington, DC: National a--y Pi--i. IMS (hereinafter, AMS Risk Management Study), p. iSL p. 153. "For representative view* of risk-aaaeaament practitioners aee, aa., Laatar B. Laos, TV Struts# ef Regulation: Decision frameworks for Policy, Washington, DC: Brookinfi, 1981; Laatar B. La**, "Method* of Riak Assessment," Chaptar in Quandtonve Risk Atonement in Regulation, Laatar B. Lav, ad., Waahiaftoa, DC: Braokanga. 1982, aap. pp. 82-64. For representative views ofrisk-management affinal* mc. t.| , William D Ruekalabaua, "Saenoe, Riak, Bad Public Policy, Vital Speeches of the Dor, Volume 49, No. SO August 1, 1983, pp. 812-615. **Saa, .f,, Howard Kuarautbar and liaa Bendixm, "Benefit* Aaaaaamaat for Ragulatory Problem*,* and Piscbbaff and Lams Anthony Con, Jr., "Conceptual Framework for Regulatory Benefit! Aaaaaamaat,* Chapter* 3 and 4, rwpoctraely, is Benefits Assessment: The State of the Art, Judith D. Beatkow, Vincent T. Covello, and Jeryl Mumpowar, ada., Dordrecht, Netherlands D. 1966, pp. 44-46. 69-61. 44 Sae U.S Office of Science and Technok>c Policy, "Chemical Carcinogen*: A Review of the Science and Ita Aaaotialad Pnnoplee,* Priaaple 29 (60 FS 10378, March 14, 1966, hereinafter, OSTP Risk Assessment Guideline*)-, US. Environmental Protaction Agency, Guideline* for Carcinogen Riak Aaaaaamaat,* 61 FR 34001 (September 34,1986, hereinafter, SPA Carcinogen Risk Assessment Guidelines); US. Department of Health and Human Servioaa, Risk Assessment and Risk Management of Tome Substances, April 1986, p. 20. SL 064744 OVERVIEW 15 fag upper bound* for these risk*. Sisk-essessaent methods with linular conservative biases are leu common elsewhere, particularly in those areas where resl-world data are available, or where the mechanism by which injury or illness occurs is better understood. A renewed commitment to the NAS recommenda tions is clearly warranted. As quantitative riak ntffwwnt plays an increasingly significant role in risk management, the need to separate science from policy becomes ever more important, if either process b to public confidence. As former EPA Administrator William D. Ruckelshaus has noted: Riak UMacacot.-Burt be baaad > ecientific < ecientifle coneenaua only Nothing will erode public eunfi- 4anw faster than the auapidon mat policy eenaiderttkaa haw bets allowed to influence the aaaaeetnent rf riak. ** ALTERNATIVE RISK-ASSESSMENT METHODOLOGIES Risk assessments of chemical substances in general (and of possible carcinogens in particular) involve a mixture of facts, models, and assumptions. Thera is considerable debate concerning the scientific merita of the models and assumptions commonly usod in risk assessments. In some cases, a scientific consensus has developed to support a particular model or assump tion. In other instances, however, certain models end assumptions are relied upon because they reflect past practices rather than the leading edge of science. Furthermore, a scientific basis for several of the moet critical models and assumptions simply does not exist Most scientists agree that these models and as sumptions impart a conservative bias: that is, they lead to risk projections that the actual (but unknown) risk is very unlikely to exceed. These "upper-bound* estimates are often useful as a screening device, to exclude from regulatory concern potential hazards that are insignificant even under worst-eaae condi tions. Unfortunately, upper-bound riak estimates are routinely employed for altogether different purpoeee, such as estimating the likely benefits of regulatory actions. Policymakers are required to act on the basis of biased representations of both the magnitude of the underlying hazard and the extant to which Govern ment action will ameliorate it. Contemporary riak assessment relies heavily upon wwiwf1 bioassay and epidemiology. Each approach theoretical advantages and disadvantages. In practice, both be misused to bolster preestablished conclu sions. The following dismission emphasizes problems in cardnogenic risk assessment, because the preven tion and cure of cancer plays such a major role in policy issues involving risks to life and health. Animal Bioaaaay Animal testir enables scientists to estimate risks ex ante, before uman health affects materialise, whereas epidemiological studies can only detect auch affects ex poet. In addition, animal tests can be conducted under tightly controlled laboratory condi tions, which provide more reliable estimates of exposure and avoid many of the confounding factors that often plague epidemiological investigations. The relatively abort lifetimes of experimental mammals (such as rata and mica) allow edentists to ascertain the possible effect* of long-term exposure in just a few years. Animal testing suffers serious limitations, how*ve|^ arising from certain critical assumptions. Despite i^P routine application, there is no accepted atientific basis for the assumption that results can be meaning- frilly extrapolated from test animals to humans. Some scientists believe that animal data should not be used in assessing human health risks." Another critical limitation is the reliance on very high dose* to generate adverse effects in teat ani mals." A mathematical model must be used to bridge the gap batween these high-dote exposures and the low-dose exposures more typically freed by people. Many different mathematical models can be con structed to fit the data at high doses. These models often vary anarmouily, however, in their predictions of risk at kiw doses. Beyond these unavoidable methodological con straints, the results of bioassays may be subject to conflicting scientific interpretation or strongly influenced by the choice of research method. * William D. Ruckelahaua, (op. etc), p. *14. OOSSTTPr GQuuidideililnime*, Guiiddetlliianfet 8, p. 108376-. "Set, t.| , Bruce Aum, Renee Magew, and lni Swiraky Geld, *RenkangPoeetble Caranogenic Haiarda,* Sdroee, Vol. 336, April 17, 161*: Gio Bette Geri, *Tht Regulation of Carcinogenic Hezerda,* Science, V*L'Z06, April 16, 1660. * OSTP GuuUhnet, Guideline 11, p. 10377. SL 064745 16 regulatory program OF THE united states government Tissue preparation and histology present obvious opportunities for error, as experts may disagree as to how slides should be interpreted.*0 This problem generally is not significant at high doses, where malignancies are often obvious. At low doses, however, pathologists often differ in how they Kativfnieh tumors from hyperplasia. Subjectivity cannot be avoided where such interpretations of the data must be made.*1 epidemiological studies often display contradictor results" Despite these constraints, properly conducted ani mal bioassays and epidemiological studies both hive useful roles to play in quantitative risk assessment. Indeed, they are complementary. The usual weak nesses of epidemiological investigations--unreliable exposure data, confounding effects--are readily avoided in laboratory experiments on animals. The weaknesses of animal bioasaays--high- to low-dose Epidemiology extrapolation, animal-to-man conversion--do not arise in epidemiological studies. Careful risk assessment Epidemiology is attractive because it largely avoids incorporates both types of analysis to ensure that the these two problems. It focuses on observable human emerging picture of human health risk it as complete health effects instead of on hypothesized outcomes as possible, and that inferences derived from this based on animal experimentation, and it relies upon picture are themselves internally consistent real-world exposures to generate empirical data. Many I of the serious problems associated with animal stadias can be avoided, allowing researchers to develop risk ISSUES IN RISK ASSESSMENTS DERIVED LARGELY FROM ANIMAL BIOASSAYS estimates that are directly related to human health. Unfortunately, epidemiological research suffers from its own set of limitations. For example, retrospective studies often have difficulty correlating morbidity and mortality with exposure to specific substances. Expo sure data are commonly lacking, incomplete, impre cise, or affected by systematic recall or selection biases. Furthermore, the risks these studies seek to detect are often very small relative to background, thus making statistically significant effects difficult to observe. When health effects are latent, correlating exposures to illness is even harder. ' Besides these unavoidable methodological Hmiteepidemiological studies often suffer from out right bias. Many studies employ scientifically ques tionable procedures aimed at demonstrating positive relationships between specific substances and human illness.13 Some researchers use inappropriate statisti cal procedures to "mine" existing databases in search of associations. One result of these practices is that Animal bioasaays tend to dominate current risk assessments. An important reason for this is that the derivation of dote-rasponse relationships is a critical regulatory motive for performing quantitative risk assessment Animal studies are ideally suited to serve this purpose by virtue of the controlled condition* under which dote and response can be calibrated. Epidemiological studies often are relegated to provid ing merely a "reality check" to ensure that the implications of animal bioaasays are plausibly consis tent with real-world experience. Because of this heavy emphasis on animal testing, the focus here is on several major problems that arise with respect to risk assessments primarily baaed on the results of nnimni bioasaays. Hie Uae of Sensitive That Animals lb enhance the power of nnitnnl tests, scientists typically rely on genetically sensitive test animal. It "la the original analyei* of the rat bioatuy used to derive the doae-raeponae fimetioa for diaria, 9 of 86 control* were wid to dtvelop hrar tumor*. Ad independent review of thi* data raauhad is 16 of the 86 control* being daaaiSed a* having aueh tumor*. Sac U.S. Environmental Protection Agency. A Cancer Ruk-Spttifte Dorn Ettimatt for 2,3,7, t-TCDD, Apptndu A, EPA/800/6-MX)7Ab, June 196S (hereinafter, Dioxin Ruk Autumtnt Appendix A), pp. 9-3. " Conn N. Park and Ronald D. Snec, "Quantitative Ri*k Aaaasement: Stata-of-the-Art for Carcmofmwia," Chapter 4 m Kttk Management of Emitting Chtmeoii, Rockville, MD: Government Institutes, 1983, p 66. " Alvan R Feinstsin, "Scientific Standard* in Epidemiological Studies of the Menace of Daily lift." Scknrr Vol. 942 December 2. 1966, pp. 1257-1263. * Linda C. Maye*. Ralph 1 Horowitz, and Alvan R. Feinstsin, *A Collection of 36 Topics with Contradictory Result* in Caee-Cmitrol Research,* International Journal of Epidtmi6U&, Vol 17, No. 3 (1968), pp. 660-665. SL 064746 OVERVIEW 17 t k e a *. e B. e ? i 1 i t It t P S: i 9 I l ^ asdear whether these spedee aecuntely mimic biologic*! responses in humxns. Some test spedes ere extremely sensitive. For tpproximetely one-third of ell male B6C3F1 miee, a common test spedes, spontaneously develop Sver tumor*.** The same phenomenon occurred in an important bioassay concerning dioxin using female Sprague-Dawley (Spartan) rats. Tumors observed in dosed were predominantly located in the liver. However, approximately one-fifth ofthe animals in the control group also developed liver tumors." The relevance of elevated liver tumors in hypersensitive spedes has been questioned by sdentists and is ne t universally considered probative evidence of earcto<> genidty. Nevertheless, cancer risk assessments often proceed on the assumption that these data sue sufficient to conclude that a substance ia indeed a carcinogen.** The reliance on sensitive test animals also biases risk assessments in a more subtle way. It astablishes powerful incentives to search for and develop increas ingly sensitive test spedes. As test animals become more sensitive, repeated testing using identical proto cols will tend to result in higher and higher estimates ef risk even if all other factor* are held constant Selective Use of Alternative Studies In their respective risk-assessment guidelines, both OSTP and EPA recommend that relevant awin*ei studies should be considered irrespective of whether they indicate a positive relationship.11 In practice, however, studies that demonstrate a statistically significant positive relationship routinely receive more weight chaw studies that indicate no relationship at all." For example, the plant growth regulator daminoxide (Alar) end its metabolite unsymmetricel 1,1-dimethylhydrexine (UDMH) recently received B2 classifications ("probable human carcinogen"). Each of these classifications was baaed on a single positive awtiwat bioassay." Overcoming such a classification requires, at a minimum, two "essentially identical" studies showing no such relationship.** In the case of Alar and UDMH, however, a more stringent test was apparently applied: Three high-quality negative stud ies showed no significant effects; these studies appear to hav received little or no wsight in the classification derision.41 Selective Interpretation of Results Bisk-assessment guidelines generally give the great- _ eat weight to the most sensitive test animals. Thus, a substance has been found to cause cancer in one^^ **Ame* tt ol., (op. at), p. 276. uDuxan Risk Assessment Appendix A, pp. 2-3. * Stf Am** tt ol., (op. eitj, p. 276 (arguing that rack data aiv torwalantk OS7T Ouidelinm GnJdttiat 9, p. 10377 (concluding that inch data *must be approached carefully"); and EPA Caninottn Risk Assessment Guidelines, p. 33995 (making the polity Judgment that such data art sufficient evidence of caraaogenesia). Liver tumor* dominated in EPA'* dionn risk aaaaasmant 8m Diem Risk Assessment, appends A, pp.2-3 r See OSTP Guidelines, Guideline 25. p 10371; EPA Carcinogen Riak Assessment Guidelines, p. 19996. "$* EPA Carcinogen Riak Assessment Guidelines, p. 33999-34000. A ringk animal taat that ahowi a positive reeult "to an unuaual decree' ip 33999) is euilident to warrant at least a B2 classification ("probable human carcinagen*), evon if this roauh occur* in a apede* known to have a bifh rate of ipontaneous turnon. A atrong animal bioaaaay or epidemiolo(ical study abowins no evidence of carcinogenic effect cannot overcome this pnaumption (p. 34000). "Sec *Second Peer Review of Demineddc (Alar) and UDMH (Unsymiaetrieal 1,1-dfanetbyIhydrarin*),* Mamorundum from John A Quest to Mark Boodoe. U.S. Environmental Protection Agency, OPTS, May 16,1999 (hereinafter, AJor/UDUH Internal Roar Review No. Si This internal OPTS pant) reviewed aovewu recant etudie* on Alar aad UDMH. One study ef Alar yielded a eutistkally .tnificanl increaee in common hag turnon in mice, but only for one ef throe doeage levels. Results were not statistically aigwidcaat at Higher and two lower dotages, and eontrola alao displayed unusually high tumor inddance. 900 of the lung tumors in doeed mice were benign, venue 89% in the controls. One study of UDMH yielded statistically significant increase* in common lung and uncommon liver turnon in mice, but only for the higher ef two dosages. 97% ef the lung turnon in doeed mice wen benign, versus 100% in the oontrol*. 29% of the liver turnon in dosed miet were benign; no tumor* wen observed in the controls. Prior studies thst purported to show a carcinogenic response had been Judged inadequate by EPA's Scientific Advisory Panel, an asternal peer review group. The Office ef Pesticide* and Toxic Substance* (OPTS) panel noted that a different internal EPA risk-assessment panel (the Carcinogen Assessment Group) considered these studies sufficient to justify B2 classifications whan it evaluated them for EPA'* Office ef Solid Waste and Emergency Response. Despite the scientific controversy, the OPTS panel interpreted these prior studio* as 'supporting evidence* under EPa'i nsk-assetsment guideline*. MSee EPA Carcinogen Risk Aaatssmanl Ousdtiintt. p 33995 (establishing the need for replicate identical studies showing no effect), and p 339991 establishing the "m* requirement of two well-designed studies showing no increased tumor incidence to warrant a "no ewdence" determination). a *'AlorlVDUH internal Peer Review No. I, pp. 6,1, 9. EPA's scheme for carcinogen elasaifimtion is itself an issue among Kientietifl See. eg . VS Environmental Protection Agency, Risk Assessment Forum, Workshop Rapert on EPA Guidelines for Carcinogen Risk Assessment. EPA/62V3-89^015, Washington, DC: March 1989, pp. 21-26. SL O64747 -iSveaiMBMMaMBMBMBuaha J'seah lea**1*I 18 REGULATORY PROGRAM 07 THE UNITED STATES QOVKRNMENT spades or fender but shown to exhibit no effects elsewhere, the results pertaining to the sensitive spedes or gender typically will be used to develop estimates of human-health risks. For example, if male mice develop cancer from a substance but female mioe and rats of both genders do not, then the results from the male mouse often will be used to derive estimates of cancer risks to humans.41 Once a positive result has been obtained in an animal bioassay, a substance often will be provision ally classified as a probable human carcinogen. 1%e statistical burden of proof then shifts to the no-affect hypothesis. Because it is logically impossible to prove a negative, however, this practice establishes a virtually irrebuttable presumption in favor of the carcinogenesis hypothesis. Severe Testing Conditions Current risk-assessment protocols require the use of very high doses. Unfortunately, high doses are often toxic for reasons unrelated to their capadty to cause cancer. A common procedure is to use what is called the maximum tolerated dose (MTD), which is the most that can be administered to s test aTMai without causing acute toxicity. At such exposure levels, substances often cause severe inflammation chronic cell killing. For example, formaldehyde causes nasal tumors in rats when administered in high doses. However, MTD administration severely iwAmbm nasal passage tissues. It is therefore unclear whether the cancers induced are caused by formaldehyde per sc or by the toxic effects of high doses. Results such as these have caused some scientists to question the validity of rodent tests performed at the MTD for estimating human health risks that arise from exposure at low doses.41 By combining very high doses with highly sensitive test subjects, some animal biossssys are predisposed to discover apparent cardnogenic effects. Belevmnoe of Animal Bioaaaay Results An Important reason why animals vary is their snaitivity is that they have different physiologies, metabolic processes, reproductive cycles, and a host of other epedas spedfic diaracteristics that largely re sult from unique evolutionary paths. Each of these factors needs to be carefully considered in evaluating tbs significance of "<***! data with respect to human health. This is recognised in both the OSTP and EPA guidelines, but it is often neglected when the guide lines are applied to specific substances.44 The most important assumption in this regard is that niwul test results can be meaningfully extrapo lated to humans. A recent study of chemicals tasted under the auspices of the U.6. National Tbrieology Program shows that this assumption can lead to the erroneous classification atmany chemicals as probable human carcinogens.41 Positive associations have been obtained In cither ruts or mice for half of 214 chemicals tasted. However, results were consistent across these two genetically similar spedes only 70 percent of the time. If it is assumed that rodent bioassays have the same sensitivity and selectivity with respect to human eardnogens as they do between rodent spedes, and it Is Airther assumed that 10 percent of all chemicals are in fact human eardno gens, than 27 of every 100 randomly selected chemicals would be misclassifted as probable human eardnogens. Only three chemicals would be misclassified as noneardnogens. Thus, "false positives" would be 9 times more common than "false negatives."41 Of course, this ratio of false positives to false negatives reflects highly conservative "upper-bound" assumptions concerning sensitivity and aelectivity. Given the high degree of similarity between rats and mice and the limited rasemblanoe between rodents and humans, the sensitivity of rodent biaassayi with respect to human eardnogenidty is probably much lower than 70 percent Furthermore, other research indicates that selectivity may be as low as 6 percent. f i 1 . * See SPA Caninogwn Ritk Aomurntni Ouidtli/m, p. 83997 (dU beat long-tom studios showing th greatest sensitivity should generally b given the greatest emphasis). 44 Set. e.g., Ames tt ai, (op. eUJ, pp. 276-277. ** OSTP Guuithntt, Guideline 25, p. 10378; SPA Cardnoptn Ritk Atmttmgnl Qviddum, p. 34003 (responding to eesnment* on the draft guideline* and affirming agreement with OSTP Guideline 26). 41 Letter B Lave, Fanny K. Ennevar, Herbert S. FneenVrani, and QQbwt 8. Oaaim, Information Value ef the Rodent Biaaay* Notwo, Vol. 336 (December 15, 1068), pp. 631-633. mFalM negatives occur when e teat fails to detect affects whan they are in ftet present Stnsitwity refers to the eepedty of test to toinimire false negatives. Folse positive* occur when a toet appears to detect effects that to feet art absent Selectivity refers to a tost't ability to min:mile false positives. Th* 9 to 1 ratio of false positives to false negativee calculated by Lava tt al. and sensitivity equal about 70%. that both selectivity SL 064748 Aftuftiiif only for this lower selectivity suggests thst ftlae pocitivo ere almost 30 times more common than fti-- negatives. This raises serious questions concern ing the practical utility of the current approach to npiw.i bioessays for the purpose of quantitative risk assessment.4* Other lectors should also be considered when nlyiaf upon nirTt*l bioaasay results as the primary h-rit for quantitative risk assessments. For example, certain substances are toxic or even cardnofenic by one pathway but not by others. Nevertheless, animal bioassay protocols often emphasize the most sensitive pathway. As long as human exposure is likely to arise the same way, then this choice may be reasonable. However, the pathwey to which the test spades is sensitive sometimes reflects an exposure route that la implausible or irrelevant for humans. For example, formaldehyde causes nasal tumors in rots at 12 timss the rate observed in the next most sensitive animal pedes. This extreme aenaitivity may be related to the foct that rats breathe only through the note. There may be important differences between animale and humans that make specific tumors irrele vant. For example, some chemicals cause cancer in the cymbal gland of the rat; because humans lack such a gland it is unclear whether these results matter in estimating human health risk. Other substances induce cancer through biochemical mechanisms not found in humans. A greater controversy surrounds the question whether the same weight should be given to benign and malignant tumors. The sdentific consensus is that benign and malignant tumors should be aggregated only when it is scientifically defensible to do so.4* In practice, however, benign and malignant tumors are routinely aggregated unless a strong case can be made against the practice.41 The difference between these default assumptions is significant: One approach counts only carcinomas that ore present, whereas the other counts tumors that might become carcinomas. In an extreme case, a substance that promotes benign tumors but never causes cancer could be classified as a probable human carcinogen simply because benign and malignant turnon are treated equally. In addition, tumor incidence is commonly pooled across sites to obtain a total aatimate of carcinogenic affects.10 This implicitly assumes that cancer induction it independent across cites and not the result of either metastasis or the same biological mechanism. Given the extreme sensitivity of test aperies and the regular use of MID administration, other explanations for turnon occurring at multiple cites appear just aa plausible. The Choice of Dose-Response Model No single mathematical model la accepted as generally superior for extrapolating from high to low doses.*1 Consequently, Federal agendas often use a variety of different models. Rather than being a scientific footnote to the risk-assessment process, however, the choice of model la actually an important policy issue. The multistage model appears to be the most *vwwnwily used method for estimating low-doee risks from chemicals, and then are two major aouroes of bias embedded in this choice; its inherent conserva tism at low doses, and the routine use of the linearised* form in which the 95 percent upper bound is used instead of the unbiased estimate. The multistage model essentially involves fitting a polynomial to a data sat, with the number of `stages* identified by the number cf terms in the polynomial. Since animal bioasaays rarely have more than three dose levels, it is unusual to see applications of the multistage model with more than two stages. Al though the multistage model enjoys some scientific support because it is compatible with multistage theories of carcinogenesis, in practice the model fails to include enough stages, due to the absence of sufficient alternative exposure cohorts. The multistage model typically yields low-dose risk estimates that art higher than most ether models. For example, when five different dose-response models were analyzed in a recent risk aasaasment of cad mium, estimates of cancer risks at moderate dotes varied by a factor of 100. This difference among c' Lav* ft oi,, (op. dt.), p. 631. Adjusting *1* tar Wo* emoitivity reduce* the ratio of tola* positive* to foist negative*. Far example. if ornsiuvity u only 10 pereem end all ether parameter* remain unchanged, than thia ratio deelinae to 9.5 to 1. However, thia impha* that hath type* of (tatiiticaJ error* are rampant, which raiaea queetaonj concerning the practimi utility at bioeaasye. lfeia ia, in fact, precisely the concern railed by Lave tt id., (op. diJ, who conclude that tuch tecta are oot-effactive investment* to information only under wnraordinary condition*. * OSTP Guidelines, p. 10376. EPA Carcinogen Ruk Ammmeat Quidtlvm, p- 33997. mli. " OSTP Guidelines, Guideline X, p. 10378; Am#* at aL, (ep. tit), p. 276. SL 064749 80 REGULATORY PROGRAM OF THE UNITED STATES GOVERNMENT l f estimates widened u doeee declined toward the very low levels within the range of regulatory concern. At eery low doses, two of the five models predicted excess lifetime cancer risks greater than one in one thousand (104), a risk oftentimes regarded by policymakers as unacceptable. However, two other equally plausible models predicted essentially no excess cancer risk at all Since none of the five models offera a scientifically superior basis for deriving low-dose risks, the choice of model is therefore a pivotal policy decision. The accepted practice under these circumstances is to develop s subjectively-derived "best" estimste while ftilly informing decisionmakers as to the extent of uncertainty surrounding it" In the cadmium case, as in most others, this practice was not followed: Estimates of the number of statistical cancers that would be prevented by regulation were presented based only on the multistage model." The linearized multistage model (IMS) is a special version of the multistage model in which the 95 percent upper confidence limit of the linear term is used instead of the unbiased estimste. That is, the model identifies the largest value for the linear term that cannot be rejected at the 95 percent confidence level and uses it in place of the unbiased estimate. Assuming that the model has been correctly specified, there is only a 5 percent chance that the true risk exceeds this level. The LMS has become the preferred statistical approach because estimates derived from it appear to be more "stable* than estimates obtained from the ordinary multistage model. The "stability1' issue origi nally arose because unbiased estimates of low-dose risks are very sensitive to the maximum-likelihood estimate (MLE) of the value of the linear term. When the MLE of the linear term is positive, it dominates estimated risks at low doses. In some instances, however, the MLE of the linear term is sero, and low-dose risk estimates decline precipitously. Using the 95 percent upper confidence limit ensures that the linear term is always positive, thus eliminating the inherent "instability" of low-dose risk estimates de rived from the multistage model. ** Another often-dted advantage ofthe LMS procedure fa that it provides a "yardstick" for comparing [ potencies across chemicals.H A uniform risk-assess- [ ment procedure such as the LMS, it is argued, enables policymakers to better understand the relative aignifi- , eance of a broad array of chemical hazards and set regulatory priorities accordingly. Finally, tilt IMS is often defended on the ground that it is prudent to err on the side of caution when uifag with potentially carcinogenic chemicals. Be cause the LMS generates upper-bound risk estimates, policymakers can be confident that actual risks are likely to be lower. None of these purported advantages of the LMS approach has a sound statistical basis. It is s fundamental axiom of statistics that unbiased esti mates are generally preferred to biased ones. Using the upper confidence limit instead of the unbiased ! estimate exaggerates underlying specification errors jinstead of eliminating them. "Instability" is overcome, but at the cost of greater errors in specification. The inherent instability of the multistage model reflects a generalized misspedfication of doseresponse--that is, the real human dose-response relationship is often very different from what the multistage model constrains it to be. The model is extremely sensitive to small differences in observed tumor incidence, which can cause dramatic changes in estimated low-dose risks. The IMS procedure elimi nates this sensitivity without remedying the underly ing specification error. Proper statistical procedure requires correcting model misspecification, not mask ing its symptoms behind biased parameter estimates. The LMS procedure inflates low-dose risk estimates by s factor of two or three when the MLE of the linear term is positive. However, it increases low-dose risk estimates by orders of magnitude when the MLE at the linear term is zero" This means that the degree of hidden conservative bias is substantially greater for what are demonstrably lower riaka. By its very nature, the IMS cannot serve si s useful yardstick for comparing the relative risk ofs variety of potential carcinogens. If s given statistics] procedure generated identical biases across substances tested, eg., OSTP Guideline*, Guideline* 27, 29, and 31, p. 10S78; SPA Carcinogen Itiek Assessment Guideline*, pp. 33999, 34003. " Occupation*} Safety sad Health Adminiatrmtion, "Occupational Expaaure to Cadmium; Pmpoead Rule," AS FR 4076 (February 6, 1990). ** Albert L. Nicboli and Richard J. Zeckhauaar, "IT* Danger* of Caution: Co&aervatiam in Awelament and th Mismanagement of Risk," Chapter 3 m Advance* in Applied Micro-Economics, Volume 4: Riek, Uncertainty, and the Valuation of Benefits and Cost*, V. Kerry Smith, ad., Greenwich, CT: JAJ Pmi, 1966, pp 55-82, aep. pp. 62-63. A nontechnical vermes of thia paper i* available by tbe tame authort a> "The Peril* of Prudence: How Conservative Riek Aeeeatmenta Distort Regulation," Rszulotion, November/December 1966, pp 13-24. **U.S. Environmental Protection Agency, A Cancer Risk-Specific Doe* Estimate for tJ,7,S-TCDD, EPA^00/6-84O07Aa, Jon* 1966 (hereinafter, Dioxin Risk Atuument), pp. 45--46. w Nickel* and Zeckhauaar, qp. at., pp. 62-63. SL 064750 OVERVIEW 21 tfaeo it would stQl yield eccurate rank-ordering of theoretical hazard*. Similarly, if the procedure added a stochastic bias from a uniformly distributed random variable, the resulting rank-ordering would still be accurate on an czpected-value basis. The problem with the IMS is that it generates biases that intensify with the dtgrt to which the multistage model miaapedfies the true dose-response relationship. Evan if the multistage model provided an accurate rank-ordering of hazards, the LMS could not do so, because it injects biases that are systematic with atatiatieal miaspedfication. The LMS procedure (and the multistage model itself) is also fatally flawed as a yardstick for regulatory priority setting because it fails to take account of human exposure in the calculation of unit risks. Regardless of the procedure's capacity to accurately rank-order hazards, failing to adjust unit risks by relative human expoturt virtually guarantees that regulatory priorities will be miaordered. Re sources tend to be focused on reducing the greatest theoretical hazards rather than the most significant human health risks* Finally, the "margin of safety" argument in favor of the LMS unequivocally contradicts the widely recog nized need to distinguish science from policy.*4 The LMS introduces into each risk assessment a con servative bias of varying but unknown magnitude. This practice fundamentally altera regulatory derisionmaking. Instead of leaving policy decisions to policymakers, the LMS disguises fundamental policy derisions concerning the appropriate margin of safety behind the veil of science. In summary, the LMS cannot be justified at a method of scientific risk assessment Tb* "yardstick* defense implicitly asserts that scientific advancements in risk-assessment methodology ehould take a back seat to the preservation of an outdated and misguided statistical procedure. The "margin of safety* argument tacitly usurps from policymakers the authority and responsibility far risk-management derisions. Finally, the statistical "instability* overcome by the LMS is an artifact of specification error, not any scientific theory ofhuman carrinogenaeii that warrants the intentional use of biased parameter estimates. The habitual reliance upon cither the multistage model or its LMS descendant cannot be supported by sound scientific principles. Alternative models are available, ofcourse, and they have been applied in many quantitative risk assess ment*. Because proper model specification is the foundation of applied statistical methodology, alterna tives to the multistage model should be expected and encouraged. Indeed, innovation is the hallmark of scientific inquiry; policies that institutionalize any particular modal specification affactively stifle scien tific advancement. Unfortunately, models other than the multistage model are often discouraged in practice.** Agendas may require substantial scientific evidence in support of an alternative model before allowing it to be used. Alternative models thus face a burden of demonstrat ing scientific plausibility that the multistage model cannot satisfy. Evan in the extraordinary ease in which this burden can be satisfied, estimates may be required from the linearized multistage model any way* The potential bur in health threat posed by dioxins provides an excelle example of the problem of model selection. Using the *ame linearized multistage model, EPA, the Centers for Disease Control (CDC), end the Pood and Drug Administration (FDA) have arrived at upper-bound risk estimates that span an order of magnitude.*1 Depending on the data and assumptions used, the linearized multistage model predicts unit risk factors that vary fay as much aa L200, with the *' Some seenusta have stumpud to devise alternative indexes of relative human health riak that sn&dtfr amount for variations in human exposure. Ames rt/., (op at.), pp. 272-273, describe m sari alternative (tha Human EspeeurWRodent Potency index, or HERP) and icpu., .slues for 36 ouboUneos. Because tho HERP index is baaed an a relative rather than abaohtta aeale, the distorting affect of eonsenattv* bias** embedded in tho underlying risk assessments has bean aifnlflcanUy reduced- Many subetanoas suspected of bemf mvirenmcntsl carcinogens rank very- law on tha HERP index, suggesting that regulatory priorities have boon seriously misdirected Sec./ . NAS Ruk Manaftmtni Study, p. 161; OSTP Risk Assessment Quiddmm, Principle 2S, p. 10378; and EPA Osttinogen Riak atrumnl Guidelines, p. 34001. "Sec. eg. Ames f ai, fop. ritj, p 276 (continued reliance on linear models despite the accumulation of evidence against linearity); and Lester B Lave. "Health and Safety Riak Analysis: InforRuucn for Battar Decisions* Science, VoL 236, April 17,1S67, pp. 291-295, P (simnet often resist modeling improvements and . a that yield lover riak estimates). EPA Caranafen Risk Assessment Guidelines, pp. 33997-&99S. "In the absence of adequate information to the contrary, the linearised multistage procedure will be employed. ... Considerable uncertainty will remain concerning rasponaae at low deeaa; therefore, in meet uses, an upper-limit risk estimate using the linearixad multistage procedure ehould also be praoantad." Dusn Ruk Assessment Appendix A p 13. Unbiased riak estimates vary by a mailer footer. SL 064751 22 REGULATORY PROGRAM OF THE UNITED STATES GOVERNMENT three risk estimates mentioned earlier clustered at the high end of the range.*2 Risk assessments based on different models have led other governments to establish unit risk factors that are a thousand times lees stringent than the most commonly used at these three; one study suggests that this particular estimate overstates the most likely risk estimate hy a factor of almost 5,000" Conversion from Animals to Humans Once risk has been extrapolated to low doses in rodents, scientists must convert them to human dose-equivalents. The two most common approaches involve the use of body-weight or surface-area conver sions, and there are scientific reasons for choosing either approach in individual eases. The surface-area approach leads to estimates of risk that are between 7 and 12 tunes greater than those based an the body-weight method, depending upon the test species. Despite the ambiguity of the underlying science, the more conservative surface-area method is often ap plied reflexively.*4 multiplicative effect of overstating risk by a factor cf two at five different points in an exposure assessment will overstate actual risk by a factor of thirty-two. Worvt-Caae Environmental Conditions Whan data are available they often relate to unusually aencitive environments or highly contami nated conditions. When estimating regional or nation wide exposures, agencies often use data from these local "hot spota" in developing more general national estimates of health risks. However, such data are never representative and estimates extrapolated from them are generally unreliable and misleading. In addition, chemicals often degrade naturally after they have been released to the environment. In eome cases, degradation occurs veiy quickly, whereas in others the process may taka many years or even decades. A common practice in exposure assessment modeling is to assume that exposures remain constant over time--that is, chemicals are assumed never to degrade, or degradation by-products are assumed to pose identical risks. ISSUES ARISING FROM HUMAN EXPOSURE ESTIMATES In addition to developing estimates of the dose-re sponse function, agencies must estimate the likely level of human exposure. This section examines some of the issues and problems that arise in conducting an exposure assessment It is a generally accepted principle of axposure nfcniiient that estimates should be based on the most likely scenario, with appropriate consideration of uncertainty.** Nevertheless, agencies often use conser vative assumptions for exposure when real-world date are unavailable. When eech of these assumptions tends to overstate likely human riaki, the multiplica tive effect of even t small overstatement at aach stage in an exposure assessment will yield a substantial overestimate of actual exposure. For example, the Hie Individual In addition to estimating the amount of a substance that may actually be present in the environment, a risk analysis must also consider the conditions under which humans may be exposed. Actual risks vary considerably depending on location, mobility, and a host of other factors. Nevertheless, estimates often are baaed on the upper-bound lifetime cancer risk to the maximum-exposed individual (MEI), the hypothetical person whose exposure is greater than all others. Sometimes, risks to the entire population are esti mated by assuming that everyone is exposed at the MEI level. Because environmental regulations are often justified using MEI-based risk assessments, actual risks may be substantially lower than what dedsionmaken and the general public perceive them to be. Diadn Riek Aimeement, pp 46-49. 1(U risk-specific does* (RaDa) derived fam the Hnearixed multistage spaa the range fan ? 1.2 picogram/kg/dey. The RaDe at EPA, CDC, sod FDA an 0.006, 0.03, and 0-06 pg/kg/day, respectively. ** Dioxin Ruk Allotment, p. 4. mEPA Caranagtn Anetement Guideline!, p. 33998. "EPA will amtiaue to uae this [surface area] !; factor unlace data on a specific agent suggest that a different scaling fates ia justified* * EPA guidance documents have historically called for unbiased estimate* of exposure. See, e.g., U.S. Environmental Protection Agency. 'Guideline* for Expoeure Aaaeacment,* 60 Fit 34042-34064 (September 24,1986, hereinafter, EPA Expoeure Allotment GuideUnit), U.S Environmental Protection Agency, Superfund Public Health Evaluation Manual, OSWER Directive 9285 4-1, October 1986, and U.S. Environment*] Protection Agency, Superfund Expoeurt Allotment Manual (Revised Draft), OSWER Directive 9285.5-1, December 1966 EPA recently abandoned the calculation of unbiased exposure estimates for Superfind sites on the ground that it we* inauffirientJy conservative. EPA'* new protocol require* the estimation of "reasonable r1"'"1""1 exposure" instead at the average and upper-bound estimates Reasonable maximum exposure constitutes a saw term of art that EPA intends to be "well above the average case* but not as extreme as the upper-bound. It provides a new opportunity for embedding conservative assumptions into expoeure assessment end exaggerating estimate* of actual human-health risk at Superfund sites Sec Riek Allotment Guidance for Suptrfund, Volume ! Human Health Evaluation Manual (Part A), Interim Final, EPA/540/1-88/002, December 1989, Chapter 6, pp. 6, 47-60. Si- 06^752 OVERVIEW 23 la developing the ME] risk level, analyte** invan* ably assume that the level of exposure is continuous ever a 70-year lifetime. This assumption overstates ictus! risks, because people are mobile, encounter a constantly ch*"rg portfolio of daily risks to life and health, and can take actions that reduce risk. Assumptions vs. Real-World Exposure Data Tbt thread that connects these exposure assessment mues is that simple constructs which overstate exposure are typically used in lieu of real-world data, often because such data ere unavailable. Hie risk estimates generated by these models depend on the validity of their assumptions; even small biases in exposure assessment assumptions can result in a substantial overstatement of risk. For example, regulatory egendes may not have statistically reliable real-world data on pestidde residues in agricultural products. They also may not know the proportion of a given crop that has been treated with a particular pestidde. A common resolu tion of these uncertainties is to assume that residues are equal to the regulatory `tolerance"--the maximum level allowed to be present in food sold in interstate commerce--and that 100 percent of the relevant crop has been treated. Both assumptions overstate actual exposure, but are encouraged by agency guidance as a way to instill conservatism in rick assessment." When data are available, however, the extent of this conservative bias becomes evident In a recent special review for the pestidde Captan, for examp!; EPA reduced its earlier upper-bound lifetime cancer risk estimate by two orders of magnitude when it replaced the original conservative assumptions with real-world data. Even with these improvements, EPA still reported that upper-bound risks were probably over stated. For example, field tests were performed baaed on applications at the maximum legal rate and aa close to harvest as the label permits. Similarly, feeding studies assumed that animal diets were dominated by feedstuff* that happened to contain high residues relative to other feedstuff's, such as almond hulls and raisin waste. As EPA noted, even if these assumptions accurately represented typical animal diets, they would do so only for portions of California where these crops are grown; nationwide extrapolation! baaed on these "hot-spots* would very likely overstate expo sure.*7 Since two of the highest product-specific risks were attributed to milk and meat, these remaining conservative biases can be expected to be significant IMPLICATIONS OF CONSERVATIVE RISK ASSESSMENT FOR RISK MANAGEMENT AND REGULATORY DECISIONMAKING Hie primary purpose of risk assessment is to provide data aa a baric for risk management decisions. Providing uaefiil data requires the synthesis of information cl coining risks and exposure levels into a coherent pa luge that can be used to develop regulatory options. Decisionmakers then can use these risk estimates in evaluating regulatory alternatives. Unfortunately, the way in which risk information is characterised tends to overstate risks, making them appear much greater than they are likely to be. As a result, decisionmakers may make regulatory choices that are very different from the ones they would make if they were fully informed. Quantification of Uncertainty In accordance with the recommendations of the National Academy of Sciences, the OSTP Guidelines explicitly call for the quantification of uncertainty, particularly aa it arises in the selection of dose-response models and exposure assumptions." Unfortu nately, Federal regulatory proposals that utilise risk assessment rarely provide this information, nor do they analyse the implications of uncertainty for decisionmaking. Instead, many risk assessments only identify a lifetime upper-bound level of risk." Hie differences between upper-bound and expectedvalue estimates may be considerable. As we indicated earlier, the upper-bound risk estimate for dioxin may be 5,000 times greater than the most likely estimate. Plausible risk estimates for pcrchloroethylene (the primary solvent used in dry cleaning) vary by a factor cf about 85,000." In tome instances, decisionmakers may not be informed that risk estimates differ because of policy choices hidden in the risk-assessment methodology. In EPA's proposed rule limiting emissions from coke m EPA Expaturt Assessment Guidelines, p $4053. "When then is unrortainty to the acmtific rids, it is Agency policy to wt on tb* d* of public safety * r See, 4 , U.S Environments! Protection A*woey, ^Captan: Intent to Cancel Registrations; Conduced of Spedsl Review," 54 FR 1127-4126 (February 24, 1989J. * OSTP Guidf/insf, (Guideline 27), p. 10378. "See. t.j . EPA Cerdrugtn Risk AiMumtni Guidtlinu, p. 33908. Nichols and Zeckhauaer, (op. dU. pp- 64-65. SL 064753 24 REGULATORY PROGRAM OT THE UNITED STATES GOVERNMENT event, for example, cancer hika were estimated bated on the LMS model--a model that it designed to yield upper-bound estimates of risk. In previous rules involving similar types of risks, however, EPA used the unbiased maximum likelihood estimate, lb the extent that decisionmakers were not informed that the higher estimate of risk was largely due to a different low-dose extrapolation procedure, regulatory decisions based on this risk assessment were likely to reflect misunderstanding rather than science.71 Plausible estimates of likely cancer risk can often be found buried in regulatory background documents. However, Federal Register rulemaking notices seldom present such estimates alongside upper-bound esti mates. This practice overstates baseline human health threats, as well as the amount of risk reduction that may be accomplished by regulation. Policymakers and the public are misled because they typically see only the upper-bound estimates of the threat The prevalent Federal agency practice is to calcu late the benefits of Federal regulatory initiatives based solely on upper-bound estimates of risk and exposure. In a recent proposal to reduce occupational exposure to cadmium, for example, the Occupational Safety and Health Administration (OSHA) developed risk estimates based on five alternative models for data, and two alternative models for human data. Across these seven data/model combinations, estimated excess lifetime cancer riak at the least stringent of the two proposed exposure standards varied from 0 to 153 cases per 10,000 workers occupationally exposed for 45 years. OSHA based its proposed exposure standards on one of these data/model combinations--the multistage model ap plied to animal data. This data/model combination predicted an excess lifetime cancer risk of 106 per 10,000 exposed workers, and was used to estimate aggregate cancer incidence and the riak-reduction benefits attributable to the new standard. Uncertain ties in the underlying risk assessment, which span several orders of magnitude, were not carried forward through the exposure assessment and benefit calcula tion stages. This analytic error effectively obscured the uncertainty surrounding the true incidence of cadmium-induced lung cancer, and resulted in benefit estimates that may exceed actual reductions is occupational by several orders of magnitude.11 disordered Priorities, Perverse Outcomes Logically, one would expect that the routine over statement of likely risks would lead to inefficient regulatory choices. Decisionmakers, convinced that a certain substance or activity poses a significant threat to public health, might well take actions that they would otherwise resist Alternatively, they might take actions that address the wrong real-life risks. lb the extent that riek assessments differ in the degree to which they adopt conservative assumptions, it is difficult to determine which activities pose the greatest risks and hard to establish reasonable priorities far regulatory action. Bectuse conservatism in risk assessment is especially severe with respect to carcinogens, it is reasonable to expect that other health and safety risks tend to receive relatively lets attention and weight. Aa a result, eodety may actually incur greater total risk, because of misordered priori ties caused by conservative biases in cancer risk assessmentn A perverse and unfortunate outcome of using upper-bound estimates based on compounded conser vative assumptions is that the practice may actually increase risk, even in situations where cancer is the only concern. Regulatory actions taken to address what are in fiict insignificant threats may implicitly tolerate or ignore better known, documented risks that are for more serious. For example, before it was banned, ethylene dibromide (EDB) was used as a grain and soil fumigant to combat vermin and molds. Vermin transmit disease, and molds harbor the natural and potent carcinogen aflatoxin B. The estimated human cancer risk from the aflatoxin contained in one peanut butter sandwich is about 75 times greater than a full day's diatary risk from EDB exposure. On this basis alone, it might have been appropriate to accept a small increase in cancer risk from EDB to reduce the much larger cancer risk from aflatoxin. By eliminating the relatively small hazard from EDB, Federal riak managers may have intensi- ''Leaer frtjrn Wendy Gnma (Administrator of tlx Office of Information and Regulatory Affair*) to Lm Thames (Administrator of the Environment*] Protection Agency), August 12, IMS, p. S. ^Occupational Safety and Health Administration, "Occupational Exposure to Cadmium; Proposed Rule,* 55 Federal Register 4076,4060, 4093, n This is precisely the policy issue raised by Nichols and Zackhauaar, (op. eft.), pp. 60-71, who note that EPA't 1985 decision to hnut bad in gasoline was threatened by concerns about potential increases in bensent exposure. Any tradeoff between lead and benscnc neks would have been biased against lead, as estimatae of benxene risks art mors conservative simply btcimte it is e eardnogen, whereas lead is not. SL 064754 OVERVIEW SJtSSi V I * ficd the relatively potent threat of aflatoxin laaodated arise in a technologically advanced society. Unfortu mith an increase in the prevalence of mold contamina nately, ft is also susceptible to hidden hjtaei that may tion.'4 Ihe emphasis on risks faced by the marimmn- exposed individual may also eause a perverse result by undermine its edentific integrity and the basis for policymakers' reliance on such information risk management derisions. For policymakers and the increasing overall population risks. For example, EPA's proposed regulation of the disposal of aewage sludge would probably create more public health risk than it eliminates. The proposal outlines a regulatory public to continue to rely on risk assessment in the development ofregulatory initiatives, a renewed effort must be made to separate eriance from policy and provide risk information that ia both meaningful and scheme that would shift disposal from generally safe leliable. practices to relatively risky alternatives. Thus, setting sludge quality standards to achieve an MEI upper- Expected Value Estimate* bound lifetime cancer risk of one in 100,000 (10-1) would prevent 0.2 statistical cancer cases resulting from monofilling and land application. However, it would cause 2.0 additional statistical cancers by forcing a shift away from these disposal approaches toward incineration.7* Perhaps the most important current need in regula tory decisionmaking is for carefully prepared and scientifically credible estimates of the likely risks involved. Relying on worst-case analysis based an extremely conservative riak assessment and exposure models leads to widespread misunderstanding on the These problems can be addressed by providing part of both Government officials and individual decisionmakers with the frill range of information on citizens. Decisionmakers at all levels need unbiased \ the rides of a substance or an activity. Thus, decisionmakers should be given the likely risks as well and impartial riak information ao they can focus their attention on significant problems and avoid being as estimates of uncertainty and the outer ranges ofthe distracted fay minutiae.*1* potential risk. Then, if regulatory decisionmakers want to choose a very cautious risk management Weight-of-Evidenoe Determinations r* strategy, they can do so and a margin of safety can be ly ie applied explicitly in the final decision. This approach Similar procedures an needed for assigning weights is superior to one in which the expected risk and an to each relevant study in the risk-assessment litera a unknown margin of safety are hidden behind the veil ture. Current practice gives undue weight to studies y of a succession of upper-bound estimates adopted at that show positive relationships. Resulting risk classi key points in the risk-assessment process. fications an thus conservatively biased estimates S : The public and effected parties also benefit from derived from samples of similarly biased baervations. I knowing both the expected risk and the margin of safety rather than being given upper-bound estimates Full Disclosure that are probably very different from actual risks. Efficient and responsible decisionmaking requins People are likely to have a better intuitive under that policymakers and the public be frilly informed 1 standing of the significance of averages than they about the implications of the regulatory alternatives > have of unlikely extremes. Tb the extent that a margin among which they must choose. Meeting this require t ef safety is appropriate--perhaps to protect unusually ment demands a canftil discrimination between I Tt% sensitive subpopulations--the magnitude of this mar gin can be more readily communicated if made science and policy. When riak estimates depend on assumptions and judgments of data, the f * explicit. In addition, providing information in this way should help improve public confidence in quantitative meaning and implications of these nonscientific pa rameters must be clearly articulated. risk assessment as the basis for decisionmaking, AVOIDING CONSERVATIVE BIASES IN RISE ASSESSMENT Risk assessment remains a powerful and useful scientific tool for estimating many of the risks that Avoiding Perverse Outcomes C* -fill attention needs to be paid to the likely nsu of regulatory alternatives, with an eye toward avoiojig choices that have the perverse effect of increasing net risk. All human activity involves risk. Am tt al., (op. at), p. 273. mU.S. Environment*] Protection Agency, ^Standard* for the Ditpoaa] of Sewage Sludge; Propoaad Rule," $4 FR 5740-6902 (F< 1 1989i. * A Nicholt and ZtckhioMit op. pp. 73-76, SL 064755 1EGULAT0RT PROGRAM OF THE UNITED STATES GOVERNMENT Decisionmakers need to be sure that ipedfic aetiona taken in the name of risk-reduction in one area do not make matters worse elsewhere. Quantitative risk assessment can help in this regard so long as the methods applied are not inherently biased in a way that undermines comparisons across alternatives, each of which entails some degree of risk. Our discussion has covered only the highlights of risk-assessment methods, yet we have identified several independent places at which conservative assumptions are commonly used. Individually, each of these assumptions might appear to be prudent responses to scientific uncertainty. In combination, however, they result in a distortion equal to the product of the individual conservative biases, lb illustrate, suppose that there are ten independent cteps in a risk assessment and prudence dictates assumptions that in each instance result in risk estimates two times the expected value. Such a process would yield a summary risk estimate that is more than 1,000 times higher than the most Hkely risk estimate. Because there are usually many more than ten steps, and many of them will incorporate conservative biases that exceed an order of magnitude, risk estimates based an such practices will often exceed the most likely value by e factor of one million or more. When risk assessments contain hidden value judg ments, their scientific audibility is inevitably compro mised. lb the extent that policymakers and the public fail to understand the magnitude of the margin of aafety embedded in quantitative risk assessments, policy choices are distorted from the course that would have been selected if decisionmakers had been better informed of the actual risks. Ironically, these policy decisions may actually increase total societal risk. Tbo much attention is focused on relatively small hazards that have been exaggerated by conservative risk assessments, leaving alone larger risk* that have been estimated using unbiased procedures. ; ' . Information as an Alternative Regulatory Strategy Federal regulation was initiated to deal with economic problems caused by monopoly and so-called 'excess competition." Subsequent events have shown that, in general, economic regulation--fixing prices, establishing restrictive terms of trade, and erecting barriers to entry--is usually inefficient and detri mental to innovation. In response to these lessons, Federal regulation of this type has been under increasing criticism. As indicated above, however, much more needs to be done to reform economic regulation and restore competition. Federal regulation has more recently been initiated to deal with what economists call externalities, situations in which participants in voluntary market transactions do not bear the full costs or capture all of the benefits of theae exchanges. Common examples of externalities include environmental pollution and traffic congestion, common property resources such as fisheries and public forests, and "public goods" such as basic scientific research. In each of these instances, regulation may be an appropriate mechanism to modify or restore distorted market processes, or to stabUsh markets where heretofore they have not existed, to maximize net social benefits (including environmental, health, and aafety benefits). The key ingredient is the determination that existing markets are, in some significant manner, failing to perform efficiently. The traditional regulatory approach to externalities has been the promulgation of standards. Because this approach often remedies existing externalities by creating new ones, economic incentive instruments are becoming an increasingly popular alternative to stand ards. The priadpsl attraction of economic incentives is thst they rely on market forces rather than attempt to suppress them. This section explores another alternative regulatory strategy--the production, provision, or mandated disclosure of information. The first subsection briefly summarizes the economics ofinformation as it relates to regulatory decisionmaking. Three points stand out in this discussion. First, because information is costly to acquire and the capacity to process it is limited, there is an optimal level of information for every market transaction. Second, differences in the amount and quality of information between buyers and sellers are normal and do not necessarily indicate market failure. Rather, these differences generally reflect variations in the costs end benefits that ere attribut able to information. Third, competitive markets pro vide powerful incentives for buyers and sellers to reveal relevant information. Market processes, not government regulations, provide the dominant motiva tion for generating, acquiring, and disclosing informa tion. The role of government regulation thus should be to supplement these processes when they prove to be inadequate, not to supplant them when they work well. The aecond subsection identifies three rationales for government intervention in the production or man dated disclosure of information. Two of these are economic--the public-good character of some types of j i . SL 064756 APPENDIX V Regulatory Impact Analysis Guidance A Regulatory Impact Analysis (R1A) should demon* *nte that a proposed ragulatory action aatisfiss the requirements of Section 2 of Executive Order No. 12291. Tb do so, it should show that; There is adequate information concerning the need for and consequences of the proposed action; The potential benefits to society outweigh *hs potential costs; and Of all the alternative approaches to the gi a regulatory objective, the proposed action will naxiinize net benefits to eociety. The fundamental test of a satisfactory RIA is ebether it enables independent reviewers to make an formed judgment that the objectives of Executive Order No. 12291 are satisfied. An RlA that includes all the elements described below is likely to ftilfill this requirement Although venations consistent with the spirit end intent of the Executive Order mey be nmnted for some rules, most RIAs should include these elements. The guidance in this document is not in the form of mechanistic blueprint for a good RlA cannot be witten according to a formula. Competent profesaonal judgment is indispensable for the preparation rf a high-quality analysis. Different regulations may all for very different emphases in analysis. For one proposed regulation, the cruciel issue may be the question cf whether e market failure exieti, and auch of the analysis may need to be devoted to that key question. In another case, the existence of a airket failure may be obvious from the outset, but (tensive analysis might be necessary to estimate the aignitude of benefits to be expected from proposed ngulatory alternatives. The amount of analysis whether scientific, statistical, or economic) that a Particular issue requires depends on how mitial that w is to determine the best alternative and on the aplenty of the issue. Regulatory analysis inevitably involves uncertainto and requires informed professional judgments. Whenever an agency has questions about such issues the appropriate analytical techniques to use or the (hematives that should be considered, it should oonwlt with the Office of Management and Budget as wriy in the analyse stage as possible. ftis document is written primarily in terms of Proposed regulatory changes. However, it is equally applicable to the review of existing regulations. In the k&er case, the regulation under review should be compared to a baseline oaee of no regulation and to reasonable alternatives. Element* of a Ragulatory Impact A&alyiis Preliminary and final Ragulatory Impact Analyses of major rules should contain five elements. They are: (1) a statement of the potential need for the proposal, (2) an examination of alternative approaches, (S) an analysis of benefits and costs, (4) tire rationale for choosing the proposed regulatory action, and (5) a statement of statutory authority. These elements are explained in Sections I-V bslow. L STATEMENT OF POTENTIAL NEED FOR THE PROPOSAL In order to establish tire potential need for the proposal, the analysis ihould demonstrate that (a) market failure exists that is (b) not adequately re solved by measures other than Federal regulation. A. Market Failure The analysis should determine whether there exists a market failure that is likely to be significant Ono* such market failure has been identified, the analysis.-* should show how adequately the regulatory alternant tivet to be considered address the specified market failure. The three mejor types of market failure are externality, natural monopoly, and inadequate infor mation. 1. Externality. An externality occurs whan one party's actions impose uncompensated benefits or coats on another outside the marketplace. Environ mental problems are a classic ease of externality. Another example is the case of common property resources that may become congested or overused, such as fisheries or the broadcast spectrum. A third example is a "public good,* such as defense or scien tific research, whoa* distinguishing characteristic is that it is inefficient, or impossible, to exclude individ uals from its benefits. 2. Natural monopoly. Natan] monopoly exists where e market can be served at lowest cost only if production is limited to a single producer. Local telephone, gas, and electricity services are examples. S. Inadequate information. The optimum, or ideal, level of information' is not necessarily the maximum possible amount, because information, like other goods, should not be produced when the cotta of doing so exceed the benefits. The free market does not SL 064757 653 664 REGULATORY PROGRAM OF THE UNITED STATES GOVERNMENT necessarily supply an optimal level of information, tell buyers about cholesterol in butter and its dmcu. because information, once generated, can be die* eonaequenoes, sellers of margarine do have geminated at little or no marginal cost, and becauac incentive). When the negative characteristic inv*w it is commonly infeasible to exclude nonpayers from health or safety hazard, the threat of foture m reaping benefits from the provision of information by net liability lawsuits may give sellers adequate inal others. Where market failure due to inadequate in* tives to iwvaal information about the potential hat. formation is the rationale for government inter ard. News media, consumer group*, public heahh vention, a regulatory action to improve the availabil agencies, and similar service* may supply infa^ ity of information will ordinarily be the preferred tion not supplied by seller*. In summary, while it a alternative. possible to identify situations in which market failure The current state of knowledge about the econom ics of information is not highly developed. Therefore, regulatory intervention to address an information problem should only be undertaken where there is substantia] reason to believe that private incentives to provide information are seriously inadequate and that the specific regulatory intervention proposed will provide net benefits for society. due to inadequate information is mort likely to iB. rant regulatory intervention, each situation must he examined on a caso-by-cas* basis. There should he a presumption against the need (or certain types of regulatory actions, except in sped*] circumstances. A particularly demanding burden of proof is required to demonstrate the potential need for any of the following types of regulations: In many circumstances, the availability of informa Fries controls in competitive markets tion, while perhaps not optimal, is reasonably ade Controls on production or salts in competitive quate, so that attempts to regulate information are as markets likely to make things worse as to make them better. Information about a particular characteristic of a product, for example, would be reasonably adequate if buyers could determine the existence of the charac teristic by inspection of the product before purchase or (in the ease of a frequently purchased product) by use of the product Even if the characteristic could not be determined by buyers, government interven tion would not be warranted where sellers have incentives to reveal the existence of the characteristic to buyers. Sellers will have substantial incentives to supply information about any characteristic thst is important to buyers end valued positively by them, Mandatory uniform quality standards for goods or services, unless they have hidden safety or other defect* and the problem cannot be adequately dash with by voluntary standard* or information disclosing the hazard to potential buyers or users Controls on entry into employment or production, except (a) where indispensable to protect health and safety (*.g., 7AA testa for commercial pilots) or (b) to manage the use of common property resources (e.g., fisheries, airwaves, Federal lands, and offshore areas). B. Alternatives to Federal Regulation particularly if the level of the characteristic variss between the products of one seller and another. In these circumstances, sellers whose products rank highly in the valued characteristic can increase their sales by informing buyers of the superiority of their products. If the level of the characteristic does not vary between the products of one seller and another, individual sellers have less incentive to inform buyers about the characteristic. Even so, the incentives of individual sellers or of a trade aasodation to supply information may be substantial. Even where a market failure exists, there msy be no need for Federal regulatory intervention if other means of dealing with the market failure resolve the problem adequately or better than the proposed Fed eral regulation would. Among the alternative means that may be applicable are the judicial system (particularly liability cases to deal with health and safety), antitrust enforcement, and workers' compen sation systems. An important alternative that may often be rele vant is regulation at the State or local level In Sellers are least likely to supply adequate informa determining whether there exists a potential need for tion about a particular characteristic of their product a proposed Federal regulation, the analysis should where the characteristic is negatively valued by con examine whether regulation at the Federal level is sumers and the level of the characterstic does not more appropriate than regulation at the State or local vary between the products of one seller and those of level This analysis may support regulation at the another (e.g., cholesterol in eggs). Even in such cir Federal level where rights of national citizenship cumstances, substantial information about the char (such as legal equality among the races) or considera acteristic may be available to buyers. For example, tions of interstate commerce are involved. If inter seller* of rival products may supply the information state commerce is involved the analysis should at (e.g., while sellers of butter may have no incentive to tempt to determine whether the burdens on SL 064758 APPENDIX V 665 torstste commerce vising from different State and to) regulations arc so great that they outweigh the rfvaatagM of diversity and local political chaise. In toe cases, the nature of the market failure may totf suggest the most appropriate governmental to) of regulatioiivFor example, pollution that spills to* stale lines (such as add rain whose precursors e transported widely in the atmosphere) is probably tot controlled by Federal regulation, while localised pollution (such as garbage truck noise) is probably to* efficiently handled fay local government ragula* b general, because demands among localities for efferent governmental aervices differ and because topetition among governmental unite for taxpayers may encourage efficient regulation, the uDest unit of government capable of correcting the aarket failure should be chosen. This must, however, b balanced against the possibility of higher coats besuse national firms would be required to comply with more than one act of regulations and because administering similar regulations in more than one premmenta) unit involves some costs of duplication. Dms, some analysis may be necessary to determine which level of government can most efficiently regubt* a specific market failure. If the analysis does suggest e potential need far a federal action, it should also consider alternatives of SDoregulatoiy Federal measures. For example, as an iheraative to requiring an action or the use of a pniculv product, it may be more efficient to subsi dise it Similarly, a fee or charge may be a preferable alternative to banning or restricting a product or ebon. An example would be an effluent discharge be, which has been recommended as an effident way to limit pollution, because it cauies pollution sources with different marginal costs of abatement to control (fluents in an effident manner. In addition, legisla tes measures that make use of economic incentives, auk u changes in insurance provisions or changes in pvptrty rights, should bt considered. H AN EXAMINATION OF ALTERNATIVE APPROACHES The R1A should show that the agency has consid er the most important alternative approaches to tie problem and must provide the agent's reasoning hr selecting the proposed regulatory change over nth alternatives. Ordinarily, it will be possible to eliminate some alternatives by a preliminary analy st, leaving manageable number of alternatives to h evaluated by quantitative benefit-cost analysis wording to the prindples to be described in Section D* The number end choice of alternatives to be elected far detailed benefit-cost analysis is unavoid ably a matter of judgment There must be aome balance between thoroughness of analysis and prac tical limits to the agency's capacity to cany out analysis. Alternative regulatory actions that should be ex plored include the following: l lforr performance-oriented standard! far health, safety, and environmental regulations. Performance standards are generally to be preferred to enginoer* ing or design standards because they allow the regu lated parties to achieve the regulatory objective in the most cost-effective way. In general, a performance standard should be preferred wherever that perfor mance can be measured or reasonably imputed. Per formance standards should also be applied ai broadly as possible without creating too much variation in regulatory benefits; for example, by tatting standards on a plant-wide or firm-wide basis rather than source fay eouree. It is misleading and inappro priate, howevsr, to characterise a standard as a performance standard if it is set so that than ia only one feasible way to moot it; as a practical matter, such a standard ia a design standard. 2. Different requirements fer different segments of the regulated papulation. For example, there might be different requirements for large and small firms. If such a differentiation is made, it should bt based on perceptible differences in the costs of compliance or in the benefits to bt expected from compliance. For example, aome worker safety measures may exhibit economies of scale, that is, lower costs per worker protected in large firms than in small firms. A heav ier burden should not bo placed on one segment of the regulated population on the ground* that it is better able to afford the higher coat; this is a sure formula for loading disproportionate costs on the most productive sectors of the economy. 8. Alternative leads of stringency. In general, both the benefits and costs associated with a regulation will increase with the level of stringency (although costs will eventually increase more rapidly than bene fits). It ia important to eonsidtr alternative levels of stringency to hotter understand the relationship between stringency and benefits and costa. This approach still increase the information available to the decisionmaker on the option that maximizes net benefits. 4. Alternative effective date of compliance. The timing of e regulation may also have an important effect on its net benefits. For example, costs of s regulation may vary substantially over different com pliance dates for an industry that requires a year or more to plan its production runs efficiently. In this^^ instance, a regulation whose requirements provide^V SL 064759 656 REGULATORY PROGRAM OF THE UNITED 8TATES GOVERNMENT auffident lead time U likely to tchievt its foals at a much lower overall cost than a regulation that is affective immediately. 6. Alternative methods of ensuring compliance. Compliance alternatives indude the appropriate an* tity Goeal, State, or Federal) enforcing compliance, whether compliance ia enforced by on-site inspection or periodic reporting, and structuring compliance penalties so that they provide the most appropriate incentives. 6. Informational measures. Measures to improve the availability of information indude government establishment of a standardized testing and rating system (the use of which could be made mandatory or left voluntary), mandatory disclosure requirements (eg., by advertising, labeling, or endosures), and government provision of information (eg., by govern* ment publications, telephone hot-lines, or public in terest broadcast announcements). If intervention is necessary to address a market failure arising from inadequate information, informational remedies will generally be the preferred approaches. As an alterna tive to a mandatory standard, a regulatory measure to improve the availability of information has the advantage of being a more market-oriented approach. Thus, providing consumers information about con cealed characteristics of consumer products gives con sumers a greater choice than banning these products (for example, consumers are likely to benefit more from information on energy efficiency than from a prohibition on sale of appliances or automobiles fall ing below a specified standard of energy efficiency). Except for prohibiting indisputably false state ments (whose banning can be presumed beneficial), specific informational measures must be evaluated in terms of their benefits and costs. Paradoxically, the current state of knowledge does not generally permit the benefits and costs of informational remedies to be measured very accurately. Nonetheless, it is essential to consider carefully the costs and benefits of alterna tive informational measures, even if they cannot be very' precisely. Some effects of informa tional measures can easily be overlooked. For exam ple, the costs of a mandatory disclosure requirement for a consumer product include not only the obvious cost of gathering and communicating the required information, but also the loss of any net benefits of information displaced by the mandated information, the cost of any inaccurate consumer interpretation of the mandated information, and any inefficiencies arising from the incentive that mandatory discloeure of a particular characteristic gives to producers to overinvest in improving that specific characteristic of their products. Where information on the benefits and costs of alternative informational measures is insufficient to provide a <!*** choice between them, as will often be the case, the least intrusive alternative, sufficient to accomplish the regulatory objective, should be chosen. For example, it will often be sufficient for government to establish a standardised tasting and rating system without mandating its use, because firms that score well according to the system will havt ampls incen tive to publicise the feet. 7. Mere market-oriented approaches. In general, alternatives that provide for more market-oriented approaches, with the use of economic incentives re placing command-end-control requirements, should be explored. Market-oriented alternatives that may ba considered include feta, subsidies, penalties, marketable rights or offsets, changes in liabilities or property rights, and required bonds, insurance or warranties (in many instances, implementing these alternatives will require legislation). m. ANALYSIS OF BENEFITS AND C08TS A. General Principles The preliminary analysis called for by Sections I and II should have narrowed the number of alterna tives to be considered by quantitative benefit-cost analysis to a workable number. Ordinarily, one of the alternatives will ba to promulgate no regulation at all, and this alternative will commonly carve as the base from which increments in benefits and costs tre calculated for the other alternatives. Even if alterna tives such as no regulation are not permissible statu torily, it is often desirable to evaluate the benefits and costs of such alternatives to determine if statu tory change would be desirable. Departments and agendas bear a similar burden when they perform environmental impact statements in which alterna tives that lie outside their statutory authority must ba considered. In tome cases, the desirability of specific alterna tives outside the scope of the agency's regulatory authority may be determined by use of basic eco nomic concepts in light of the prindples enumerated in Section L In other instances, however, only a quantitative benefit-cost analysis resolve the question, and such alternatives will need to be in cluded in the analysis of this section. In addition, alternative forms of agency regulation will need to be evaluated by quantitative benefit-cost analysis. 1. Evaluation ofAlternatives. Except where prohib ited by law, tha primary criterion for choice among alternative! is axpected net benefit (benefits minus costs). Other ariteria may sometimes produce equiva lent results, but they must be used with care to avoid SL 064760 APPENDIX v 667 Ibe potentially aerious pitfalls to be explained in Part I of thi> aection and in Section IV. Both benefit# end ootti abould be expreaaed in discounted constant fcllan. Appropriate discounting proceduree are dia* aaoad in the following aection. lbe distinction between benefits and costs in bcneflt-cost analysis is somewhat arbitrary, since a pon gee benefit may be conaidered a negative cost, and eiee versa, without affecting the net benefit (benefits inni costa) decision criterion. This implies that the csidcrations applicable to benefit eetimatas also apply to costs and vice versa. Hie different issues arc aoosidcred separately under benefita or ooata in SecCons B and C below according to when they moot dts> arise. If the proposed regulation is composed of a number rf distinct provisions, it is important to evaluate the benefits and costa of the different provisions sepa rately. The interaction effects between separate previ ous (such that the existence of one provision affects da benefits or coats arising from another provision) ay complicate the analysis but dots not eliminate the need to examine provisions separately. In such a sue, the desirability of a specific provision may be appraised by determining the net benefits of the proposed regulation with and without the provision in question. Where the number of provisions is large and interaction effects are pervasive, it is obviously Unpractical to analyze all possible combinations of provisions in this way. Some judgment must be used select the most significant or suspect provisions for such analysis. 1 Discounting. The monetary values of benefits sou cvns occurring in different years should be discounted to their present values so that they are comparable. This is not the same as correcting for i&fiSliuu. An inflation adjustment is made with a price index, whereas discounting to present value is done with a discount rate. Benefita and coats ex pressed in constant (La., unaffected by inflation) dol lars must further be discounted to present values before benefits and costs in different yean can be added together to determine overall net benefits. As an equivalent alternative to discounting nonontihr: benefits, the RIA may use the discount me to annualize (amortize) costs over a period that corresponds to the occurrence of the benefita. Regardfas of the discounting procedure selected, the RIA *ust contain a schedule indicating when the benefits and costs occur. Discounting takes account of the fact that resources (goods or services) in a given year are worth more than identical resources in t later year. The underlyg reason for this is that resources can be invested so as to return more resources later. Partly because of this productivity of investment, individuals value consumption in earlier yean higher than consump tion in later yean. Modem analysis of discounting for public programs stresses the distinction between two rates of return: The befbn-tax rete, also known as the opportunity coat of capital This is the teal rate of return to marginal private investments. Estimates of the opportunity cost of capital in the U.S. economy vary substantially. The 10 percent discount rate specified by OMB Circular A-94 for uae in evalu ating government programs is intended to repreoont the opportunity ooet of capital The o/tsr-tax rata, also known as the consumption rate of interest This represents the rate at which consumers would be willing to exchange present for fUturs consumption, that ia, the rate at which eonaumera must be compensated for pw*p**<"g their consumption. As with the opportunity coet of capital alternative estimates of the consumption rate of interest vary significantly. A rate of 4 percent is reasonably representative of the range of alternative estimates and consistent with a 10 percent bafora-tax rate of return. The basic concept underlying the academic litera ture on public-sector discounting is that economic welfare is ultimately determined by consumption and only indirectly by investment Therefore, the value of investment must be measured by the value of the subsequent increase in consumption it permits. Any effect that a government program has on investment must be converted to an equivalent time-stream of consumption before being discounted. In practice, this results in a complex procedure that uses the beforetax and after-tax discount rates, a "shadow price of capital" and the impacts of benefits and costs on investment It ia recommended that agendas continue to uae the well-understood procedure of discounting by a single rate (as specified by OMB Circular A-94) and, when appropriate, perform additional analysis using the more complex shadow-price-of-capital meth odology. There are two circumstances when it is important to perform sensitivity analysis using the shadow price of capital approach: (a) Where the costs of the regulation are almost entirely current coats boras by consumers. In such circumstances, a low rate close to 4 percent is called for. (This assumes, as is normally the case, that the benefits art all in the form of disposable income or other benefita directly to individuals.) (b) Where some of the costs are capital costs financed out of saving and there is a long period between the time when moet costs are incurred end the time when most benefits accrue. In general, the SL 064761 668 REGULATORY PROGRAM OP THE UNTIED STATES GOVERNMENT smaller the fraction of costs that arc capital costs financed out of saving and the longer the time period between costs and benefits, the greater the likelihood that the shadow price of capital approach will be uiiiect. It is conceptually incorrect to adjust the discount rate as a device to account for the uncertainty of expected ftiture benefits and costs. This procedure will virtually never lead to a correct adjustment at benefits and costs. Therefore, risk and uncertainty should be dealt with according to the principles in Section 8 below and not by changing the discount rate. 8. Treatment ofRiek and Uncertainty. Where uneer* tainties exist about important parameters affecting the expected benefits or costs of an alternative under consideration, it is essential to cany out a eeneitivity anelyti* to determine the effect on net benefits of plausible variations in the value of the parameters. One form of sensitivity analysis involves calculation of the 'switch-point* value of the parameter under examination, that is, the value of the parameter at the break-even point at which the net-benefit decision criterion switches over from favoring one alternative to favoring another. When this break-even point of the parameter value is determined, the analysis may then consider the probability that the true parameter value is above or below the break-even value. For example, if the major uncertainty about a proposed regulation were its cost, the analysis could calculate how high the cost would need to be in order to reduce the net benefit of the proposal to tero. If it is judged to be highly unlikely that the actual cost would be that high or higher, it may be concluded that the choice of the proposed alternative is not sensitive to uncertainties about its cost A primary objective of sensitivity analysis is to identity where additional analysis may be moat needed. If the choice at a specific regulatory action is sensitive to alternative parameter values that are about equally likely to be true, more research to better determine the true parameter value could be very valuable. Wherever parameter estimates are uncertain, for either benefits or costs, expected-value estimates should be presented. Hypothetical best-case or worstcase estimates may be presented as alternatives for sensitivity analysis. Where possible, information about the probability distribution of the parameter estimate should be presented. A common situation that arises in estimating both benefits and costs is that a number of different studies may exist which together provide a range at different estimates for a particular parameter. In general, it is not appropriate to use the midpoint of the range at extreme values provided by the studies. Such a technique ignores the information provided fay all studies except those providing the extreme values, which may be the least reliable. The preferred ap proach to deriving an expected-value estimate of a particular parameter in this situation would be to derive it as a weighted avenge of the estimates of the individual studies, with the weight of each esti mate being baaed on the reliability (in the best judgment of the agency) of the study that produced it Where expected ftiture benefits or costs are un certain, their value to those who receive them may be different from their value if they were certain. (Often, but not always, a certain ftiture benefit is worth more to people than an uncertain ftiture benefit with the same expected value.) As noted in the previous section, it is incorrect to adjust the discount rate as a device to account for the riskiness of future benefits or costs. Any allowance for risk should be made by adjusting the monetary values (for the year in which they occur) of the uncertain benefits and costs so that they are expressed in terms of their "certaintyequivalents* For an uncertain benefit in ftiture year X the certainty-equivalent is the number of certain dollars in year X that the uncertain benefit is worth to its recipient. For example, suppose that a particular regulation reduces the probability of fire in a particu lar type of facility. As part of a benefit-cost analysis for this regulation, the dollar value of the expected reduction in fire loss would be calculated. Hie owners of the protected facilities place a higher dollar value on the risk of a fire than the expected dollar value of the loss. This is demonstrated by their willingness-to- pay for fire insurance. Therefore, their relative net cost (the percentage difference between insurance premiums and insurance company payments) for fire insurance can be used to increase the ex pected dollar value of the reduction in fire loss to its certainty-equivalent value. In the example of the preceding paragraph, the adjustment for risk would involve an increase in the value of the benefit, whereas uncertainty of a benefit is normally thought to reduce its certainty-equivalent value. Hie reason is that even though this benefit fay itself is uncertain, it acts to reduce the overall level of risk that would prevail in the absence of the regulation. This illustrates the important principle that what matters is not the variability or riskiness of a regulation's net benefits by themselves but the regulation's effect on risk and uncertainty overall. While an adjustment to account for risk may be called for in the fire-risk example given, a similar adjustment for the value of reductions in fatalities and injuries would not be appropriate. Assuming that SL 064762 AFFSNDXXV 669 g* values of fatalities and usuries have been derived ly the willingness-to-pay methodology recommended to Section B.2 below, they would already represent the certaintyequivalent value of the uncertain risk. Ihia is because the estimated dollar values represent the certain dollar amounts that individuals would aerifies to reduce these risks. probably, in most cases, it will not be advisable to adjust for risk and uncertainty. As a theoretical matter, no adjustment for risk is necessary wherever the net benefit* are widely dispersed among many individuals and are not correlated with disposable income. And in cases where this does not apply, risk may be relatively unimportant or may already be taken into account by use of the willingnees-to-pay methodology. In other eases, there may be no practi cal way to quantify the value of changes in risk. LAuumptioru. Where benefit or cost estimates are heavily dependent on certain assumptions, it is es sential to make these assumptions explicit and, vbere alternative assumptions are plausible, to cany cut sensitivity analyses based on plausible alterna tive assumptions. If the decision criterion proves to he eensitive to alternative plausible assumptions, this ay necessitate further research to develop more evidence on which of the alternative assumptions la the most appropriate. Because the adoption of a particular estimation methodology sometimes implies major hidden assumptions, it is important to analyze estimation methodologies carefully to make hidden assumptions explicit. 5. Internationa! Trade Effect*. In calculating the benefits and costs of a proposed regulatory action, generally no explicit distinction needs to be made between domestic and foreign resources. I& for example, compliance with a proposed regulation re quires the purchase of specific equipment, the oppor tunity cost of that equipment is ordinarily best repre sented by its domestic cost in dollars, regardless of whether the equipment is produced domestically or imported. The relative value of domestic and foreign resources is correctly represented by their respective foliar values, as long as the foreign exchange value of foe dollar is determined by a free exchange market. Nonetheless, an awireness of the role of international trade may be quite useful for assessing the benefits and costs of a proposed regulatory action. For exam ple, the existence of foreign competition usually makes the demand curve facing a domestic industry more elastic than it would be otherwise. Elasticities ef demand and supply frequently can significantly affect the magnitude of the benefits or costa of a regulation. A regulation that diacriminatea unjustifiably against foreign exporters is a form of economic pro tectionism. Tha economic loss to the UiS. due to the foci that protectionism ia economically inefficient will be reflected in the net benefit estimate of any prop erly conducted benefit-coat analysis. However, a bene fit-cost analysis will generally not be able to measure the potential U.S. loos from the threat of fotun retaliation by foreign governments. Therefore, special attention should be given to any possibility that a regulation would unjustifiably discriminate between domestic and foreign producer* and consumers--both discrimination against foreigners and discrimination in favor of foreigners. The fact that a regulation has a differential affect on foreigners as compared to Americans does not necessarily constitute discrimination. I for example, an automobile safety standard could be complied with leas expansively fay large can than by email cars, such a standard would be more favorable to American car producers, who produce relatively more large cars compared to the fleet mix of foreign producer*. None theless, such a differential affect would not b* dis criminatory if the difference in compliance ooat between large and small can was aaossaaiy to achieve legitimate regulatory ofajectivas in tbs most affidastway. If a regulation has an advene differential affect on foreign produccn or consumers relative to domestic producen end consumen that is not necessary to realise regulatory goal* efficiently, then e discrimina tory effect on foreign trade exists, lb* RIA should identity any substantial differential affect an interna tional trad* and explain why it ia necessary to achieve legitimate regulatory goals in the moat effi cient way. One means for reducing the likelihood of international discrimination would be for a U.S. prod uct standard for an internationally traded good to be baaed on an international standard, wherever an international standard exists and is compatible with the health, safoty, or environmental needs of the UB. International harmonization can b* beneficial for reg ulations directly setting standards for internationally traded goods or services. For example, H would be appropriate to consider international harmonization in setting safety standards for automobiles. Then is no similar advantage to international harmonization whan a regulation does not directly affect the quality of an internationally traded good or service, even if it indirectly affects its costs (*4., environmental con trols for automobile plants). 6. Distributional Effect*. Those who bear the costa of a regulation and those who eiyoy its benefits often an not the same penons. Benefits and costs of regulation may also be distributed unevenly over time, perhape spanning aevaral generations. Then is no generally accepted way to monetize potential SL 064763 660 REGULATORY PROGRAM OF THI UNITED STATES GOVERNMENT distributional effects. Attempt* to incorporate distributional concern* in benefit-cost analysis require tb* establishment of unequal weight* for different group* in society. Because positive economics treats equally the willingness-to-pay of all individuals, any alternative weighting would undermine the objective character of the analysis. Policymakera may wish, however, to take account of the distributional effects et various regulatory alternatives. Therefore, where there are potentially important differences between those who stand to gain and those who stand to loos under alternative regulatory options, the RIA should identify these groups and indicate the nature of the differential effects. The RIA should also present infer* nation on the streams of benefits and costs over time as well as present value estimates, particularly where intergeneratianal effects are concerned. B. Benefit Estimates The RIA should state the beneficial effects of the proposed regulatory change and its principal alterna tives. In each case, there should be an explanation of the mechanism by which the proposed action is ex pected to yield the anticipited benefits. An attempt should be made to quantify all potential real incre mental benefits to socisty in monetary terms to the TriMT-imnm extent poasible. A schedule of monetized benefits should be included that would show the type of benefit and when it would accrue; the numbers in this table should be expressed in constant, undiacounted dollars. Any expected incremental benefits that cannot be monetized should be explained. The RIA should identify and explain in detail the data or studies on which benefit estimates are based. Where benefit estimates are derived from a statistical study, the RIA must provide sufficient information so that an independent observer can determine the rep resentativeness of the sample, whether it was extras dated from properly in developing aggregate esti mates, and whether the results are statistically significant For regulations addressing health and safety risks, the calculation of potential benefits should derive from the agency's estimate of the mean expected value of the reduction in riek attributable to the standard. Estimates of the prevailing level of riak and of the reduction in risk to be anticipated from a/ proposed standard should be unbiased expected-value estimates rather than hypothetical worst-case este-/ mates* Extreme safety or health results should be weighted (along with intermediate results) by the probability of their occurrence to estimate the ex pected result implied by the available evidence. In addition, to the extent possible, the distribution of probabilities for various possible results should be presented separately, so as to allow for an explicit margin of safety, where required, in final decisions. If a margin of safety is to be provided, the proper pi--t for it is the final stage of the decision-making pr cess, not fay adjusting tha riak or benefit estimates in a conservative direction at the information-gathering or analytical stages of tha process. Conservative esti mates should bs presented as alternatives to best estimates for sensitivity analysis but should not sub stitute for them. It Is important to guard against double-counting of benefits. For example, if a regulation improved the quality of the environment in a community, the value of real estate in tha community might rise, reflecting the greater attractiveness of living in the improved environment It would ordinarily be incorrect to in clude the rise in property values among the benefits of the regulation. Ordinarily, the value of environ mental benefits (e.g., reduced health risks, efic improvements) will already he included among the benefits. The rise in property values reflects the capitalized value of these improvements. Therefore, to count as benefits both the value of the environ mental improvements and the corresponding increase in property values is to count the same benefits twice. Only where a direct estimate of the benefits has not been included would it be appropriate to include the increase in property values among the benefits. 1. General Considerations. The concept of "opportu nity cost" is ths appropriate construct for valuing both benfits and costa. The principle of "willingnestto-pay* captures the notion of opportunity cost by providing an aggregate measure of what individuals are willing to forgo so as to enjoy a particular benefit Market transactions provide the richest database for estimating benefits based on willingness-to-pay, so long as the goods and services affected by a potential regulation are traded in markets. Estimation prob lems arise in a variety of faetewa--, of course, where prices or market transactions are difficult to monitor. Markets may not even exist in some inatanrrs, for cing regulatory analyst* to develop appropriate prox ies that simulate market exchange. Indeed, the ana lytical process of deriving benefit estimates by simulating markets may suggest alternative regula tory strategies that create such markets. Willingness to pay always provides the preferred measure of benefits. Estimates of willingness-to-pay based on observable and replicable behavior deserve the greatest level of confidence. Considerably leas confidence should be conferred on benefit estimate* that are neither derived from market transactions nor baaed on behavior that is observable or replica ble. Of course, innovative benefit estimation method- SL 064764 APPENDIX V <61 ologiM may be necessary in aome cams, and should be encouraged. However, reliance upon such methods intensifies the need for quality control to ensure that estimates derived conform as closely as possible to what would be observed if markets existed. 2. Principle* for \hluing Directly Observable Bene fits, Ordinarily, goods and services are to be valued at their market prices. However, in aome instances, the market value of a good or service may not reflect its true value to society. If a regulatory alternative involves changes in such a good or service, its mane* tsiy value for purposes of benefit-cost analysis should be derived using an estimate of its true value to society (often called its "shadow price"). For example, Suppose a particular air pollutant damages eops. One of the benefits of controlling that pollutant will be the value of the mjp saved as a result of the controls. If the price of that crop is held above the fine-market equilibrium price by a government pricesupport program it will overstate the value of the benefit of controlling the pollutant if the crop saved were valued at the market price established by the support program- The social value of the benefit should be calculated using a shadow price for crops subject to price supports. The estimated shadow price should reflect the value to society of marginal uses of the crop (e.g., the world price if the marginal use is for exports). If the marginal use is to add to very large surplus stockpiles, the shadow price would be the value of the last units released from storage minus storage cost. Therefore, where stockpiles are large and growing, the shadow price is likely to be low and could well be negative. 3. Principles for Valuing Benefits that art Indirectly Traded in Markets. In some important Instances, a benefit corresponds to a good or service that is indirectly traded in the marketplace. Important ex amples include reductions in the health-and-safety risks, the use-value of environmental amenities and scenic vistas, and savings in time. 1b estimate the monetary value of such an indirectly traded good, the willingness-to-pay valuation methodology is still con ceptually superior, because the amount that people are willing to pay for a good or service is the best measure of its_value to them. As noted in Sections 4 and 5 immediately following, alternative methods may be used where there are practical obstacles to the accurate application of direct willingness-to-pay methodologies. A variety of methods have been developed for estimating indirect benefits. Generally, these methods apply statistical techniques to distill from observable market transactions the portion of willingness-to-pay that can be attributed to the benefit in question. Examples include estimates of the value of environ mental amenities derived from travel-cost studia^fc hedonic price models that measure differences changes in the value of land, and statistical studies of occupational-risk premiums in wage rates. Contingent-valuation methods have become in creasingly popular for estimating indirect benefits, but they suffer from the foot that survey instruments have a limited capacity to simulate real-world market behavior. Benefit estimates derived from contingentvaluation studies thus have a greater burden of analytical care to ensure that they represent in an unbiased manner what actually occurs in the market place. 4. Principles and Methods fir Valuing Benefits that are Not Traded Directly or Indirectly in Markets. Same types of goods, such as the eocial benefit of preserving environmental amenities apart from their use and direct enjoyment by people, are not traded directly or indirectly in markets. The practical obsta cles to accurate measurement are similar to (but generally more asvsre than) those arising with re spect to indirect benefits, principally because there are not market transactions to provide data for willingnesa-to-pay estimates. Contingent-valuation methods provide the only an alytical approaches currently available for estimating the benefits of such untraded goods. The absence^fo observable and replicable behavior with respect to IV benefit in question, combined with the difficulties of avoiding bias in contingent-valuation studies, argues for greet cere end circumspection in the use of such methods. This means, for example, that estimates of willingness-to-pay must incorporate the variety of alternative means individuals have of expressing value for untraded goods. Moreover, analyses must foithfoUy capture individuals' budget constraints, which restrict their willingness-to-pay for untraded aa well as traded goods and aarvioas. Benefit analyses derived from contingent valuation and sfrnflw meth ods thus require considerable analytic rigor in design and carafol execution. Absent such efforts, analyses based heavily on the benefits of untraded goods and services ordinarily would foil ths test of a satisfactory RIA. 6. Methods fir Valuing Health and Safety Benefits. Tor health and safety benefits, a distinction should be made between risks of nonfatal illness or injury end fatality risks. (a) Nonfatal illness and injury. Although the willingneae-to-pay approach is conceptually superior, the current state of empirical research in -he area is not sufficiently advanced to assure that estimates derived by this method are necessarily superior to direct-c^t valuations of reductions in risks of nonfatal illnei^B injury. Any injury-value estimate from a willingn^^ SL 064765 662 REGULATORY PROGRAM OP THE UNITED STATES GOVERNMENT to-pay study is necessarily an average over a spadfic appropriate to value reductions in risk from that high combination of injuries of varying severity. If the level more highly than equal risk reductions at lower average injury severity in such a study is greatly risk levels. In general, the greater the risk that an different from that for the regulatory action under individual bears, the higher will be the value the study, then the study's estimated injury value nay individual plaoaa on marginal changes in risk. On the not be appropriate for evaluating that action. Accord* ether hand, where a high risk is chosen voluntarily ingly, the agency should use whichever approach it thoee manning the risk tend to be persons who place considers most appropriate for the decision at hand. a relatively low value on averting safety risks. Empir The primary components of the direct-cost approach * ical studies of risk premiums in high-risk occupations are medical costs and the value of loet production. suggest that reductions in voluntarily assumed high Poesibly important costs that may be omitted by the risks should be valued lass than equal risk reductions nee of the direct-cost approach are the value of pain at ordinary risk levels. and suffering and the value of time loot from leisure end other activities that are not economically directly productive. Estimates of the value of fatality risks refer only to changes in an uncertain risk of death. They have no application to the certain prevention of the death of (b) Fatality. Reductions in fatality risks are best an identifiable individual. monetized according to the willingness-to-pay ap proach. The value of changes in fatality riisk is sometimes expressed in terms at the Value of life." This is something of a misnomer since the value of a life really refers to the sum of many small reductions in fatality risk. For example, if the annual risk of death is reduced by one in a million tor each at two million people, that represents two "statistical lives* savad per year (two million x one millionth two). If the annual risk of death is reduced by one in 10 million for each of 20 million popple, that also repre sents two statistical lives saved. The conclusion that the fatality risk reductions in these two cases are equivalent implies an assumption. The. implidt as sumption--that equal increments in risk are valued equally--allows different risk increments to be added together and compared directly. As a different exam ple, suppose there are two alternative reductions in the annual risk faced by an individual: 6. Alternative Methodological Frameworks far Esti mating Health and Safety Benefits. Several alterna tive ways of incorporating fatality risks into the framework of benefit-cost analysis may be appropri ate. These may involve either explicit or impUtit valuation of totality risks. One acceptable explicit valuation approach would be for the agency to select a single value for reduc tions in totality risk at ordinary risk levels (below l<r* annually) and use this value consistently for evaluating all its programs that affect ordinary fatal ity risks. Another acceptable explicit valuation ap proach would be to use a range of values for reduc tions in totality risk and apply sensitivity analysis as with other parameters that hare alternative plausible values. The range of alternative values should be a reasonable one, not one that includes the most ex treme upper and lower values of fatality risk reduc tion that have been estimated. Extreme values are A: from .10 x 10** to .09 x Iff4 * .01 x Iff4 B: from 1.00 x Iff4 to x Iff4 .01 x Iff4 more appropriate for instances of extraordinarily high risks (above Iff4 annually), with the extreme low valuee being appropriate whan voluntary assumption of high risk leads to self-selection and the extreme Since in both eases the reduction in "nni1 risk is high values being appropriate where the high risk is the tame (.01 x Iff4), the value of A and B should be involuntarily fumed, considered the same. Th# Mvnption that equal increments in fatality risk are of equal value is a legitimate one, so long aa the level of fatality risk is below Iff4 annually. There is evidence that the willingnesa-to-pay value for increments in fatality risk does not change signifi cantly over a wide range of risk exposure below Iff4 annually. Where the analysis usee a range of alternative values for reductions in fatality risk, it may be useful to calculate break-even values, as in other sensitivity analyses. This requires calculating the borderline value of reductions in fatality risk at which the net benefit decision criterion would switch over from favoring one alternative to favoring another (Le., the value of fatality risk at which the net benefits of the For levels of annual risk exposure of Iff4 and above two alternatives are equal). This method will fre it cannot be assumed that equal increments of risk quently be infeasible because of its computational are valued equally. At these higher risk levels, it is demands or because alternative* are continuous particularly important to distinguish between situa rather than discrete (e.g., alternative stringencies for tions of voluntary risk assumption and those of invol exposure levels), but where appropriate, it is a useful untary risk. Where the high risk is involuntary, it is supplement to the sensitivity analysis. SL 064766 A to to to im i i Si mil APPENDIX V 063 As implicit valuation approach could astail calcula tions of the cost per unit of reduction in fatality risk (cost per 'statistical life saved*), with costs defined as costs minus monetized benefits. This must be used with care since there is a serious potential pitfall: It to mot comet to choose between two mutually axdufeve alternatives by selecting the alternative with hwswt cost per ststisticel life saved. Ihe alternative eritfa higher cost per life eaved may nonetheleaa be the alternative with the higher net benefit to eoriety. The way to avoid this pitfall whils retaining the implidt vehiatdon approach is to make all ealcula* turns of cost per life saved in terms of increments between alternatives. Alternatives should be arrayed in order of their total reduction in expected fatalities end the incremental cost per life saved calculated between each adjacent pair of alternatives. In eon* trast to sxplidt valuation approaches, this avoids the necessity of specifying in advance a value for reduc tions in fatality risks. However, a range of values will be implied by the final selection of an alternative. This range should be consistent with estimated val ues of reductions in fatality risks calculated according to the willingness-to-pay methodology. Another way of expressing reductions in fatality risks is in terms of life-years saved. For example, if a regulation protected individuals whose average re maining life expectancy was 40 yean, then a risk reduction of one fatality would be axpraased aa 40 hie-years saved. Such a refinement may be desirable for regulations that disproportionately protect young people (e.g., motor vehicle safety regulations) or el derly people (e.g., regulations controlling cardno- drive the value of a life-year saved from an astimate of the value of life, first determine the average remaining life expectancy of the sample pop ulation in the study from which the estimate was drawn. Assuming that the average age of the sample population is known, the average remaining Ufa ax* pectaacy may be derived from actuarial tables giving life expectancy in relation to age. Using standard compound interest tables, the value of a life-year saved can then be determined aa the estimated value of life annualized over e period equal to the number ill ;eu> uf remaining average life expectancy. C. Cost Estimates 1. General Considerations. The opportunity cost of an alternative is the value of the benefits foregone as a consequence of that alternative. For example, the opportunity cost of banning a product (e.g., a drug, food additive, or hazardous chemical) is the foregone net benefit of that product It is measured by changes In producers' and consumers' surpluses. (Producers* surplus is the difference between the amount a producer is paid for a unit of a good and the mini mum amount the producer would accept to supply that unit It is measured fay the distance between the price end the supply curve for that unit Consumers* surplus is the difference between what a consumer pays for a unit of a good and the amount the would be willing to pay for that unit It is measured by tbs distance between the price and the curve for that unit) Aa another example, van if a resource required by regulation does not have to be paid for because it is already owned by the regulated firm, nonetheless, tha use of that resource to meet the regulatory requirement has an opportu nity cost equal to the net benefit it would have provided in the absence of the requirement Any such foregone benefits for an alternative should be mono* tized wherever possible and either added to the costs or subtracted from the benefits of that alternative. My costs that are averted as a result of an altema* tive should be monetized wherever possible and ei ther added to the benefits or subtracted from the cost* of that alternative. All coats calculated should be incremental, that is, they should represent changes in costs that would occur if the regulatory alternative is chosen compared to costs in the base case (ordinarily no regulation or the existing regulation). Future costs that would incurred even if the regulation is not promulgated, uW well aa costs that have already ban incurred (sunk costs), art not part of incremental costa. If margin*! cost is not constant fir any component of crate, incremental costs should be calculated aa the 'a under the marginal cost curve over the reK it range. Costa indude private-sector compliance coets, govemment administrative costs, and coets of reallocat ing workers displaced as a result of the regulation. Coets that are not monetary outlays must be included and should be attributed a monetary value wherever possible. Such costs may indude the value (opportu nity cost) of benefits foregone, losses in consumers' or producers* surpluses, discomfort or inconvenience, and lota of time. A schedule of monetized costs should be indudsd that would show the type of cost and when it would occur; the numbers in this table should be expressed in constant, undiscounted dol lars. Any expected incremental costs that cannot be monetized should be explained. An important type of cost that often cannot be quantified is slowng in the rate of innovation or of adoption of new t* ool ogy. For example, regulations requiring a cost nd time-consuming approval process for new produ or new facilities may have such costs, as may regui^^ tions setting much more stringent standards for fodlitiss than existing ones. SL 064767 664 REGULATORY PROGRAM OP THE UNITED STATES GOVERNMENT Two accounting coat concepts that should not be counted as costs in benefit-cost analysis are interest and depredation..The time value of money is already accounted for by the discounting of benefits and costs. Depreciation is already taken into account fay the time distribution of benefits and costa; the only legiti mate use for depredation calculations in benefit-cost analysis is to estimate the salvage value of a capital investment 2. Seal Costs versus Transfer Payment*. An impor tant but sometimes difficult problem in cost estima tion is to distinguish between real costs and transfer payments. Transfer payments are not genuine costs but payments for which no real good or service is received in return. Several examples of problems that may arise from the confusion between transfer pay ments and real costs (or benefits) may help to iden tify situations in which further analysis of the prob lem may be warranted. Monopoly profits, insurance payments, government subsidies and taxes, and dis tribution expenses are four potential problem areas. (a) Monopoly profit*. If, for example, aales of a competitively produced product were restricted by a government regulation so as to raise prices to con sumers, the resulting monopoly profits are not a benefit of the rule, nor is their payment by consumers a cost The real benefit-cost effects of the regulation would be represented by changes in producers* and consumers' surpluses. (b) Insurance payments. Potential pitfalls in bene fit-cost analysis may also arise in the case of insur ance payments, which are transfers. ^Suppose, for example, a worker safety regulation, by decreasing employee injuries, led to reductions in firms* insur ance premium payments. It would be incorrect to count the amount of the reduction in insurance pre miums as a benefit of the rule. The proper measure of benefits is the value of the reduction in worker injuries, monetised as described previously, plus any reduction in real costs of administering insurance (such as the time of insurance company employees needed to process claims) due to the reduction in worker insurance^ claims. Reductions in insurance premiums that are matched by reductions in insur ance claim payments are changes in transfer pay ments, not benefits. (c) Indirect taxes and subsidies. A third instance where special treatment may be needed to deal with transfer payments is the case of indirect taxes (tariffs or excise taxes) or subsidies on specific goods or services. Suppose a regulation requires firms to pur chase a $10,000 piece of imported equipment, on which there is a $1,000 customs duty. For purposes of benefit-cost analysis the cost of the regulation for each firm ordinarily would be $10,000, not $11,000, since the $1,000 customs duty is a transfer payment from the firm to the Treasury, not a real resource cost This approach, which implicitly assumes that the equipment is supplied at constant costs, should be used except in special circumstances. Where the taxed equipment is not supplied at constant cost, the technically correct treatment ia to epilate how many of the unite purchased as a result of the regulation are supplied from increased production and how many from decreased purchases fay oty- buyers. The former units would be valued at the price without the tax and the latter units would be valued at the price including tax. This calculation is usually difficult and imprecise because it requires estimate* of supply and demand elasticities, which ire often difficult to obtain and inexact Therefore, this treat ment should only be used where the benefit-cost conclusions are likely to be sensitive to the treatment of the indirect tax. While costa ordinarily should be adjusted to remove indirect taxes on specific goods or services es described here, similar treatment is net warranted for other taxes, such as general sales taxes applying equally to moat goods and services or in come taxes. (d) Distribution expense*. The treatment of distri bution expenses ia also a source at potential error. For example, suppose a particular regulation raises the cost of a product by $100 and that wholesale and retail distribution expenses are on average SO percent of the factory-level cost It would ordinarily be incor rect to add a $50 distribution markup to the $100 coat increase to derive a $150 incremental coat per product far benefit-cost analysis. Most real resource costs of distribution do not increase with the price of the product being distributed. In that ease, either distribution expenses would be unchanged or, if they increased, the increase would represent distributor monopoly profits. Since the latter are transfer pay ment*, not real resource costs, in neither case should additional distribution * be included in the benafit-oost analysis. However, increased distribution expenses should be counted as costs to the extent that they correspond to increased real resource costs of the distribution sector as a result of the change in the price or characteristics of the product D. Expenditure Rules Regulations establishing terms or conditions of Federal grants, contracts, or financial assistance call for a different form of regulatory analysis thin do other types of regulation. In tome instances, s foilblown benafit-oost analysis may be appropriate to inform Congress and the President more folly shout the desirability of the program, but this would not ordinarily be required in a Regulatory Impact Analy se 06476s APPENDIX V 665 . the primary fonction of the RIA for this type of probably is not sensitive to likely alternative param regulation should be to verity that the terms or eter values. If the ratio is only slightly greater conditions are the <* necessary to achieve the an*, the conclusion probably is sensitive. The benefit- purpoeea for which the fond* were appropriated. cost ratio may aometimas be acceptable as a rough They should not contain conditions in pursuit of goals aubstitut* for genuine sensitivity analysis where it is that are not germane to the purpose for which the not feasible to cany out a full sensitivity analysis foods were authorized and appropriated. Beyond eon* (e.g., if tbs number of regulatory parameter! to be fools to prevent abuse and to ensure that foods tested by sensitivity analysis is large). Whan so used, * appropriated to achieve a specific purpose are ehan- tit* benefit-cost ratio should be recognised as only a | Baled efficiently toward that end, msrimmn discre- crude approximation to a genuine sensitivity analysis i tion should be allowed in the use of Federal fonda, and the analyst should be aware of its limitations \ particularly whan the recipient is a State or local (*4., the benefit-cost ratio is sensitive to the arbi | government trary classification of an item as a benefit or an l IV. RATIONALE FOR CHOOSING THE PROPOSED REGULATORY ACTION averted root). Where the benefits of proposed regulatory alterna tives include reductions in fatality risks, an accept able alternative to direct calculation of net benefits is The RIA should include an explanation of the the indirect approach of calculating incremental coats masons for choosing the selected regulation. Ordinar per life saved between adjacent alternatives. This is ily, the regulatory alternative selected should he the done by ranking all tha alternatives according to tit* one that achieves the greatest net benefits. If legal number of lives they save and than calculating the constraints prevent this choice, they should he identi change in eocto and th* change to lives saved be fied and explained, and their net cost should be tween each alternative and the one with the next estimated. highest number ef fives saved. If tbs alternative Where uncertainties are substantial or a large proportion of benefits cannot be monetized, other methods of summarizing the benefit-cost analysis may sometimes be appropriate. When alternative selected is tha one whose incremental cost par life saved is closest to the wQlingness-to-pay value of fife, this decision criterion is analyticdly equivalent to that of maximizing not bmeflt forms of presentation are used, the objective must In cases where important benefits cannot ba as continue to be the maximization of net benefits (ex signed monetary values, eoat-effactivanass analysis cept where prohibited by lew). Alternative criteria should be used where possible to evaluate alterna must he used with care because of the potential for tives that generate equivalent nonmcnetizabl* bene errors or misinterpretation. fits. Coats should be calculated net of monetized Agencies need not calculate the internal rate of return for e regulation. The internal rate of return is often difficult to compute and is problematical when multiple rates exist. It must not be used as a crite rion for choosing between mutually exclusive alterna tives. As s criterion for choosing between alternative* benefits. Between two alternatives with equivalent nonmonetizable benefits, the alternative with the lower net costs should be selected. Cost-effectiveness analysis should also be used to compare regulatory alternatives to eases whan tha lave! of benefits is specified fay statute. that are not mutually exclusive, it has no advantages aver the criterion of twvitniring the present value of V. RAXUTORY AUTHORITY net benefits. Benefit-cost ratios, if used at all, must be used with care to avoid a common pitfall. It is a mistake to ehoose among mutually exclusive alternatives by se The RIA should include a statement of determina tion and explanation that tha proposed regulatory action is within tha agency's statutory authority. lecting the alternative with the highest ratio of bene fits to costs. An alternative with a lower benefit-cost Further ratio than another may have the higher net benefits. Whether a regulation's benefits are greater (or lees) Edith Stokey and Richard Zeckhauser, A Primer for titan its costs can be determined by whether its Policy Analytic. Chapters 9 and 10 provide a good benefit-cost ratio is greater (or less) than one. The introduction to basic concepts. benefit-cost ratio may be used as a very simplified E. J. Mishan, Economics for Social Decuicna: Ele indicator of the likely sensitivity of the result' If the ments 0/ Cost-Benefit Analysts. Assumes some knowl benefit-cost ratio is much greater than one, the con edge of economic*. Chapters 5-6 should be helpful ong^ clusion that the regulation's benefits exceed its costs the important subjects of producers' and cantumen^P SL 064769 666 REGULATORY PROGRAM OF IBS UNITED STAIRS GOVERNMENT surpluses (not discussed extensively in this guidance document). W. Kip Vucusi, Risk By Choice. Chapter 6 is a good starting point for the topic of valuing health and safety benefits. Other more technical sources are 'ven in the bibliography. Robert Cameron Mitchell and Richard C. Careen, Using Surveys to Value Public Goods: The Contingent Valuation Method. Provides a valuable discussion on the potential pitfalls associated with the use of con* tingent-valuation methods. V. Kerry Smith, Ed., Advances in Applied Microeconomics: Risk, Uncertainty, and the 1hluation of Benefits and Costs. Judith D. Bentkover, Vincent T. CoveDo, and Jeryl Mumpower, Eds., Benefit* Assessment: The State of the Art. SL 064770