Sir, Roberts et al. (BDJ 2008: 204: E7) describe a simple method of dental age assessment which they claim provides a rapid and accurate estimate of age in children and emerging adults. This topic is now a hot potato politically, as the Home Office UK Border Agency is keen to apply the method to estimate the ages of asylum seekers, to decide whether they are a child or an adult based on a cut-off of 18 years. In the light of this political sensitivity the robustness of the method needs examining.

The method uses a previously assembled database of children whose ages and tooth stages are known. For a new subject, each of their teeth in turn is associated with the mean age of database children with the same tooth stage, and these 'tooth' ages are averaged across the child's teeth to estimate their chronological age. Roberts et al. show that the resulting mean age has a narrow confidence interval (CI) and hence high precision. However, what they do not say is that this CI applies to the population mean age, not to the age of the index child. To obtain the CI for the child, which is the more relevant statistic than the mean, and which is up to five times wider than the population mean CI, requires an additional calculation which they do not discuss. Thus the precision claimed for the method is exaggerated.

Another weakness of the method is its inability to handle mature stage H teeth. A mean age cannot be ascribed to such teeth (as the age range is unbounded upwards), so they are simply omitted from the calculation. As a result the estimate of the child's age is biased downwards. For example a young woman with 13 teeth at stage H and three at stage G has her age estimated from just the stage G teeth, ignoring the evidence of her many mature teeth.

A third and related problem with the method relates to the phrase 'emerging adult' in the title. Once children are over 16 years or so, all their teeth bar the third molars are mature and hence ineligible. Thus for emerging adults the ageing relies on the third molars, of which there are between none and four (as against 16 for a younger child), and as a result the precision is much reduced. Stages G and H together bracket the age of 18 years, stage G having a mean age of 17.5 years and an associated 95% CI of ±2.8 years, with stage H somewhat later. Yet stage H can be seen in emerging adults as young as 15 years, while stage G is seen as late as 23 years. Thus the age range of stages G and H is too wide to provide useful evidence of an individual being under or over 18 years.

In summary, the methodological weaknesses of the method described by Roberts et al. make it unsuitable for assessment of dental age in individual children, and particularly in emerging adults.

The authors, Graham Roberts, Aviva Petrie and Victoria Lucas, respond: We thank Professor Cole for his interest in our paper and respond to his comments in order. We share some of his concerns and ask him and your readers to be aware that this is 'work in progress'. We draw attention to the section in our Discussion which stated that ' other investigators are encouraged to repeat and extend this work to determine whether there is a general applicability of the method'. We also stated that there will be ongoing research aimed at extending and improving the method.

Professor Cole pointed out that we omitted to say that the confidence interval for the population mean age is substantially narrower than the relevant interval for the age of an individual child. We apologise for any misinterpretation of the results that may have arisen as a consequence. Readers should be reassured that we recognised this shortly after the publication of the paper and so the confidence interval for the mean has not been used in practice in relation to an individual child's estimate for some time. Furthermore, there is work in progress to provide appropriate measures of precision which relate to the estimated age of a particular child.

Professor Cole's letter does not make any reference to the section entitled 'Testing the method for accuracy'. We remind readers that in this section we applied our method by reusing, with blind assessment, the radiographs of 50 patients of known chronological age. These cases did not form part of the database. We found that the maximum likely difference between chronological age and our estimated dental age was 1.65 years and, on average, the DAA overestimated the age by 0.29 years.

This brings into question Professor Cole's comment that the omission of Stage H biases age assessment downwards. Because tooth growth is an ordered process, genetically controlled, the omission of Stage H will not influence the estimation of age from the remaining teeth which have not yet completed growth. As an extreme example, consider the LL6 which, as determined by our current database, will finish its growth (ie reach Stage H) at the latest by 11.07 years. Therefore using only age information on a LL8 at Stage G, which has a minimum age of 14.41 years, will not underestimate dental age. It is interesting to note that Professor Cole, in effect, supports not using Stage H by stating in the penultimate paragraph of his letter that ' once children are over 16 years or so, all their teeth bar the third molars are mature and therefore ineligible'.

We agree with Professor Cole that the age ranges of Stages G and H of third molars make it difficult to provide useful evidence of an individual being under or over 18 years of age. This has led to further work in this contentious area which enables us to estimate the probability that an individual is over 18 years of age. For example, the probability of an individual being over 18 years of age with LL8G is 26.6% or with LL8H is 67.0%.

We are grateful to Professor Cole for his attention to detail. Such a critical approach can only lead to improved reliability of age estimates. It is important to recognise that, to date, our approach has provided one of the most accurate estimates of age. This is supported by the excellent results obtained when we submitted our DAA method to a 'blind' test of subjects of known chronological age. No other investigator has submitted their results to such robust scrutiny. Our database is available for any investigator who wishes to further his/her research on the subject: please contact the first author on graham.j.roberts@kcl.ac.uk.

Declaration of interests. Graham Roberts uses income from the DAA clinic at King's College Hospital to fund research in this area. He also has a pecuniary interest in this funding. He is co-supervisor, with Professor Cole, of Susan Parekh's PhD. Victoria Lucas conducts research that is funded by the DAA income and also has a small pecuniary interest in the DAA income. Aviva Petrie has no competing interests to declare.