Elsevier

Forensic Science International

Volume 257, December 2015, Pages 512.e1-512.e7
Forensic Science International

Forensic Anthropology Population Data
Age estimation in the living: Transition analysis on developing third molars

https://doi.org/10.1016/j.forsciint.2015.07.049Get rights and content

Abstract

A radiographic assessment of third molar development is essential for differentiating between juveniles and adolescents in forensic age estimations. As the developmental stages of third molars are highly correlated, age estimates based on a combination of a full set of third molar scores are statistically complicated. Transition analysis (TA) is a statistical method developed for estimating age at death in skeletons, which combines several correlated developmental traits into one age estimate including a 95% prediction interval. The aim of this study was to evaluate the performance of TA in the living on a full set of third molar scores. A cross sectional sample of 854 panoramic radiographs, homogenously distributed by sex and age (15.0–24.0 years), were randomly split in two; a reference sample for obtaining age estimates including a 95% prediction interval according to TA; and a validation sample to test the age estimates against actual age. The mean inaccuracy of the age estimates was 1.82 years (±1.35) in males and 1.81 years (±1.44) in females. The mean bias was 0.55 years (±2.20) in males and 0.31 years (±2.30) in females. Of the actual ages, 93.7% of the males and 95.9% of the females (validation sample) fell within the 95% prediction interval. Moreover, at a sensitivity and specificity of 0.824 and 0.937 in males and 0.814 and 0.827 in females, TA performs well in differentiating between being a minor as opposed to an adult. Although accuracy does not outperform other methods, TA provides unbiased age estimates which minimize the risk of wrongly estimating minors as adults. Furthermore, when corrected ad hoc, TA produces appropriate prediction intervals. As TA allows expansion with additional traits, i.e. stages of development of the left hand-wrist and the clavicle, it has a great potential for future more accurate and reproducible age estimates, including an estimated probability of having attained the legal age limit of 18 years.

Introduction

The correlation between the degree of tooth mineralization and chronological age can be used for estimating chronological age in individuals where the age is unknown [1], [2]. In most countries, the chronological age of 18 years mark the difference between being a minor and child as opposed to an adult [3]. Above the chronological age of approximately 14.0–16.5 years, third molars are the only teeth left to mineralize [4], [5]. Although highly variable, a radiographic assessment of third molar development is essential for differentiating between minors and adults in young asylum seekers without documents to prove their age [2], [3].

The majority of reference studies investigating dental age report the correlation between chronological age and the stages of third molar development as means and standard deviations for the third molars individually: upper right (UR), upper left (UL), lower left (LL), and lower right (LR) [6], [7], [8]. Unfortunately, mean ages are affected by the age composition of the reference sample. This bias is known as age-mimicry [9]. Although small, when used for age estimation purposes this bias may affect whether the examinee is assessed as a child or an adult. For example, the mean age at stage R3/4 in UR has been reported as both 17.0 years and 18.3 years [8], [10]. Thus, simply taking the mean of means or using discrete age intervals may become attractive, although actually methodologically wrong.

A few published studies provide age estimates from combinations of a full set of third molar scores [1], [11], [12]. Unfortunately, such statistical models are complicated due to the correlation between the developmental stages of third molars, i.e. the developmental stage of one third molar carries information about the developmental stage of the other third molars. Consequently, correlation between biological traits limits the number which can be integrated in a classical regression model. Such examples can be found in Bassed et al. [13] and in Gunst et al. [1]. Moreover, due to a systematic bias (attraction of the middle), traditional regression models systematically overestimate the age of younger individuals and underestimate the age of older individuals [11]. The direction of this bias risks that minors are assessed as adults, which should be avoided [11].

Regressing biological traits on age, i.e. treating age as the explanatory variable, removes the bias of age-mimicry and the tendency to overestimate the age of young [14]. However, these estimates have to be manipulated to produce estimates of age [9]. For this purpose, the use of Bayes theorem has been proposed [9], [11], [14]. Bayes theorem states that the prior probability of a phenomenon should play a role in evaluating its posterior probability. Methods using Bayesian statistics for estimating age in the living have been published for third molars [11] and clavicles [15].

In lack of perfect correlation between a biological trait and age the Study Group on Forensic Age Diagnostics (AGFAD) and Forensic Anthropology Society of Europe (FASE) recommend to use as much information as possible, i.e. to combine several traits in order to achieve more accurate age estimates with appropriate levels of uncertainty [2], [3]. Although this multi-factorial approach is recommended, no current method combine the traits used most in forensic age estimation cases (hand-wrist, teeth and clavicles) to obtain an estimate of age including a 95% prediction interval [16]. While the Bayesian method proposed by Thevissen et al. showed promising results, it is computationally too complicated to allow for further traits to be added [11]. Transition analysis (TA) may be an important step towards a solution to this problem. TA is a Bayesian approach developed for assessing age-at-death in skeletons with the aim of avoiding age-mimicry [9]. It combines several developing age-related traits into one point age estimate including a 95% prediction interval. TA assumes conditional independence, i.e. that any correlation between traits is purely attributable to age. This assumption simplifies computations, allowing probabilities to be multiplied. Even when this assumption does not hold, TA still provides age estimates which are unbiased. However, confidence intervals appear narrower than they really are. This bias is corrected ad hoc. According to the originators, TA is applicable in any trait that changes unidirectional with age in non-overlapping stages and always from stage i to stage i + 1. Restricting this approach to forensic odontology, this paper presents the first attempt to apply TA on radiological stage assessments of third molar development with the aim to provide age estimates including a 95% prediction interval from a full set of third molar scores.

Section snippets

Materials and methods

Statistical analyses were performed using data from a cross-sectional sample comprising 854 panoramic radiographs of subjects in the chronological age range 15.0–24.0 years. All subjects comprised a full set of third molar scores. The sample was archived panoramic radiographs used for diagnosis and treatment planning collected at the Department of Oral Medicine and Oral Diagnosis, School of Dentistry & Dental Research Institute, Seoul National University, Republic of Korea. For each radiograph

Descriptive analysis

An equal stage was assessed in the right and left maxillary third molars in 70.6% (316/425) of the males and 75.1% (322/429) of the female subjects. In the mandible, the figures were 77.4% (347/425) in males and 72.0% (309/429) in females, respectively. In the remaining 414 right–left third molar pairs, a maximum stage difference of 5 was found in one case. A stage difference of 1 was predominant in 73.91% of the cases. The remaining pairs ranged from 2 to 4 (2 stages in 68 pairs, 3 stages in

Discussion

As confirmed by our results, the correlation, conditioning on age, between the 4 third molar stages are high [3], [7], [21], and stronger between molars located in the same arch (left/right) than opposite (maxillary/mandibular). Hence, more information about age may be anticipated from molars located in opposite arches than within. The slight left–right stage difference, although only significant in the mandible, may indicate that all four molars potentially carry information about age. The

Conclusion

This study evaluated the applicability of the reproducible and well-described statistical approach of TA on stages of third molar development in the age range 15.0–24.0 years on a cross sectional sample. At any combination of third molars, the described procedure provides age estimates (maximum likelihood) including a 95% prediction interval as well as a probability of being above or below a certain age limit. While accuracy (the difference between estimated age and actual age) did not

References (28)

  • H.M. Liversidge et al.

    Estimating age and the likelihood of having attained 18 years of age using mandibular third molars

    Br. Dent. J.

    (2010)
  • C.F. Moorrees et al.

    Age variation of formation stages for ten permanent teeth

    J. Dent. Res.

    (1963)
  • H.H. Mincer et al.

    The A.B.F.O. study of third molar development and its use as an estimator of chronological age

    J. Forensic Sci.

    (1993)
  • J.L. Boldsen et al.

    Transition analysis: a new method for estimating age from skeletons

  • Cited by (33)

    • Using data from the US Korean War Dead and the Terry Collection to demonstrate problems of the common “overlap methods”

      2021, Remodeling Forensic Skeletal Age: Modern Applications and New Research Directions
    • Transition analysis applied to third molar development in a Danish population

      2020, Forensic Science International
      Citation Excerpt :

      It was concluded that the method performed accurately and that the risk of assessing a minor as an adult was minimized. Traditionally, TA has been fitted by applying a model only allowing for linearity in the form of the generalized linear model (glm) [10,18]. In the modelling of binary outcomes, like those used for TA, glm corresponds to a linear logistic regression model.

    • Age estimation of living persons: A coherent approach to inference and decision

      2020, Statistics and Probability in Forensic Anthropology
    View all citing articles on Scopus
    View full text