Understanding GLP Carcinogenicity Studies: OECD 451 & 116

The Organisation for Economic Co-operation and Development (OECD) is an inter-governmental organisation which aims to facilitate global trade and economic development. A key area is the development of global standards. The OECD’s Good Laboratory Practice (GLP) scheme is an example of such standards: It provides guidance on all aspects related to  non-clinical health and environmental studies which includes toxicity studies.

One aspect of a toxicity study is the analysis of carcinogenicity – the propensity for a chemical to cause cancer in a subject. As a GLP-licensed statistical provider, Quantics has been involved in carcinogenicity studies for a range of products. We want to outline some of the understanding we’ve gained over this experience by examining OECD 451 and OECD 116, giving you a sense of what is required in your next carcinogenicity study from a statistical point of view.

The OECD outlines guidance for conducting a carcinogenicity study in guidance document OECD 451, which is supported by guidance document OECD 116. OECD 451 outlines the full procedure of a carcinogenicity study, including study preparation, dosage and endpoints. OECD 116 expands on this information, providing more detailed information for these aspects as well as, importantly from our perspective, guidance on the statistical analysis of data from a carcinogenicity study.

According to OECD 116, these guidances are relevant to “a wide range of chemicals, whatever their application, including industrial chemicals, pesticides, and pharmaceuticals”. It is noted, however, that pharmaceutical products might have further testing requirements outside of the scope of guidances produced by the OECD.

At the outset, it is important to note that the study outlined in OECD 451 is an in vivo assay, meaning it involves the use of animal testing, typically using rats or mice. Both documents emphasise the importance of ensuring that test animals are treated humanely and that the “3 Rs” – Replacement, Reduction, and Refinement – should form an integral part of the considerations of any carcinogenicity study.

The goal of a carcinogenicity study is to understand the relationship between prolonged exposure to a substance and cancer in test subjects. The study consists of groups of animals, each of which receives a different dose of the substance in question. These dose groups are run in parallel with a control group. Each of these should contain sufficient animals included to allow for a thorough statistical evaluation – OECD 451 recommends at least 50 for each sex.

The method of exposure to the test substance will vary depending on how it is likely to be encountered by humans. This is often an oral dose mixed with the subjects’ food or water, but aerosol or dermal dosage are also recognised, though requiring of more complex apparatus. In the case of oral dosage, the subjects are given the test substance once a day, typically for a period of 24 months, with some triggers for early termination such as if the number of survivors in the lowest dose group or the control group decreases below 25% of the original population.

During the trial period, each subject is regularly weighed and examined for signs of ill health or visible cancers. Should any of the latter appear, then information about the growth, location, and spread of the cancer is also recorded. Should any subject die naturally or be euthanised during the study, then a full necropsy is performed to further examine for any internal cancer growth. This is performed for all subjects at the end of the trial period.

The process for a statistical analysis of data from a carcinogenicity study is outlined in OECD 116. Here, we find a key distinction between two classes of tumours. The first and most obvious are those tumours which themselves killed a subject, known as fatal tumours. Then, there are tumours that are detected when a subject is euthanised at the end of the study or when the subject died of another cause. These are known as incidental tumours, since the death of the subject is not attributable to the tumour. The classification of the tumour is known as its context of observation (COO).

In such trials, there are several different methods of statistical analysis available. The simplest is a binary analysis: did the subject die or survive the study? While this used to be standard practice, it is no longer recommended. Compared to other, more statistically advanced methodologies, a binary analysis tends to be less statistically powerful, meaning more animals are required to achieve necessary precision. In addition, such an analysis can even lead to incorrect conclusions. Animals in lower dose groups might live longer, meaning the incidence of certain cancer types might be higher.

A more commonly used method, and one which is recommended in the guidance, is known as Peto analysis, named for its British inventor, the statistician and epidemiologist Richard Peto. He devised an analysis method to avoid the problem mentioned above. In this methodology, the COO of a tumour is of the utmost importance.

That’s because, for a fatal tumour, we can perform a time-to-death analysis, where the endpoint of interest is how long it takes for a subject to die from a cancer. But, since an incidental tumour is, by definition, non-fatal, the death of the subject is caused by factors which are independent of the tumour. This means that a time-to-death analysis is not appropriate.

We could perform these analyses separately, but this would not give us a strong overall picture of the carcinogenicity of the test substance. At the end of the day, we need to answer the question whether a certain dose of a substance increases the probability of developing cancer, taking into account evidence from fatal and from incidental tumours.

There is also a time component for prevalence: an incidental tumour discovered later should be evaluated differently from an incidental tumour discovered earlier, because an incidental tumour at a later stage also means that the animal survived until then. To accommodate this, the study period is typically divided into a few larger time intervals, and animals which die in a given period are used in a prevalence analysis for incidental tumours in that selected time interval.

A Peto analysis allows us to combine a time-to-death analysis and a prevalence analysis into one, providing an overall assessment of the carcinogenicity of the substance when compared to the control. This assumes that the control and treatment animals are equally likely to be killed at any stage in a tumour’s development, that any animals which die of other causes are representative of all animals which have survived to that point, and that the pathologist examining the subjects is prepared to make a distinction between fatal and non-fatal tumours.

An important consideration when using the Peto analysis is that its results are very sensitive to misclassification of tumours. Specifically, the authors state that if fatal tumours are misclassified as incidental, then the substance can appear less carcinogenic than it actually is. And, conversely, recording incidental tumours as fatal will make the substance appear more carcinogenic.

Peto et al’s 1980 paper gives an example of an experiment where the classification of tumours was the difference between observing an effect and not. Namely, a test of whether cancer developed in the pituitary glands of rats dosed with N-nitrosodimethylamine (NDMA) was found to only be interpretable if the tumours were properly classified. While NDMA is known to be highly carcinogenic in the liver, an appropriate analysis would show that it has no effect when the pituitary gland is specifically in question.

However, if a prevalence methodology was used for all tumours, then the test resulted in a statistically significant indication that NDMA reduced the onset rate of pituitary tumours. And, if a time-of-death analysis was used for all tumours, then the test returned a significant indication that it caused pituitary tumours.

As with most problems, that of the statistical analysis of carcinogenicity data is one with many solutions. OECD 116 suggests that multivariate regression techniques such as poly-k testing can also be used to analyse such data. Nevertheless, when compared directly to Peto testing, it is recommended that a Peto analysis is preferred so long as the pathologist can accurately differentiate between incidental and fatal tumours. If this is not possible, then an alternative such as a poly-3 test is more appropriate.

Quantics has been involved in a wide range of GLP studies for more than 20 years. We are the only independent statistics consultancy in the UK to be licensed as a GLP test site. You can find out more about our services here, or visit our contact us page to get in touch.