Skin Deep currently contains information on 8,892 personal care product ingredients, culled from ingredient labels on products and from the scientific and industry literature on personal care products. We assign a standardized name to each ingredient in the Skin Deep database, generally taken as the International Nomenclature for Cosmetic Ingredients (INCI) standard, with some exceptions where alternate names are more easily recognized by consumers. Each of these ingredient names is associated with a unique ingredient identification number in our database. The processing steps for ingredients are described below.

Skin Deep's ingredient database is constructed from the sources listed below:

• Product ingredient listings. EWG researchers have parsed ingredient lists from products contained in the Skin Deep product database to construct a database of all unique ingredients listed on the product labels. We have reviewed each ingredient, corrected misspellings in ingredient names, and combined ingredients that are synonyms into a single unique chemical that is assigned a unique chemical identification number in Skin Deep. The database currently contains 152,580 unique chemicals. This means that each ingredient is shown an average of 23 different ways (various spellings and synonyms) on the labels of the various products containing it. Through this work we see that companies routinely ignore FDA guidance for standard ingredient nomenclature.
• Industry ingredient listings. The International Cosmetics Ingredient Dictionary and Handbook (Tenth Edition, 2004, published by the Cosmetic, Toiletry, and Fragrance Association, Inc.) contains "name monographs" for 491 ingredients listed under their International Nomenclature for Cosmetic Ingredients (INCI) designations. EWG researchers include these ingredients in the Skin Deep ingredient database.
• Industry-reviewed ingredients. The personal care product industry's (Personal Care Product Council's) internal safety panel, the Cosmetic Ingredient Review, had assessed the safety of _ ingredients as of their latest compendium publication (Cosmetic Ingredient Review (CIR), 2009 CIR Compendium, Washington, DC). EWG researchers entered each of these ingredients (and accompanying safety findings made by the panel) into Skin Deep's ingredient database.
• Ingredients from toxicity, regulatory, and study availability databases. We imported additional ingredients into the database from the nearly 60 data sources we have compiled on the toxicity, regulatory status and study availability of chemicals in personal care products. These sources are listed below under a section titled "Data Sources."
• Manufacturer-entered ingredients. Skin Deep contains information on ingredients in products entered into the site by manufacturers via data entry tools we make available to companies that have signed the Compact for Safe Cosmetics via the Campaign for Safe Cosmetics. Most of these ingredients are also found in other ingredient sources listed above; some are unique, found in none of our other data sources.

Creating electronic ingredient lists from product labels

EWG parses (electronically separates) individual ingredients from the ingredient list on each product. The parsing tools we have designed do not yet take into account every variance in labeling and will likely never be able to account for the wide range of errors in spacing and delimiting found on ingredient labels. Because of these factors, EWG staff members carefully review parsed ingredient lists and manually correct ingredients that are not accurately separated, and routinely update the parsing program to improve accuracy prior to the manual review.

Assigning unique chemical name and identification number to ingredients

Each ingredient is assigned a unique name and identification number within Skin Deep's ingredient database, in a multi-step process that involves resolving chemical synonyms and names that are misspelled or that do not follow standard industry naming conventions:

• For ingredients on product labels or company entries that exactly match the spelling of one of our core database ingredients or its synonyms, we assign that ingredient the existing chemical name and identification number from our ingredient database.
• For ingredients that do not exactly match an existing ingredient name, we identify if the new ingredient is a misspelling of an ingredient in our core database. We process the new ingredient with a program that identifies possible name matches, first through a match of Chemical Abstract Service (CAS) identification numbers when they are available, and secondly, through a custom, chemical-name spell checker. Then through a manual review of these tentative matches, we either match the new ingredient with one already existing in our database, or we determine that it is a new, unique ingredient, and assign it a new ingredient identification number. This manual review draws from a variety of sources, including our reviews of cosmetic industry literature (such as the Cosmetic Ingredient Review articles (CIR 2009), the International Cosmetic Ingredient Dictionary and Handbook (CTFA 2004), or one of several other chemical databases. We also routinely contact companies with questions about their ingredients.

Additional data stored with a product's ingredient list

In the Skin Deep ingredient database EWG also stores information how each ingredient is used in each product - for instance, its status as an "active ingredient" in the product; its listing under "may contain," or "organic" under USDA standards; or its association with modifiers that indicate manufacturing methods, like USP for United States Pharmacopeia standards or NF for National Formulary standards.

EWG created a core, integrated database of chemical hazards, regulatory status, and study availability by pooling the data of nearly 60 databases and sources from government agencies, industry panels, academic institutions, or other credible bodies. Collectively, these data sources detail more than 1,535 unique chemical classifications. EWG uses these databases to assess potential health hazards and data gaps for cosmetic ingredients. Individual toxicity, regulatory, and study availability data sources we compiled are listed below.

Primary references - Known and probable carcinogens, reproductive and developmental toxicants

• ACGIH (American Conference of Governmental Industrial Hygienists) 2004. ACGIH cancer classification system. www.acgih.org. [1]
• California EPA (California Environmental Protection Agency). 2006. Office of Environmental Health Hazard Assessment. Safe Drinking Water and Toxic Enforcement Act of 1986. Chemicals known to the State to cause cancer or reproductive toxicity. [2]
• ECB (European Chemicals Bureau). 2006. Classification and Labelling: Chemicals: Annex I of Directive 67/548/EEC. [3]
• EPA (U.S. Environmental Protection Agency). 2006. Integrated Risk Information System (IRIS). Evidence for human carcinogenicity based on 2005 guidelines. [4]
• IARC (International Agency for Research on Cancer). 2006. Overall Evaluations of Carcinogenicity to Humans, as evaluated in IARC Monographs Volumes 1-88 (a total of 900 agents, mixtures and exposures). [5]
• NTP (National Toxicology Program). 2005. Eleventh Report on Carcinogens, Eleventh Edition; U.S. Department of Health and Human Services, Public Health Service, National Toxicology Program. [6]

Secondary references - Known and probable carcinogens, reproductive and developmental toxicants

• CHE (The Collaborative on Health and the Environment). 2006. Toxicant and Disease Database. [7]
• ED (Environmental Defense). 2006. Scorecard _ The Pollution Information Site. http://www.scorecard.org. [8]
• EPA (U.S. Environmental Protection Agency). 2006. EPA Water Disinfection By-Products with Carcinogenicity Estimates. Last updated: 10 April 2006. [9]
• Grandjean P and PJ Landrigan. 2006. Known human neurotoxins from: Developmental neurotoxicity of industrial chemicals. Lancet. 2006 Dec 16;368(9553):2167-78. [10]
• NIOSH (National Institute for Occupational Safety and Health). 2006. NIOSH Carcinogens List (Potential occupational carcinogens). http://www.cdc.gov/niosh/npotocca.html. [11]
• NTP (National Toxicology Program) 2006. Chemicals Associated with Site-Specific Tumor Induction in Mammary Gland. [12]

Other health endpoints (neurotoxicity, immunotoxicity, etc.)

• AOEC (Association of Occupational and Environmental Clinics). 2006. Asthmagen compilation - AEOC exposures codes.[13]
• CIR (Cosmetics Ingredient Review). 2005. CIR Compendium, containing abstracts, discussions, and conclusions of CIR cosmetic ingredient safety assessments. Washington DC. [14]
• ECB (European Chemicals Bureau). 2006. Classification and Labelling: Chemicals: Annex I of Directive 67/548/EEC. [3]
• ED (Environmental Defense). 2006. Scorecard _ The Pollution Information Site. http://www.scorecard.org. [8]
• EPA (U.S. Environmental Protection Agency). 1987 & 2005. Office of Pesticide Programs. Inert (other) Pesticide Ingredients in Pesticide Products - Categorized List of Inert (other) Pesticide Ingredients. [15]
• FDA (U.S. Food and Drug Administration). 1993. Center for Food Safety and Applied Nutrition. Unsubstantiated Claims and Documented Health Hazards in the Dietary Supplement Marketplace. Illnesses and Injuries Associated With the Use of Selected Dietary Supplements. [16]
• FDA (U.S. Food and Drug Administration). 2000. FDA Center for Food Safety and Applied Nutrition. Cosmetics Compliance Program. Domestic Cosmetics Program. Coverage of Bovine-Derived Ingredients for Bovine Spongiform Encephalopathy (BSE). Issued July 31, 2000. [17]
• Illinois EPA (Illinois Environmental Protection Agency), 2000. Preliminary list of chemicals associated with endocrine system effects in animals and humans or in vitro. In EPA (U.S. Environmental Protection Agency), 2000. Handbook for Non-Cancer Health Effects Valuation, Appendix C. [18]
• NLM (National Library of Medicine). 2006. HazMap - Occupational Exposure to Hazardous Agents. [19]
• NTP (National Toxicology Program). 2006. NTP Center for the Evaluation fo Risks to Human Reproduction (CERHR). NTP-CERHR Reports and Monographs. [20]
• Occupational Safety and Health (NIOSH), currently maintained by Elsevier/MDL. [21]
• Colborn T, D Dumanoski, JP Myers. 2006. Widespread Pollutants with Endocrine-disrupting Effects. Updated from original listing in "Our Stolen Future" (1996). [22]

Restrictions and warnings on safe use of ingredients in cosmetics

• CIR (Cosmetic Ingredient Review). 2005. CIR Compendium, containing abstracts, discussions, and conclusions of CIR cosmetic ingredient safety assessments. Washington DC. [14]
• CTFA (Cosmetic, Toiletry and Fragrance Association). 2006. International Cosmetic Ingredient Dictionary and Handbook, 11th Edition. Color Additive Information. Washington, DC. [23]
• EC (European Commission of the European Union). 1999-2006. Enterprise Directorate-General Pharmaceuticals and Cosmetics. The rules governing cosmetic products in the European Union, Volume 1, "Cosmetics legislation." [24]
• FDA (U.S. Food and Drug Administration), 2006. Ingredients Prohibited and Restricted [in Cosmetics] by FDA Regulations. FDA Center for Food Safety and Applied Nutrition, Office of Cosmetics and Colors. Downloaded from http://www.cfsan.fda.gov/~dms/cos-210.html. [25]
• FDA (U.S. Food and Drug Administration) 2006. Center for Food Safety and Applied Nutrition. Color Additive Status List. September 2006. [26]
• Health Canada. 2007. List of Prohibited and Restricted Cosmetic Ingredients. Canada's Cosmetic Ingredient Hotlist. March 2007. [27]
• IFRA (International Fragrance Association). 2006. Codes and Standards. [28]
• Japan Ministry of Health, Labour and Welfare. 2006. Standards for Cosmetics. Evaluation and Licensing Division. Pharmaceutical and Food Safety Bureau. [29]
• SCCPNFP (Scientific Committee On Cosmetic Products And Non-Food Products). 1999. Opinion Concerning Fragrance Allergy In Consumers. . SCCNFP/0017/98 Final, December 1999; and SCCPNFP (Scientific Committee On Cosmetic Products And Non-Food Products). 2000. An Initial List Of Perfumery Materials Which Must Not Form Part Of Fragrances Compounds Used In Cosmetic Products. SCCNFP/0320/00, final May 2000. [30]

Persistent Bioaccumulative Toxics

• Aarhus LRTAP. 1998. Aarhus Protocol on Persistent Organic Pollutants (POPs). Geneva Convention on Long-range Transboundary Air Pollution, United Nations Environment Program. [31]
• EC (Environment Canada). 1994. Persistent Bioaccumulative and Toxic (PBT) Chemical Program. [32]
• EC (Environment Canada). 1994. Accelerated Reduction/Elimination of Toxics (ARET). ARET substance list of persistent, bioaccumulative and toxic chemicals. [33]
• EPA (U.S. Environmental Protection Agency). 1998. Resource Conservatin and Recovery Act (RCRA) Waste Minimization Program - priority chemicals for elimination or reduction. [34]
• EPA (U.S. Environmental Protection Agency). 1999. Toxics Release Inventory Program. PBT Chemical Rule. [35]
• EU (European Union). 2006. Persistent, bioaccumulative chemicals identified in PRIO database - a tool for risk reduction of chemicals. Secondary PRIO database - a tool for risk reduction of chemicals. Secondary Kemi. Place Published, Kemi-Swedish Chemicals Inspectorate. [36]
• Great Lakes BTS (Binational Toxics Strategy). 1997. Canada-United States Strategy for the Virtual Elimination of Persistent Toxic Substances in the Great Lakes. Appendix I - Level 1 and Level 2 substances. [37]
• OSPAR (Oslo-Paris). 2002. OSPAR List of Substances of Possible Concern. Secondary OSPAR List of Substances of Possible Concern. Secondary OSPAR. Place Published, OSPAR Convention for the Protection of the Marine Environement of North-East Atlantic. [38]
• UNEP (United Nations Environment Programme). 2001. Stockholm Convention on Persistent Organic Pollutants (POPs) - POPs Treaty. [39]

Other resources relevant to consideration of human health risks

• EPA (U.S. Environmental Protection Agency). 2005. Office of Air. The 112(b)1 Hazardous Air Pollutants List (as modified). Last modified: 12 Dec 2005. [40]
• EPA (U.S. Environmental Protection Agency). 2005. Office of Water. National Water Quality Standards Database. Clean Water Action priority pollutants. Eighth release. September 2005. [41]
• EPA (U.S. Environmental Protection Agency). 2005b. Priority PBT (Persistent, Bioaccumulative, Toxic) Chemicals. EPA Office of Prevention, Pesticides and Toxic Substances. [42]
• EPA (U.S. Environmental Protection Agency). 1998. Waste Minimization Program (RCRA) - persistent, bioaccumulative, and toxic chemicals. [43]
• EPA (U.S. Environmental Protection Agency). 1999. Toxic Release Inventory - persistent, bioaccumulative, and toxic chemicals. [44]
• EU (European Union). 2000. Water Framework Directive (Directive 2000/60/EC) - integrated river basin management for Europe. List of priority substances. [45]
• FDA (U.S. Food and Drug Administration) 2006. Food Additive Status List. Downloaded from http://www.cfsan.fda.gov/%7Edms/opa-appa.html, Oct 16, 2006. [46]
• FDA (U.S. Food and Drug Administration) 2006. Center for Food Safety and Applied Nutrition. Color Additive Status List. September 2006. [47]
• FDA (U.S. Food and Drug Administration). 2006. EAFUS [Everything Added to Food]: A Food Additive Database. FDA Office of Food Safety and Applied Nutrition. [48]

Availability of hazard and safety data

• CIR (Cosmetic Ingredient Review). 2005. CIR Compendium, containing abstracts, discussions, and conclusions of CIR cosmetic ingredient safety assessments. Washington DC. [14]
• EPA (U.S. Environmental Protection Agency). 1998. Chemical Hazard Data Availability Study. High Production Volume (HPV) Chemicals and SIDS Testing. Master Summary for the Chemical Hazard Data Availability Table. [50]
• FDA (U.S. Food and Drug Administration). 2005. EAFUS [Everything Added to Food]: A Food Additive Database. FDA Office of Food Safety and Applied Nutrition. [51]

Human exposure factors

• EWG (Environmental Working Group). 2007. National Tap Water Quality Database. EWG compilation of tap water testing data from 42 states and more than 40,000 communities nationwide, 1998-2003. Available online at http://www.ewg.org/sites/tapwater/. [52]
• EWG (Environmental Working Group). 2007. Biomonitoing findings from studies across the U.S. Human Toxome Project: Monitoring the Pollution in People. Available online at http://www.ewg.org/sites/humantoxome/. [53]
• CDC (Centers for Disease Control and Prevention. 2005. National Exposure Report. Biomonitoring data collected under the National Health and Nutrition Examination Program. [54]
• NanoWerk. 2007. Nanomaterial Database. Available online: http://www.nanowerk.com/phpscripts/n_dbsearch.php. [55]

Study/assessment availability sources

• CIR (Cosmetic Ingredient Review). 2005. CIR Compendium, containing abstracts, discussions, and conclusions of CIR cosmetic ingredient safety assessments. Washington DC. [14]
• NLM (National Library of Medicine). 2012. PubMed online scientific bibliography data. http://www.pubmed.gov. [56]

Chemical nomenclature databases

• CIR (Cosmetic Ingredient Review). 2005. CIR Compendium, containing abstracts, discussions, and conclusions of CIR cosmetic ingredient safety assessments. Washington DC. [14]
• CTFA (Cosmetic, Toiletry and Fragrance Association). 2006. International Cosmetic Ingredient Dictionary and Handbook, 11th Edition. Color Additive Information. Washington, DC. [23]
• NLM (National Library of Medicine). 1994. ChemID Plus. [57]
• EPA (U.S. Environmental Protection Agency). 2006. Substance Registry System. [58]
• USDA-NRCS (U.S. Department of Agriculture, National Resource Conservation Service). 2007. The PLANTS Database, National Plant Data Center, Baton Rouge, LA 70874-4490 USA. [59]

We cross-linked the chemicals in our ingredient database with the compounds contained in the toxicity, regulatory, and study availability databases we compiled. These pairs form the basis for the hazard assessment ratings and data availability ratings shown in Skin Deep.

For an initial estimate of the chemical pairings between these two groups of databases, we linked together chemicals when names matched perfectly between data sources; when Chemical Abstract Service (CAS) numbers matched, when they were available from our in-house CAS database; or when strings of letters matched between ingredients.

This initial process created over 195,752 initial name pairings, which we reviewed manually. We rejected pairs when corresponding ingredient names were not, in fact, the same compound. For many pairings we conducted a detailed review of the validity of the match between an ingredient and a hazard category, storing data on hazards that apply only to a particular route of exposure, for instance, for use in the hazard scores we developed for ingredients.

Introduction

Skin Deep relies on a dual rating system to inform consumers about the potential hazards and health concerns of thousands of cosmetics and personal care products:

• Hazard (concern) rating. We developed a hazard rating that reflects known and suspected hazards associated with ingredients and products. This rating considers potential health hazards but does not account for exposure or individual susceptibility, factors which will drive health risks, if any, but which are generally not available for assessment. Hazard ratings within Skin Deep are shown as low, moderate, or high concern categories, with numeric rankings spanning those categories that range from 1 (low concern) to 10 (high concern).
• Data availability rating. We developed a data availability rating within Skin Deep, primarily to describe the extent to which low hazard scores associated with some ingredients or products are based on definitive data demonstrating health, hazard, and safety or, at the other extreme, on a near absence of data either demonstrating or disproving hazard. The data availability rating - none, limited, fair, good or robust - is a combination of two factors: the scope of ingredient health, hazard, and safety data contained in Skin Deep, and the number of studies available in the open scientific literature. The rating reflects how much scientists know - or don't know - about an ingredient. Not all cosmetics chemicals have been thoroughly studied. Some may rank low for hazards but only because little research has been done. The lower the data availability, the less we know.

Database processing for dual rating system

Both of Skin Deep's rating factors (hazards and data availability) are calculated from information drawn from the nearly 60 integrated toxicity, regulatory, and study availability databases. We divided these databases into 260 individual categories ranging from "known human carcinogen according to EPA" to "skin irritant identified by the Cosmetic Ingredient Review panel." We then mapped each of these categories into one of 205 "score categories" that are detailed in the table below. We ranked each score category according to a subjective assessment of its relative importance to public health or regulatory compliance, and then assigned a weight to each score category, also described in tables below.

The hazard score on Skin Deep is calculated in steps.

We categorize the studies and data contained in Skin Deep into 17 general hazard categories: cancer, reproductive/developmental toxicity, neurotoxicity, endocrine disruption potential, allergies/immunotoxicity, restrictions/warnings, organ system toxicity, persistence/bioaccumulation, multiple/additive exposure, mutations, cellular/biochemical changes, ecotoxicity, occupational hazards, irritation, absorption, impurities, and miscellaneous.

We then assign each study/data source a score between 0 and 100 based on the weight of evidence. For example, a "known human carcinogen" is assigned a 100 for the cancer category while a "probable human carcinogen" is given a much lower score (55). The breakdown for each category may be found in tables below.

Next, we calculate overall ingredient scores in each of the 17 categories:

• For restrictions/warnings, multiple/additive exposure, impurities, and miscellaneous, we simply add up all the scores assigned in step 2. This means that the "restrictions/warnings" score for a chemical banned in Europe & Japan will be higher than for a chemical only banned in Europe.
• For the remaining categories, we apply the highest score found to that category for that ingredient. This means that if a chemical was determined to be a "known human carcinogen" by EPA and a "probable human carcinogen" by the EU, it will only receive the score for being a "known human carcinogen".

Next, we weight each category score (except absorption) and take the sum of the weighted factors as the raw hazard score. The weighting factors are listed in a table below.

We then weight the raw hazard score by the absorption category score. The absorption category score takes into account the particle size and the ability of an ingredient to enhance penetration.

Products

We return to the 17 categories to calculate a product score. For each category (except absorption), we add the highest scoring ingredient to the average score for the rest of the ingredients. For example, if a product contained water, coal tar, and sodium chlorite, the cancer score would be 100 (from the known carcinogen score for coal tar) + 15 (from the average of water [0] and sodium chorite [30 - limited evidence]). The resulting raw cancer score would be 115.

Just as with an ingredient, these category scores (except absorption) are weighted and then summed to produce a raw score. The weighting factors are listed in a table below.

We then weight the raw product score by the absorption category score. The absorption weighting factor includes an assessment of penetration enhancers, and nano-scale ingredients.

Final scaling of hazard scores

In the final step for deriving product and ingredient hazard scores, all scores are scaled from 1 to 10, with 10 corresponding to ingredients and products with the greatest concern, and 1 to ingredients and products with the least concern. We assign a score of 10 to the top 5% most hazardous products and ingredients, and then scale down uniformly to 1 for scores lower than this.

Major categories of concern and their weighting factors

We categorized every piece of toxicity and hazard information in our integrated database into one of 17 categories. We developed these categories based on our review of available data, and modeled them after a variety of toxicity classification systems developed by government, industry, and academic organizations. We assigned to each of these categories a weighting factor representing a judgment on their relative importance to and impact on human health. We assigned higher weighting factors to categories of health concern for which studies provide evidence for effects at low doses, for permanent effects stemming from exposures during development, for toxicity endpoints that tend to impact multiple biological systems in the body or to impair reproduction. These endpoints include cancer, reproductive and development effects, immunotoxicity, neurotoxicity, and effects stemming from exposures to endocrine (hormone) disruptors. Organ systems that are more localized, such as gastrointestinal, kidney, respiratory, etc. are weighted in the mid-range. Toxic endpoints that measure adverse effects at the cellular level, which may or may not have implications for human health (such as mutations or biochemical changes) are weighted the least. This scoring system does not account for individual sensitivities or differences between the severities of different health endpoints within a particular category.

Table 1: Hazard categories and weighting factors

Category	Weighting factor	Description
Cancer	1.0	linked to cancer in government, industry, or academic studies or assessments.
Developmental/reproductive toxicity	1.0	linked to developmental and reproductive toxicity, a broad class of health effects that can range from infertility and reproductive organ cancers to birth defects and developmental delays for children.
Endocrine disruption	1.0	the body's natural hormones, the chemicals that carry messages across the body to manage growth, tissue repair, and reproduction.
Allergies/immunotoxicity	1.0	linked to immunotoxicity, or harm to the immune system, a class of health problems that manifest as allergic reactions or an impaired capacity to fight disease and repair damaged tissues in the body.
Miscellaneous	1.0	Includes toxicity endpoints that didn't fit in another category, efficacy scores (scores that might counteract toxicity scores), and scores for unidentified ingredients.
Neurotoxicity	1.0	linked to neurotoxicity, or harm to the brain and nervous system, a class of health problems that can range from subtle developmental delays to chronic nerve degeneration diseases.
Use restrictions	0.9	prohibited for use in cosmetics, or subject to concentration, use, or manufacturing method restrictions, according to industry safety guidelines and government requirements and guidance from the U.S., E.U., Japan, and Canada.
Organ system toxicity (non-reproductive)	0.5	linked to toxicity of one or more biological systems in the body (cardiovascular, stomach and digestive trace, respiratory system, etc.) through laboratory studies or studies of people.
Biochemical or cellular level changes	0.3	the ability to affect the body at a cellular or biochemical level that may have larger, but poorly understood health implications.
Multiple, additive exposure sources	0.3	also found as contaminants in tap water and food, as ingredients in other kinds of consumer products, or in people in biomonitoring studies that measure chemicals in blood, urine, and other fluids and tissues.
Mutations	0.3	linked to both cancer and developmental defects. Includes government, industry, or academic assays, studies and assessments.
Persistence and bioaccumulation	0.3	persistent and/or bioaccumulative, resisting normal chemical breakdown in the environment; building up in wildlife, the food chain, and people; and lingering in body tissues for years or even decades after exposure.
Ecotoxicology	0.2	linked to toxicity of wildlife that may include fish, wildlife, plants, or other wild organisms.
Occupational hazards	0.2	linked to hazards for workers exposed on the job, including acute dangers from chemical handling, or longer term health effects from routine occupational exposures.
Irritation (skin, eyes, or lungs)	0.1	linked to irritation of the skin, eyes, or lungs according to government assessments, industry reviews, and peer-reviewed studies.
Enhanced skin absorption	0.0	an enhanced capacity to absorb through this skin by virtue of chemical properties like penetration enhancing abilities or small particle size (including nanoparticles), or by virtue of where it is applied on the body (on infant skin, lips, or damaged skin).
Decreased skin absorption	0.0	an decreased capacity to absorb through this skin by virtue of chemical properties like penetration enhancing abilities or large particle size (including nanoparticles), or by virtue of where it is applied on the body (on infant skin, lips, or damaged skin).
Data gaps	0.0	linked to data gaps that constitute the absence of basic toxicity studies and hazard and safety assessments in Skin Deep's core databases, or that reflect findings of data deficiencies in government or industry assessments.
Contamination concerns	0.1 for ingredients 0.01 for products	may be contaminated with toxic impurities, many of which are linked to cancer, according to government and cosmetic industry ingredient safety assessments or peer-reviewed studies.

We assigned numeric hazard scores for each scoring category based on professional judgment of the relative importance of each with respect to potential health concerns. These scores were informed by a number of factors, including the weight of the evidence associated with each scoring category (e.g. whether the chemical categorization is derived from a full government assessment or from a single peer-reviewed study), and by other hazard classification systems, such as the Nordic Substances Database.

For most types of hazards, we assign scores as a function of the lowest known harmful dose where that information is available, the weight of the evidence (limited, moderate, and strong evidence), and the source of the data (individual study; literature review, industry review panel, or major government study; and comprehensive government assessment). We use the scores shown below in our calculations of final hazard (concern) scores for ingredients and products, as described in subsequent sections. The tables below detail the hazard scoring system.

Table 2: Hazard scoring framework for cancer, developmental and reproductive toxicity, endocrine disruption, immunotoxicity, and multiple organ system toxicity

The scores we assign for these hazard categories range from a maximum of 100 for chemicals known to be toxic to humans in a given category as determined by a definitive government assessment, to 20 for chemicals showing limited evidence for toxicity in a non-academic review, down to 0 for chemicals determined not likely to be human toxicants.

Hazard score category	Data source	Hazard Score
Known human toxicant	Government assessment	100
Possible human toxicant	Government assessment	55
Limited evidence of human toxicity	Government assessment	30
Strong evidence for human toxicity	Literature review, industrial panel, or major government study	55
Moderate evidence for human toxicity	Literature review, industrial panel, or major government study	30
Limited evidence for human toxicity	Literature review, industrial panel, or major government study	20
One or more animal studies show effects at very low doses	Individual scientific or peer-reviewed study	30
One or more animal studies show effects at low doses	Individual scientific or peer-reviewed study	20
One or more animal studies show effects at moderate doses	Individual scientific or peer-reviewed study	10
One or more animal studies show effects at high doses	Individual scientific or peer-reviewed study	5
Not likely to be a human toxicant	Government assessment, literature review, industrial panel, or major government study	0

Table 3: Hazard scoring framework: mutations

We assign scores for mutation data essentially the same as for other hazard classifications (as in Table 2), but make modifications to account for the unique range of tests available to define mutation, as shown in this table. Scores we assign for mutation range from 100 for a known mutagen as determined by a definitive government assessment, to 10 for ingredients for which one or more studies on micro-organisms show positive mutation results, to 0 for ingredients determined not likely to be mutagens in humans based on a definitive government review.

Hazard score category	Data source	Hazard Score
Known mutagen	Government assessment	100
Possible mutagen	Government assessment	55
Strong evidence for mutagenity in human cells	Literature review, industrial panel, or major government study	55
Moderate evidence for mutagenity in human cells	Literature review, industrial panel, or major government study	30
Limited evidence of mutagenity in human cells	Government assessment	30
One or more studies on mammalian cells show positive mutation results	Individual scientific or peer-reviewed study	30
Limited evidence of mutagenity in human cells	Literature review, industrial panel, or major government study	20
One or more studies on non-mammalian cells show positive mutation results	Individual scientific or peer-reviewed study	20
One or more studies on micro-organisms show positive mutation results	Individual scientific or peer-reviewed study	10
Not likely to be a mutagen	Government assessment, literature review, industrial panel, or major government study	0

Table 4: Hazard scoring framework: biochemical and cellular level changes

We assign scores for biochemical and cellular level changes, where the human health impact may be unclear. The framework is essentially the same as for other hazard classifications (as in Table 2) with some modifications to account for the unique range of tests available to define biochemical changes and because more significant reviews generally are undertaken only on concrete human health effects. Scores we assign for biochemical and cellular level changes range from 100 for a reactive oxygen species that are beginning to be linked up to definitive health effects to 5 for ingredients for high-dose studies showing biochemical changes.

Hazard score category	Data source	Hazard Score
Produces excess reactive oxygen species that can interfere with cellular signaling, cause mutations, lead to cell death and may be implicated in cardiovascular disease.	Individual scientific or peer-reviewed study	100
Interferes with gene expresion	Individual scientific or peer-reviewed study	30
One or more animal studies show effects at very low doses where the human health implications are not yet well understood	Individual scientific or peer-reviewed study	30
One or more animal studies show effects at low doses where the human health implications are not yet well understood	Individual scientific or peer-reviewed study	20
One or more animal studies show effects at moderate doses where the human health implications are not yet well understood	Individual scientific or peer-reviewed study	10
One or more animal studies show effects at high doses where the human health implications are not yet well understood	Individual scientific or peer-reviewed study	5

Table 5: Hazard scoring framework: Assigning categories based on the Nordic Substances Database hazard potency framework

A scoring framework implemented in the 2007 update of Skin Deep accounts for detailed toxicity study findings. These findings are reviewed by EWG staff, and are drawn from the open scientific literature and from a government database containing over 6000 peer-reviewed references.

We assign hazard scores to each recorded lowest toxic dose according to weighting factors used in the Nordic Substances Database classification system, as shown in Table 5. No attempt was made to evaluate the length of time or dosing regimen (such as differentiating between chronic and sub-chronic studies) for repeat doses or accounting for route of exposure. In cases of acute studies where the LOEL (Lowest Observed Effect Level) instead of the LD50 (Lowest Dose producing 50% mortality of test animals) was reported, the doses were multiplied by a factor of 10 to estimate the LD50.

Potency	Acute Studies1 (LD50 - mg/kg of body mass)	Repeat Dose Studies2 (LOAEL - mg/kg day of body mass)
Very highly toxic - very low dose	<25	<2.5
Highly toxic - low dose	25<=Dose<200	2.5<=Dose<20
Moderately toxic - moderate dose	200<=Dose<2000	20<=Dose<200
Low toxicity - high dose	Dose>=2000	Dose>=200

• 1. LD50 refers to the chemical dose at which 50% of the animals died.
• 2. LOAEL is the Lowest Observed Adverse Effect Level which is the lowest dose of a chemical at which a harmful effect is observed in a lab animal.

Table 6: Hazard score framework: Occupational hazards

Potency	Occupational 8 hr TLVs and PELs (mg/m3)
Very highly toxic - very low dose	<3.47
Highly toxic - low dose	3.47<=Dose<27.9
Moderately toxic - moderate dose	27.9<=Dose<279
Low toxicity - High dose	>=279

Note: We derived the occupations exposure factors shown above by converting the animal study doses define potency in the Nordic Substances Database (see Table 6) into equivalent concentrations in workplace air assuming a 70 kg male, 15 breathes/minute, 0.7 liters/breathe over an 8-hour workday.

Table 7: Hazard score framework: Government or industry restrictions or guidelines

Scores in this category range from a maximum of 100 for a violation of a ban to 1 for caustic chemicals used as a pH balancer in a specific product.

Finding	Data source	Hazard Score
Banned or found unsafe for use in cosmetics	Government assessment	100
Violations of restrictions and safety warnings	Government assessment	90
Not safe in cosmetics for specific use	Literature review, industrial panel, or major government study	40
Safe use with an industry determined concentration limit	Literature review, industrial panel, or major government study	25
Safe use with specific consumer instructions	Literature review, industrial panel, or major government study	10
Safe use as a pH adjuster	Literature review, industrial panel, or major government study	1
Approved for use	Government assessment, literature review, industrial panel, or major government study	0

Table 8: Hazard score framework: persistence and bioaccumulation

Many chemicals persist in the environment and bioaccumulate in people and the environment. EWG has compiled listings of persistent, bioaccumulative compounds from authoritative bodies, for compounds considered potentially hazardous to humans or the environment.

For the hazard score category titled "Ingredients not fully identified," we assign a score of 100, flagging ingredients with unknown identity, such as fragrance and unidentified essential oils.

Finding	Level of finding	Hazard Score
Persistent, bioaccumulative in wildlife and humans	Government assessment	100
Persistent, bioaccumulative in wildlife	Government assessment	50

Table 9: Hazard score framework: multiple exposure routes

Finding	Level of finding	Hazard Score
Used in food or as an additive with limited or no toxicity information available	Government assessment	100
Contaminant in tap water	Government assessment	100
In other consumer products besides personal care products	Government assessment	50
High production volume chemical	Government assessment	50
Environmental releases by industry	Government assessment	50
Used as an inert ingredient in pesticides	Government assessment	50
Designated as safe for use in food	Government assessment	0

Adjusting hazard scores for skin absorption potential

In the penultimate step for deriving product and ingredient hazard scores, we adjust preliminary scores to account for increased potential for an ingredient to penetrate the skin. We account for the potential enhanced absorption of a product stemming from the presence of penetration enhancing ingredients, and known or potential nano-scale ingredients (NanoWerk 2007).

For nano-scale and potentially nano-scale ingredients, the ingredient scores are scaled as in Equation 3 with scaling factors of 1.5 and 1.25, respectively. The overall product scores are not adjusted. In cases where the skin absorption potential is lowered, we also scale downwards for the individual ingredients. Scores are multiplied for 0.5 for reduced or limited absorption, and 0.25 where the ingredient has been shown not to absorb into intact skin. Only for ingredients shown not to absorb into intact, damaged, infant, and thinner skin, is the factor reduced to 0.

Skin Deep's Health Concern Summary Bars

Hazard scores are reflected as bars at the top of each ingredient, product, brand, and company page. These bars represent the relative hazard rating for that ingredient or ingredient list. Using the general hazard category scores described above, the bars are filled in from left to right with a score of 0 on the left and the maximum on the right (described below). Each page in the Skin Deep online database includes a bar for overall hazard; cancer; developmental/reproductive toxicity; allergies/immunotoxicity; and use restrictions.

Ingredient Pages
The maximum value for each bar is 100. For example, a "known human carcinogen" is assigned a 100 for the cancer category while a "probable human carcinogen" is given a much lower score (55). (See more about the scoring above.) Other concerns are separated into three categories:

• "strong" - category score>60; the bar is 100% full
• "moderate" - category score>10; the bar is 66% full
• "lesser" - category score>0; the bar is 33% full

Product Pages
The calculation of category scores is the sum of the maximum value for its ingredients + the average of the other ingredients. For "cancer" and "developmental/reproductive toxicity", the bars are scaled from 0 to 100. For "use restrictions" and "allergies/immunotoxicity", the bars are scaled from 0 to 150. Other concerns are listed only if the score is >15. The bar is then scaled based on the number of other concerns that are listed with a scale from 0 to 15 possible extra concerns.

Company and Brand Pages
Hazard category scores for companies and brands are taken as the average value for ingredients used by the brand/company. Category bars displayed in Skin Deep are scaled from 0 to 100. Other concerns are listed only if the average score is >15. The bar is scaled based on the number of other concerns that are listed, with a scale from 0 to 15 possible extra concerns.

New to the April 2011 release of Skin Deep is a data availability rating that gives site users a measure of the availability of health, hazard, and safety data for any particular ingredient or product. The data availability rating - none, limited, fair, good or robust - is a combination of two factors: the scope of ingredient health, hazard, and safety data contained in Skin Deep, and the number of studies available in the open scientific literature. The rating reflects how much scientists know - or don't know - about an ingredient. Not all cosmetics chemicals have been thoroughly studied. Some may rank low for hazards but only because little research has been done. The lower the data availability, the less we know. We recommend that consumers buy products with lower hazard ratings AND at least "fair" data availability.

Products - data availability rating

The data availability rating for products is the average of the data availability ratings of individual ingredients contained in the product.

Ingredients - data availability rating

The data availability rating for each ingredient considers the scope of ingredient health, hazard, and safety data contained in Skin Deep. To reflect data availability for ingredients with relatively little information in Skin Deep, we also account for the number of studies available in the open scientific literature. In these cases, 50 percent of the data availability rating for an individual ingredient is based on the scope of Skin Deep information. The remaining 50 percent reflects the number of studies listed for that ingredient in the government's PubMed scientific research index (www.pubmed.gov).

Breakdown of the PubMed and Skin Deep portions of the data availability rating

Ingredient studies in PubMed
The method EWG uses to calculate an ingredient's data availability rating takes into account the number of ingredient studies contained in the government's index of peer-reviewed scientific journals (www.pubmed.gov). EWG determines the total study count returned by PubMed by searching the site's index for the ingredient name and all of its synonyms listed in Skin Deep. This factor constitutes up to 50% of the final data availability rating for an ingredient, with a numeric rating assigned as follows:

Table 10: DATA AVAILABILITY RATINGS BASED ON OPEN SCIENTIFIC LITERATURE

Ingredient studies contained in PubMed	Data availability for PubMed portion of data availability rating
0 or 1 study	0
2 - 100 studies	1
> 100 studies	2
> 1,000 studies	3
> 10,000 studies	4

Ingredient studies in Skin Deep
Another portion of the data availability rating reflects the quantity of ingredient studies in Skin Deep. This portion accounts for the significance of the health concern (reflected by the health hazard category weight from Table 1), and the data robustness for each piece of information in Skin Deep (weighted as listed in Table 11 below).

To calculate the score for each ingredient:

• Each piece of data in Skin Deep is assigned a data rating using the values in Table 11.
• For each health hazard category outlined in Table 1, the maximum data rating from all relevant findings is assigned.
• These values are then squared to give more weight to higher levels of data availability. The squared values then weighted with factors listed in Table 1. These weighted factors are added together.
• Finally, the square root is taken of the sum, and the resulting number is scaled between 0 and 4 for all rated ingredients. In this scaling, well-studied ingredients are set at 4 (formaldehyde, benzene and other ingredients with higher ratings than these two). Ratings for less-studied chemicals are scaled from that maximum value. The resulting scaled score from 0 to 4 is taken as the overall data availability rating for the Skin Deep data.

Final data availability rating

The final rating for the data rating is either the Skin Deep rating or the average of the Skin Deep rating and the PubMed rating -- whichever is higher. Values are rounded to produce integer values (with values >0 and <0.5 being assigned a 1 to ensure that a "no data" rating is reserved for cases where no data at all has been identified, in Skin Deep data sources or the open scientific literature. The value 0 is assigned the text "none", 1 => "limited", 2 => "fair", 3 => "good", and 4 => "robust."

Product data ratings are simply an average of the data ratings of the ingredients. The rounded values are used to assign text in the same fashion as ingredients. Values greater than 0 but less than 0.5 are assigned a value of 1 for "limited" data.

Table 11: DATA AVAILABILITY - RATINGS FOR EACH DATA SOURCE CONTAINED IN SKIN DEEP

Database	Data availability
European Union - Classification & Labelling	1-2
Amer Conf of Gov't Industrial Hygienists - Carcinogens	2-4
FDA Food Additive Status	0-3
Association of Occupational and Environmental Clinics	0-3
U.S. - Ingredients Prohibited and Restricted in Cosmetics by FDA Regulations	3-4
FDA Ingredients with Bovine Spongiform Encephalopathy Concern	1-2
Canada - Prohibited and Restricted Cosmetics Ingredients	3-4
CHE Toxicant and Disease Database	1-2
Cosmetic Ingredient Review Assessments	0-4
FDA Color Additive Status	1-2
EPA Clean Water Act - Priority Pollutants	3
EPA Water Disinfection By-Products Carcinogenicity	1-3
FDA Everything Added to Food	1-2
Environment Canada Domestic Substance List	1-2
Illinois EPA Chemicals Associated with Endocrine System	2-4
EPA Integrated Risk Information System (IRIS)	1-4
European Union - Banned or Restricted in Cosmetics	3-4
European Commission on Endocrine Disruption	1-3
EU Banned and Restricted Fragrances	4
European Union - Water Framework Directive	2
EWG Assessment of Open Scientific Literature	1
PCPC International Cosmetic Ingredient Dictionary and Handbook	2-4
IFRA Fragrance Ingredient List	1
EPA Hazardous Air Pollutants	2
FDA Dietary Supplement Concerns	2
National Library of Medicine HazMap	0-2
Int'l Agency for Research on Cancer (IARC) - Carcinogens	2-4
International Fragrance Association Codes & Standards	2-3
Impurities - Open Scientific Literature	1
EPA Categorized List of Inert Pesticide Ingredients	2
Japan's Standards for Cosmetics	2-3
Nanomaterial Database	0
Chemicals known to be neurotoxic to humans	2
NIOSH Occupational Carcinogens	3
NTP Report on Carcinogens, 11th Edition	3-4
NTP - Tumor Induction in Mammary Gland	3
NTP - Risks to Human Reproduction	4
Our Stolen Future Endocrine Disruptors	2
Aarhus Protocol on Persistent Organic Pollutants	4
Canada PBTs - Accelerated Reduction/Elimination of Toxics (ARET)	2-4
EU PTBs - PRIO database & Secondary PRIO database	3-4
Great Lakes BTS (Binational Toxics Strategy) PBTs	3-4
OSPAR PBTs - Substances of Possible Concern	3-4
Emerging PBTs from peer-reviewed literature	1
EPA PBTs - Waste Minimization Program (RCRA)	4
United Nations Environment Programme/POPs Treaty	3-4
EPA Toxic Release Inventory PBTs	4
Open scientific literature	1
California EPA Proposition 65	4
Registry of Toxic Effects of Chemical Substances	1
Scorecard.org Toxicity Information	1-2
Silent Spring's Mammary Carcinogens Reviews Database	3