/
The reliability of toxicological data decides what a cosmetic product safety report is worth. A margin of safety can be calculated to three decimals and still be wrong if the value it starts from comes from a study nobody checked, or from a summary table that never named its source. This article goes through what Regulation (EC) No 1223/2009 and the SCCS Notes of Guidance say about that starting value, and what the person reading a CPSR (a Responsible Person, a brand owner, a buyer’s quality team) can reasonably ask of it.
What the Regulation says about the reliability of toxicological data
Three provisions carry most of the weight. Annex I, Part A, point 8, asks for the toxicological profile of each substance and for a margin of safety (MoS) based on the no observed adverse effect level (NOAEL); leaving out the MoS has to be duly justified. The same point adds two sentences that matter here: any read-across between substances must be documented and justified, and the source of the information must be clearly identified. Article 10(1)(b) makes the responsible person ensure that an appropriate weight-of-evidence approach is used to review data from all existing sources. And Article 10(3) requires the non-clinical safety studies referred to in the safety assessment and carried out after 30 June 1988 to assess the safety of a cosmetic product to comply with EU legislation on good laboratory practice (GLP) in force when the study was run, or with other international standards that the Commission or ECHA recognises as equivalent.
There is no scoring scale in the Regulation, no ranking of study types and no rule for choosing between two conflicting NOAELs. The Regulation sets the frame and leaves the choice inside it to the safety assessor, which is where a weak CPSR and a strong one part ways.
What the SCCS Notes of Guidance are, and what they are not
The Notes of Guidance (SCCS/1647/22, 12th revision) were written for the safety dossiers on Annex substances that companies submit to the SCCS. Their introduction says they were not written for substances outside the Annexes, that they can nevertheless be of practical use when compiling the Product Information File of a finished product, and that they are not a prescriptive procedure. We use them in that spirit: as a detailed technical reference on how the Committee looks at data quality, not as a requirement of the Regulation. Where a point below is professional judgement and not the text of the guidance or of the Regulation, we say so.
The starting value: NOAEL, LOAEL or BMDL
The guidance calls the value from which extrapolation to low doses begins the point of departure (PoD) and notes that in most SCCS opinions it has been a NOAEL. For substances with a threshold it can also be a LOAEL, the lowest observed adverse effect level, or a BMDL, the lower limit of a benchmark dose. For its own opinions the SCCS considers the benchmark dose (BMD) approach preferable where usable in vivo data exist: it uses the whole dose-response curve and allows the uncertainty to be quantified. A CPSR built on a NOAEL is not wrong for that reason, since Annex I itself names the NOAEL. What is wrong is a bare figure: a value in mg/kg/day with no study, no endpoint and no document behind it.
Age and origin matter too. The guidance recalls that the ban on testing finished cosmetic products has applied since 11 September 2004 and the ban on testing ingredients since 11 March 2009, with the marketing ban applying from 11 March 2009, or from 11 March 2013 for repeated-dose toxicity, reproductive toxicity and toxicokinetics. It also says that animal data generated before those dates, or for other legislation such as food, medicines or biocides, may still support the assessment of a cosmetic ingredient. Many values in circulation therefore come from studies run for another purpose, so ask where a given value was obtained.
Scoring the quality of a study
To assess the quality of a toxicological study, the guidance says the Klimisch score could be used, for example, and points to ToxRTool, a software tool developed by the EU Reference Laboratory for the Validation of Alternative Methods (EURL ECVAM) that determines the Klimisch score in a systematic way and also works for in vitro studies. The guidance names the score but does not reproduce its scale, which comes from the 1997 paper it cites. It also says that when data are taken from the scientific literature, a systematic search and review following pertinent standards (PRISMA is the example given) should be performed and documented: search engine, key words, and the reason each study was included or excluded.
A CPSR does not need a full systematic review for every ingredient. The underlying idea is the best answer to the question «how do you know there is no better study?»: a search that another person could repeat. Two further points are our own criteria, not rules. A study scored as unreliable should not, on its own, carry a signed margin of safety. And because Article 10(3) applies to studies run after 30 June 1988 to assess the safety of a cosmetic product, the report should state how compliance with good laboratory practice was checked, and not leave the reader to assume it. On deviations from GLP, the guidance is explicit for the data it receives: they must be indicated, explained and scientifically justified.
Read-across and in silico predictions: support, not substitute
Models are useful and the guidance is demanding about them. A (Q)SAR model is considered appropriate for regulatory use only if it was developed under the OECD validation principles: a defined endpoint, an unambiguous algorithm, a defined applicability domain, appropriate measures of goodness-of-fit, robustness and predictivity, and, where possible, a mechanistic interpretation. The QSAR Model Reporting Format (QMRF) exists to summarise that information for each model. The guidance also quotes the 2014 report of the International Cooperation on Cosmetics Regulation (ICCR), which concluded that in silico approaches for cosmetic ingredients were largely limited to internal decision-making and not yet a mainstream alternative to testing.
Read-across gets similar treatment. Structural similarity alone is not enough: the guidance, citing ECHA, asks that similarity in physicochemical properties, toxicokinetic and toxicodynamic behaviour and toxicological effects also be considered, and that analogues be selected impartially and not by picking a few on personal judgement. Put next to Annex I, Part A, point 8, the practical reading is that a prediction or an analogue is one piece of the weight of evidence, documented and justified, and not the only foundation of a margin.
Four questions to put to any CPSR
The reliability of toxicological data in a report comes down to whether, for each substance with a calculated margin of safety, you can answer four things. Which value was used and of what type: NOAEL, LOAEL or BMDL. Which document exactly it comes from, because «SCCS» on its own is not a source and Annex I asks for the source to be clearly identified. How reliable the study was judged to be and by what method. And what the position on good laboratory practice is. That value is then set against the systemic exposure the intended use produces, which Article 10(1)(a) requires to be taken into account, and Annex I, Part B, point 3, asks for the margins of safety, where relevant, to be assessed and discussed in the reasoning. An error in the input travels through all of it without any formula noticing. If there is no valid value at all, the report has to justify the absence, and that is a different thing from filling the gap with the number of a similar substance.
Frequently asked questions
Does the SCCS guidance apply to my CPSR?
Not as a legal requirement. It is written for SCCS dossiers, and it describes itself as an approach to adapt case by case. It is nonetheless a detailed public account of how the Committee weighs data, so it is a sound benchmark for the reliability of toxicological data in a report signed by a safety assessor.
Can I use the NOAEL of a similar substance?
Only as support inside a weight-of-evidence analysis, with the similarity justified and the source identified. The assessor should not present it as the value of the substance being assessed.
What if a REACH registration dossier lists several studies?
Then the report should say which study carries the assessment and why, instead of defaulting to the first one found. Which one to choose is a professional judgement; the Regulation only asks that it be justified.
When does a value need to be revisited?
When new relevant information appears. Article 10(1)(c) requires the safety report to be kept up to date with information generated after the product is placed on the market, and a new SCCS opinion on an ingredient is exactly that.
Who answers for the value chosen?
The responsible person ensures the safety assessment is done (Article 10(1)), and Part B is carried out by a person with the qualifications set out in Article 10(2), whose name, proof of qualification, date and signature appear in the report (Annex I, Part B, point 4).
If you place cosmetics on the EU market and want to check the reliability of toxicological data behind your safety report, write to us through our contact page. You can also read our guide to the Cosmetic Product Safety Report, our note on the Responsible Person for cosmetics, our cosmetics advisory service, or the Spanish version of this article.