Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
B Analytical Methodology for Estimating Oncogen~c Risks of Human Exposure to Agricultural Chemicals in Food Crops JOHN P. WARGO The charge to this committee required the ability to characterize and analyze dietary oncogenic risk associated with exposure to pesticides through food crops, totaled by individual pesticide, by crop, and by pesticide-crop combinations. The capability of responding to various assumptions and questions was also necessary; for example, · How many pesticide residue tolerances would be affected if a regulatory threshold Of 10-6 were established for a specific pesticide-crop combination? · What if the risk threshold were changed, or if the threshold were applied only to risk from residues in processed foods? · What if regulatory thresholds were established by setting a limit based on total allowable risk by crop, or by pesticide? · How is risk distributed among types of pesticidesfor example, apple fungicides or corn herbicides and how might risk be reduced as a result of alternative regulatory scenarios? DATA MANAGEMENT SYSTEM No computerized data management system existed prior to this study that could respond to these questions. The conceptual framework for this effort was derived in part from the U.S. Environmental Protection Agency's (EPA's) Tolerance Assessment System (TAS), which is main- frame based. The TAS joins pesticide-commodity tolerance data with 174
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS 175 food consumption data to estimate possible chemical intake levels. Pesticide intake levels can then be converted into oncogenic risk esti- mates if reliable oncogenic potency estimates are available. Also, the TAS permits the comparison of possible daily pesticide intake levels with Acceptable Daily Intakes. The TAS represents a remarkable breakthrough in analytical capability by the EPA, yet it is difficult and expensive to operate and update. And because it is mainframe based, it is accessible only to those with the computers and the expertise in mainframe operating systems and data analysis software. The microcomputer-based system designed for this committee report incorporates TAS food consumption estimates and tolerance data, but it differs radically from the TAS in several critical respects. It operates on IBM-compatible microcomputers and uses common statistical (SAS), data-base management (dBASE III), and spreadsheet (Lotus) software. Not only does this dramatically increase the types of analyses that can be conducted, it substantially reduces the cost, time, and expertise required to perform the analyses. The new system also contains simulation models that permit the user to change assumptions regarding tolerances, commodity consumption lev- els, percentage of acres treated, and oncogenic potency factors, instantly recalculating risks while graphing the results. In other words, these variables are linked by formulas in the system so that if any component in the risk equation is changed, the net effect on oncogenic risk is instantly demonstrated. The effects of as many as a dozen different tolerance- setting strategies can be forecast and graphed in 15 minutes. The structure of the new system differs significantly from the TAS in the variety of new data fields. For example, the most recent EPA oncogenic potency estimates are incorporated into the system, permitting the transformation of estimates of chemical intake into estimates of oncogenic risk. Raw commodities, processed commodities, and specific pesticide-crop combinations are uniquely coded for separation into fields, enabling the calculation of risk by each field. Economic and pesticide-use data fields were added so that economic effects of different regulatory scenarios could be forecast. Also, new fields were created by mathematically transforming existing fields. For example, the chemical intake field is calculated as the product of the Code of Federal Regulations (CFR) published tolerance field and the mean consumption field, and adjusted by the standard error field in a manner dependent on the desired confidence interval. If desired, the risk field could be adjusted by the percentage of acres treated. Total risks by pesticide, by crop, and by pesticide-crop combination also exist in separate data fields, having been calculated from the individual pesticide and commodity risk fields. .
76 APPENDIX B The new system can be updated easily. Tolerance, toxicological, ecological, and residue data are constantly being submitted to the EPA. These new data can be added to the system simply by typing the data into appropriate boxes on the screen. Finally, a system for rapid electronic transfer of data between the mainframes and the IBM-compatible microcomputers was designed. DATA BASE CONSTRUCTION Tape-to-Mainframe Transfer and Data Transformations Several transformations of data files were necessary to conduct the analyses presented within this report. First, several TAS files formatted in SAS (Statistical Analysis System) were transferred from magnetic tapes to an IBM 4043 disk, using IBM's Job Control Language. They included the Mean Consumption file, containing estimates of consumption of the 376 distinct food types by 23 population groups along with associated standard errors, and similar mean consumption data broken into 691 separate food forms. Two additional files- the TAS Tolerance file and the TAS Preamble file were transferred to the mainframe disk in raw (rather than SAS) format. The commodity codes in the TAS Tolerance file match the commodity codes in the TAS Mean Consumption file, but the commodity codes in the CFR Published Tolerance file do not match commodity codes in the Mean Consumption file. A code conversion was therefore necessary to relate current CFR tolerances to TAS-formatted consumption statistics. Once on the mainframe, these files were transformed into a SAS data set. The Mean Consumption file, the Preamble file, and the TAS Tolerance file were then merged so that individual records included pesticide names and codes, commodity names and codes, published tolerance levels, mean consumption estimates, standard errors of these estimates, and a summary of toxicological data. This merged file was then sorted, first by commodity code and second by pesticide code, and transformed into raw (ASCII) format to prepare it for transfer to a microcomputer. Mainframe-to-Microcomputer Transfer The ASCII data files were then electronically transferred to a micro- computer that is hardwired to the mainframe. A utility called YTERM was used to transfer the files and to compare the original data and the copied version, highlighting any transfer errors. The data were stored in the directory containing the analytical software. (However, they can be stored in any desired subdirectory on the microcomputer hard drive.)
METHODOLOG Y FOR ESTIMA TING ONCOGENIC RISKS 177 Raw-to dBASE III Format dBASE III was used to analyze the master data file. It not only stores data within individual records but contains its own programming language which permits mathematical operations such as sums of risks only for specified criteria, and other operations such as merging, rapid sorting, and indexing of large files. Before the raw data set could be loaded into a dBASE file, the dBASE file structure had to be created. Field names, types (numeric, character, date, or memo), and lengths had to be specified to precisely match the location of the data in the stream of numbers and characters lying in raw data files. When the dBASE file format was established, the raw data set was appended to the dBASE file "shell." (The only limitation to the size of the data base is the available disk memory, since dBASE files do not reside in Random Access Memory fRAM]. A 20-megabyte hard drive was sufficient for the analyses performed for this study.) dBASE III-t - Lotus 1-2-3 Transfer One of the major limitations of dBASE III is the fact that mathematical transformations of entire data fields must be accomplished within the context of programs. dBASE is not interactive in a way that allows the user to change a set of assumptions and to immediately see the effects on the mathematically related fields. Lotus 1-2-3 is similar to dBASE III in that data are aligned within records and fields in a matrix format. The major difference between the two systems lies in the ability of Lotus to establish formulate relationships between cells of the data matrix. For example, if column 1 contains data on pesticide type, column 2 contains data on commodity names, column 3 contains data on published tolerances, and column 4 contains data on mean food consumption adjusted by standard errors, it would be possible to place a formula in column 5 that tells Lotus to multiply the value in column 3 by the value in column 4 to obtain an estimate of pesticide intake (Theoretical Maximum Residue Contribution, or TMRC). Once this relationship is established by formula, any change in one of the cells on which the formula is dependent will change the visible value or result of the formula. Lotus was therefore ideal for analyzing the sensitivity of pesticide-crop risk to changes in assumptions regarding variables such as tolerances, residues, consumption estimates, potency factors, and acres treated. Once the risk formulas were established within the spreadsheet, any changes in assumptions were instantly converted into changes in esti- mated risk. The primary limitation of Lotus lies in the fact that the entire package
178 APPENDIX B resides in RAM, which is generally limited to 640 kilobytes (kb) by IBM's Disk Operating System (DOS) unless a special expansion board is installed to increase memory. This memory constraint limits the size of analytical spreadsheets. For example, the master dBASE file used in this study was too large to fit within a single Lotus spreadsheet and, therefore, had to be broken into smaller unitscommonly, groups of pesticides or crops. For the analyses conducted for this report, the RAM residency of Lotus was not a problem, since most analyses were performed on relatively small files. DATA DESCRIPTION AND SOURCES Diverse types of data are contained in the microcomputer-based system used to calculate oncogenic risks for this study. These data, their sources, and their relevance to the analyses presented are briefly described below. For clarity, a data file can be thought of as a simple electronic matrix, with data being stored within individual "cells," defined by the horizontal and vertical location of the data. The horizontal rows of the file are known as "records," and the vertical columns are known as "fields." Data can be entered in character format (e.g., "BENOMYL") or numeric format (e.g., ".0000435"~. Most computer-based analyses are performed on data within the cells, through such operations as sorting or merging based upon specified criteria, statistical analyses, or mathematical transformations dependent upon user-specified formulas. Chemical Identification Data Chemicals are identified by various codes and names. Each chemical considered in this study was assigned three different alphanumeric codes: (1) a Chemical Abstracts Service code; (2) a Caswell code (CASWL); and (3) a Shaughnessy code. Pesticides are also identified by preferred name, and by as many as five alternate names. These data were derived almost exclusively from the Preamble file from the EPA's Tolerance Assessment System. Additional data fields were added to these identification codes to indicate the primary uses of each pesticide for example, fungicide, herbicide, or insecticide. Chemical Tolerance Data Tolerance data were derived from two sources: the TAS Tolerance file and the CFR Published Tolerance file, which lists all tolerances in the Code of Federal Regulations. The TAS Tolerance file contains essentially the same data as the CFR
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS 179 Published Tolerance file, but there are several important distinctions. The TAS file contains 16,526 pesticide-commodity records, as contrasted with 8,477 records'in the CFR file. This difference in file size can be explained as follows. In the CFR file, for example, one tolerance might be listed for captan on fresh tomatoes and another for captan in processed tomato products. In the TAS file, however, each possible processed-food form- catsup, juice, puree, and paste, for example is listed as a separate record. This expansion was accomplished to more accurately represent dietary exposure to the various food forms and to enable the estimation of average daily pesticide intakes by food form. The TAS Tolerance file contains 342 different pesticides (each assigned a CASWL code by the EPA) and 434 different commodity or food forms (each assigned a Raw Agricultural Codecalled an EPARAC by the EPA). (See Pesticide-Commodity Codes, below, for a description of these codes.) All tolerances are expressed in parts per million. For this study, use cancellations or suspensions were identified by adding a new data field to each record. Canceled or suspended uses were deleted before any risk calculations were performed. The CFR Published Tolerance file mirrors the data contained in the Code of Federal Regulations. The 1985 file includes 351 distinct chemi- cals, 32 of which are listed as "Exempt" (40 CFR Part 180, Subpart D), and 19 of which are listed as Generally Regarded As Safe (GRAS) (40 CFR § 180.21. Together, the Exempt and GRAS categories account for 97 of the 8,477 records.2 Each record contains the pesticide name and unique pesticide code, the commodity name and unique code, the 1985 tolerance, the CFR citation, and an EPA-assigned petition number. Food Consumption Data The food consumption (Mean Consumption) file was transferred di- rectly from the TAS, and was based on a survey conducted during 1977-1978 by the Nutrition Monitoring Division of the Human Nutrition Information Service (HNIS) of the U.S. Department of Agriculture (USDA).3 Data were collected from 30,770 individuals who were asked to recall food eaten the previous day and to record food eaten during the day of the interview and the day that followed. Each individual reported the food ingested, an estimate of the amount, the eating "event" (such as breakfast or lunch), water consumption, general health status, height and weight, and socioeconomic and geographic characteristics.4 The primary sample was a multistage, stratified probability sample of all households in the coterminous United States. Within this sample, four independent, interpenetrating samples were drawn in four successive seasonal quarters between April 1, 1977, and March 31, 1978. A similar survey is currently
i 80 APPENDIX B being conducted by the FINIS; however, its results will not be available until 1989 (Bruce Gray, personal communication, 19861. The Research Triangle Institute (RTI) transformed the USDA con- sumption survey into a Raw Agricultural Commodity Data Base, defining 376 unique food types.5 These foods included commodities for which the EPA would be most likely to establish pesticide residue tolerances, thereby permitting the estimation of pesticide intake by commodity. It was not intended to include all edible foods. The list was developed from various sources, including the EPA's Food Factor list, the CFR listing of then-current tolerances, and the USDA food consumption survey codebook. Although the list includes 376 unique food types, only 273 of these foods were mentioned by the USDA survey respondents. That survey did not include food components such as spices, estimated to constitute less than 0.1 percent of a given food consumed. Estimates of consumption for the 103 food types not reported by survey respondents were derived from studies by the USDA,6 Magness et al.,7 the National Academy of Sciences,8 and the International Tariff Commission.9 Finally, an "arbitrary consumption value" (lo-6 g/kg body weight/day) was assigned to foods for which tolerances do not exist, or in cases where information could not be obtained.~° RTI also developed a set of "food form" codes that distinguish between raw, cooked, fresh, frozen, canned, dried, baked, broiled, fried, pickled, corned, or salt-cured forms of the 376 food types. The Food Form file includes 691 records, yet does not include any new food types. Four records exist in the Food Form file for each of the 30,770 survey respondents. The first record contains descriptive information (such as age, sex, weight, census region); the second, third, and fourth records for each respondent indicate food consumption for each of the three days surveyed. Consumption is recorded in grams of food form per day. Basic transformations of data conducted by RTI include · Converting consumption from grams/day to grams/kilogram of body weight/day; · Aggregating consumption over foods; and · Averaging consumption over days and/or individuals. The USDA Average Food Consumption file was used as the primary basis of this analysis. It was derived from the 90,000-record Person-Day Food Consumption file which was then averaged over the three days to produce a 30,000-record Average Daily Individual Food Consumption file. This file was then averaged over individuals to provide mean consumption estimates for the 273 food types reported by survey respon- dents. 12 Since this mean consumption estimate is essentially an arithmetic mean
METHODOLOG Y FOR ESTIMA TING ONCOGENIC RISKS I 8 1 derived from the sample, it was adjusted, or weighted, on the basis ot sample size to derive an estimate of the population mean consumption for each food type.~3 Standard errors of the mean consumption estimates for each of the commodities were computed by RTI using a first-order Taylor-series approximation of the deviations of the estimates from their expected values. (Formulas used by SESUDAAN, a statistical package developed by RTI, have also been reported by RTI.~4) These data were used to calculate various confidence intervals that bound the estimated popula- tion mean consumption. For this study the high end of the 95 percent confidence interval, or two standard errors above the population mean consumption, was chosen by the committee to represent the U.S. average commodity consumption for each food type. Thus, if a similar-sized sample were taken from the population of the contiguous United States, there is a 95 percent proba- bility that the population mean consumption estimate derived from the second sample would lie within the same confidence interval as the first. This does not mean, by contrast, that 95 percent of all individuals will consume less than the computed estimate, since the statistic was calculated from the standard error of the population mean consumption estimate, not from the standard deviations of individual consumption data. Pesticide Residue Data - In an ideal world, accurate estimates of pesticide residues in foods and in water would be available as the basis for predicting average chemical intake (commonly described by the EPA as a chemical's Theoretical Maximum Residue Contribution, or TMRC). The FDA is responsible for enforcing tolerances for all pesticides used in food. To do this, it samples 7,000 domestic and 5,000 imported shipments of food each year. There are several potential problems with relying on the FDA's sampling results to estimate food residues, most of which are associated with the agency's declining budgetary resources. First, the number of samples tested for each pesticide-crop combination is not large enough to develop popula- tion consumption estimates with meaningful confidence intervals. Second, detection of residues in raw agricultural commodities is often a meaningful indicator of residues in processed-food forms. Drying, oil extraction, cook- ing, and other processing techniques can dramatically alter pesticide residue levels. Third, the EPA is currently requiring a battery of residue tests prior to pesticide reregistration. Complete residue data are available for only a small percentage of the pesticides that are currently registered, leading to gaps and significant variation in quality of data on pesticides. Fourth, the multichemical detection technology used by the FDA cannot detect residues
I 82 APPENDIX B of all registered pesticides. Fifth, this multichemical detection technology is not sufficiently sensitive to simultaneously detect residues of all the chemi- cals that can be detected individually. Together, these conditions led the committee to base its estimate of the human pesticide intake (TMRC) on the assumption that the chemical is in or on the food at the current tolerance level. This choice was made to standardize the quality of data on chemicals, even though it will likely overestimate chemical TMRCs since most FDA samples contained resi- dues at levels substantially below established tolerances.~5 Toxicological Data The pesticide oncogenic potency factors in this study were estimated by the EPA's Hazard Evaluation Division. In no instance did the staff of the Board on Agriculture or the authors of this report interpret the EPA potency estimates. The potential for tumor induction is indicated by a "potency factor" or Q*. The Q* is an estimate of the number of additional tumors that can be expected to develop within a human population, based upon the dose response results of animal bioassays. Laboratory animals are generally exposed at much higher doses over their lifetimes than the average human would normally encounter in his or her lifetime. Because animal life-spans are much shorter than human life-spans, dose response data must be extrapolated to predict human tumor incidence at the lower doses that humans are likely to encounter in food or water. With only several exceptions, the 28 potency estimates used in these analyses were derived from the linearized multistage model of low-dose extrapolation. The potency factor used by the EPA is the slope of the line at the 95 percent upper confidence limit representing the dose response relationship(change in lifetime probability of extra tumor incidence)/ (unit of exposure of dose). The expression of the oncogenic potency factor as a linear extra tumor incidence/dose ratio enables the prediction of tumor incidence based on estimates of human chemical exposure. The potency factor (estimated extra tumor incidence/dose) is simply multi- plied by the pesticide intake estimate (dose) derived from the food consumption data and assumptions regarding the pesticide residue level in the food at the time of consumption. Regulatory Status Data Tolerances are listed in Titles 21 and 40 of the Code of Federal Regulations and in the TAS Tolerance file for pesticide-commodity combinations. A separate data field was created in the data base to indicate the status of tolerances. However, risk calculations were based
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS I 83 only on active tolerances, under the assumption that residues of canceled or suspended pesticides would not be relevant to analyses designed to measure the impact of theoretical regulatory policies. These data were provided by the EPA in January 1986.~6 Food Processing Data The TAS Tolerance file is an expanded version of the CFR Published Tolerance file, as described above. Risk estimates based solely on the CFR file would underestimate the risk for commodities for which toler- ances are listed only for the raw form, even though residues could also be expected in processed forms. The expansion program was developed by the EPA to more acurately reflect probable sources of exposure to chemicals in processed-food forms. When the TAS expansion program is run on the CFR Published Tolerance file, a record listing a tolerance for raw tomatoes, for example, is automatically expanded to separate consumption records for tomato paste, juice, catsup, and puree. Each food form carries with it the tolerance and CFR code from the original raw commodity. Where processed-food tolerances existed in the CFR file, the correct tolerance and CFR citation are carried into the TAS Tolerance file. Duplicates created by the expansion are automatically deleted. Since neither the TAS nor the CFR Tolerance files code individual commodities by their raw or processed state, a new data field had to be created. A new code Phantom was designed to distinguish CFR toler- ances from TAS tolerances and to distinguish between processed forms of commodities (where residue concentration might occur) and raw agricul- tural commodities. The following Phantom codes were assigned to each chemical-commodity record that existed in the TAS Tolerance file: 0 = New TAS raw-commodity tolerance created by TAS expansion; 1 = New TAS processed-commodity tolerance created by TAS expan- s~on; 2 = 01d CFR processed-commodity tolerance; 3 = 01d CFR raw-commodity tolerance. The Phantom codes were assigned to the food type codes known as EPARACs, which are uniquely assigned to all food types contained in the food consumption survey (see Pesticide-Commodity Codes, below, for more information). The proper form (raw or processed) of each of the 376 commodities listed within the survey was determined by matching EPARAC codes (explained below) in the CFR Tolerance file with EPARAC codes in the TAS Tolerance file and through information
I 84 APPENDIX B provided by the Residue Chemistry Branch of the Office of Pesticide Programs (OPP). The development and assignment of the Phantom code made possible the estimation of oncogenic risk based on the raw or processed form of the commodity. The code also made possible the distinction between risk associated with processed-commodity tolerances in the current CFR Published Tolerance file and risk associated with processed-commodity tolerances in the expanded TAS Tolerance file. Chemical Use and Cost Data For eight distinct crop-pesticide combinations, data on acre treatments, percentage of planted acres treated, and expenditures per acre were entered as new fields. These combinations are corn- and soybean- herbicides, cotton-insecticides, and apple-, tomato-, potato-, peanut-, and grape-fungicides. The primary sources for these data included the Eco- nomic Analysis Branch of the Office of Pesticide Programs and the USDA Economic Research Service. The addition of these data fields facilitates the analysis of pesticide- commodity risk by acres treated, acre treatments, and expenditures. Comparison of expenditures per acre for likely substitute pesticides allows the development of analyses that estimate the cost per unit risk reduction, assuming various patterns of pesticide substitution. Pesticide-Commodity Codes The TAS data sets include unique codes assigned to unique pesticides (CASWL codes) and to unique commodities (EPARAC codes). Because most analyses relevant to this study involved calculation of risk by pesticide-commodity combination, it was necessary to develop a code that was also unique to pesticide-commodity combinations. The CASRAC code was designed for this purpose and is a combination of the CASWL code (for unique pesticides) and the EPARAC code (for unique commod- ities). The EPA-designed EPARAC code was not suitable for the committee's analyses. Each code has a five-digit numeric prefix and a two-digit character suffix. For most commodities, all numeric suffixes are identical for basic commodity forms, such as tomatoes or apples. In several cases, such as peanuts, this was not the case, and the codes were redefined so that all peanut records began with the same numeric prefix. This was absolutely crucial to the development of a methodology to sum risks by pesticide-crop combination. Now all distinct crop groups (apples, pea- nuts, corn, grapes, etc.) have distinct EPARAC codes (renamed
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS ~ 85 EPARAC 11. The CASRAC field, which is unique to each distinct pesticide-crop combination, permitted the calculation of total pesticide- commodity risks that might be affected if, for example, tolerances for each pesticide-commodity combination exceeding a predefined risk threshold were revoked. ANALYTICAL METHODS Risk Calculation The method of oncogenic risk estimation used in this study is a minor variant of the Routine Chronic Analysis used by the EPA and described in the documentation of the Tolerance Assessment System. )7 The oncogenic risk associated with any individual chemical can be calculated only if a reliable estimate of oncogenic potency (Q*) has been developed for that chemical. The committee's risk estimates are derived from 28 of 30 compounds for which oncogenic potencies were provided by the EPA. In these cases, risk estimates were calculated for all distinctive food types in which residues could be anticipated. For example, a separate risk estimate was calculated for the residues of alachlor in raw corn as well as alachlor in corn oil. Of the 16,500 tolerances that exist in the TAS Tolerance file, risk estimates were derived for only 2,306 pesticide- commodity combinations. The limited number of potency factors reflects findings of non-oncogenicity for many compounds and the absence of valid oncogenic or chronic feeding studies for numerous other pesticides. For several of the scenario analyses presented below, calculation of the number of pesticides and crops that would be affected by different regulatory thresholds was based on a pool of 53 chemicals that the EPA believes to be oncogenic, despite the absence of potency factors for 25 of them. The critical variables that are components of the risk calculation are briefly described below and more thoroughly described under Data Description and Sources, above, and Uncertainty in Oncogenic Risk Estimates, below. CHEMICAL RESIDUES The current tolerances and the residue estimates obtained through the TAS expansion of the CFR Published Tolerance file were used as the basis for estimating "worst-case" pesticide residues in the commodities. Although a far more accurate representation of likely exposure might be developed through statistically valid commodity and residue sampling
I 86 APPENDIX B techniques, these data were not available for this study. Tolerances are expressed in parts per million. CONSUMPTION ESTIMATES Commodity consumption estimates were based on the mean consump- tion data developed from the 1977-1978 food consumption survey of individuals within the 48 contiguous states. The consumption estimate is a U.S. average statistic and could vary significantly beyond the mean for individuals. This estimate was used to calculate the high end of the 95 percent confidence interval. Consumption estimates are expressed in grams of commodity per kilogram of body weight per day. ESTIMATE OF CHEMICAL INTAKE An estimate of mean pesticide intake was developed by multiplying the tolerance and the mean consumption estimate for each pesticide- commodity record. This exposure estimate is called the Theoretical Maximum Residue Contribution. The TMRC assumes that residues are present at the tolerance level on every crop that has a tolerance, and that all acres of all crops with tolerances are treated. Exposure to these residues is expressed in milligrams of chemical per kilogram of body weight per day. This method of calculating exposure to residues overes- timates actual dietary exposure across the whole population, but it is preferred by the EPA as an initial step in risk assessment because it incorporates a prudent safety factor into the risk assessment process. POTENCY FACTOR OR Q* The potency factor, as calculated by the EPA, is the slope of the dose response curve or line from animal oncogenic tests. This slope represents the change in Y (tumor incidence) over the change in X (dose). Potency therefore increases with the steepness of the slope. The units of the potency factor are tumors per milligram of pesticide per kilogram of body weight per day. The potency factor assumes that this average level of exposure over a 70-year human life span is necessary for tumor induction. The Qua used by the EPA represents the upper bound of the 95 percent confidence interval surrounding the potency estimates. RISK ESTIMATES The estimate of risk is derived as the product of the estimate of pesticide intake and the estimate of the potency factor. Thus,
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS ~ 87 TMRC x (ma pesticide/ kg body wt/day) Q* or potency factor (excess tumor x incidence/mg pesticide/ kg body wt/day) excess tumor incidence/unit dose excess tumor incidence/unit dose. This amount is commonly very small, i.e., 0.000001 or 10-6, assuming daily exposure at this level for a 70-year period. This number means that an individual would have a 1 in 1 million risk of additional tumor induction above normal probability, assuming lifetime exposure to pesticide resi- dues at the level indicated. PERCENTAGE OF ACRES TREATED In the crop-level analyses in the scenarios below, these risk estimates were adjusted by an additional estimate of the percentage of total acres of any single crop treated with a pesticide. The percentage of acres treated represents a national average which does not take into account regional and local pesticide application and food distribution patterns. Ideally, critical components of exposure analysis would include where the pesti- cide is applied as well as the distributional pattern of produce within the country. Scenario Analyses The analysis of changes in the distributions of risks and benefits associated with alternative regulatory scenarios is a critical component of this report. Four scenarios are considered and distinguished by various risk levels that trigger regulatory prohibitive action. The threshold risk level may vary between tolerances for raw and processed foods within any scenario. The scenario analyses were performed using dBASE files containing all pesticide-commodity combinations for the 53 compounds identified by the EPA as potential oncogens. Scenarios dependent on quantitative risk levels were conducted only on the 28 compounds with Q*'s. As noted in Chapters 2 and 3 of the report, it is virtually impossible to distinguish which portion of the raw commodity will be sent to the fresh produce market and which portion will be processed. In designing these scenarios, the committee adopted the EPA's assumption that a regulatory strategy is impractical if it denies processed-food tolerances while allow- ing raw-commodity tolerances, in the expectation that residues in the raw foods will somehow not make their way into the processed-food forms.
I88 APPENDIX B Scenario 1 revokes all tolerances for all oncogenic pesticides. Scenarios 2, 3, and 4 require the identification of combined raw- and processed- commodity risks associated with any individual pesticide under carefully defined conditions that vary among scenarios. For example, Scenario 2 requires the cancellation of all processed-commodity tolerances with a risk greater than zero, along with the cancellation of all raw commodity tolerances that are associated with the canceled processed-commodity tolerances. In this case, if the risk from pesticide X in apple juice exceeded zero, then all tolerances for pesticide X in all apple products would be canceled. Scenario 3, by contrast, requires the cancellation of both raw- and processed-commodity tolerances for any individual pesticide, if the combination of raw-commodity and processed-commodity risks exceeds a threshold probability of 1 x 10-6. Finally, Scenario 4 requires the revocation of both raw- and processed-commodity tolerances for any individual pesticide if and only if the risk associated with the processed commodity tolerances exceeds the threshold probability of 1 x 10-6. The ultimate purpose of these scenarios is to estimate and compare the amount of risk; the number of pesticides, crops, and tolerances; and the percentage of total pesticide expenditures that would be affected by the application of the regulatory thresholds described above for each sce- nario. The calculation of these estimates required the development of several new data fields in the dBASE files: · Chemical-Crop Risk (RISKCCAJ, which is the summation of tolerance-specific risks for all raw- and processed-commodity tolerances associated with any specific pesticide-commodity combination (for exam- ple, all tolerances for captafol on apple products); · Chemical-Crop Processed Risk (RISKCCPJ, which is the summation of tolerance-specific risks for all processed-commodity tolerances for any specific pesticide-commodity combination; · Tolerances Affected (TOLAFFJ, which is the summation of toler- ances affected by applying the regulatory standard in each scenario; and · Crops Affected (CROPAFFJ, which is the summation of crops affected by applying the regulatory standard of each scenario. The creation of these data fields required the design of a new uniquely defined chemical-commodity code (CASRAC), which is a combination of the unique chemical code CASWL and the commodity code EPARAC. An example of the file structure is shown in Table B-1. To perform the scenario analyses, the file is sorted on the CASRAC field and the records affected by a particular scenario are displayed. Affected risks, tolerances, and crops are summed within this file, based
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS al S&9 TABLE B-1 Sample of Fields Created to Analyze Distribution of Risks Associated with Alternative Regulatory Scenarios File structure used to determine effects of regulatory threshold changes TOL TOL CROP CASRAC CHEM COMMODITY TYPE RISKCCA RISKCCP AFF AFF 04001075 CHEMX APPL RAW RAW .000008 .000007 3 1 04001075 CHEMX APPL JUI PROC .000008 .000007 3 1 04001075 CHEMX APPL DRY PROC .000008 .000007 3 1 11004093 CHEMY SOY RAW RAW .000006 .000004 2 1 11004093 CHEMY SOY OIL PROC .000006 .000004 2 1 File sorted on CASRAC field CASRAC CHEM COMMODITY 04001075 CHEMX 11004093 CHEMY TOL TYPE RISKCCA RISKCCP .000007 .000004 APPL RAW RAW .000008 SOY RAW RAW .000006 TOL CROP AFF AFF 3 2 on the specific conditions defined by the scenarios. For example, scenario 2 would require RISKCCA, TOLAFF, and CROPAFF to be summed conditionally for instances where processed-commodity risk (RISKCCP) exceeds zero. In the example in Table B-1, since both cases meet the condition specified, total affected risk would be 0.000014, total number of tolerances affected would be 5, and total crops affected would be 2- apples and soybeans. Scenario 3 would require that RISKCCA, TOLAFF, and CROPAFF be summed conditionally for cases where total pesticide-commodity risk exceeds the threshold probability of 0.000001. Scenario 4 would require that RISKCCA, TOLAFF, and CROPAFF be summed conditionally for cases in which RISKCCP (processed- commodity risk) exceeds the threshold probability of 0.000001. The scenario criteria were applied to three subsets of the master dBASE file: 1. Chemical type analyses. The effects of applying each regulatory scenario to types of pesticides (herbicides, fungicides, and insecticides) were estimated by performing the calculations described in the previous paragraph with the added condition that only the specific chemical class of interest be used as the basis of calculation. 2. Individual chemical analyses. The effects of the various regulatory scenarios on individual pesticides were estimated by first creating a file of unique pesticide-commodity records. Then, within this file, RISKCCA (total pesticide-commodity risk) was summed conditionally based on the scenario-specific criteria wherein RISKCCP > 0 (scenario 21; RISKCCA
190 APPENDIX B ~ 0.000001 (31; RISKCCP > 0.000001 (41. Again, affectedrisk, affected tolerances, and affected crops were all calculated for each pesticide. 3. Crop analyses. Finally, the effects of applying the regulatory sce- nario were estimated for eight crop groups of particular interest: apples, corn, cotton, grapes, peanuts, potatoes, soybeans, and tomatoes. The scenario-specific risk thresholds were again applied to calculate affected risk and affected numbers of tolerances, as in the previous two cases. In addition, percent affected acre treatments, and percent affected total pesticide expenditures were calculated for each crop. UNCERTAINTY IN ONCOGENIC RISK ESTIMATES This report includes numeric estimates of dietary oncogenic risk based primarily on tolerance, consumption, oncogenic potency, and percentage of crop acres treated data. This section briefly describes the types and ranges of uncertainty that surround these estimates. All risk estimates other than the crop-level estimates adjusted by the percentage of crop acres treated represent a conservative upper-bound calculation of the additional oncogenic risk across the U.S. population from exposure to any oncogenic agent. Conservative upper-bound esti- mates or "worst-case" estimates are used primarily to allow for uncer- tainties in the independent variables that determine the risk estimate (residues/tolerances, consumption, acres treated, and chemical potency). Residue Estimates All risk estimates in this report assume that chemicals exist in com- modities at tolerance levels when consumed. The results of the FDA's Market Basket surveys indicate that residues very rarely occur in raw commodities at the tolerance level; they are more commonly at levels of less than 50 percent of the tolerance. Similarly, residue levels can be affected by the method of food preparation; for example, boiling certain vegetables can volatilize water-soluble chemical residues. Therefore, it seems reasonable to conclude that the assumption of exposure at full tolerance is a highly conservative, very-low-probability event. However, this is not always the case. An October 1986 report by the U.S. General Accounting Office (GAO) found that 3 to 4 percent of all foods sampled by the FDA contained violative residues.' Further, some pesticide metab- olites and conversion products are known to increase during food storage and cooking. One of two assumptions could have been made when calculating risks that residues exist in commodities at a level detected by the FDA's
METHODOLOGYFOR ESTIMATING ONCOGENIC RISKS 191 sampling surveys, or that residues exist at the tolerance level in commod- ities. If one were primarily interested in calculating probable past exposure and resulting risk, use of a residue survey, if statistically valid, would seem reasonable. A problem then arises, however, when attempting to use past residue data to project future risk, particularly for pesticides with tolerances far in excess of residues actually detected. Use of such survey data could underestimate risk in some cases; further, residues could theoretically rise to the tolerance level without triggering a regulatory response. In contrast, the assumption that residues will exist at tolerance levels will likely overestimate risk, except for new compounds for which tolerances have been set close to anticipated residue levels. The method of risk estimation adopted for this study assumes that residues occur at the tolerance levels, which the committee deems reasonable in the absence of other comprehensive and validated data sets on actual residue levels. Food Consumption Estimates In all cases, the U.S. mean consumption estimate has been used to calculate risk in this report. The mean consumption estimate has been adjusted by the standard error, so that it actually represents the outer bound of the 95 percent confidence interval for the population mean consumption estimate. One can therefore conclude that there is a 95 percent probability that if a similar-sized sample were surveyed from the same population at the same time, the estimate of the population mean consumption level would not exceed the original outer-bound estimate. However, one cannot conclude that 95 percent of the individual consump- tion reports will fall below this level. That outer bound would likely be far higher. It is clear that using the U.S. average consumption estimate alone will inaccurately estimate food consumption for many population subgroups. For example, infants have a low level of diversity in their diets, and their consumption of fruits and fruit juices (in grams/kilogram body weight/day) is far higher than the U.S. average consumption estimate. Table B-2 demonstrates these differences for fresh apple and apple juice consumption. As can be seen, some population groups may consume eight times the U.S. average estimate for certain foods if measured on a milligram/ kilogram body weight/day basis. The certainty that U.S. mean consump- tion will rarely be exceeded therefore seems quite low. Specific age group-commodity combinations may exceed the U.S. mean consumption estimate by as much as an order of magnitude.
~ 92 APPENDIX B TABLE B-2 Apple Consumption by Age Class (g/kg body weight/day) U.S. Nursing Nonnursing Children Children Average Infants Infants Ages 2~ Ages 7-12 Fresh apples 0.457 2.203 2.854 1.228 0.762 Apple juice 0.222 2.517 3.464 0.994 0.198 NOTE: Differences in mean consumption estimates between subpopulations will result in differences in chemical intake estimates or TMRC. SOURCE: USDA 1977-1978 food consumption survey. On the other hand, the exposure model used for risk assessment (the TMRC method described above) also assumes that every individual consumes some portion of every food form every day for 70 years. Benomyl, for example, has nearly 100 food tolerances, and the exposure model assumes that each individual is exposed to benomyl as a residue in each of these 100 different foods every day for an average lifetime. While it is prudent to consider this a possibility, its probability is extremely remote, since most people do not eat all these foods and most chemicals are used for far less than 70 years. Finally, the estimate of outer-bound consumption used in this study is based on a standard error adjustment of the mean consumption data. This procedure is appropriate for estimating the 95 percent outer-bound level for population consumption means, but the estimate will be far lower than would an estimate of 95 percent outer-bound consumption based on means of individual consumption data adjusted by the standard devia- tions. For this reason, the consumption estimates used here for individual foods are likely close to the mean. Given a sample size of over 30,000 people, something approaching a normal distribution can be assumed, suggesting that roughly 45 percent of the population will be consuming higher levels than estimated in this study. Acreage Treated An additional source of uncertainty in risk estimation is the fact that pesticide exposure estimates are not adjusted for likely geographical patterns of pesticide application and food distribution. For all but the above-described analyses of chemical levels in crops, the committee assumed that all acres of all crops were treated with all pesticides for which tolerances were available. In most cases this method of risk estimation is also used by the EPA. In certain cases, however, an average percentage of total acres treated is incorporated into the risk estimate and
METHODOLOG Y FOR ESTIMA TING ONCOGENIC RISKS 193 risk assessment process. For example, there may be instances in which a particular apple fungicide is used only in one part of the country, and the apples from this area are distributed only within a several-state region surrounding the production site. By incorporating the percentage of acres treated into exposure and risk estimates and by assuming a random national distribution of these treated apples, the EPA would arrive at theoretical average exposure estimates that would tend to overestimate fungicide exposure to the portion of the population living outside the distribution region but underestimate exposure to the portion living within the distribution region. The percentage-acres-treated statistic was used to adjust the risk estimate for the eight crop analyses described above. These percentages represent an average of three years, generally between 1981 and 1984 but occasionally including 1985 use data. For soybeans, cotton, and corn, 1983 data were not used because of acreage reductions under the Payment-in-Kind program. The following method was used for these analyses: if 500,000 acres were planted with apples and 20 percent of those acres were treated with benomyl, the apple-benomyl risk estimate would be reduced by 80 percent (that is, the risk would be multiplied by 0.2~. To assume that the risk to all individuals is actually reduced by 80 percent requires an assumption that all benomyl-treated apples are evenly distributed throughout the population. This is obviously a gross oversim- plification of probable exposure. If percentage-of-acres-treated data are incorporated, a theoretical U.S. average oncogenic risk can be estimated, but it will disregard the high probability that regional populations will be exposed at far higher levels. For example, if the acres-treated estimate for an apple fungicide is 0.1 and this estimate is used to adjust the risk estimate, it is probable that risk to individuals who eat the treated apples will be underestimated by an order of magnitude. If it is assumed that all acres are treated, however, then risk will be overestimated for a large percentage of the population. Crop-level risk estimates adjusted by the acres-treated data underes- timate upper-bound risk for some percentage of the U.S. population. Individual chemical risk estimates and pesticide group risk estimates for herbicides, insecticides, and fungicides assumed that 100 percent of the acres were treated and will probably overestimate risk for some per- centage of the population. The degree of over- and underestimation will be directly related to (though only partially controlled by) the degree that pesticide use and foods are evenly distributed around the country. This dispersion will vary considerably among individual pesticide-crop combinations. In all cases, the degree of uncertainty is inversely related to the percentage of acres treated; for example, if 100 percent of all corn acres are treated with
194 APPENDIX B atrazine, the uncertainty associated with the percentage-acres-treated statistic is negligible. Toxicological Certainty Oncogenic potency factors (Q*) were used as primary indicators of toxicity in this report. They were derived primarily from the linearized multistage low-dose extrapolation procedure used by the Office of Pesti- cide Programs, yielding upper- and lower-bound estimates of excess tumor incidence/unit of dose statistic. In order to introduce a margin of safety into the risk assessment process that will in part compensate for (1) uncertainties in characterizing the oncogenic response, (2) the existence of sensitive individuals in the population, and (3) possible synergism of pesticides or metabolites, the upper bound of the 95 percent confidence interval, or Q*, is used as the potency factor by the EPA and in all cases for this study. Most potency factors used to estimate risk in this study are averages derived from the results of several positive oncogenicity studies. The Q*'s derived from individual studies on the same compound can vary by an order of magnitude or more. Interspecies extrapolation modeling is a predictive device based on the best available evidence. Perhaps its greatest value lies in the ability to compare relative risks associated with individual chemicals and among clusters of possible chemical substitutes, provided the chemicals being compared were all tested in a similar manner. Conclusion Worst-Case Scenario The certainty surrounding oncogenic risk estimates is directly related to the uncertainty associated with components of the risk equation: tolerance x consumption factor x potency factor x percent acres treated = risk. In drawing conclusions based upon this methodology, it is useful to remember a fundamental principle of probability: the probability of any outcome is the product of probabilities of independent variables that are believed to influence that outcome. Consider the case in which we are 95 percent confident that each component of our risk equation is an upper-bound estimate (i.e., 95 percent of the cases will be less than the level cited). The resulting probability of our risk estimate is then (0.95)T x (0.95)C x (0.95)A= (0.81)RISK. In summary, of the four components of the risk estimate, the use of the tolerance rather than residue levels is the factor most likely to overesti-
METHODOLOGY FOR ESTIMATING ONCOGENIC RISKS 195 mate risk since residues are commonly far below tolerance levels. The food consumption estimates used may cause risk to be underestimated because of differences in diet among subpopulations. When acreage- treated adjustment is applied as a U.S. average, exposure of regional populations can be underestimated, depending upon chemical use and food distribution patterns. Finally, it is extremely difficult to characterize the certainty surrounding the oncogenic potency factor (Q*) other than to recognize that it represents the conservative upper bound of the number of excess tumors per unit dose at the 95 percent confidence interval. NOTES 1. Code of Federal Regulations. 1986a. Title 21, Part 193. Tolerances for pesticides in food administered by EPA. Washington, D.C.: U.S. Government Printing Office. 2. Code of Federal Regulations. 1986b. Title 40, Part 180. Tolerances and exemptions from tolerances for pesticide chemicals in or on raw agricultural commodities. Washington, D.C.: U.S. Government Printing Office. 3. Research Triangle Institute (RTI). 1985. Documentation of Analysis and Statistical Methods Used in the Tolerance Assessment System. RTI Publ. RTI/2751/04-09F. Research Triangle Park, N.C. 4. Rizek, Robert. 1985. Nationwide Food Consumption Survey: 1977-78. Hyattsville, Md.: U.S. Department of Agriculture. 5. White, S. B., E. Crouch, and M. Cirillo. 1983. The Construction of a Raw Agricultural Commodity Data Base. Research Triangle Institute Project 252U-2123-7. Research Triangle Park, N.C. 6. U. S. Department of Agriculture. 1981. Agricultural Statistics. Washington, D.C.: U.S. Government Printing Office. 7. Magness, J. R., G. M. Marble, and C. C. Compton. 1971. Food and Feed Crops of the United States Bulletin 828. JR-4. New Jersey Agricultural Experiment Station. 8. National Research Council. 1979. 1977 Survey of Industry on the Use of Food Additives. 3 vols. Washington, D.C.: National Academy Press. 9. International Tariff Commission. 1984. Spices and herbs. In summary of Trade and Tariff Information. Publication 288. From personal communication with John Reeder, Washington, D.C. 10. Research Triangle Institute, 1985, Appendix D. 11. Research Triangle Institute, 1985. 12. Id., p. 22. 13. Id., p. 25. 14. Id., Appendix D. 15. U.S. Food and Drug Administration. 1986. Annual Report of Food Residues in Domestic and Imported Commodities: 1985. Washington, D.C. 16. U.S. Environmental Protection Agency. 1986a. Guidelines for estimating exposures. Federal Register 51: 185. 9-24-86. 17. U.S. Environmental Protection Agency. 1986b. Guidelines for carcinogen risk assess- ment. Federal Register 51:185. 9-24-86; Research Triangle Institute, p. 50. 18. U.S. General Accounting Office. 1986. Pesticide Need to Enhance FDA's Ability to Protect the Public From Illegal Residues. GAO/ACED 87-7. Washington, D.C.: U.S. General Accounting Office.