Document N2z4YdgRennJzXxLEbb2dRGVb

ASSES S>--KT or CAP.CIKOCENiC RISKS FROM PCBs IK POOD Prepared for United States Congrats Office of Technology Atactawent Under Contract No. 933.1350.0 Kenny S. Crutap, Ph.D. Marjory D. Hasteraan, B.S. science ksssazch iystuis, ikc. Box 7356 ftuston, LA 71272 April. 1979 MONS 221314 TABLE OF CONTENTS Pa^c 1. Chlaical Propartita .................................................................................. 1 2. Food and Environmental Contamination.............................................. 5 2.1 Routes of Centanination and Environmental Diatribution ................................ 6 2.2 Levala in Food from the FDA Total Diet Program ... 9 2.3 Levala in Flab.................................................................................. 11 2.4 Human Exposure from Consuming Lake Michigan Sport Fish............................................................................................ 16 2.3 FCB Levels in Human Milk............................................................ 18 3. Carcinogenic Effects...................... 21 3.1 Human Data............................................................................................ 21 3.2 Data from Lifetime AnimalFeedingExperiments. ... 23 3.2.1 The Experiment of Kimbrough tt al. (1975) on the Effect of Aroelor 1260 on Female Sherman Rets.................................................................... 24 3.2.2 The Rational Cancer Institute (1978) Bloassay of the Carcinogenic Effect of . Aroelor 1254 in Fisher 344 Rats............................. 26 3.2.3 Industrial Bio-Test Experiment with Charles River Rats....................................................................... 28 3.3 Synergistic Effects.......................................................................... 30 4. Quantification of Carcinogenic Risk............................................... 31 4.1 Calculation of Virtually Safe Dotes from Different Mathematical Models........................................................... 31 4.2 Estimates of Risk from PCBa in Diets of U.S. Adults..................................................................................................... 35 4.3 Estimates of Risk to Breast-FedInfanta............................... 39 Appendix: Review and Evaluation of Methods of Determining Risks From Chronic Lov-Lcvcl Carcinogenic Insult................... 46 References................................ .... ......................................................................... 80 MOMS 221315 X. Chtaiical Properties Polychlorinated biphenyls (PCBs) art a family of compound* produced by the total or partial chlorination of biphenyl. There are ten posalble hydrofen bonds vhlch can acccsaedate chlorine substitution in a biphenyl molecule. A specific chloroblphenyl molecule may be referred to by naming the positions of the chlorine atoms accordlnf to the dia gram in Figure 1 (National Institute for Occupational Safety and Health (NIOSH), 1977). In naming a particular molecule, the lowest possible numbers are assigned and the ring with the fewest chlorine substitutions is assigned primed numbers. See Figure 2 for examples of the naming convention. When two compounds have the same number of chlorine substituents on the biphenyl molecule but the substitutions occur In different locations, the two compounds are called isomers. The class of chlorobiphenyls consisting of one type of Isomers is referred to using a prefix denoting the number of chlorine substituents, s.g. pentachlorobiphenyls. There are ten possible chloroblphenyls and 209 possible individual isomers. The term PCBs is used to refer to a complex mix ture of different chloroblphenyls and their Isomers (Nisbet, 1976a). The synthesis of a pure sample of a specific chloroblphenyl in the laboratory Is quite difficult because the chlorination process of biphenyl naturally forms mixtures of many FCBs. Because some chlorine substitution patterns are more likely than others, some Isomers are more likely to occur than others and this In turn causes some of the 1 MOMS 221316 32 6' 5' Figure 1; Biphenyl molecule and ring numbering system Cl Cl 3'3\4-trlchl or obi phenyl Figure 2: Examples of the numbering system for chloroblphenyls HONS 22131? 5 chlorobiphenyls to be more abundant than others. For example, a ehlorobiphenyl completely chlorinated on or.e ring and unchlorinated on the other la not likely to appear (Environmental Protection Agency (EPA). 1976). PCBs are produced commercially by the chlorination of biphenyl with anhydrous chlorine in the presence of a catalyst, usually iron filings or ferric chloride (EPA, 1976). The chlorination process is terminated at a particular step In order to produce a mixture of chlorinated biphenyls with an average percentage of chlorine. These crude isomeric mixtures are further refined to remove color, traces of hydrogen and the catalyst by a distillation process (EPA, 1976). PCI products made by Monsanto are called by the trade name "Axoelor". Varying kinds of Aroclors are designated by a four-digit number, e.g. Aroclor 1254 or Aroclor 1242. The first tvo digits refer to the fact that blphanyl is made up of twelve carbon atoms, and the second tvo digits refer to the approximate percent by weight of chlorine in the mixture. Thus, Aroclor 1254, a mixture of chlorinated biphenyls, would contain by weight, 542 chlorine and 462 carbo'n, hydrogen and trace elements. Aroclor 1254 would not alvaye contain the exact same amount of a particular compound since commercial mixtures will vary from batch to batch, but the relative amounts of the ten isoner classes will stay the same from batch to batch. One exception to the usual Aroclor nomenclature la Aroclor 1016, which vas Introduced in 1971 to gradually replace Aroclor 1242. Aroclor 1016 la conprised mainly of tri- and tetrachlorobiphcnyl compounds and contains 412 chlorine by weight. It differs from Aroclor 1242 in that MQNS 221318 4 the latter contain* a greater percentage of the more highly chlorinated compounds (KIOSK, 1977), In Japan other PCB product* were marketed under the trade names "Santothers*' and "Kaneehors" 300, 400, 500, and 600, In Cernany PCS product* were marketed under the trade name "Clophens" A50 and A60, and in Prance under the nar>e* "Phenoelor*" and "Pyralenes" (KI0SH, 1977). Studies of the commercial preparations with PCS* have turned up two interesting fact*. First, about half of the 209 possible isomers do not occur in the coraercial mixtures. Zn fact, Aroclor 1248 contains less than 50 separable Isomers. Second, trace contaminants such as polychloro naphthalenes and polychloro dibenzofurans (PCDF) heve been found in some samples of PCS*. The contamination say occur during the manufacturing process (K1CSK, 1977). PCS* have several physical and chemical properties of both In dustrial and environmental, interest. Even through the basic properties of the different eoaaerelal PCB mixtures depend on the particular mixture of ehlorobiphenyls in the specific preparations, the mixtures are enough alike that the Aroelors can be used ns a basis for discussion of these properties (EPA, 1976). Commercial PCB mixtures can be mobile oils, viscous liquids, sticky resins or crystalline solids at room temperature. Their color ranges from clear to yellow to yellow-green to black. A more complete listing of physical properties is given in EPA (1976). p. 40. Chemically, PCBs are very stable organic compounds, exhibiting a high heat capacity, low electrical conductivity, a resistance to oxidation, low vapor pressure and excellent dielectric properties. HOMS 221319 5 PCBs are quite stable to long heating and those PCI liquids with tour or taort chlorine ator.a are non-flarrj-jble as are their vapors. Al though PCSs are not very soluble in water, with this solubility decreas ing with increasing clorine confant, PCSs are quite soluble in hydro carbon .solvents, lipids and in oils. Also the adsorption rate of PCBs onto plastic, glass, dust particles, silt and sand is high. Because of these "attractive" chordeal properties, industry has ssdi exten sive use of PCSs, mostly in "closed" or "senl-closed** systems such as electrical transforcers and capacitors, and in heat transfer systees (ZPA, 1976). When it became evident in the lace 1960's that PCBs bad become an environoental pollutant, Monsanto Co. voluntarily phased out pro duction of Aroelors 1232, 1248, 1260, 1262 end 1268 in 1971. Konsento slso began selling PCSs for use in "elcsed" electriesl systems only end, ss of 1975, was producing only Aroelors 1221, 1016, 1242 sod 1254 (Nisbct, 1976s). 2. Pood and Ir.viror.mental Centar.lnstlon The particular combination of phyaical and chemical properties in ehloroblphenyls that led to such s widespread usage of PCBs in industry is also an important factor in the process of PCBs becoaing environmental pollutants (IPA, 1976). Some of the more significant properties include low solubility in water, high solubility in lipids and high specific gravity. To deteralne the impact of PCBs ss en vironmental pollutants and food contaminants, it is necessary to determine first how they ere released into the environment and, secondly, how they are dispersed throughout the environment. HONS 221320 6 There art basically three ways that PCBs can be releasee into the environment. First, there is intentional disposal of PCBs and products containing PCBs, mostly into duttp* and landfills. Second, there ia disposal into water, including losses of industrial fluids and deliberate dumping of industrial vastes. Third, PCBs are released into the air, including vaporization fros plasticised products, releases from incomplete burning in dumps or incinerators, and re leases from burning serap products (Klsbet, 1976b). There are tvo major methods of transport once the PCBs are in the environment. PCBs are transported aerially in the fora of a vapor or because they are trapped on airborne dust particles (CPA, 1976), and eventually return to earth in rain or dustfall. PCBs are trans ported through vster in several vays including the movement of sediment containing PCBs, dumping sewage sludge in the ocean and direct leaks from coastal industries (Klsbet, 1976b). Due to the low solubility of PCBs in water and/or the high specific gravity of PCBs, most of the PCBs discharged into the water will either be ab sorbed into the sediment at the bottom of the body of vster or evaporate into the air. The latter possibility is evidenced by a co-evaporation study on Arodor solutions, and since each Aroelor is made up of dif ferent isomers of several chlorobiphenyls, the vapor pressure or solubility of a particular Aroelor really represents an average over its various isomeric components (EPA, 1976). The former possibility Is evidenced in a study of drainage basins in the United States and Puerto Kico where it was apparent that bottom sediments contained much higher concentrations of PCBs than did overlying water (EPA, 1976), MONS 221321 7 presumably because of the high specific gravity and low solubility In vatar of PCBs (Dennis, 1976). PCBs are available for introduction into food chains through the above methods of release into the environment. The PCBs are taken in by biota directly, through absorption, or indirectly, through a food chain. That biological magnification can occur has bacn ob served in the vide ranging accumulation of PCBs within the biota, where extremely large loads are found in the highest members of the food chain (EPA, 1976). It has been demonstrated la the laboratory that fish can accumulate PCBs to concentrations as much as 276,000 tines the concentration of PCSs in water (Nlsbet, 1976a). Food nay be eontaninated by PCBs also through machine leaks, such as heat transfer equipment leaking during pasteurization, contamination of packaging materials (recycled paper) and contamination of animal feed which leads to contamination of neat and poultry, Mothers' milk has also been found to contain PCBs. Polychlorinated biphenyls are very persistent in the environment. The store highly chlorinated biphenyls are more resistant to degradation, resulting in a "differential persistence'' of PCBs. The selective degrada tion of PCBs has an effect on which of the Aroelors are most persistent in air, water, fish, and in humans. Moat of tht PCBs found in dry HONS 221322 e fallout and in the air resemble Aroclor 1254 (Kisbct, 1976c), In a study don* by tht U.5.Department of Interior's ecological Survey on PCSs contained in vhole water samples and in bottom sediments in the major drainage basins of the United States and Puerto Rico for 1971 1974, the predominant PCB detected vas Aroclor 1254 (Dennis, 1976). In other studies though, combinations of PCSs are reported: in samples of both bottom and suspended sediment taken from the upper Chesapeake Bay Aroclors 1254, 1262, 1242 and 1246 vere found (Munson, t, ,*1,., 1976). A study by Sanborn _et_ el. (1976) on the green sunfish demonstrated that in fish a main difference in metabolism comes between trichlorobiphenyls and tetrachlorobiphenyls, with a relatively efficient storage of tetrachlorobiphenyls. This is also evident in data from the 1970 Kational Pesticide Residue Monitoring Program in which substantial fractions of Aroclor 1248 occurred in fish taken from the Ohio and Hudson Rivers (Valker, 1976). McDermott, et^ al. (1976) found that Aroclors 1254 and 1242 were present in Dover sole taken near discharge points of southern California vastewater plants in the ratio of 2:1 while the ratio in the wastewater was about 3:8. This last exaspl* would imply that other factors sueh as th* sediment load or solubility of s particular PCS are involved in determining the levels of PCSs in fish. Another factor is the phenomenon of bioaccumulation, the con centration of PCBs in the tissues at much higher levels than in the surrounding water. This is due primarily to the fact that PCBs are highly soluble in lipids and not very soluble in water. In humans there is differential storage of the more highly chlorinated biphenyls, and the less chlorinated isomers are either metabolised or excreted. Results of studies on residues in human HONS 221323 9 adipose tissue indicate that the chiorobiphenyl compounds most often found vara the penta-, hexa-, and heptachlorobiphanyl cor,pounds, in amounts most comparable to those prevalent in Aroclors 1254 and 1260 (Kutt and Strassr.an, 1976). But human food monitoring indicates that most hucans are exposed to ?CBs through diet, and that most PCBs in human food are in fish. This would apparently contradlet the fact that substantial quantities of tetrachloroblphenyls are found in fish. What apparently happens is that humans may be exposed to these less highly chlorinated biphenyls but do not retain them (Kisbet, 1976c). 1.2 Levels in rood from the FDA Total Piet Program In the FDA Total Diet Frograo (Johnson and Manske, 1977), a mar ket basket of food representing the basic 2-week diet of a 16-to-19year-old nele is collected in each of several geographic areas. The various foods are prepared in the manner in which they would normally be eaten and are then analysed for the presence of various substances including PCSs. The lowest concentration at which it is possible to quantify the level of ?C2s present is .05 ppm. Levels detected below this level are recorded as "trace". The results of these studies perteining to PCBs for the fiscal years 71-76 are summarised in Table 1. HONS 221324 10 ;:>)( 1: Total Dint Studies-American ttonag* m*l Percent of composites containing PCB's Pood class composites fiscal Year Oils , Dairy r.eat, Grain legume Root fats Sugars pro- fish & & cereal vege- vege- Garden & short and ducts poultry products Potatoes tables tables fruits ening adjuncts 1S71 47 1972 6 46 13 6 6 3 3 17 1973 10 33 17 3 3 1974 43 1975 list half) 1976 40 13* * Manske (1979), personal communication Sourct: Jelinek and Corncliussen (1976) 6 3 (Ko entries signify zeros) Table 2: Estimates of daily PCS intakes (Total Diet Study-teenage male) Fiscal year Averaoe daily intale of PCS'sa Total diet (pg/day) I'ieat-fish-poul try food (pg/day) 1971 1972 1973 1974 1975 (1st half) 1976 15.0 12.6 13.1 8.8 8.7 3.3* 9.5 9.1 8.7 8.0 8.7 3.3* *lower limit of quantitative reporting 0.05 ppm with analytical method employed. * Hanske (1979)(personal communication Source: Jelinek and Corncliussen (1976) MONS 221325 II The only food clast in vhich PCSs ware detected in 1975 or 1976 was neat, fish and poultry and all of the positive observations were trace values. It is liV.ely that these positive finds are all due to PCSs in fish samples. Jelinek and Corneliussen (1976) estimated the microgras per day Intake of PCSs baaed upon the FDA Total Diet Program. These values which are listed in Table 2, must be viewed as only vary crude estieates since, in order to obtain them, numerical values had to be assigned to trace observations. For those levels which were reported as trace, the decision was made to consider them to be at one-half the quantitative lower level of detection, i.e., .025 ppm (Kisbet, 1976a). 2.3 level* in Fish Sated upon the FDA Total Diet Study (Johnson and Manske, 1977), It appears that, for the U. S. population as a whole, the most signif icant exposure to PCSs from food is through fish. It does not seem possible to evaluate adequately at this' time the overall distribution of exposures to PCBs through eating fish. The distribution of total fish consumption and consumption of tuna can be calculated from the Seafood Consumption Study (National Marine Fisheries Service, 1976), but the distribution of consumption of other species is not available. Moreover, the concentration of PCSs in fish is highly variable, both between species and within a single species. Fish caught further offshore tend to have smaller amounts of PCBs than estuarine fish, snd MOMS 221326 12 frh water fish caught in of high l`CL pollution tend to liav the highest concentration# oi ?CB$. Sutler end Schutsmann (1?7S) reported on ?CB residue* in juvenile estuarine fish as part of the National Pesticide Monitoring Program, This report is based upon 1524 analyses collected in 144 estuaries nation-wide during the period 1972 to 1976. The annual incidence of PCS residues snd the aver.ge residue discovered are given in Tsble 3. These data indicate a possible gradual dacline in both the msximu residues observed and average concentration*. Thi* i* likely due to the general curtailment in production *nd u*e of the*e compounds. It appear* from Tabla 3, however, that PCS* will continue to con taminate e*tu*rine fish for many year* to cone. Table 4 indicate* the geographical distribution of PCS residue* in estuarine fi*h. V, n<fte that average concentrations vary widely froa ct*te to state. Although these dat* are indicative of geographical distribution and time Crania of PCB concentration*, thay only represent a lower bound to concentra tions of PCS* in fish caught in the*e waters for human consumption* This is baesuse tha fish sampled were generally no mora than one year old. Pish esught for human consumption are usually older and eon.* sequently more heavily contaminated with PCBa. Valker (1976) sunaariaed tha PCB lavels in freshwater fish a* determined by the National Pish and Wildlife Monitoring Program. Geographically, the higher concentrations *??cat to be associated with certain river systems having industrisl activity .... PCB residues expressed as Aroclor 1254 were found in five major river systess in the Atlantic coastal region, with residues exceeding 5 mg/kg. Pour of these stations had residues exceeding 10 mg/kg during the lost 5 years. Pish in four cf the Crest bakes stations had PCB concentrations exceeding the 5 ng/kg level and all stations reported concentration* MOMS 221327 3 Tabic 3: Annuel incidence of IDCT, PCS, end dieldrin residues in juvenile whole fish samples, 1972- 76 Vie* "ibll l*M Itl* ISIS A4:KM|tK K t % P8brtc*f 111 41 Ml >4 III 44 lie M 1*9 it 8> fKt fllAfiril OUT M*eeei.-ee Atitewr 44 T IM on ! Al>l 43 SI 4t SI i SlsiHtl, et/it ren % i v*wt >4 01 19 99 9* mi 9* >* 19 11*0 *>fh r*IMt iriBi A *14iCC ltnvi * M * m 4* >si Source; Butler end Schutzasnn (1978) fief |+*l* b M* ClntU4t SrtiM X y l4 31 t M a HI 13 i IS *4 *-- Table 4: Frequency end everege concentration of PCS residues In Juvenile estuerine fish by coestel eree, 1972-76 Ami r*>*w*cY Of ftMWII. AM* A*M*I CtW!*ueMt*, #(/(' COMHOMM SbMf lbJ*A4 Kt- Yet OlU* Vie 110408 t4t ClWtoua n>4i * Tkh Atoktw Vie#4* H<|*4 him 0(*|84 Cw|h l4ui>iini Al#t4 ( Maeiatai^M SI TS o St u 9* J] >1 >c 91 It IT v< r# * 1 ) I WMheAftM 89#44 v*p**i8 Vr^e IblteSl Crcn Ms|UM K Tot l#uiai|g| fbltl# S4S HmiU K*tb C**h*4 C^44mm| Otr*4 S^maC******* AliKan Ttilt r >* Till m 04 m 990 111 ) >04 H) 9M >< >44 >4? 99* m in i) ns 19 HOTI' C"*fi****4*ii #| JaSkMM* too M#e aMr neipe*#* ! S4U Sm # io*td4e| pqeee^b (r#t| Ml toeH^', 8*4 Ifetifi *1 (** 4>K Jr rub twdii *Af*millHn<neieMgftieMe,ter* f MTIM si SUllMl m i* Source: Butler end Sehuttmenn (1978) MOMS 221328 14 exceeding 0.3 mg/kg. In the Mississippi River system, the Allegheny arid Ohio 'ere the hot spots, with seven cut of the eight stations reporting residue concentrations in excess of 5 ng/kg. Thirty-one of thirty-five stations in this river system reported residues in excess of .13 ng/kg in the 1970-73 sapling programs. The highest residues, often exceeding 10 ag/kg were found in the Allegheny, Ksnasha, Cumberland, Tennessee, and Ohio Rivers along with stations on the Mississippi River at Memphis, Tennessee, and the Mis souri River at Kernin, Missouri. Other monitoring cations that were found to have residue levels ex ceeding 5 og/kg during the sampling periods 1970-73 included: the Villia^ette River on the Columbia system; the Rouge River in the Pacific coastal drainage; the Sacramento River in California; the Chena River tributary of the Yukcn in Alaska; and the Rio Grande, Alabama, and Mississippi Rivers in the Gulf States region. Only in two sample periods of 1972-73 and in the current monitoring simples; which are still yet to be fully analysed, has thert bean a downward trend, but this occurs only in those saeples where residues are not being detected. The stations where high residues hive been noted in the past still remain relatively contaminated with PCS. Unlike the decline of DDT in Great Lakes fishes, PCS concentrations do not shew significant changes and may trend upward in salmonlds. An analysis of PCS Icvala in frtah-watcr fiah in New York State fros 1971 through 1973 by the New York State Department of Environ mental Conservation (Spagnoli and Skinner, 1977) Indicated that Hudson River fiah contained tha highest known PCS concentredens with in th United States. Levels in individual fish were often found to exceed 100 ppm. The highest individual concentration recorded was 359.23 ppm la a large eal. Lake Michigan la another body of water that has a high level of contamination by PCSs. Tablt 3 shows data from a thrcc-year study of two sptciea of Lake Michigan fish by tha U. S. fish and Wildlife Servlet Laboratory (Humphrey, 1977). As shown, the PCS levels in lake trout rose during the three year period snd levels in Coho salmon did not change appreciably. HONS 221329 15 Table 5: PCB contamination in whole fish* Collection Length Sar.pl e Kean PCB Species 1 Location Year (inches) Size Value (ppm) Lake Trout near South Haven 1972 1973 197M 20 - 28 20 - 28 20 - 28 9 12.86 i M.7S 30 18.93 t 2.08 30 22.91 t 3.7S Coho Salmon near Ludington 1972 1973 197M 70 - 32 20 - 32 20 - 32 10 10.93 t 2.12 29 12.17 a 0.77 30 10.HS t 0.92 *Data from the Great Lakes Tishery Laboratory, U.S. Bureau of Sport Fisheries and Wildlife. Source: Humphrey (1977) HONS 221310 16 2.4 Human Exposure from Cer.surjnr. Lake Hlchlr.an Sport Fish A two year study (1972-1974) vn made under sn FDA contract (Humphrey, 1977) of persons who regularly consumed PCB-contaminated Lake Michigan sport fish and randomly selected persons who did not consume such fish. A total of 160 adults participated in the study including 91 who consumed more fish than recoemended by the Michigan Department of Publie Health (no more than one meal per week or 24 pounds per year), 21 who consumed less than 6 pounds per year, another group who consumed an intermediate amount, and 19 fish eaters who consumed fish primarily from Lake St. Clair'which are not as heavily contaminated with PCBs as fish from Lake Michigan. A medical record, a dietary record, and blood specimens were obtained for all participants. Summaries of mean PCS levels of participants arranged by city of residence and level of fish consumption is shown in Table 6. In 1973 those who annually consumed 24 or more pounds of sport fish from Lake Michigan had a mean blood PC3 level of 0.073 ppm; persons annuelly eating 6 pounds or leas had a mean level of 0.020 ppm; persons eating no fish had averaged 0.017 ppm; and persons eating 24 or sore pounds from Lake 5t. Clair (who resided in Algonac) had an average blood level of .023 ppm. A comparlaon of PCI blood levels in persons in the group with highest consumption of Lake Michigan fish to PCB levels in those who ate no fish re\ealed a highly significant difference in mean PCB levels (p << ,001), A similar comparison between persona who occasionally ate Lake Michigan fish (1-6 pounds per year) with those who ate no fish also Indicated a highly significant difference (p << .002). The level of PCBs found in the blood of participants did not change significantly from year to year, nor did it diminish HONS 221331 'tit 6: Year to year comparison of mean PCS blood levels for consumers of Lake Michigan fish City Kean Total PC3 Value (ppm) 1973 1 Participant Group Cont rol Inter red iate Exposed Kean Total PC3 Value (ppm) 19 71< Participant Group Inter Control mediate Exposed Inverse City 0.017 0.026 0.060 0.019 0.036 0.061 ^r.istee 0.026 0.066 0.109 0.026 0.029 0.120 Ludingt on 0.021 0.0M8 0. 062 0.02M 0. 066 O.O&S South Haven 0.028 0.036 0.0B3 Algonac - 0.02U 0.023 Croup Hean f i rst 1 Cit i t* 0.020 "UT*e; Humphrey (1977) 0.0UM 0.073 0.073 0.0M1 0.075 HONS 221332 IB significantly when fish consumption was eliminated for up to nine months. PCS levels measured during the study in cool:ed Lake Michigan fish ire exhibited in Table 7. A comparison with Table 5 reveals that levels in cooked fish appear to be only 1/6 to 1/4 the levels in raw fish. The quantity of PCB ingested iron eating Lake Michigan fish averaged 46.5 og/yemr and ranged from 14.17 to 114.31 mg/year for the 91 participants eating more than 24 pounds per year. Vithln this saoc heavily exposed group, the PCB dose averaged 1.7 wg/kg/day <microgress PCB per kilogrea body weight per day) and ranged from .490 to 3.940 vg/kg/day. One participant in the study gave birth to a child in January of 1979. A milk specimen fros this individual contained 4 ppm PCB (fat basis--'the specimen hed 2Z fat content), whereas a blood speci" men collected et the setae tie* contained ,053 ppm total PCB. As a group, the exposed participants had no health problems or medical conditions that could be correlated with PCB blood levels, exposurs to Lska Michigan fish, or known syeptoms of PCB poisoning. 2.5 PCB Levels in Human Milk Savage (1979) reported on levels of PCSs in human milk sampled from 1036 individuals residing in 44 different states. Of the 1036 samples, PCBa vert detected in all but 9 samples. Trace values were separated for 729 samples and tha concentrations in the other 309 samples ranged from a low of 0.3 ppm (fat basis) to 16.92 ppm (fat basis). Eighty-on* or 7.62 of the samples contained residues in excess of 2.5 ppm (fet basis) which was the temporary tolerance sec by FDA for PCB levels in commercial milk. HONS 221333 Title 7: ltl level* In cooked Kichi&an fih ]9 pik. VhiteHeh, Smelt, Chub*. Menominee n* fereh Source: Humphrey ( 1977) HONS 22133* 20 To eetiraate the average concentration of PCSs In human milk, w* took natural logarithms of the reported concentrations in whole milk.. Estimates v and o2 of the mean and variance of the log-concentration of ?C5s in whole milk were calculated using the method of Cohen (1961) for estimating the mean and variance of a normal distribution from censored data. We then estimated the mean concentration of PCBs -in whole human milk to be exp(p + 62/2) exp(-3.4126 + .6337/2) .0653 ppm. Since the average percent lipid content in the 309 samples was cal culated as 4.497Z ve estimated the average concentration of PCSs in human milk to be .0453/.04497 * 1.007 ppm (fat basis). By way of con trast, the temporary tolerance set by FDA for Infant and junior food is 0.2 ppm. Based upon our use of the log-normal distribution for PCB concentration in whole milk samples, we estimate that approximately 952 of the 1036 human silk samples had PCB concentrations in excess of 0.2 ppn (fat basis). We, likewise estimate that the percent of human ilk samples having more than 2.5 ppm, 5.0 ppm, 10 ppm is 7Z, 1Z, .053, respectively. It would be difficult to estimate precisely the proportion of PCB contamination of human milk that is due to FCBs in food, but most of the contamination is likely due to dietary exposure. The single measurement of 4 ppm (fat basis) of PCB in human milk of a consuMr of Lake Michigan fish in the FDA study (Humphrey, 1977 ) indicates that consumers of PCB contaminated sport fish may have significantly higher levels of PCBs in their milk than women in general. To corroborate this single measurement, we considered comperisons of PCB levels in human blood plasma and milk made by Poliahuk, e_t el. (1977) in a study involving 29 Israeli women. Average PCS levels measured were .0193 - .0126 ppm in plasma, end .0442 - .0412 ppm In whole HONS 221335 nilk. Consequently, the concentration of FCEs in whole milk was 71 .0442/.019 2.33 timet the average concentration in plasma. If this ratio holds for the most heavily exposed group of fish eaters then the average concentration of PC5s in whole milk of hunans who consume PCBa from fish at the same level as this heavily exposed group should be about (.073)(2.33) " .170 ppm. If further, we assume the average lipid content of whole milk is 4.497* (Savage, 1979), then the average concentration of PCBs in milk from thesa heavily exposed individuals should be about .170/.04497 3.78 ppm (fat basis). The level of uncertainty associated with this calculated value if fairly high. However, it does agree nicely with the value of 4 ppm measured in the single sample obtained in the FDA study (Humphrey, 1977). Since high consumers of Lake Michigan fish were purposefully re cruited for the study, it is not possible to esrimate from the study the number of people in Michigan who consume such high levels of contaminated fish. However, based upon the information in tht study, it seem* reasonable to assume there ia a sizable population of women who consume fairly high levels of PCB contaminated sport fish end who consequently have PCB levels in milk which average about four times * (* 4 ppm vs. 1 ppm) the levels in the general population. 3. Carcinogenic Effects 3.1 Human Pete In the early part of 1968 the accidental contamination of edible rice-bran oil led to an epidemic poisoning of the Japanese families 'A* consumed this oil. The disease later became known as Yusho or Mce-oil disease. The contamination of the rice-oil was due to a l*k in a heat-exchange unit while the ell was being heated to remove *o*e of its oderous components (KIOSK, 1977). The chief symptoms of Y*ho disease were ehloracne and eye discharges while other symptoms MOMS 221336 2: Included discoloration of the skin, headaches, fatigue, abdo:iun,il pain, menstrual changes and liver disturbances. Sable* born to mothers who consumed the rice-oil were sr.all-fer-dat* and had temporary skin discoloration. Tha first symptom* of Vusho disease war* registered on June 7, 1968 and 1291 cases had been reported ea of Kay, 1975 (N10SH, 1977). The concentration of PCBe in the oil varied according to date of production end method of shipment. Tor canned oil produced on February 5, 1968, the PCB concentration ranged from 2000 to 3000 ppm, while the bottled oil produced on February 10, 1968 contained 134 pps and only a trace v*s detected in samples of oil produced after February 19, 1968 (Cordle, *t_ al. , 1978), The main contaminant of the rice-oil was Kanechlor 400, but one important "minor" contaminant vat polychlorinated dibanaefuran, present at about 1/200 the concentration of PCEa in tha rice-oil (KIOSK, 1977). An analysis of 146 of the total 1291 reported cases of Yusho disease was dona in 1971. It was estimated that tha lowest dose pro ducing overt effects was 0.5 g (Cordle, et_ l., 1978) while the average ingestion producing these same symptoms was 2.0 g and tha average volume consumed was 800 ml (K10SH, 1977). But tha question arises as to whether the effects noted are du* in feet to the PCBs alone. The ratio of PCBs to FCDFs in the Yusho oil (containing "used" Kanechlor 400) vae 200:1 whereas tha ratio of PCBs to PCDFs in "unused" (unhested) Kanechlor 400 la 50,000:1. Thus with respect to PCBs the ratio of PCDFs in Yusho oil to PCDFs in rice-oil is 250:1. Also, the toxicity of PCPFs ranges from 200 to 500 times thst of PCBs (Cordle, et al. 1978). Thus for cqutl.amounts of rice-oil and pure Kanechlor 400, the HONS 221337 toxicity of th rice-oil would range from 2 to 3.5 times that expected from its PCD content alone. Because of the uncertainty about the confounding of effects be tween PCBs and PCDFs, it is difficult to deternine from the Yusho data exactly what effeet(s) exposure to PCBs alone could have on humar.s. Careful records of the 1291 Yusho patients have been kept in an ef fort to determine possible long term effects. At least 9 out of the 29 deaths that occurred as of >`ay, 1975. were attributed to malignant neoplasms (NIOSH, 1977), but a causal relationship between PCBs and cancer can necessarily be inferred because of the high concentration of PCDF in the oil. The Yusho study, nevertheless, had two important results: first, the information established that PCBs can be trans ferred from mother to fetus and from mother to child through breast feeding, and second, highly chlorinated PCS compounds are excreted more slowly from the body than the less chlorinated ones (K10SH, 1977). Zn e study of chemical workers (Bshn, et_ al_., 1976, 1977), two malignant melanomas were diagnosed in 31 workers heavily exposed to Arofhlor 1254 (and also exposed to other chemicals that could possibly cause cancer). It was estimated that .04 malignant melanomas would have been expected from this group of individuals. Among 41 other workers also, but less heavily, exposed to Aroclor 1254, one additional melanoma was diagnosed. 3.3 gats frsm Lifetime Animal Fecdlm Experiments There have bean a number of studies on neoplastic effects resulting from feeding PCBs to experimental rats or mice (Zto, 1973, Linder, _t l., 1974, Kimbrough, at al,-, 1973, 1974, 1975, Calandra, 197G, lid, 1978)- Although each of these studies provides sons evidence that HOMS 221338 24 FCBs induce neoplastic lesions in rodents, only three of these experi ments vere of sufficient duration to be classified as lifetime feeding experiments. These three long-term experiments will be summarized below. 3,2.1 The Experiment of Kimbrough t al. (1975) on the Effect of Aroclor Four hundred weanling Sherman strain COBS female rats 21-26 days old were divided randomly into two groups of 200 animals each. One group was maintained as controls and the animals In the other group were fed libitun e diet containing 100 ppm Aroclor 1260. Exposure was continued until 6 veeks prior to the termination of the experisent. When the enicals were 23 months old ell surviving animale were sacrificed and autopaied. A total of 184 dosed rata and 173 controls survived to the end of the experiment. Pathologic findings arc susssariced in Table C. Of the three studies discussed in this section, this study provides the most convincing svidence of the carcinogenicity of PCBs> This study involved e relatively large number of animals and the increased incidence of hepatocellular carcinoma* in the treated group la highly significant. Following arc excerpts from th* analysis of this study in the Criteria Document for PCBs (Kiabet, 1976a). This was a well-designed and well-conducted study, deviating from recommended protocols in only one respect: Its limitation to female rata. . . . Two other limitations may be noted: (a) the rats vara not exposed prior to weening, although the authors . . . had indicated the rata exposed to PC&'s during gestation and nursing showed early liver changes at much lover dietary levels; (b) the rats vere killed st 23 months, relatively early in eomperison to the 26-30 month expected life span. Each of these design features of the experiment, although com mon practice in carcinogenesis bioascays, is likely to have reduced its sensitivity in demonstrating carcinogenic effects. . . . MOMS 221339 1 25 >]* $: Incidence end type of liver lesions end tumors of other orgens exemined histologicelly txlut Uu TS/ruid fiend Adr*Ml iHnrf ftlvillP)' (Uni Vl.nu CnM7 feUddtr Mtmmkiy .iM iljTIIT fluid L*l ASimm iWm Bruit 0*'7 Htir-iU^w lit Kt^iwcilivW nrtlc<u S:!ul nc^ala r*fi *t %nu W rvi*. plumk l:#nt46 <te| Clir:r..Tit . lat.KRi ;(!).inJ f-1yp lixI ttriemt4 tl ndBiintl T n:\iU :e*.4j n ptpi'Umi rib^i^HRi ftbrttilttlM Adtwm lijiflk ClllAI Cfa^wWi diti nt] turner BtpilUry tdineme CiwWytK Uueimift KV*f TWrenu n(tail * laatfsa* >)tR)L4.{)(l9A Tb) KrO/T> ACrnemi F.W=* gfw diwsitss *lil Centruk Lrptri BM4J 1/173 35/154 0/173 1U/1S4 35/173 1I3/1S4 37/t 7/133 41/151 0/151 15/14* 0/14* 3/14* 1/1*7 17/173* 4/173* 1/1*3* 3/173 0/173* 0/173 5/14* 1/14* 1/171 0/173 0/173 1/173* 0/173 0/171* 1S/155 1/1*7 3S/H* 1/13* 24/153 3/153 7/10 0/lt* 11/IM* 1 /1S* 0/154* 3/154 3/154* 3/1C4 0/IO 3/153 0/154 3/1*4 1/1(1 0/lSi* 3/154 1/154* Source: Kie<brou*h, et el.. (1975) HONS 221340 26 3.2.2 The National Cancer Institute (197S? Bio.irsav of the Careincfcnlc Effect of Aroclor 1234 in Fisher 344 Eats Croups of 24 fisher 344 rats of each sex were administered Aroclor 1254 at one of three doses, either 25, 50, or 100 ppm ad libltuo for 104-105 weeks. Matched controls consisted of groups of 24 untreated rats of each sex. All animals were 53 2 days of age when placed on study. All surviving animals were sacrificed 104-105 weeks alter the aninals were placed on study. It was concluded that "under the condi tions of this bioassay, Aroclor 1254 vss not carcinogenic in Fisher 344 rats; however, a high incidence of hepatocellular proliferative lesions in both aale and female rats v*as related to the administration of the cheaical. In addition, the carcionnaa of the gastrointestinal tract say be associated with the administration of Araclor 1254 in both sales and fassies." The observed incidences of leukemias, lymphomas, and hepatocellular carcinoses are listed in Table P. Although Aroclor 1254 was not shown to be carcinogenic in Fischer 344 rats by the NCI Bioassay, it must be kept in mind that this was a relatively small experiment utilising only 24 animals per dose group. The usual number of animals per group for an NCI bioaasay appears to be 50. This study was part of a larger study designed to assess the cotfeined effects of a group of chemicals which is likely the reason this bieassay is smaller than most. Overlooking for the moment the fact that the protocols were different and different Aroclors were tested, the results of the NCI bioassay and the Kimbrough study are not all that inconsistent. A x3-tet that the rick of hepatocellular carcinoma at 100 ppm was the sane for Kimbrough study as for the female rats in the NCI study was barely significant st the .05 level. The corresponding HONS 221341 27 Table 9: Incidence* of leur.kniss and hptoee)Ju]ar carcinoma* In rats Iron1. KC2 (1978) Lioassay of Aroclor 1254. Data Is reported as no. animals vith condition/no. animals examined for condition. Dose (ppo) 0 25 50 100 Leukemias Kale 3/24 2/24 5/24 8/24 Female 4/24 6/24 5/24 4/24 Hepatocellular Carcinomas Hals Female 0/24 0/24 0/24 0/24 1/24 0/24 2/24 0/24 MONS 221342 j^-tcst applied to the Kimbrough study and the male rate in the KCI study was not significant (p > 0.3). Thus, because this experiment involved relatively few animals, it is simultaaeously consistent with both no carcinogenic effect as well as an affect approaching that observed in the Kimbrough (1975) study. 3.2.3 Industrial Bio-Test Experiment with Charles River Wats This experiment was described in an unpublished report (Industrial Bio-Test Laboratories, 1971) and a brief susaary of results was presented at the Kational Conference on Polychlorinated Biphenyls (Celandra-, 197 6). The following comments on the experiment are taken from the EPA Criteria Document on PCBs (Kisbet, 1976a). One thousand Charles River strain albino rats vere divided into 10 treatment groups. One hundred rats (50 male and 50 female) served as controls end 100 rats (50 sale and 50 female) were included in each of nine exposed groups which were fed diets containing 1, 10, and 100 ppm of Aroelors 1242, 1254 and 1260 respectively. Dosage started when the animals were about 4-6 weeks old and continued for 24 months. The liver slides from this study have been examined twice by the original pathologist, ones in the original re port (Industrial Bio-Test Laboratories, 1971) and in a later report (Monsanto, 1975). The diagnoses in these two examinations were markedly differenc. For example, in che animals dosed at 100 ppm the earlier examination diagnosed one hepatoma and two animals with nodular hyperplasia, whereas the later examination diagnosed eleven animals with hepatomas and twenty-eight animals with nodular hyperplasia. A summary of the results of the latter pathological examination aa given by Kisbet (1976a) it listed in Table 10. HONS 221343 10: Liver pathology reported in Industrial 3io-Test rat experiment: 1975 re-diagnosis < r<tc (pro) ; of anieal* 1 J Iiruolsr ehsnge I seal hypertrophy Control Aroelor 1242 ____________ 1 10 100 23 -- 31 30 20 1 78 9 0 23 8 <Vular hyperplasia 1 02 8 ;.i?at eras 0 00 2 hittulsr hyperplasia 5 33 3 fnola ngio-hepa t oma 0 00 1 itpatoeellular necrosis 1 110 Aroelor 1254 1____ 10 100 31 28 27 8 10 35 03 00 63 00 31 13 13 13 4 4 2 2 Aroelor 1250 1 10 100 25 23 27 58 3 10 09 00 65 00 42 10 11 7 5 14 2 8 Source: Nisbet (1976) MOMS 221344 30 Tor some reason that was net explained, there was unusually high mortality among the rats in the experiment. The numbers of rats were further reduced by interim sacrifices during the course of the experi ment. Tor example, only 6--21 ar.icale out of the initial 100 in each treatment/sex subgroup fed 100 ppm survived to the terminal sacrifice. Even so, the re-diagnosis of the liver pathology (Table 13) Indicates a Significant tumorigenle effect, The incidence of noduler hyperplasia is significantly elevated in the group fed 10 ppo Aroclor 1260 over the Incidence in the control group (p .005, Tisher1a exact test). There are a total of 9 hepatomas in the groupa fed 100 ppc of one of the three Aroclors, but none in the groups fed lesser concentrations of PCBs. 3.3 Svnercirtic Effects The synergistic effaces of PCBe acting in conjunction with other chemicals may be quite significant in some instances, although quantita tive estimates of these effects st environmental levels of exposure, would be quite tenuous. Ito _t al. (1973) reported on experiments in which groups of 12 30 dd mile miee were exposed to various levels of PCBs (Kaneehlors XC-300, XC-400, KC-500 at 100-300 ppm in the diet) alone or in combina tion with one of three isomers of BHC (hexechlorocyclohexene 6-BHC, p-BUC, y-BMC st 30-250 ppm in the diet). In the groups exposed to PCBs alone, only the group exposed to 500 ppm KC-500 developed hepato cellular eercinomas (5 animals out of 12 expostd). In the groups ex posed to BMC elone, only the group exposed to 250 ppm o-BWC developed hepatocellular carcinomas (S animals out of 30). however, csrcinor.as HOMS 221345 3] wr* induced within 24 ueeV.s by joint feeding of PCBs end BHC at levels below thoae which had no effect when fed independently. The combina tions of 100 ppa c-BHC end 230 pps KC-500 produced 1 hepatocellular carcinoma in 25 animals; 50 ppa c-HC and 250 ppm KC-500 produced 2 hepatocellular carcinomas in 30 animals, 250 ppm each of g-BHC and KC-500 produced 6 hepatocellular carcinomas in 29 initials. The authors concluded that PCS* promoted tha induction of tumors by o-KHC and 6-BHC. Other experiments have produced negative and even antagonistic results. In an experiment with mice (Uchiyaaa, et_ al_., 1974}, PCBs (KC-4C0) fed in the diet did not appear to accelerate the develrpment of cervical carcinoma induced by methyleholanthrene. In another study in rats (I'aViura, 1974}, feeding with PCBs for 24 weeks appeared to inhibit the induction of hepatocellular carcinoma by three other car cinogens, VCi (vinyl chloride) is a known human liver carcinogen and is also a nutagen, but both its carcinogenicity and mutagenicity appear `to depend on metabolism in the liver to an active metabolite (Rannug, _t al., 1974). The mutagenicity of VOI has been demonstrated only in the presence of liver microsomal preparations, including those stimulated by PCBs (Kiebet, 1976a). This strongly suggests that there nay be a synergistic affect between PCBs and VCK in tha production of tumor*. 4. Quantification of Careinoccnic PJsk 4.1 Calculation of Virtually Safa Poses from Different Mathematical Models To illustrate tha computation of virtually safe doses (VSPs) by extrapolation of the result* of animal experiment* to low doaa level*, MONS 221346 32 wt hive selected the three don sets listed In Table 11. To this date m hive applied three dose response nodels and related extrapolation procedures; the probit itodel (Mantel, it il., 1575). ,a+bLogJ0d P(d) - C + (1-0 I (2*)-1/29xp(.y2/2)dy [1) the one-stage or one-hit model (Cruap ii', 1977) P(d) - l-xpf-(qe + qjd)}, - 0, [2) ind the multistage model (Crump, _!., 1977) P(d) - l-xp{-(qo + qjd + q2d}+...+qj.d )), qQ,...,qk 2 0 (3) The gimi multihit model (Rai end VanRysin, 1978) vis ilso considered. This modal requires it least tvo positive experimental doses and could not be applied to the first two data sets. Vhen applied to the third data set, the computer algorithm for computing the estimates failed to converge (Van Rysin, personal communication). For each of these models, P(d) represents the probability of a particular tuaorignic response in a single animal fed a test substance at a dose-rate d during the course of a lifetime feeding experiment. A discussion of these nodels and their theoretical rationale is given in the Appendix. The results of applying these extrapolation procedures to the data sets in Table 11 are rummerited in Table 12. In this table are computed maximum likelihood (best-fit) estimetes of the dose required to produce extre risks over background risk of 10", 10"4, 10*5, respectively. Lover 95Z confidence bounds for each of these doses ere also computed from the various models. Maximum estimates are not recorded for the method baaed upon the probic model sinee this method was not designed to fit date. As explained in the Appendix, the NOMS 221347 33 Table 11: Date seta used for calculating virtually safe doses In Table 12. Data Set l--Kimbrough et il. (1975) rat study vith Aroelor 1260-- Hepatocellular carcinomas Dietary level (ppm) 0. 100 No. of animals 173 184 No. animals with Hepatocellulor carcinomas 1 26 Data Set P2--Kimbrough It il. (1975) rat study with Aroelor 1260-- Liver neoplastic nodules Dietary level (ppm) 0 100 No. of animals 173 164 No. animals with Neoplastic Nodules 0 144 Data Set 3--Industrial Bio-Test rat experiment vith Aroelor 1260-- Liver neoplastic nodules Dietary level (ppm) 0 1 10 100 No. of animals 23 25 23 27 No. animals vith Neoplastic Nodules 1 0 9 7 MOMS 221348 3- Table 12: Virtual! Safa Doses computed Iron data in Table Y Data Sac Analytic Method Maximus livelihood estimates of dose in ppb. corresponding to extra risk of io- io-6 io-s Virtually safe doses (lower 95* confidence bounds lor dose) in ppb. corresponding to extra risk of 10"a IO"* io-s 1 Probit 1.96 14.2 43.6 1 One-Hit and .0069 .666 6. 86 .0051 .511 5.11 Multistage II Probit .025 .180 .552 II One-Hit and .00063 .063 .634 .00055 .0551 .351 Multistage III Probit .178 1.29 3.97 III One-Hit .0046 .465 4.65 .00235 .235 2.35 III Multistage . 0046 .465 4.65 .00205 .205 2.05 HONS 221349 35 parameter b appearing in this modal is set equal to unity regardless of the data. This designation precludes the model from providing an adequate fit to most data sets. Likewise, the gamma multihit model is applied only to data set 111 since this model cannot be applied to data containing only one experimental dose. These models yield quite different values for the lover confidenc bounds on dose. These divergent results are predicted by and, to some extent, explained by the discussion in the Appendix. The probit procedure will typically estimate virtually safe doses larger than those predicted by the multistage models for extra risks below about 10"*. Virtually safe doses based upon the multistage model are identical with chose based upon the one-hit model when there is only one positive experimental dose. This is Illustrated with data sets I and II. The multistage procedure also yields virtually the same values as the one-hit model for data set III since introduction of the parameters qj, qj, etc. into the model cannot improve the fit to this data, however, as explained in the Appendix, the multistage can yield higher values for the VSD than the one-hit model whenever the data exhibit upward curvature and are Inconsistent with the one-hit model. To compart with the VSDs in Table 12, we estimate from Table 2, that the average consumption of PCBs in the United States in 1976 was about 3.3 ug/day t 1357 gra/day 2.1 ppb of PCB, assuming an average dally food intake of 1557 grams. 4.2 Estimates of Risk from PCEs in Diets of U.S. Adults Next we shall estimate levels of risk for the general U.S, popula tion and for the population of consumers of fresh water fish. The MOMS 221350 36 risks from consuming human milk in Infancy present some special dif ficulties and will be dealt with separately. The data of Kimbrough at al. (1975) on the increased incidence of hepatocellular carcinomas in rats fad Arodor 1260 provides the most convincing evidence of carcinogenicity of ?CBs. Our estimates of risks to hirr.an populations will therefore be based upon this data. The fact that only one dosa level of PCBs was used in this experiment means that the one hit is the only one of the models discussed previously that can be utilized. The Mantel at al.(1975) procedure based upon the probit model [I] is an operational procedure for calculating a virtually safe dose. Since the parameter b in the model is not chosen based upon the data, the method should not be used to calculate "best" estimates (e.g., max imum likelihood estimates) or risk at a given dose. Exoept for the one-hit model the other models require at least two positive experi mental doses for their utilization. In Table 15 are exhibited lifetime risks of cancer based upon extrapolating the Kimbrough date from the experimental doses to PCB exposure levels in humans. This process involved two steps. First, exposure levels in huesns were converted to equivalent exposure levels in rats. Results are given for two methods for making this conver sion: 1) on a ppm of PCBs in total diet basis and 2) on a mg intake of PCBs per kg body weight oer day basis. The next step is to cal culate the extra risk of hcpatoeellular carcinoma at this dose based upon a one-hit model fit to the animal data. The risks exhibited in Table 13 are based upon the one-hit model which fits the animal data best. Upper confidence bounds on these risks arc not listed but an upper 952 confidence bound is given by multiplying all corresponding risks in the table by 1.5. MOMS 221351 36 Th* estimates of dose from the total diet study (Table 2) ere tenuous. The estimates for the years 1973 and 1976 are based upon 8 and 3. respectively, trace findings of PCBs in 20 samples from the meat, fish, and poultry food class and no findings at all in samples from ocher food classes. The PCS levels in Table 2 were estimated by assuming that .023 ppm PCS (1/2 of .05 ppm, the lower limit of PCB quantification) were present in all food samples in which trace amounts of PCBs were found (Nlsbat, 1976a). la computing the risk estimates, no allowance was made for the composition of PCBs in food to which humans art txposed being different from the composition of Aroelor 1260 used in the experiment. As in dicated in an earlier section, however, it appears thst the higher clorineted biphenyls predominate in the environment end consequently the composition of PCBs in food should not be too different from Aroelor 1260. Also, no allowance was mads for tha fact that PCB lavals in human adlpost tissua are likely much greater chan corresponding levels in rats for a given level of dietary exposure. Tha risk estimates art computed for a hypothetical human popula tion axposad throughout thair lifetime co current levels of PCBs in food. In actuality, PCB levels appear to be decreasing. Based upon the estimates in Tibia 13 the expected number of extra cages of hepatocellular carcinoma par year resulting from s nationwide exposure at the dietary level detected in the 1976 Total Diet Study (3.3 vg/day) would rangs between 220,000,000/(238,000 * 70) *11.1 cases/ yaar and 220,000,000/(764,000 * 70) - 4.1 casea/yaar. Using the assumptions of s one-hit model, figures on total catch of Lake Michigan fich (Humphrey, 1977) and estimates of PCB levels in HOMS 221352 LjV.o Michigan fish, we can estimate tha total expect number of hepatoeclluiar carcinomas among residents of Michigan from consuming PCBs In Lake Michigan fish. From Table 2 in Humphrey (1977) ve estimate ,that in 1974 Michigan sport fishermen caught .4,065,920 lbs, of lake trout, 6,979,500 lbs. of salmon, and 2,930,220 of other sport fish from Lake Michigan. From Table 7 it is estimated that cooked portions of these fish contained an average of 3.26 ppm, 2.61 ppm, and 1.0 ppm PCBs, respectively. Assuring that 1/3 of the total of weight of these fish were injested by humans, It follows that a total of approximately [4,065,930 * 3.28 + 6,979,500 2.88 + 2,930,220 * 1.16] 10"6 * .333 - 12.28 lbs. of PCBs were injested by humans from eating these fish. Since, according to Table 12 injesting 127 ug/dty of PCI (127 x 10~* * 365 * 70 - 3.2 gn in a Ufa tine) would produce on the average about 1/8,000 tumors, 12.28 lbs. " 5575 gss of PCBs would produce (1/6000) (5575/3.2) .22 hepatocellular csrclnosas/year. 4.3 Estimates of Risk to Breast-Fed Infants The population of infants yhose diet during early life consists wholly or at least in large part of human milk has a much highar ex posure to PCBs through food thsn the genersl populstion. As estimated in Section 2.5, huaen milk in the United States contains an average of 1 ppm (fat basis) PCBs with 72 of tha samples containing more than 2.5 ppm (fat basis) PCBs. Further, it was estimated that there is a population of consumers of fresh water fish whose milk samples could contain on tha average four times tha levels In the general population. Risks from exposures at such an early age are difficult to quantify. Studies on similar exposure patterns In animals have apparently not been done. HONS 221353 There ere e number of tnimal studies in whleh various acute effacta vrc detected with 1-10 ppm exposure to PCBs in food. Linder et al. (1974) exposed rats for two fenerations to 1, 5, 20 and 100 ppa PCBs. Significant Increases in liver weights were observed in 21-d*yold F} male weanlings at the 1 ppm level of Aroelor 1254 and in either rex of Fj and Fj weanlings at 5 ppm or higher levels of both Aroelor 1254 and 1260. It must be kept in mind, however, that because of bioaagnlfieation, concentrations of PCBs in milk may have been much greater than the levels in feed. Rhesus monkeys ere considerably more sensitive to PCBs than rodents: Rhesus monkeys fed 3 ppm FCBs all died after 8 months (MeKulty, 1976). Female Rhesus monkeys given 2.5 end 5.0 ppm Aroelor 1248 in their diets developed facial edema, swollen eyelids, erythema, loss of hair and acne within two months (Barsotti, 1976). Following 6 months of FCB exposure the female monkeys were bred to control males. There was a significant reproduc tive effect of the PCB in the diet as shown in Table 14 . The six Infant monkeys vere permitted to nurse their mothers for 4 months. The mothers' milk contained 0.15-0.40 ppm of PCBs (whole milk basis). Within 2 months focal areas of hyperpigmentation, swollen lips and eyelids, loss of eyelashes, and acne form lealona of the face developed. Within four months, 3 of the 6 Infants died due to PCB intoxication. From the data of Savage (1979) it ia estimated that approximately 32 of human milk samples in the United States have PCB levels in excess of ,15 ppm.(whole milk basis), Any estimates of carcinogenic risk from exposures to PCBs in human milk on the basis of existing animal data must be regarded es extremely tenuous. Wot only does one heve to contend with the moms 22135* T*bl 14 : Modification in reproduction in primate* that vara expoaad to Aroclor 1246 in tha diet Control 2.S ppm 5,0 ppm Total impregnated {no./no. animals) 12/12 6/B 6/8 ftesorptions or abor tions {no./ no. animals) 0/12 3/8 4/8 Stillborn (no./no. animals) 0/12 0/8 1/8 Normal births {no./ no. animals) 12/12 5/8 1/8 Source: Allan and Morback (1976) NOUS 221355 uncertainties of extrapolation from high doses in animals to lov dotes in humans, it is also necessary to make assumptions regarding the effect of a time-varying dosage. The effeet of dosages early in life on carcinogenicity could be much different from equivalent doses later in life. To Illustrate what such an effect night be, we shall calculate some hypothetical risks of hepatocellular carcinoma by relating the eancer incidence observed in the Kimbrough ^ l. (1975) study to human PCS exposure levels in human nilk. The calculation will be based upon the multistage nodel of cancer. Peto (1977b) indicates that this model may be particularly applicable to carcinomas. The awdel is capable of explaining observed high rates of Increase of esneer with age without having to specifically posit an aging mechanism In applying the model v must decide which stages PCSs might effect and the nature of the effeet. Let us assume in the absence of evidence to the contrary, that PCSs affect only the first stage and that the effect is to increase the rate at which calls pass through this stage by *n amount proportional to the quantity of PCSs present in the diet at the tine. This assumption is a "worst case" assumption because with the particular multlstagt modal wa hav in mind, cells pats through tha stages in s prescribed order on their way to becoming cancerous. Consequently, exposures at infancy will have the greatest overall effect if they act upon the first stsge. With such a model, it can be argued that when exposure to PCSe is at a level D from birth to en age which represents an age-fraction f of the normal life span, with no exposure for the remainder of the lifetime, then the extra ilfc-tlmc risk of hepatocellular carcinoma HONS 221356 over risk st tiro dose should bo approxitnatcly of the form P(D,f) - l-pt-SD[l-(l-f)*]) (5} where V. is th numbsr of states in the multistage model. U sstimsts 6 .147 * 10~S fros the Kimbrough st si. (1975) data by tsklng f * 1 and O " 100 ppm. This method of sstimating 0 should tend to under estimate the carcinogenic risk sines the rats in this experiment were exposed only after weaning and prior to 21.5 months of age, end were sacrificed st 25 months of age. With 8 estimated in this fashion ve can now calculate theoretical extra lifetime risks of hepatocellular carcinoma from [5] for different values of 0, f and k. The results of some such calculations are tabulated in Table 14, These estimates arc based upon urverifiable assumptions and should be regarded sa hypothetical. They do illustrate, however, the carcino genic effect that might result from doses of a carcinogen applied early in life, such as PCBs in human milk. Some of the assumptions made in constructing there estimates ware conservative in that other assumptions which are equally reasonable would predict even higher caneer risks. For example, if we had converted dose from rats to humans on a ng/kg/day basis rather than on a ppm basis, the estimated risks would likely be greater. Cells in Infants are Going through mitosis much more often than cells in adults which might imply that they ere at a greater risk per unit time from a carcinogenic insult then cells in adults, tlovcver, this possibility was not reflected in the model. It should perhaps be pointed out that with this model (as well ss possibly In real life), even though carcinogenic dose la vary early MOMS 2213s7 in life, a cancer eventually resulting iron title dose can occur at any age. The risks tabulated in Table'15 are perhaps more useful for com parative purposes rather than in an absolute sense. By comparing these values vith those in Table 13, it appears that the subpopulation of breast-fed infants could be at nruch higher levels of risk of hepato cellular carcinoma from FCSs than the U.S. population in general. HOMS 22135ft 45 Table lit Tneoreiieal extra lifetime risks of hepatocellular carcinoma la breast-fed Infants based upon Kimbrough t_ a^. (1975) rat data. A k-stage modal of canetr Is assumed ulth dose affecting the first stage only (Equation (5}). Conversion of dose from rats to humans is by assuming ppm of PCB in rat diet is equivalent to ppm (fat basis) in human milk. Extra risk to Infant In ganeral population (0 - 1.0 ppa) Age at Veaning K 6 months 12 months (f-1/140) (f-1/70) 1 1/95000 1/46000 3 1/32000 1/16000 5 1/19000 1/9600 Extra risk to offspring of consumers of aore then 24 lbs./ year of Lake Michigan sport fish (D 4.0 ppa) Age at Veanlng 6 months 12 months f-1/140) f*l/70) 1/24,000 1/12,000 l/COOO 1/4000 1/4600 1/2500 MOMS 221359 37 Table 13: Estimates of lifetime extra rial: to humane of hepatocel)ular carcinoma based upon applying a one-hit model (2] to the KleSrough rt al. (1973) rat study. Hunan PCB Dosnee 8.7 vg/day (1973 Total Diet Study, Table 2) 3.3 vg/day (1976 Total Diet Study. Table 2) 127 vg/day (Avg. intake of people consuming more than 26 lbs./yr. Lake Michigan fish. Humphrey (1977) Risks calculated from converting human dose to animal dose on the basis of ppm in diet ec/kt/dav 1/123.000 1/288,000 1/328,000 1/764,000 1/8,000 1/20,000 HOMS 221360 APPENDIX 46 AND EVALUATION OF METHODS OF DETERMINING RISXS FROM CHRONIC LOW-LEVEL CARCINOGENIC INSULT Basie Principles and Considerations To aid in determining a proper regulatory action regarding ;:r.?;en that is present in man's environment, whether it be iidditive, industrial polutant, or otherwise, it is helpshave some knowledge about the number of extra cancers ct likely to be caused by the presence of the carcinogen (environment. It is also helpful to have some knowledge i likely change in number of extra cancers that would seme projected increase or decrease in the level of exposure occuring either as a result of regulatory action :tion. This kind of information is usually impossible to : directly from human data. For this reason it is often :*;y to use data from animal feeding experiments to estimate Usk. This procedure involves two difficult steps, 1) re * the animal risk at high dcses to doses very near to zero '* 'elating the animal risk to risk in humans. typically, animal experiments use on the order of 100 ** *t each experimental dose. If a particular experimental Cluses a lifetime increase in cancer risk of 1/10, this HONS 221361 ?ctse can b measured with a sin* 11 degree of accuracy using ianimals. But if the increased cancer risk is less than 1/100 "is increase will often not even be detectable by an animal ':ting experiment. For example, if the true risk is 1/100 it ,,-jld require that over 400 animals be tested at that dose In :ier to be 99% sure of detecting any carcinogenic response at (i.e., for there to be a probability of .99 that at least one ^inal gets caneer). If background or spontaneous carcinogenesis j present even larger numbers of animals will be required. On -t other hand, the extra human risk that we may want to estiMtc resulting from environmental exposure is usually (and hope?:lly) much smaller than 1/100 for any given chemical, perhaps s the order of 1/1,000,000. It is clear that it would not be nctical to conduct an experiment with enough animals to measure -irtctly an increase in risk this small. For these reasons the procedure has been developed of ''ducting lifetime animal feeding experiments using, in addition a control dose of zero, several doses at which the projected cancer risk may be 1/10 or larger. This high dose data is *--tn used to estimate the extra risk at a dose where the extra `I* may be no larger than, say, 1/1,000,000. An equally import*** variant to this problem is the calculation of the so-called f**' dose, that is, a dose for which there is some measure of :**tistical assurance that the extra risk at that dose is no <r* than, say, 1/1,000,000. These problems are often referred 4 cllectively in the literature as the "low dose extrapolation HOMS 221362 Imoortance of the Mathematical Modal Performing a low-dose extrapolation involves the choice = a mathematical function to medal the dose-carcinogenic respo.-.s* relationship and the choice of statistical procedures to apply the mathematical function. The choice for this mathematical function turns out to be extremely crucial to the outcome of lev dose risk estimation. Zf the assumed relationship between turrr occurrence and dose does not apply in the regions to which the extrapolation is being made, a serious overestimate of the 'sa! dose may result (Mantel and Bryan, 1961, p. 458). Chand and Kc (1974) compared five standard dose response models and observe! that they could differ by many orders of magnitude at low dcse levels for which extra risks are on the order of 1/100,000,0CC. It might be supposed that it should be possible to dis criminate among the various potential dose response functions :: the basis of experimental data but, unfortunately, two differer.* dose response functions can often fit experimental data equally well but still differ by several orders of magnitude at very I-doses. Moreover, even if a particular dose response function were to give a significantly better fit to data than several others this would still not furnish assurance that this functi would necessarily correlate in any way with the true dose response at very low doses where it is not feasible to measure the true extra risk disectly. As a consequence of the great disparity of dose response functions at low doses it is impera tive that the dose response function be selected, neither arbi trarily nor solely on the basis of hew well it can be made to MOMS 221363 txperimental data, but, insofar as is possible, it should reflect known or at least plausible information regarding the biological ptchanisms through which a chemical induces or promotes cancer. What Shape Should be Expected for the Dose Response Curve at Low Doses? Tumors of so many different types arise in such a diver* city of different tissues, their etiology is so little understood, and the agents that cause tumors affect a subject in such diverse ways, that it might seem that no general conclusions can be drawn. However, for a certain broad class of "directly-acting" chemical carcinogens the range of uncertainty associated with the shape of the dose response curve at low doses can be greatly narrowed. As used in this paper, the term "directly-acting carcinogen" encom passes (Guess, Crump and Peto, 1977) carcinogenic agents for which either the agent itself or a metabolite acts directly at the cellular level and produces a hereditable change which eventually leads to the formation of a tumor. Carcinogens which are carcino genic by reason of their mutagenicity should fall into the cate gory of "directly-acting carcinogens." Accordingly, carcinogens which test positively using the Ames mutagenic! ty screening test for carcinogenicity are very likely to be directly-acting. (See HcCann and Ames, 1976). In a recent study (McCann, Choi, Yamasaki and Ames, 1975) in which about 300 carcinogens and non carcinogens were tested using the Ames Test, 90\ (157 out of 175) f the carcinogens were mutagenic including almost all of the MOHS 221364 50 known human carcinogens. This indicates that the class of directly-acting carcinogens may encompass most of the known carcinogens. A partial solution to the low-dose extrapolation problem for the case of directly-acting chemical carcinogens has been given in Peto (1977), Crump, fioel, Langley and Peto (1976), and Guess, et al. (1976). The key result is that, at least as long as background carcinogenesis is present, we should expect the dose response curve not to be absolutely flat at zero dose, What this means is simply that when risk is plotted against dose response on ordinazy linear scales, the tangent line to the dose response curve at zero dose should have a positive slope. When a dose re sponse function has this property we will say it is linear at lev- dose. This simple property can have far-reaching consequences on low-dose extrapolation. For example, consider the two potential dose response functions 1) d + JJJ d } and 2} .Id4 for the dose interval 0 ^ d ^ 3. Both of the curves give a risk of 1/10 at a dose of d 1 and are practically indistinguishable at higher doses. However, at a dose of d 1/1,000 1) predicts a risk of 1/100,000 and 2} predicts a risk of 1/10,000,000, a dif ference of two orders of magnitude. We note also that 1} has a tangent line with a positive slope at d * 0 whereas 2) does not. One explanation of why the dose response function should ks linear at low dose when background is present may be found in Crump, et al. (1976) and Peto (1977) and will be briefly outlined here. When background carcinogenesis is present, the cellular mechanism through which the test agent produces cancer should HONS 221365 Si already be operative in producing background tumors. When this is true the effect of the test agent is to add to an already engoing process. The result of this additive effect is illustrated in rigure 1. The dose response curve is for all tumors produced through the mechanism through which the test agent acts. Back ground carcinogenesis is allowed for in the Figure by an effec tive background dose dQ. Ve see that/ in this case/ the added risk caused by a dose d of the test agent should be expected to increase approximately linearly near d * 0 (i.e., the tangent line at d - 0 will have a positive slope), Implicitly assumed by the way Figure 1 is drawn is the fact that an added dose of a carcinogen acting through this mechanism does not produce a smaller risk. If background carcinogenesis is allowed for as in Figure 1 by positing an effective background dose dQ which is estimated from the data then the wide range of risks obtained using different models effectively disappears (Peto, 1977). We note that the existence of a tangent line with a positive slope at zero dose does not/ in itself, imply any lower bound for extra risk at low doses since the slope of the tangent line could possibly be very small. The evidence given above for a positive slope to the dose response function at zero dose applies particularly to the case In which background carcinogenesis is operative. This does not inply that we expect the dose response curve not to be linear at low dose in the absence of background carcinogenesis. Tor exam ple, the multistage models of cancer (Armitage and Doll, 1961} *rt a fairly broad elass of models in which it is assumed that MONS 221366 Dose rigure 1. Illustration of why the doss response curve should be linear at low dose in the presence of background ca einogenesis. d is the effective background dose and d is t.w. dose of the carcinogen of interest. MONS 221367 53 a number of events are required to occur at the cellular level to initiate cancer. Although all models in this class are linear at low dose provided background carcinogenesis is present, a sizable subclass of them are linear at low dose in the absence of back ground carcinogenesis {Crump, et al., 1376). Watson (1377) has recently proposed a more specialized model for cancer induction and promotion based upon reversible epigenetic cellular changes. Watson concludes that his model supports the low-dose-linearity hypothesis and states "as sug gested by a different argument of Crump, et el. . . . , it is reasonable to essess the risk due to an additional carcinogen at low constant dosage by a lintar relation. " The evidence for low-dose-linearity given above applies mainly to directly-acting carcinogens. An indirectly acting car cinogen might be one which causes some gross physiological chance such as suppression of ovulation which could predispose the sub ject to cancer. For such careincgens tht shape of the dose re sponse curve at low dose is highly speculative. There could possibly be a threshold dose below which the agent has no car cinogenic effect at all on an individual. However even if a threshold mechanism is operative, there is likely to be con siderable variation in individual thresholds in a large popula tion. Consequently the dose response curve for the entire population could still exhibit a linear trend at risks as low as 1/1,000,000 or lower. The effects of metabolic activation and detoxification upon carcinogenic dose response have been recently considered by MOMS 221368 34 Cornfield (1977) through a kinetic modal that encompasses free toxic substance, metabolite, deactivator, and the interactions of these substances. Only a steady state situation is studied in that variation over tine of the concentrations of these agents is not considered. The model predicts e threshold dose below which there is no carcinogenic risk under the assumption that the de activator is 100% efficient in deactivating the carcinogen. How ever, in a naturally occurring process it is likaly that deacti vation would not be perfect and would be less than 100% effective in always combining with 100% of the carcinogen before an amount of the active metabolite reaches a cancer target site. Any of a number of modifications to the model to allow for non-perfect deactivation would rule out a threshold and would lead directly to a model for which carcinogenic response varies linearly with dose at low doses. Cornfield's own modification of perfect de activation, that of allowing the deactivating reaction to be reversible, leads, as Cornfield points out, to a model which is linear at low dose. This occurs regardless of how slowly the reverse reaction takes place, as long as the possibility is not eliminated entirely. Furthermore, even in the extremely unlikely case of perfect deactivation, an otherwise realistic model should still imply low-dose-linearity since the theoretical time re quired for perfect deactivation would not be zero and would likely be infinite. For most, perhaps all, carcinogens, the mechanisms through which cancer is produced is not sufficiently understood so that the shape of the carcinogenic response curve can be theoretically MOMS 221369 55 predicted with certainty. As pointed out earlier, neither can experiments of sufficient sire be conducted that would permit direct experimental investigation of the dose response curve at low dose, we have noted that there are plausible arguments that the dose response curve is linear at low dose for many carcino gens. On the other hand, this author Knows of no serious pro posal of a mechanism that would lead to a more conservative dose response relationship such as the risk varying approximately as the square root of dose at low dose. In view of these uncertain ties it would seem reasonable to base estimates of added risk of cancer upon a mathematical model that encompasses low-doselinearity Unless, of course, the mechanism through which the carcinogen operates is sufficiently understood so that low-doselinearity can be conclusively ruled out. Once the principle of low-dose-linearity is accepted the problem of estimation of risks at low dose is nearly solved. This is because the dis agreement between the upper statistical confidence bounds on risk at low doses based upon a model that incorporates low-doselinearity and one that does not is typically several orders of Btgnitude whereas the corresponding disagreement between two reasonable models both of which incorporate lcw-dose-linearity U usually much less than this. MOMS 221370 *x Specific Methods for Low Dose Risk Estimation A. Manttl-aryan The Mantel-Bryan procedure (Mantel and Bryan, 1961 and Mantel, Bohidar, Brown, Ciminera and Tukey, 1975) nas been elected by the Food and Drug Administration as part of their procedures for evaluating assays for carcinogenic residues in edible products of animals. Before the Mantel-Bryan procedure HONS 221371 57 itself I* discussed, the "Sensitivity of Method" (SOM) document (federal Register, vol. 42, No. 35 - Tuesday, Feb. 22, 1977, p. 1041?) will be reviewed briefly so that the role of the Mantel-Bryan procedure in official FDA policy can be understood and appreciated. The so-called Delaney clause of Section 409 of the Federa Food, Drug and Cosmetic Act of 1958 prohibited the approval of any food additive that "is found to induce cancer when ingested by man or animal, or if it is found, after tests which are ap propriate for the evaluation of safety of food additivies, to induee cancer in man or animal .... This language was inter preted as forbidding. FDA to approve the use of a carcinogenic substance as a feed additive for food-producing animals whether of not the compounds might leave any residues in the edible pro ducts of the animals. However, in 1962 Congress modified this prohibition as part of the Drug Amendments of 1962. Section 409 (c)(3)(A) now reads in part; . . . [N]o additive shall be deemed to be safe if it is found to induce cancer when ingested by man or animal, or if it is found, after tests which are appropriate for the evaluation of the safety of food additivies, to in duce cancer in man or animal, except that this proviso shall not apply with respect to the use of a substance as an ingredient of feed for animals which are raised for food production, if the Secretary finds (i) thatT uncer the cbncitions of use end feeding specified in pro posed labeling and reasonably certain to be followed in practice, such additive will not adversely affect the animals for which such feed is intended, and (ii) that no residue of the additive will be found (by methods of examination prescribed or approved by the Secretary by regulations, which regulations shall not be subject to subsections (f) and (g)) in any edible portion of such animal after slaughter or in any food yielded by or derived from the living animal ... ' WHS 221372 it Originally, tha FDA applied this ao-callad "DCS proviso" on a casa-by-casa basis without publishad critaria. Howavar tha FDA on Fabruary 22, 1977 publishad tha "Sansitivity of Method" doeumant which established objective critaria for aecaptanea of in assay mathod for datacting rasiduas of a carcinoganic faad idditiva and for astablishing post-administration withdrawal per iods for thasa additivas. Thus tha doeumant furnisbod an opera:ional dafinition of tha no-rasidua requirement of tha DCS iroviso. This oparational dafinition involvas tha application if tha Mantel-Bryan procadura to animal axparimantal dosa ra dons a data. Tha tast-animal spacias/strains to ba utilirad ara > hava "tha graatast possibla suscaptibility to tha tast comvr.d" whila at tha sama tiraa baing "appropriata medals for man." tdaral Register, vol. 42, 1977, p. 10418). Tha Mentel'/an procadura is applied to tha animal tast data to determine residue level, designated SQ, corresponding to a risk of ',000,000 in tast animals (i.e., SQ is tha "safe" dosa level mined by tha Mantel-Bryan procedure). Tha level SQ is messed as a fraction of tha total diet (i.e., parts par lion). This level is adjusted to account for tha respective Portions of tha human diet that is represented by tha various ! products containing rasiduas of tha carcinogen baing tasted, transfer from animals to man baing made on a fraction of '1 diet basis. Tha resulting dosa level s_ is "the level of 1 residues of carcinoganic concern that can ba operationally 'ed as satisfying the no-rasidua requirement of tha act for fie tissues" (Federal Register, Vol. 42, 1977, - MMS 221373 39 The dose level Sn thus represents the upper bound to the lowest limit of reliable measurement that an approved assay method must satisfy. The Mantel-Bryan procedure is thus an integral part of the "Sensitivity of Method" document. The procedure as originally set forth {Mantel and Bryan, 1961) and "improved" (Mantel, et al, 1975} is for the purpose of conservatively choosing a "safe" dose of a carcinogen, a "safe" dose being defined as one for which it can be expected that, with a given level of statistical assur ance (e.g., 99*), the true dose producing a preassigned "safe" level (e.g., 1/1,000,000) of risk will lie above the "safe" dose. In the Mantel-Bryen procedure the mathematical model used for the dose response model is the probit function P(d) - C (1-C) ^*+bLog,d(2jrJ-l/2 #xp {-x2/2} dx [Al). where d represents the dose of the carcinogen and ?(d) represents the probability of a cancerous response in an animal subjected to a dose d. The parameters in the model are an intercept parameter a, a probit slope parameter b, and C, which represents the probability of a response in untreated animals. The parameter b is not estimated from the data but rather is arbitrarily set equal to 1. This choice is stated as being conservative (Mantel and Bryan, 1961), the argument for this being that typical dcse data exhibit a probit slope in the experimentally observable region above 1* incidence that is greater than one. In Mantel and Schneiderman (1975) it was observed that a set of DSS data (Gass, G. H., Coats, D., Graham, N., l9, C3H females) exhibited MOMS 221374 60 a probit slop* of 1/2, but th* general us* of a probit slope of bl was still suggested. A slop* of b-1 was adopted in the SOM document unless the experimental date exhibit a smaller slope, in which case the smaller value would be used. With the probit slop* parameter fixed at bl th* remaining parameters, a and C, are estimated from the data and then ad justed so as to produce a higher level of risk as a given dose that corresponds to a upper 99% statistical limit on the true risk at a given dose. The value 99% is the on* used in the SOM document but the choice is arbitrary. The safe dose is then de termined to be the one producing a given low risk (e.g., 1/1,000,000} based upon th* adjusted values of a and C. The risk value of 1/1,000,000 was also adopted in the SOM document. As pointed out in Mantel, et al. (1975), the Mantel-3rvan procedure rewards larger and better experiments in that the more evidence there is of safety, the higher the calculated safe dose will b*. However this advantage should be shared by any extra polation method that uses reasonable statistical procedures. Some have considered the Mantel-3ryan procedure as adopted in the SOM document to be too conservative (Federal Register, Vol. 42, 1977, p. 10419) in that it involves three conservative choices (99% statistical assurance, lifetime risk of 1/1,000,000, and probit slope set equal to 1) and that any one of these assumptions alone could provide adequate protection to the public. The first two of these choices are regulatory decisions that would have to be made with any extrapolation procedure. However, the trhitrary selection for the slope parameter seems to be peculiar NONS 221375 / 61 to the Mantel-Bryan procedure. To investigate its offset upon the extrapolation procedure e typical fit is presented in Figure 2 (from Crump, 1977b) of the Mantel-3rvan probit curve fAl) to experimental carcinogenicity data when the probit slope para meter is fixed at b - 1. As can be readily seen the probit curve typically provides a very poor fit, curving downward even when the ..t-i of the data is toward an increasingly upward curvature. This typically bad fit to data of the probit curve raises serious questions regarding the validity of statistical procedures asso ciated with the Mantel-3ryen method. (See Salsburg, 1977 and Crump, 1977b.) On the other hand, the Mantel-Bryan procedure utilizing the choice b 1 was put forth as conservative procedure and it gives that appearance in Figure 1 since the probit curve appears to lie far above the trend, of the data at the lowest doses. How ever as mentioned earlier and also pointed out by Mantel (Mantel and Bryan, 1961, p. 458} e procedure may, while appearing con servative at experimental dose levels, at the same tin* seriously overestimate the "safe" dose (i.e., be seriously anticonservative) if the assumed dose response relationship does not apply at the low risk levels to which extrapolation is being made. Thus, before the degree of conservation can be evaluated for any pro cedure, the properties of the dose response curve at very low doses must be evaluated. As described earlier, there are streng arguments that indi cate the dose response curve should be 'linear at low dose* (see p.50 ior the definition of this concept) particularly for HOMS 221376 62 o indicates data point (no. responders/no. animals tested) .6 - Mantel Bryan maximum likelihood curve a> o o Mantel Bryan u "safe" dose curve o -O O 3 aE. 0 2 4 6 Dose Fi?1-- 2. TVpical fit of y*anttl-2ryan curve to trcerinental data (frcra Crjrp, 197Tb). ' MOMS 22137? 6: directly acting carcinogens in the presence of background carcino genesis. This has led Peto (1974) to recommend extrapolation pro cedures using only dose response functions fren a class containing only dose response functions which are linear at low dose. At the very least# however# it would seen prudent not to go to the opposite extreme and use a dose response function which rules out linearity at low dose by assumption. However, the K#ntel-Bryan procedure, through its use of the probit curve (All rules out linearity at low dose in favor of a "flatness property" (see Hartley and Sielken, 1977, Mantel, 1977, and Crump, 1977b) at low dose which is anticonservative to the extreme. This property im plies that mathematical derivatives of all orders of the probit curve approach zero (through positive values) as the dose approaches zero. This unusual property is most often discussed within the context of mathematical oddities rather than in con nection with a scientific investigation. It implies that if the true dose response curve comes from an extremely broad class of functions known as analytic functions and which pervade scienti fic applications of mathematics, then the probit curve will eventually underestimate the true risk at low doses. Further more, at low enough doses, the degree to which the risk will be underestimated will be arbitrarily large (i.e., the ratio of the true risk to the probit estimate will grew arbitrarily large). It was emphasized earlier that it is important when, extra polating to low doses for the assumed dose response function to incorporate known or at least plausible facts about the mecha nisms of carcinogenesis, in neither the original paper MOMS 22137# 6 (Mantel and Bryan, 1961) nor in the paper outlining the improved version is biological justification given for the selection of a curve having the above described "flatness property." It should be mentioned at this point that the incorporation of background carcinogenesis into the Mantel-3ryan probit model [Al] using the parameter C implies that the mechanism through which the test carcinogen produces cancer is independent c the mechanisms through which all of the background cancers are produced (Crump, et al, 1976). In keeping with the discussion in the last section it would seem z raoror: proper to incorporate background into the probit model by positing an effective background dose dQ which adds to the dose d of the test carcinogen. If background is incorporated in this way the probit curve no longer has the "flatness property" and becomes linear at low dose (Guess, et al, 1977). In fact, with background incorporated in this way, the probit curve assumes a shape similar to the one-hit model, sometimes referred to as the most conservative of all procedures (e.g., Mantel, 1977). Sven though the "flatness property" implies the probit curve should at suitably lew doses be anticonservative to the extreme, the "flatness" property holds only for doses approaching ter and the feature of arbitrarily fixing the prebit slcpe at 1 mitigates the anticonservativeness implied by the "flatness" i property at any given low dose (although the property itself' will hold for all choices of the parameters a, b, c). The cancer risks that are typically extrapolated to are in the risk ranees 1/10,000 to 1/100,000,000. We will examine the outcome of MONS 221379 65 Mantel-Bryan extrapolations to these risk levels in a later sec tion when we compare several different extrapolation procedures. B- Linear extrapolation ' The technique for linear extrapolation was recommended by Hoel, Caylor, Xirschstein, Safiotti and Schneidexman (1975) for use on an interim basis until better procedures could be devel oped. The procedure is straightforward and is based upon an assumed linear relationship between dose and response at low dose. The procedure utilises only the data for the group of control ani mals and a single other dose group, usually either the highest dcse that elicits no response or else the lowest dose that elicits some response. In the case there are no cancers in the control animals the 'safe* dose, based upon a maximum risk of 1/1,000,000 and 99% statistical assurance, is calculated as follows: An upper 99% confidence bound is calculated for the cancer risk in the dose group of animals. From this risk and dose one extrapolates back toward zero dose and zero risk using a straight line relationship. The dose corresponding to a risk of 1/1,000,000 on this straight line is taken to be the "safe" dose. If there are cancers in the control animals this procedure is modified to allow for the sta tistical treatment of the response in the control group while re taining the straight line relationship. When data at other experimental doses are available this method of linear extrapola tion has the obvious shortcoming of not fully utilizing the available data. A linear dose response curve is linear at low dose but the converse is not necessarily true. A curve can be linear at HOMS 221360 66 low dcse and still have a high degree of nonlinearity at higher doses. Linear extrapolation is viewed by some es e very conserva tive procedure. For example, comments were made during the decision on which extrapolation procedure to incorporate into the SOM document to the effect that linear extrapolation is the most conservative of all procedures. Crump, et al (1976) examined the extent of the conservatism of a linear dose response function when compared with a multistage dose response model (Armitege and Doll, 1961). The multistage model assumes that a cell must go through a number of different stages before cancer is initiated in that cell and the model can encompass a high degree of nonlinearity. It was determined that the maximum pos sible degree of conservatism of a linear model relative to a multistage model depended rather heavily upon the incidence at the experimental dose relative to the background incidence. (This is consistent with the general relationship between back ground carcinogenesis and linearity at low dose as discussed sarlier.) For example, when the incidence at the experimental dose is four times the incidence at zero dose the extra incidence at low doses derived from the linear dose response differs from the incidence derived frem the multistage model by, at most, less than a factor of 2.S regardless of the number of stages in the multistage process. Thus, when background carcinogenesis is present, the linear dose response curve is not overly conserva tive relative to the multistage dose response curve, in fact the linear dose response curve is anticonservativo when compared to HONS 221311 67 the one-stage or one-hit models. Linear extrapolation has long been proposed for use in radiation carcinogenesis (see Brown, J. M., 1976, for a review of the relevant reports). The BIZH (1972) report on radiation risks from the National Academy of Science recommended linear extra polation as a "best estimate" approach as opposed to a conserva tive approach. Brown reviewed arguments both for and against linearity and concluded that "linear extrapolation of* human data from high doses of low LI? radiation cannot be said to overesti mate the risk at low doses. In faot, there is some doubt as to whether the risk is not underestimated." Certainly much remains to be learned about both radiation and chemical carcinogenesis. However, if both radiation and chemicals cause cancer through similar mechanisms then it should be expected that there would also be similarities between the respective carcinogenesis dcse response functions. Direct damage to DNA by the carcinogenic agent has been implicated as one cancer . initiating mechanism for both radiation and chemicals (Brovn, J. M. 1976 and McCann and Ames, 1976). Thus, the findings related to the potential linearity of the dose response function for radia tion has implications for chemical carcinogens as well, particu larly for "directly-acting" carcinogens. The results of linear extrapolation will be compared with other methods in a later section. HONS 221382 it C. Extrapolation Methods Based tlaon the .yultlstace Modal Two methods of low dose extrapolation which are alterna tives to the Mantel-Bryan or linear procedures have recently been proposed independently by Cues** Crump, and Deal (Guess and Crump, 1976, 1978 and Crump, Guess and Deal, 1977) and Hartley and Sielkin, 1977. Both of these methods utilize a multistage dose response function of the form P(d) l-exp{-(q0 + q^d + q2d2 + + qkdk) ) If- (A*: where qQ, q^, **, q^ are all nonnegative parameters to be esti mated from the data. This dose response function is general enough to yield a considerably wide range of responses at low dose. On the one hand, if q^>o and q^o for i>2 the dose response function (A2J becomes the one-stage model which yields risks at lew doses comparable to what would be obtained with linear extrapola tion. On the other hand, the model can produce risks even as low as the probit curve (All down to any fixed positive low dose. Thus this model is capable of fitting both highly linear and highly nonlinear dose response relations. Since the model has the property of "linear at low dose" if q^>o and does not have this property if q^o, use of this model does away with having to make the arbitrary but crucial decision of having to either assume linear at low dose as in linear extrapolation or else assume a highly nonlinear dose response relationship t low dose as in Mantel-Bryan extrapolation. Thus "safe" doses computed using this model should provide a more realistic measure of the true un certainty of low dose extrapolation than would "safe" doses basod HOMS 221383 69 on either an assumed linear curve shape or an assumed highly nonlinear curve shape. The dose response relation fA2] contains all of the Armitage and Dell (IS61) multistage dose response models as special cases but also contains curves which are much flatter at low dose than any of the multistage curves. The two extrapolation procedures based upon GtiJ have some features which are different. When computing *oet likely" esti mates the procedure of Guess, Crump and Deal uses an infinite dimensional maximization procedure so that it is not necessary to specify a value of k, the degree of the polynomial in [A2]. How ever, the two methods differ chiefly in the way the statistical conficence intervals are computed. There have not yet been suf ficient comparative calculations made to determine how safe doses may differ using the two approaches. Mantel (1977) has made a critical review of the statistical procedure used by Hartley and SielJcin for calculating the 'safe* dose. Both the Hartley and Sielkin and the Guess, Crump and Deal (as extended by Crump, 1977) procedures can utilize times at which cancer is detected in the experimental animals rather than just the dichotomous information of whether or not an animal contracted cancer before it died of some other cause or before the termination of the experiment. The utilization of time data In low dose extrapolation is important for at laast two reasons: 1) The age at which cancers occur should be important in assessing the magnitude of the harmful effect of a carcinogen upon man It.g., cancers that occur early in life should be viewed as more MONST ^213*4 70 serious than those which occur in extreme old age). 2) In many animal carcinogenicity experiments the response data at the highest doses lies below the trend of the lower dose data. This sometimes appears to be due to the fact that at the highest doses some of the animals are being poisoned by the chemical before they have a chance to develop cancer. When this occurs the high dose data is often just deleted from the analysis. Jtowever if the times at which the animals die is properly used the high dose data might not appear anomolous. More research needs to be dene on the proper utilization of animal time of death data to assess the harmful effects of chemicals to man. D. The Garena*Multlhlt Carcinogenesis Pose Response Model Rai and Van Ryzin (1978) have proposed basing risk estima tion on the gamma multihit model P(d> - C + (l-C) jid " du k20 - C + U-C){i-e'*dJl + Id + + ... + * 1,2#.-. IA3J where ?(d) is the lifetime probability of cancer in a tissue when subjected to a constant dose-rate d of the carcinogen. This model is obtained by assuming that cancer due to the carcinogen occurs when the tissue has baen hit k times, these hits occurring randomly according to a Poisson distribution. The manner in which background carcinogenesis is incorporated into the model is MOMS 221995 71 equivalent to assuming that the event "cancer occurs due to the action of the carcinogen" is independent of the event "cancer occurs spontaneously." This assumption would not apply, for example, to processes in which the effect of the carcinogen is to speed up the rate at which the "spontaneous" events occur which lead to the background cancers. At low dose rates, the response is approximately given by Consequently, this dose response model is linear at low dose rates when and only when k 1. Rai and Van Ryzin calculate confidence intervals for added risk at a given dose and for the dose producing a fixed added risk using asymptotic maximum likelihood theory. Although In the theoretical development k must be an integer, in the applications k is allowed to assume any positive value. Upper statistical confidence intervals on extra risk at a given low dose computed using this procedure can be compared with those computed from the multistage model [A2} by considering two general classes of data. If the data exhibit a general downward curvature as illustrated in Figure la, upper confidence bounds on extra risk computed from the gemma multihit model [A3] should generally be greeter than or equal to corresponding upper confidence bounds computed from the multistage model [A2). In certain instances the gamma multihit upper bounds could be much greeter than the corresponding multi* stage upper bounds. This could occur when the data is consistent HONS 22136 7: with Jc * 1 in the gamma multihit modal (prasunably correspond; to a fraction of a hit}. On tha othar hand, if the data axhibit a general upward curvature ae illustrated in Figure 3b the reverse situation vi. hold; gamma multihit upper confidence bounds on extra risk wii: generally be less than or equal to corresponding multistage bounds. Gamma multihit bounds will not in general share the low-dose-linearity of multistage bounds and because of this may be smaller than the multistage bounds by orders of magnitude a low doses. The reason these large differences can occur is as fdllows. The multistage family of models contains members whir, are simultaneously linear at low dose and exhibit upward curvet', at moderate doses. For example, the particular multistage model qi,q*.> 0 is linear at low dose since qx > 0 and still can exhibit upward curvature at moderate doses since qa > 0. This means there are dose response curves in the multistage class which are both linear at low dose and can adequately describe data of the type exemplified by Figure 3b. On the other hand, this will generally not be true of gamma multihit models. All dose-respcr.s curves in the gamma multihit class which are linear at lew dose must exhibit downward curvature and conseeuently would generally not be consistent with the data in Figure 3b. when this is true gamma multihit confidence upper bounds will be sublinear (e.g., quadratic) at low dose and consequently much smaller then the multistage confidence bounds. MOHS 22138? I responses 73 * * * i T>b gur* -4*. Do** Exanpl* of Cat* Exhibiting Upward Curvatur* JJ t responses X X '. DOS* Example of Data Exhibiting Downward Curvacur* * MONS 221388 74 Confidence bounds based upon the gamma multihit model will be approximately correct whenever this model is the correct medt A similar statement could be made for the probit model, multi stage model, or any other modal to which reasonable statistical methods are applied. However, there stay be considerable uncer tainty as to what the true model may be in a particular situatic The multistage model not only reflects some reesonable assunptic regarding the carcinogenic process which dovetail nicely with epidemiological data for many cancers (Pete, 1977b), but it also reflects some of the uncertainty with regard to the true model by virtue of encompassing a relatively large class of doseresponse functions. For example, as noted earlier, the multista class contains dose response functions which are linear at low dose and also exhibit upward curvature at moderate doses. On the other hand, the gamma multihit class is more restrictive at this point in that it does not permit such behaviour. Is this extra restrictiveness of the gamma multihit model justified? To help answer this question, consider the following modifica tion to this model. Suppose that the hits (phenomenological events which are required to oeeur in a tissue in order that a cancer appear) can possibly oecur spontaneously in the absense of the carcinogen. Part of the effect of the carcinogen would then be to speed up the races at which the spontaneous hits are occurring. For example, if one of the "hits" is an in correct base substitution in DNA during mitosis, the carcinogen could speed up the rate at which these "hits" are occurring in HONS 221389 73 *n ineorfect substitution, with this modificstion to the model, the gamma multihit upper confidence bounds will no longer be sublineer et low dose end will likely be very close to corre sponding bounds calculated front the multistage model [A2J. On the other hand, versions of the gamma model which incorporate such an assumption as this could still be consistent with data with strong upward curvature as exemplified by Figure 3b. Thus, in order to obtain sublinear upper confidence intervals with the gamma multihit model a modification such as the one described above must be ruled out, not on the basis of data, but by assumption. By was of summary, confidence intervals based upon the multi stage model will always be linear at low dose. Confidence inter vals based upon the gamma multihit model may ba aithar "super linear" (corresponding to k * 1) or "sublineer" (corresponding to k > 1). Superlinearity is achieved by making the model too broad in that a fraction of a hit is allowed which has no bio logical basis. On the other hand, sublinearity is achieved by making the model possibly too restrictive in that models which ere reasonable from e biological viewpoint are ruled out by assumption. Comparisons and Discussion To compare low dose extrapolations using the Mentel-arvan pro'oit model (All with those using the multistage model {A2} we present figure < based upon the same data as figure 2. Zn this MOMS 221390 Dose 76 Figure 4. Comparisons of 'safe* doses computed from the MantelBryan procedure and from a procedure based*upon the multistage model (from Crump, 1977b}. MONS 221391 77 Figure are plotted on a log-log scale the Mantel-Sryan 'safe' dose as well as both the multistage 'most likely' curve based upon li.2) and the 'safe' dose based upon U2j and computed as outlined in Crump, et al. (1977). A 99% statistical assurance was used for both 'safe' dose curves, we note that the Mantel-3ryan 'safe' dose lies above the multistage safe dose curve at values of added risk below 5 x lo"4. The Mantel-Bryan "safe" dose curve lies above the multistage 'safe' dose curve by a factor greater than 20 for an added risk of 10"* and by a factor greater than 300 for an added risk of 10 Because of the 'flatness' property of the Mantel-Bryan probit function 1*1) described earlier, the Mantel-Bryan 'safe* dose curve will lie above the multistage safe dose curve by arbitrarily large factors at extreme low doses. Guess, et al. (1977) have compared the Mantel-Bryan 'safe* dose curves to the multistage 'safe' dose curves and found this to be a typical situation. Thus it is clear that if the true dose re sponse curve could be similar to the multistage dose response function hi) then the Mantel-Bryan procedure could not be Justi fiably called conservative. (See also Crump 1977 for further discussion of this point.) On the other hand, we have seen that there are quite plausible arguments for the true dose response curve to have the same shape at low dose (linear) as the estimated multistage eurve. we note that both the multistage 'safe' dose curve and `most likely' curve have a slope 1 in Figure a as plotted on the log-log scales which is equivalent to the curves being linear at low dose. The fact that the 'most likely' curve has slope 1 is MOMS 221392 due to the fact that with this particular data sat tha line*? coefficient in (A2] was astimatad to ba positiva. However t; safe* doss eurva using tha multistaga model (A22 will always hi linear at low dose regardless of whether or not the linear eeef ficient is estimated to be positive. This property should 1 shared by any valid statistical procedure based upon a dose re sponse function that does not rule out linearity at low dose by assumption as Mantel-Bryan does. Just as it is not possible t? prove statistically the existence of a threshold, it is lihewis not* possible to rule out the possibility that the true dose response curve is linear at low dose on the basis of statistics analysis. (See Guess, et al. (1977) for a thorough discussion this important point.) The Mantel-Bryan obtains 'safe' dose estimates which are considerably higher 'than those obtained usi the multistage model because it assumes away linearity at low dose, an assumption that we have seen is probably unwarranted i that majority of carcinogens which are classified as 'directlyacting1 carcinogens. Since extrapolation based upon a model such as the multi stage model t2J must always be linear at low dose, the question arises as to how different the result will be from simple lines extrapolation. Tor some data the difference will be minimal. For example, for the data upon which Figure 2 is based, 'safe doses' computed using the multistage model is almost identical with 'safe' dosas based upon linear extrapolation. For some data sets, however, the difference could be considerable. Tor example, with the Cass, et al. (196-SJ DES usinr? th~ rvj fr-- HONS 221393 79 mice, the 'safe' dose based upon linear extrapolation is lewer than the 'safe' dose based upon the multistats model by a factor of about five and there are doubtless eases where this difference could be greater than an order of magnitude. MOMS 221394 to REFERENCES Alien, J. F. and NorbseV. (1976). Pathobiological response of privates to polychlorinated biphenyl exposure. Proceedings of the !'atirr.a'. Conference on Polychlorinated Biehenvls (Nov. 19--21, *'1975, Chicago, 111.) E?A-560/6-75-004. Armltage, P. and Doll, R. (1961). Stochastic models for carcinogenesis. In Proceedings of the Fourth Berkeley Svmaosium on >iathemiticil Statistics and Probability, Vol. 4, pp. 19-38, University of California Press, Serkeley and Los Angeles. Bahn, A. K., Rosenvaika, I., Herrmann, N., Grover, P,, Stellman, J. and O'Leary, K. (1976). Melanoma after exposure to PCB's. Kev End and Journal of Medicine 395, 450. Bahn, A. K., Grover, ?., Rosenvaike, I., O'Leary, K. and Stellman, J. (1977). PCS? and melanoma. Kev England Journal of Medicine 296, 108. Baraottl, D. A., i-Jarlar, R. J,, and Allen, J. X. (1976). Reproductive dysfunctions in rhesus nonkeys exposed to low levels of poly chlorinated biphenyls (Arochlor 1248). Food and Cosmetics Toxicology. Bier (1972). The effects on populations of exposures to low levels cf ionising radiation, Kational Academy of Sciences. National Research Council (Vashington, D.C., 1972). Brown, J. M. (1976). Linearity vs. nonlinearity of dose response for radiation carcinogenesis. Health Phvsics 31. 231-245.. Butler. P. A. and Schultrmann, R. L. (IS'78). Residues of Pesticides end PCBs in Estuerine Fish, 1972-1976 - National Peaticide Monitoring Program. Peaticide Monitoring Journal 12. 51-59. Calandra, J. C. (1976). Suraary of toxicological studies on coraercial PCB'a. Proceedings of the Kational Conference on Polvchlorine*.td Biphenyls (Kov. 19-21, 1975, Chicago, 111.) EPA-560/6-75-004, 35-42. Chand, N. and Hoel, D. C. (1974). A comparison of models for determining safe levels of environmental agents. In Reliability and Eiomttrv. ed. F. Prochan and R. J. Serfling, pp. 362-401, Philadelphia: SIAM. Cohen, A. C. (1961). Tables for maximum likelihood estimates: singly truncated and singly censored samples. Technomctrlca 3. 535-541. Cordle, F., Corneliussen, P,, Jelinek, C. Hacklcy, B.t Lehman, R., McLaughlin, J., Rhoden, R. and Shapiro, R. (1978). Human ex posure to polychlorinated biphenyls and polybreainsted biphenyls. Environmental Health Perspectives _2. 157-172. MOWS 221395 81 Cornfield, Jerome (1977). Carcinogenic risk asse::snt*nt. Scivnrc 193: 693-699. Crump, K. S., Hoel, D. C.. Langley, C. H., and Feto, R. (3976). Fundamental carcinogenic proccstes and their implication for low doa* risk miisrien;, Cancer Research 36:2973-29J9. Crump, K. S., Cueaa, H. A. and Deal, K. L. (1977). Confidence inter vals and teats of hypotheses concerning dose response relations inferred fros animal carcinogenicity data. Biometrics 33:437-451. Crump, K. S. (1977). Response to open query: Theoretical problems in the modified llantel-Sryan procedure. Biometrics 33:752-755. Deffer, P. 2., Cruap, K. 5. and Masteraan, M.D. (1979). Asymptotic theory for analyzing dose response survival data with application to the low-dose cctrspolatlon problem (submitted). Dennis, D. S. (1976). Polychlorinsted biphenyls in the surfsce water and bottom sediments of the csjor drainage basins of the United States. Proceedings of the National Conference on Polychlorinated Biohenvls (Kov. 19-21, 1975, Chicago, 111.) EPA-360/6-75-004, 183-194. environmental Protection Agency (1976). FCBs in the United States industrial use and environmental distribution, task 1. EPA 560/6-76-005. Food Safety Council (1976). Proposed system for food safety assessment. Food end Cosmetic Toxicology (December). Cess, C. H., Ccsts, D. and Crahas, K. (1964). Carcinogenic doseresponse curve to oral diethylstilbcstrol. J. Natl. Cancer Inst. 33: 971-977. Cuess, H. A. and Cruap, X.S. (1976). Lev-dose rate extrapolation of data froa animal carcinogenicity experiaents--analysis of s new statistical technique. Mathematical Biosciences 30, 15-36. Cuess, H. A. and Crisp, K. S. , (1978). Maximus likelihood estimation of dose-response function-, subject to absolutely aonctonie Constraints. Annals of Statistics 6. 101-111. Cusss, H. A-, Crump, K, S., and Pcto, R,, (1977). Uncertainty estimates for low-dose rate-extrapolations of animal carcinogenicity data. Cancer Research 37;3475-3483. Hartley, H. 0. and Sielken, R. L. (1977). Estimation of "safe doses" in carcinogenic experiments. Biometrics 33:1-30. f-s lj Hoel, D. C., Caylor, D, U., Kirschstcin, R. L., S.iffiotti, U. and Sehncidcrman, >1. A. (1975). Estimation of risks of irreversible, delayed toxicity. J. of Toxirolorv and Env(rt>nnwwtil Health 1:133-151. MOMS 221396 Humphrey, H. E. B. (1577). Evaluation of changes of the level of polychlorinated biphenyl* (PC3) in human tissue. Final Report on FDA Contract 233-73-2209, 1ARC (1976) Polychlorinated biphenyl* and polybromated biphenyl*. 1AKC (Intarnaticn*l Agency for Research on Cancer) Monograph* on the Evaluation of the Carcinogenic Risk of Chemical* to Humana, Vol. 16; VHO Publication Centre, Albany, K.Y. Industrial Bio-Teat Laboratories, Inc. (1971). Reports to Monsanto Company. Tve-year chronic toxicity vith Arcchlor 1242, 1254, and 1260 in albino rats. Unpublished reports, November 12, 1971. 1BT No. B7298. Zto, N., Nagasaki, R., Arai, M., MaV.iura, S., Sugihera, S., and Hirao, K. (1973). Hisopachologic studies on liver tumorigenasis induced by mice by technical polychlorinated biphenyls and its promoting effect on liver tumors induced by benezene hexachloride. Journal of the National Cancer Institute 51. 1637-1646. Jellnek, C. and Corneliussen, P. E. (1976). Levels of FCBs in.tbe. U.S. food supply. Proceedlmrs of the National Conference on Polychlorinated Slabenvla (Nov. 19-21, 1975, Cnicago, 111.) EPA-560/6-75-004, 147-154. Johnson, R. D. and Mask*, D. D. (1977). Pesticide and Other Chemical Residues in Total Diet Samples (XT) . Pesticide Monitoring Journal 11, 116-131. Kla&rough, R. D., Linder, R. E., Burse, V. V. and Jennings, R. V. (1973). Adenofibrosis in the rat liver, with persistence of polychlorinated biphenyls in adipose tissue. Archives of Environmental Health 27. 390-395. Kimbrough, R. D. and Linder, P.. E. (1974). The induction of adenoflbrosia and hepatomas of the liver in mice of the BALB/cJ strein by polychlorinated biphenyls (Aroclor 1254). Journal of the National Cancer Institute 53, 544-552. Kimbrough, P.. D., Squire, R. A., Linder, R. E., Strandburg, J. D., Mont all, R. J., and Burse, V. W, (1975). Induction of liver tumors in Sherman strain female rats by polychlorinated biphenyl Aroclor 1260. Journal of the National Cancer Institute 55. 1453-1459. * Ruts, F. W, and Straasnan, S. C. (1976). Residues of polychlorinated biphenyls in the general population of the United States. Proceeding of the National Conference on Polvchlorinated Blphcnvla (liov. 19-21, 1975, Chicago, 111.) E1*a"5G0/6-75-004 , 139-143. Linder, R. E., Caines, T. B., and Kimbrough, R. D. (1974). The effect of polychlorinated biphenyls on rat reproduction. Pood and Cosmetic Toxicolorv 12. 63-77. M0NS 221397 63 Kflkiurot Ji,, Aoi( H., Sugihara, 5** llirjp, K* Arii '* Ito, h. (1S7A). Inhibitory effects of polychlorinated biphenyls on livr tuaoriganesis in rsrs treated with 3'-ntthyl-4-diBeth)lamino*- cobinrens, lf-2-fluorcnylaeetanide, and diethylnitrosamine. Jourr.sl of the Nation*) Cancer Institute 33, 1253-1257. Mentel, N. snd Bryen, V. R. (1961). Safety testing of csrcinogenic agents. J. Natl. Cancer Inst. 27:455-670. Mantel, K. and Sehnaiderman, M. (1975). Estimating 'safe* levels, a hazardous undertaking. Cancer Research 35:1379-1316. Mantel, N., Bahidar, N. R., Brown, C. C., Cidnera, J. L. and Tukey, J. V. (1975). An Improved Mantel-Bryan procedure for the "safety" testing of carcinogens. Cancer Research 35:665-872. Mantel, N. (1977). Aspects of the Hartley-Sielken approach for setting "safe doses" of carcinogens. In Origins of Human Cancer Booh C. Human Risk Assessment. H. H. Hiatt, J. 0. Watson, and J. A. Winston (eds.) Cold Springs Htrbcr, K.Y., Cold Springs Harbor Laboratory, 1397-1401. McCann, J., Choi, E., Yamasaki, E. and Aces, B. N. (1975). Detection of carcinogens as mutagens in the Salmonella/microsome teat: Assay of 300 chemicals, Proc. Kst. Acad. Sci. USA 72:5135-5139. McCann, J. and Ames, . X. (1976). Detection of carcinogens as mutagens in the Salmonella/microsone test: Assay of 300 chemicals: Dis cussion. Proc. Nat. Acad, Sci. USA 73:950-954. McDermott, D. J., Young, D. R., and Heeaen, T. C. (1976). PCI con tamination of southern California marine organisms. Proceidincs of the Rational Conference or Polvchlorinsted Einhenvls (Kov. 19-21, 1975, Chicego, 111.) EPA-560/6-75-004, 209-217. McNulty, V. P, (1976). Dee of the rhesus monkey ee e model for human toxicity with polychlorinated biphenyls. Proceedinte of the Rational Conference on Pelvchlcrlcated Blohcnvls (Nov. 19-21, 1975, Chicego, 111.). EPA-560/6-75-004, 332-333. Monsanto (1975). Industrial Rio-Test T.rhevatorias, Inc. Reports. Mstopathologlcal evaluation of additional liver sections. Msreh 24, 1975. Unpublished reports, Monsanto Co., St. Louis, Missouri. Munson, T. 0., Palmer, H. D., and Forns, J. M. (1976). Transport of chlorinated hydrocarbons in the upper Chesapeake Bay. Pro ceedings of the National Conference on folvchlorinatcd Blnhenvls (Nov. 19-21, 1575, Chicago. 111.) EPA-360/6-7~5-004, 216-219. Rational Marine Fisheries Service (1976). Compendium of PCB data, N0AA, U.S, Dept, of Commerce, Washington, D.C. MOMS 221398 64 Kational Institute for Occupational Safety and Health 0?77). Criteria for * recommended standard - . . occupational exposure to poly chlorinated biphenyls ()>*). DHSVT (N10SK) Publication };0. ` 77-225. NCI (Kational Cancer Institute) (1978). Bioassay of Arocler 1254 for possible carcinogenicity, Carcinogenesis Technical Report Series No. 36- CAS No. 27323-16-6 NC1-CC-TK-38. Risbet, I. C. T. (l?76a). Protection Agency. PB-255-397. Criteria document for PCBs. environmental Kational Technical Information Service Klsbet, 1. C. T. (1976b). Introductory remarks, session III. Pro ceedings of the Kational Conference on Polychlorinated Birhcnvls (Nov, 19-21, 1975, Chicago, 111.) EPA-560/6-75-004, 137-136. Kisbet, 1. C. T. (1976c). Environmental transport and occurrence of PCBs in 1975. Proceadincs of tht National Conference on Poly chlorinated Eiphenvls (Nov. 19-21, 1975, Chicago, 111.) EPA-560/6-75-004, 254-256. Pete, R. (1974). (Paper delivered at the KIIHS Vrightsvilit Beech conference on Lov-Dose Risk Estimation). Peto, R. (1977). Tha carcinogenic affects of chronic exposure to very low levels of toxic substances. Presented at an K1EKS con ference on cancer risk estimation, Pinehurst, K.C., March 10 - 12, 1976. Environmental Health Perspectives. Pete, R. (1977b). Epidcmology, multistage models, end short-term mutagenicity tests. In Origins of Hunan Cancer Book C, Human Risk Assessment. H. H. Hiatt, J. 0. Watson, end J. A. Winsten (eds.) Cold Springs Harbor, K.Y., Cold Springs Harbor Laboratory, 1397-1401. Polishuk, 2. W., Ron, M., Vasserman, w., Cucos, Simi, Vasserman, Dora, and Lemeseh, C. (1977). Organoclorine compounds in han blood plana and milk. Pesticides Konitorine Journal 10, 121-129. ------------- Rai, K. end Van Ryxin, J. (1976). Risk assessment of toxic substances based on e generalised multihit dose response model. Troc. SIMS Conference, Alta, Utah, June 26-30. Rannug, U., Johansson, A., Rand, C., and Vachtmeister, C. A. (1974). The mutagenicity of vinyl chloride after metabolic activation. Ambio 3, 194-197. Salsburg, D. S. (1977). Open query: Theoretical problems in the modified Mantsl-Bryxn procedure. Biometrics 33:419-421. Sanborn, J. R., Childers, V. F., and Mctcslf, R. L. (1976). Uptake of three polychlorinated biphenyls, DDT and DDE by the green sunflsh, Lcpccis Cyanelluf. Raf. Proceedings of the National Conference on rnlvchlorlnnted Biphenyls (Nov. 19-21, 1975, Chicago. HI.) El'A-560/6-75-004, 236-242. MOMS 221399 Savage, t. P. (1979). (Personal comnunication). Sp.ignoli, J. J. and Skinner, L. C. (1977). I'CE'e in fish from selected waters of Uev York State. Pesticide Honltorlnr Journal 11. 69-17. Uchiyaaa, M., Chiba, T., and Node, K. (1974). Co-carcinogenic affect of DDT and PCS feedings on methylehloanthrene-induced chemical carcinogenesis. Bulletin of Environmental Contamination and Toxicology 12. 6S7-693. Valher, C. X. (1976). The occurrence of PCS in the National Fish and Wildlife Monitoring Program. Proceedings of the National Con ference on Polychlorinated Biphenyls (Kov. 19*21, 1975, Chicago, 711.) EPA-560/6-75-0047 161-176. Vateon, C. S. (1977). Age incidence eurvea for cancer. Proc. Nat. Acad. Scl. USA 74, 1341-1342. -- t60M$ 221400