Forensic Sci Med Pathol DOI 10.1007/s12024-015-9656-x
ORIGINAL ARTICLE
Is Demirjian’s original method really useful for age estimation in a forensic context? Jose´ Luı´s Carneiro • Ineˆs Morais Caldas • Ame´rico Afonso • Hugo Filipe Violante Cardoso
Accepted: 16 January 2015 Ó Springer Science+Business Media New York 2015
Abstract Purpose The suitability of Demirjian’s method for forensic age estimation has been systematically questioned. The aim of this study is to further assess the reliability of Demirjian’s original method in forensic age estimation using a sample of Portuguese children. Methods 564 panoramic radiographs of Portuguese boys and girls between 6 and 16 years of age were evaluated using Demirjian’s method. Dental age (DA) was determined using the 50th percentile for the maturity score obtained for each age group. The mean difference between chronological age (CA) and dental age (DA) and the mean absolute difference between CA and DA were calculated for each age group. Paired t tests were used to test the statistical significance of mean differences between CA and DA. For each individual, a 94 % confidence interval was calculated for estimated DA, using the 3rd and 97th percentiles in Demirjian’s conversion tables. Results Chronological age was overestimated in boys, in every age group; mean differences between CA and DA
were statistically significant, expect for age 7. In girls, chronological age was overestimated in the 10–15 year-old age group. The difference between CA and DA was highest in the 12 years olds for both sexes. The 94 % confidence intervals did not include the true chronological age in all 6, 13, and 15 year-old girls, and all 14 and 15 year-old boys. Only a small portion of the individuals in the remaining age groups had their true chronological age falling within the probable age interval. Conclusions Results show a systematic bias and consistent inaccuracy in estimating age from dental development using Demirjian’s original method, making this methodology unsuitable for age estimation in the study sample. These results add to published evidence which suggests that Demirjian’s method is not suitable and should be abandoned altogether for forensic age estimation purposes. Keywords Forensic odontology Age estimation Dental development Demirjian’s method
Introduction J. L. Carneiro Departamento de Medicina Legal e Cieˆncias Forenses, Faculdade de Medicina da Universidade do Porto, Alameda Prof. Hernaˆni Monteiro, 4200-319 Porto, Portugal I. M. Caldas (&) A. Afonso Faculdade de Medicina Denta´ria da Universidade do Porto, Rua Manuel Pereira da Silva, 4200-393 Porto, Portugal e-mail: [email protected] I. M. Caldas A. Afonso CENCIFOR, Coimbra, Portugal H. F. V. Cardoso Department of Archaeology, Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6, Canada
One of the most widely used methods for dental age estimation is that of Demirjian and co-workers, first described in 1973 and based on a large sample of French-Canadian children [1]. Demirjian’s method establishes age by obtaining a maturity coefficient for each of the lower left seven permanent teeth, using eight stages of development. A maturity score is obtained from the sum of the eight maturity coefficients and is translated into a dental age using a table that converts maturity scores to ages [1]. This method was first developed for clinical purposes, i.e., to assess the suitability of therapeutic approaches and their estimated time length [2, 3], according with the estimated
123
Forensic Sci Med Pathol
dental maturity, and not for forensic age estimation, as it is now frequently used. Despite this important drawback, Demirjian’s method has been systematically used and tested in several samples of children from various parts of the world [4–10]. Not surprisingly, Demirjian’s method does not seem to be very accurate and in fact a systematic bias has been identified, where age is consistently overestimated. The inaccuracy of Demirjian’s method has been documented by systematically comparing the estimated age with the real chronological age of the individual and statistically testing the differences. However, this comparison only provides a partial understanding of the accuracy or inaccuracy of Demirjian’s method since only point age estimates are being compared. Because Demirjan and coworkers were not concerned with age estimation, the method does not provide a way to obtain a confidence interval or a measure of uncertainty for the estimated age. However, obtaining a confidence interval for the estimated age is crucial, because age estimation in a forensic investigation cannot be provided as a point value but rather as a probable range. Thus, it seems that Demirjian’s method is considered inaccurate only when point estimates are being compared, whereas it may actually perform well (or worst) when true ages are contrasted against a probable age range. Although closeness between estimated and true age is important for establishing the reliability of a certain method, there is currently no information as to whether Demirjian’s method can provide reliable ranges for estimated age. Since Demirjian’s method was not devised for age estimation, it seems unlikely that it can provide a measure of variation in age for a certain maturity score that can be used as an estimated age interval. Although not precisely a confidence interval, age estimates using Demirjian’s method can be given a range by using percentiles. More specifically, ages in the 3rd and 97th percentiles can approximate a 95 % confidence interval, but currently no study has explored this possibility in more detail. This paper wishes to assess Demirjian’s method reliability in estimating the age of a sample of Portuguese children, by comparing true chronological age with an estimated age range using the 3rd and 97th percentiles in Demirjian’s conversion tables. This was further contrasted with the traditional approach, where the point age estimate is compared to the chronological age in the sample. In the end, the paper wishes to determine whether Demirjian’s method can still provide reliable estimated age ranges that do not systematically exclude the true chronological age.
Methods
were females, and 260 were males. The radiographs belonged to randomly selected healthy patients, attending the clinical services at the Faculty of Dental Medicine of the University of Porto, Portugal. Age ranged between 6 and 16 years old (female mean = 10.44 years, standard deviation = 2.55; male mean = 10.75, standard deviation 2.63), and age and sex distribution is shown in Table 1. The criteria for inclusion were: (1) absence of clinical medical history that could affect the development of permanent teeth; (2) seven left mandibular teeth present without gross pathology. Two types of observer agreement were quantified: (1) Reproducibility or intra-observer error: assessing agreement between 20 randomly selected radiographs examined twice by the same examiner 1 month apart; (2) Repeatability or inter-observer error: assessing agreement between 20 randomly selected radiographs examined by two different examiners. Reproducibility and repeatability were assessed using Cohen’s kappa (k), and the quality of the agreement was evaluated as suggested by Landis and Koch [11]. The stage of development of the seven left lower permanent teeth was determined by the first author using Demirjian’s method. Age was estimated by using the age in the 50th percentile for the maturity score obtained for each individual. Mean estimated and mean chronological age and their respective standard deviations, as well as the mean difference (MD) between chronological age and dental age and the mean absolute difference (MAD) between chronological age and dental age were calculated for each 1-year age group. Paired t tests were used to test the statistical significance of mean differences between chronological and estimated age. All tests were performed at a 95 % confidence level. Finally, for each individual a 94 % confidence interval was calculated for estimated dental age, using the 3rd and 97th percentiles in Demirjian’s conversion tables. The percentage of individuals whose chronological age falls within the 3rd–97th percentile interval was calculated. All statistical analyses were performed using SPSS 20.0 (SPSS, Inc., Chicago, IL). Table 1 Age and sex distribution of the studied sample
Age (years)
Boys
Girls
6–6.9
20
20
7–7.9
19
46
8–8.9
41
38
9–9.9
41
44
10–10.9 11–11.9
32 17
40 26
12–12.9
21
29
13–13.9
29
30
14–14.9
25
16
15–15.9
A total of 564 panoramic radiographs of Portuguese boys and girls of known chronological age were selected: 304
123
Total
15
15
260
304
Forensic Sci Med Pathol
Results
Discussion
Cohen’s kappa coefficient results showed that there was good reproducibility and good repeatability (k [ 0.80 for both cases). Reproducibility was lower (\0.9) in stage assessment of tooth 31, 33, 34, and 35 (kappa values of 0.810, 0.843, 0.0865, and 0.865, respectively); for repeatability lower values were found in stage assessments of tooth 34, 35, and 37 (kappa values of 0.863, 0.803, and 0.873, respectively) (Table 2). Mean chronological age, mean estimated age using Demirjian’s method, MD, and MAD between estimated and chronological age are shown in Table 3 for boys, and in Table 4 for girls. In boys, the use of Demirjian’s age standards led to an overestimation of age in every age group and mean differences in age are all statistically significant, expect for age 7 (Table 3). The difference between mean estimated age and chronological age was highest for the 12 year olds. Comparatively, in girls the use of Demirjian’s age standards led to an overestimation of age in the older age groups, namely the 10, 12, 13, 14, and 15 years-olds, and in the remaining groups, age was underestimated. The age difference between mean estimated age and chronological age was also highest for the 12 year olds. However, for girls, there were no statistically significant differences between chronological and estimated age in all age groups (Table 4). MAD results show that, on average, the chronological and the estimated age differ by 0.5–2 years and the difference tends to increase with age. When assessing the accuracy of age estimates using the Demirjian’s 3rd–97th percentile interval, only 6, 13, and 15 year-old girls had all estimated ages falling outside the percentile range, whereas in boys that only occurred in the 14 and 15 year-olds (Table 5). Except for these ages, the percentage of individual whose estimated age fell within the 94 % percentile range could be as low as 7.7 % in girls and 27.6 % in boys.
Results show that Demirjian’s method provides probable age intervals that are very likely to exclude the true chronological age of the individual. This is particularly noteworthy given that the differences between estimated and chronological ages in girls were found to be systematically insignificant. In fact, the lack of a significant difference between estimated and true chronological age in girls results simply from the overestimation in the older age groups being canceled out by the underestimation in the younger age groups. Consequently, although the point estimates seemed overall a good approximation of chronological age, in fact they were not. As such, Demirjian’s method can be further shown to be inappropriate for age estimation. This is but a confirmation that the method was devised specifically to measure biological maturity, where chronological age and dental maturity was expressed as a sigmoid curve [12]. This curve expresses the relationship between the maturity score and chronological age and presents several limitations, namely: (1) it provides dental maturity score ranges from just above 10 % at age 2.5–98.4 and 100 % at 16 (males and females, respectively), excluding very young children (younger than 3 years-old); (2) the relationship expressed by the curve applies to individual teeth but not to sequential development [12]. As a consequence, Demirjian’s method may not be able to satisfactorily minimize the differences between chronological and dental age. Because the method is not designed for age estimation it also does not provide adequate error statistics for the calculation of confidence intervals for dental age estimates. Additionally, Demirjian’s method was based on a mixed-longitudinal sample, and it is probable that the age at which individuals enter a stage is seen as occurring earlier than in a cross-sectional sample. This might be an explanation as to why numerous studies report a systematic positive difference (overestimation of age) with the Demirjian’s method [12]. Despite these limitations, Demirjian’s method has been routinely used and tested as a method for forensic age estimation [4, 5, 7, 9]. For example, Flood et al. [13] compared the accuracy of the following Demirjian’s four dental development methods: the original 7-tooth technique; the revised 7-tooth system; the 4-tooth method; and the alternate 4-tooth approach. Their results revealed significant overall differences between mean chronological and estimated age in both sexes, and each method consistently overestimated chronological age. Blenkin and Evans [14] obtained similar results, stating that the use of the Demirjian standards resulted in consistent overestimates of chronological age in children under the age of 14. Goya et al. [15] also reported an overestimation of age in most groups studied, stating that the Demirjian standards were not suitable for their
Table 2 Kappa and p values reproducibility and repeatability Tooth
Reproducibility k value
Repeatability p value
k value
p value
31
0.810
\0.001
1.000
\0.001
32
1.000
\0.001
1.000
\0.001
33
0.843
\0.001
1.000
\0.001
34
0.865
\0.001
0.863
\0.001
35
0.865
\0.001
0.803
\0.001
36
1.000
\0.001
1.000
\0.001
37
1.000
\0.001
0.873
\0.001
123
Forensic Sci Med Pathol Table 3 Descriptive statistics for mean chronological age (CA) and mean estimated age (EA) using Demirjian’s method, for boys (SD—standard deviation)
Mean difference (MD) and mean absolute difference between chronological and estimated age are also shown, together with p values for paired t tests Table 4 Descriptive statistics for mean chronological age (CA) and mean estimated age (EA) using Demirjian’s method, for girls (SD—standard deviation)
Mean difference (MD) and mean absolute difference between chronological and estimated age are also shown, together with p values for paired t tests
Age (years)
n
Mean CA (SD)
6–6.9
20
6.57 (0.26)
7–7.9
15
7.54 (0.25)
8–8.9
39
9–9.9 10–10.9
MD
MAD
p values
7.18 (0.54)
0.61
0.66
\0.001
8.16 (1.33)
0.62
1.25
0.115
8.54 (0.31)
9.97 (1.35)
1.43
1.62
\0.001
40
9.62 (0.28)
11.30 (1.44)
1.68
1.82
\0.001
32
10.42 (0.24)
12.16 (1.54)
1.74
1.92
\0.001
11–11.9
17
11.55 (0.27)
13.76 (2.04)
2.21
2.41
\0.001
12–12.9
21
12.53 (0.24)
14.93 (1.96)
2.40
2.87
\0.001
13–13.9
29
13.38 (0.26)
15.48 (0.83)
2.10
2.11
\0.001
14–14.9
25
14.52 (0.27)
15.59 (1.00)
1.07
1.41
\0.001
15–16
15
15.53 (0.27)
15.99 (0.18)
0.46
0.49
\0.001
Age (years)
n
Mean CA (SD)
Mean EA (SD)
MD
MAD
p
6–6.9
20
6.64 (0.24)
6.55 (0.76)
-0.09
0.57
0.563
7–7.9
44
7.55 (0.29)
7.36 (1.27)
-0.18
0.99
0.318
8–8.9
32
8.51 (0.26)
8.00 (1.24)
-0.51
1.04
0.014
9–9.9
31
9.50 (0.28)
9.29 (1.13)
-0.21
1.46
0.243
10–10.9
32
10.48 (0.30)
10.53 (1.60)
0.05
1.99
0.855
11–11.9
22
11.42 (0.31)
11.41 (1.57)
-0.01
1.63
0.978
12–12.9
26
12.51 (0.30)
13.31 (1.91)
0.80
2.72
0.040
13–13.9
30
13.57 (0.28)
13.63 (1.03)
0.06
1.53
0.698
14–14.9
25
14.44 (0.28)
14.53 (1.13)
0.09
1.43
0.762
15–16
15
15.52 (0.39)
15.87 (0.52)
0.35
0.52
0.024
Table 5 Number and percentage (%) of individuals whose estimated age fell within the 94 % Demirjian’s percentile range (3rd–97th) Group ages (years)
Females
Males
6–6.9
0
6 (30.0)
7–7.9
12 (26.1)
13 (68.4)
8–8.9
12 (31.6)
28 (68.3)
9–9.9 10–10.9
13 (29.5) 19 (25.0)
33 (80.5) 10 (31.3)
11–11.9
2 (7.7)
8 (47.1)
12–12.9
18 (62.1)
12 (57.1)
13–13.9
0
8 (27.6)
14–14.9
6 (37.5)
0
15–16
0
0
population. Clearly, Demirjian’s method is not appropriate for forensic age estimation and should be abandoned altogether. Regardless of the important limitations in Demirjian’s method, there have been recent attempts to improve and/or correct the method to increase its success rate in estimating age. These studies cannot be considered revised, corrected or improved versions of the method per se, but instead new
123
Mean EA (SD)
methods in their own merit that are based on Demirjian’s general approach, in particularly the dental scoring system. For example, Willems [16] studied 2523 orthopantomograms from a Belgian population, and reported an overestimation of chronological age using Demirjian’s original method. However, their revised version of the approach using an adapted scoring system resulted in a higher accuracy in the sample. Similarly Chaillet et al. [6] compared the accuracy of Demirjian’s original method with a revised version using differently weighted scores, polynomial functions and a Belgian sample. Their results showed that the polynomial functions were highly reliable, and that the percentile method using population-specific weighted scores was very accurate, supporting the notion that a revised Demirjian approach (not revised method) can provide better results. In 2004, Chaillet et al. [17] also tested Demirjian’s original method with a revised model using polynomial functions and a Finn sample, and reported a significant increase in accuracy. Demirjian’s original method was deemed most useful for dental health clinicians. As a comparison, Nolla and Haavikko’s method underestimated age [18], and seems to be unreliable in a forensic context as well. Although these studies do support how useful Demirjian’s general approach can be for devising new methods for age estimation, they actually
Forensic Sci Med Pathol
reinforce the conclusion that Demirjian’s original method is indeed not suitable for forensic age estimation. In addition, despite the merits of these new methods, they have yet to report their performance when age is estimated as a probable interval. A clear understanding dental maturity and how different methods establish a relationship with chronological age is of the utmost importance to forensic scientists since it can provide, not only the means for age estimation in a variety of scenarios, but also the information that allows them to decide which methods to apply in which circumstances. Although different methods have been proposed to estimate age from assessments of dental maturity [1, 10, 19– 29], the accuracy of these methods has yet to be determined in detail. Particularly because little attention has been given to the calculation of confidence intervals for the estimated age, which is the only valid approach in any forensic investigation. Consequently, methods that have been considered reliable based on point age estimate comparisons may actually be revealed as inaccurate and unreliable. In addition to testing the available methods further, several authors suggest that differences between estimated and chronological age in test samples are due to populationbased variation [4, 7–10, 18, 30, 31], arguing for the use of specific-population standards. However, Liversidge [12] has argued that the existence of population differences is still not clear, since many studies have different age ranges and uneven age distributions, which may explain some of the differences found. According to Liversidge [12], dental maturity curves are similar in different regions and population specific data do not reduce age variation in dental maturation, suggesting that population differences may actually not be important and population specific methods are unlikely to improve the accuracy of age estimation.
Conclusions Our conclusions from this study can be summarized are as follows: A systematic bias was found in estimating age from dental development using Demirjian’s original method, particularly in boys. In addition to the systematic bias, the true chronological age almost always fell outside of the 3rd–97th percentile range of dental age, particularly in girls. Globally, Demirjian’s original method was found unsuitable for age estimation in the study sample and results suggest it is a method that should never be used for age estimation in a forensic context. Other methods must be systematically tested or developed specifically for the purpose of age estimation.
Key Points 1. 2. 3.
4. 5. 6. 7.
Demirjian’s method was not developed for forensic age estimation purposes. The purpose of this study was to test Demirjian’s method when a probable age range is obtained. A systematic bias and consistent inaccuracy was found in estimating age from dental development using Demirjian’s method. Age overestimation occurred in boys in every age group. In girls, age was overestimated in the 10–15 year-old age groups. Estimating age as a 94 % confidence interval proved unreliable. Demirjian’s method was deemed unsuitable for age estimation.
References 1. Demirjian A, Goldstein H, Tanner JM. A new system of dental age assessment. Hum Biol. 1973;45(2):211–27. 2. Krailassiri S, Anuwongnukroh N, Dechkunakorn S. Relationships between dental calcification stages and skeletal maturity indicators in Thai individuals. Angle Orthod. 2002;72(2):155–66. 3. Sahin Saglam AM, Gazilerli U. The relationship between dental and skeletal maturity. J Orofac Orthop. 2002;63(6):454–62. 4. Bagherpour A, Imanimoghaddam M, Bagherpour MR, Einolghozati M. Dental age assessment among Iranian children aged 6–13 years using the Demirjian method. Forensic Sci Int. 2010;197(1–3):121 e1–4. 5. Celikoglu M, Cantekin K, Ceylan I. Dental age assessment: the applicability of Demirjian method in eastern Turkish children. J Forensic Sci. 2011;56(Suppl 1):S220–2. 6. Chaillet N, Willems G, Demirjian A. Dental maturity in Belgian children using Demirjian’s method and polynomial functions: new standard curves for forensic and clinical use. J Forensic Odontostomatol. 2004;22(2):18–27. 7. Koshy S, Tandon S. Dental age assessment: the applicability of Demirjian’s method in south Indian children. Forensic Sci Int. 1998;94(1–2):73–85. 8. Lee SS, Kim D, Lee S, Lee UY, Seo JS, Ahn YW, et al. Validity of Demirjian’s and modified Demirjian’s methods in age estimation for Korean juveniles and adolescents. Forensic Sci Int. 2011;211(1–3):41–6. 9. Tunc ES, Koyuturk AE. Dental age assessment using Demirjian’s method on northern Turkish children. Forensic Sci Int. 2008;175(1):23–6. 10. Willems G, Van Olmen A, Spiessens B, Carels C. Dental age estimation in Belgian children: Demirjian’s technique revisited. J Forensic Sci. 2001;46(4):893–5. 11. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33:159–74. 12. Liversidge HM. The assessment and interpretation of Demirjian, Goldstein and Tanner’s dental maturity. Ann Hum Biol. 2012;39(5):412–31.
123
Forensic Sci Med Pathol 13. Flood SJ, Mitchell WJ, Oxnard CE, Turlach BA, McGeachie J. A comparison of Demirjian’s four dental development methods for forensic age assessment. J Forensic Sci. 2011;56(6):1610–5. 14. Blenkin MR, Evans W. Age estimation from the teeth using a modified Demirjian system. J Forensic Sci. 2010;55(6):1504–8. 15. Goya HA, Satake H, Maeda T, Tanaka S, Akimoto Y. Dental age in Japanese children using a modified Demirjian method. Pediatr Dent J. 2009;19(1):82–8. 16. Willems G. A review of the most commonly used dental age estimation techniques. J Forensic Odontostomatol. 2001; 19(1):9–17. 17. Chaillet N, Nystrom M, Kataja M, Demirjian A. Dental maturity curves in Finnish children: Demirjian’s method revisited and polynomial functions for age estimation. J Forensic Sci. 2004; 49(6):1324–31. 18. Maber M, Liversidge HM, Hector MP. Accuracy of age estimation of radiographic methods using developing teeth. Forensic Sci Int. 2006;159(Suppl 1):S68–73. 19. Foti B, Lalys L, Adalian P, Giustiniani J, Maczel M, Signoli M, et al. New forensic approach to age determination in children based on tooth eruption. Forensic Sci Int. 2003;132(1):49–56. 20. Gleiser I, Hunt EE Jr. The permanent mandibular first molar: its calcification, eruption and decay. Am J Phys Anthropol. 1955;13(2):253–83. 21. Gustafson G, Koch G. Age estimation up to 16 years of age based on dental development. Odontol Revy. 1974;25(3):297–306. 22. Haavikko K. The formation and the alveolar and clinical eruption of the permanent teeth. An orthopantomographic study. Suom Hammaslaak Toim. 1970;66(3):103–70.
123
23. Hagg U, Matsson L. Dental maturity as an indicator of chronological age: the accuracy and precision of three methods. Eur J Orthod. 1985;7(1):25–34. 24. Harris MJP, Nortje CJ. The mesial root of the third mandibular molar. A possible indicator of age. J Forensic Odontostomatol. 1985;2:39–43. 25. Hunt EE Jr, Gleiser I. The estimation of age and sex of preadolescent children from bones and teeth. Am J Phys Anthropol. 1955;13(3):479–87. 26. Kullman L, Johanson G, Akesson L. Root development of the lower third molar and its relation to chronological age. Swed Dent J. 1992;16(4):161–7. 27. Liliequist B, Lundberg M. Skeletal and tooth development. A methodologic investigation. Acta Radiol Diagn. 1971;11(2): 97–112. 28. Moorrees CF, Fanning EA, Hunt EE Jr. Age variation of formation stages for ten permanent teeth. J Dent Res. 1963;42: 1490–502. 29. Nolla CM. The development of the permanent teeth. J Dent Child. 1960;27:254–66. 30. Frucht S, Schnegelsberg C, Schulte-Monting J, Rose E, Jonas I. Dental age in southwest Germany. A radiographic study. J Orofac Orthop. 2000;61(5):318–29. 31. Chen JW, Guo J, Zhou J, Liu RK, Chen TT, Zou SJ. Assessment of dental maturity of western Chinese children using Demirjian’s method. Forensic Sci Int. 2010;197(1–3):119 e1–4.
ORIGINAL ARTICLE
Is Demirjian’s original method really useful for age estimation in a forensic context? Jose´ Luı´s Carneiro • Ineˆs Morais Caldas • Ame´rico Afonso • Hugo Filipe Violante Cardoso
Accepted: 16 January 2015 Ó Springer Science+Business Media New York 2015
Abstract Purpose The suitability of Demirjian’s method for forensic age estimation has been systematically questioned. The aim of this study is to further assess the reliability of Demirjian’s original method in forensic age estimation using a sample of Portuguese children. Methods 564 panoramic radiographs of Portuguese boys and girls between 6 and 16 years of age were evaluated using Demirjian’s method. Dental age (DA) was determined using the 50th percentile for the maturity score obtained for each age group. The mean difference between chronological age (CA) and dental age (DA) and the mean absolute difference between CA and DA were calculated for each age group. Paired t tests were used to test the statistical significance of mean differences between CA and DA. For each individual, a 94 % confidence interval was calculated for estimated DA, using the 3rd and 97th percentiles in Demirjian’s conversion tables. Results Chronological age was overestimated in boys, in every age group; mean differences between CA and DA
were statistically significant, expect for age 7. In girls, chronological age was overestimated in the 10–15 year-old age group. The difference between CA and DA was highest in the 12 years olds for both sexes. The 94 % confidence intervals did not include the true chronological age in all 6, 13, and 15 year-old girls, and all 14 and 15 year-old boys. Only a small portion of the individuals in the remaining age groups had their true chronological age falling within the probable age interval. Conclusions Results show a systematic bias and consistent inaccuracy in estimating age from dental development using Demirjian’s original method, making this methodology unsuitable for age estimation in the study sample. These results add to published evidence which suggests that Demirjian’s method is not suitable and should be abandoned altogether for forensic age estimation purposes. Keywords Forensic odontology Age estimation Dental development Demirjian’s method
Introduction J. L. Carneiro Departamento de Medicina Legal e Cieˆncias Forenses, Faculdade de Medicina da Universidade do Porto, Alameda Prof. Hernaˆni Monteiro, 4200-319 Porto, Portugal I. M. Caldas (&) A. Afonso Faculdade de Medicina Denta´ria da Universidade do Porto, Rua Manuel Pereira da Silva, 4200-393 Porto, Portugal e-mail: [email protected] I. M. Caldas A. Afonso CENCIFOR, Coimbra, Portugal H. F. V. Cardoso Department of Archaeology, Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6, Canada
One of the most widely used methods for dental age estimation is that of Demirjian and co-workers, first described in 1973 and based on a large sample of French-Canadian children [1]. Demirjian’s method establishes age by obtaining a maturity coefficient for each of the lower left seven permanent teeth, using eight stages of development. A maturity score is obtained from the sum of the eight maturity coefficients and is translated into a dental age using a table that converts maturity scores to ages [1]. This method was first developed for clinical purposes, i.e., to assess the suitability of therapeutic approaches and their estimated time length [2, 3], according with the estimated
123
Forensic Sci Med Pathol
dental maturity, and not for forensic age estimation, as it is now frequently used. Despite this important drawback, Demirjian’s method has been systematically used and tested in several samples of children from various parts of the world [4–10]. Not surprisingly, Demirjian’s method does not seem to be very accurate and in fact a systematic bias has been identified, where age is consistently overestimated. The inaccuracy of Demirjian’s method has been documented by systematically comparing the estimated age with the real chronological age of the individual and statistically testing the differences. However, this comparison only provides a partial understanding of the accuracy or inaccuracy of Demirjian’s method since only point age estimates are being compared. Because Demirjan and coworkers were not concerned with age estimation, the method does not provide a way to obtain a confidence interval or a measure of uncertainty for the estimated age. However, obtaining a confidence interval for the estimated age is crucial, because age estimation in a forensic investigation cannot be provided as a point value but rather as a probable range. Thus, it seems that Demirjian’s method is considered inaccurate only when point estimates are being compared, whereas it may actually perform well (or worst) when true ages are contrasted against a probable age range. Although closeness between estimated and true age is important for establishing the reliability of a certain method, there is currently no information as to whether Demirjian’s method can provide reliable ranges for estimated age. Since Demirjian’s method was not devised for age estimation, it seems unlikely that it can provide a measure of variation in age for a certain maturity score that can be used as an estimated age interval. Although not precisely a confidence interval, age estimates using Demirjian’s method can be given a range by using percentiles. More specifically, ages in the 3rd and 97th percentiles can approximate a 95 % confidence interval, but currently no study has explored this possibility in more detail. This paper wishes to assess Demirjian’s method reliability in estimating the age of a sample of Portuguese children, by comparing true chronological age with an estimated age range using the 3rd and 97th percentiles in Demirjian’s conversion tables. This was further contrasted with the traditional approach, where the point age estimate is compared to the chronological age in the sample. In the end, the paper wishes to determine whether Demirjian’s method can still provide reliable estimated age ranges that do not systematically exclude the true chronological age.
Methods
were females, and 260 were males. The radiographs belonged to randomly selected healthy patients, attending the clinical services at the Faculty of Dental Medicine of the University of Porto, Portugal. Age ranged between 6 and 16 years old (female mean = 10.44 years, standard deviation = 2.55; male mean = 10.75, standard deviation 2.63), and age and sex distribution is shown in Table 1. The criteria for inclusion were: (1) absence of clinical medical history that could affect the development of permanent teeth; (2) seven left mandibular teeth present without gross pathology. Two types of observer agreement were quantified: (1) Reproducibility or intra-observer error: assessing agreement between 20 randomly selected radiographs examined twice by the same examiner 1 month apart; (2) Repeatability or inter-observer error: assessing agreement between 20 randomly selected radiographs examined by two different examiners. Reproducibility and repeatability were assessed using Cohen’s kappa (k), and the quality of the agreement was evaluated as suggested by Landis and Koch [11]. The stage of development of the seven left lower permanent teeth was determined by the first author using Demirjian’s method. Age was estimated by using the age in the 50th percentile for the maturity score obtained for each individual. Mean estimated and mean chronological age and their respective standard deviations, as well as the mean difference (MD) between chronological age and dental age and the mean absolute difference (MAD) between chronological age and dental age were calculated for each 1-year age group. Paired t tests were used to test the statistical significance of mean differences between chronological and estimated age. All tests were performed at a 95 % confidence level. Finally, for each individual a 94 % confidence interval was calculated for estimated dental age, using the 3rd and 97th percentiles in Demirjian’s conversion tables. The percentage of individuals whose chronological age falls within the 3rd–97th percentile interval was calculated. All statistical analyses were performed using SPSS 20.0 (SPSS, Inc., Chicago, IL). Table 1 Age and sex distribution of the studied sample
Age (years)
Boys
Girls
6–6.9
20
20
7–7.9
19
46
8–8.9
41
38
9–9.9
41
44
10–10.9 11–11.9
32 17
40 26
12–12.9
21
29
13–13.9
29
30
14–14.9
25
16
15–15.9
A total of 564 panoramic radiographs of Portuguese boys and girls of known chronological age were selected: 304
123
Total
15
15
260
304
Forensic Sci Med Pathol
Results
Discussion
Cohen’s kappa coefficient results showed that there was good reproducibility and good repeatability (k [ 0.80 for both cases). Reproducibility was lower (\0.9) in stage assessment of tooth 31, 33, 34, and 35 (kappa values of 0.810, 0.843, 0.0865, and 0.865, respectively); for repeatability lower values were found in stage assessments of tooth 34, 35, and 37 (kappa values of 0.863, 0.803, and 0.873, respectively) (Table 2). Mean chronological age, mean estimated age using Demirjian’s method, MD, and MAD between estimated and chronological age are shown in Table 3 for boys, and in Table 4 for girls. In boys, the use of Demirjian’s age standards led to an overestimation of age in every age group and mean differences in age are all statistically significant, expect for age 7 (Table 3). The difference between mean estimated age and chronological age was highest for the 12 year olds. Comparatively, in girls the use of Demirjian’s age standards led to an overestimation of age in the older age groups, namely the 10, 12, 13, 14, and 15 years-olds, and in the remaining groups, age was underestimated. The age difference between mean estimated age and chronological age was also highest for the 12 year olds. However, for girls, there were no statistically significant differences between chronological and estimated age in all age groups (Table 4). MAD results show that, on average, the chronological and the estimated age differ by 0.5–2 years and the difference tends to increase with age. When assessing the accuracy of age estimates using the Demirjian’s 3rd–97th percentile interval, only 6, 13, and 15 year-old girls had all estimated ages falling outside the percentile range, whereas in boys that only occurred in the 14 and 15 year-olds (Table 5). Except for these ages, the percentage of individual whose estimated age fell within the 94 % percentile range could be as low as 7.7 % in girls and 27.6 % in boys.
Results show that Demirjian’s method provides probable age intervals that are very likely to exclude the true chronological age of the individual. This is particularly noteworthy given that the differences between estimated and chronological ages in girls were found to be systematically insignificant. In fact, the lack of a significant difference between estimated and true chronological age in girls results simply from the overestimation in the older age groups being canceled out by the underestimation in the younger age groups. Consequently, although the point estimates seemed overall a good approximation of chronological age, in fact they were not. As such, Demirjian’s method can be further shown to be inappropriate for age estimation. This is but a confirmation that the method was devised specifically to measure biological maturity, where chronological age and dental maturity was expressed as a sigmoid curve [12]. This curve expresses the relationship between the maturity score and chronological age and presents several limitations, namely: (1) it provides dental maturity score ranges from just above 10 % at age 2.5–98.4 and 100 % at 16 (males and females, respectively), excluding very young children (younger than 3 years-old); (2) the relationship expressed by the curve applies to individual teeth but not to sequential development [12]. As a consequence, Demirjian’s method may not be able to satisfactorily minimize the differences between chronological and dental age. Because the method is not designed for age estimation it also does not provide adequate error statistics for the calculation of confidence intervals for dental age estimates. Additionally, Demirjian’s method was based on a mixed-longitudinal sample, and it is probable that the age at which individuals enter a stage is seen as occurring earlier than in a cross-sectional sample. This might be an explanation as to why numerous studies report a systematic positive difference (overestimation of age) with the Demirjian’s method [12]. Despite these limitations, Demirjian’s method has been routinely used and tested as a method for forensic age estimation [4, 5, 7, 9]. For example, Flood et al. [13] compared the accuracy of the following Demirjian’s four dental development methods: the original 7-tooth technique; the revised 7-tooth system; the 4-tooth method; and the alternate 4-tooth approach. Their results revealed significant overall differences between mean chronological and estimated age in both sexes, and each method consistently overestimated chronological age. Blenkin and Evans [14] obtained similar results, stating that the use of the Demirjian standards resulted in consistent overestimates of chronological age in children under the age of 14. Goya et al. [15] also reported an overestimation of age in most groups studied, stating that the Demirjian standards were not suitable for their
Table 2 Kappa and p values reproducibility and repeatability Tooth
Reproducibility k value
Repeatability p value
k value
p value
31
0.810
\0.001
1.000
\0.001
32
1.000
\0.001
1.000
\0.001
33
0.843
\0.001
1.000
\0.001
34
0.865
\0.001
0.863
\0.001
35
0.865
\0.001
0.803
\0.001
36
1.000
\0.001
1.000
\0.001
37
1.000
\0.001
0.873
\0.001
123
Forensic Sci Med Pathol Table 3 Descriptive statistics for mean chronological age (CA) and mean estimated age (EA) using Demirjian’s method, for boys (SD—standard deviation)
Mean difference (MD) and mean absolute difference between chronological and estimated age are also shown, together with p values for paired t tests Table 4 Descriptive statistics for mean chronological age (CA) and mean estimated age (EA) using Demirjian’s method, for girls (SD—standard deviation)
Mean difference (MD) and mean absolute difference between chronological and estimated age are also shown, together with p values for paired t tests
Age (years)
n
Mean CA (SD)
6–6.9
20
6.57 (0.26)
7–7.9
15
7.54 (0.25)
8–8.9
39
9–9.9 10–10.9
MD
MAD
p values
7.18 (0.54)
0.61
0.66
\0.001
8.16 (1.33)
0.62
1.25
0.115
8.54 (0.31)
9.97 (1.35)
1.43
1.62
\0.001
40
9.62 (0.28)
11.30 (1.44)
1.68
1.82
\0.001
32
10.42 (0.24)
12.16 (1.54)
1.74
1.92
\0.001
11–11.9
17
11.55 (0.27)
13.76 (2.04)
2.21
2.41
\0.001
12–12.9
21
12.53 (0.24)
14.93 (1.96)
2.40
2.87
\0.001
13–13.9
29
13.38 (0.26)
15.48 (0.83)
2.10
2.11
\0.001
14–14.9
25
14.52 (0.27)
15.59 (1.00)
1.07
1.41
\0.001
15–16
15
15.53 (0.27)
15.99 (0.18)
0.46
0.49
\0.001
Age (years)
n
Mean CA (SD)
Mean EA (SD)
MD
MAD
p
6–6.9
20
6.64 (0.24)
6.55 (0.76)
-0.09
0.57
0.563
7–7.9
44
7.55 (0.29)
7.36 (1.27)
-0.18
0.99
0.318
8–8.9
32
8.51 (0.26)
8.00 (1.24)
-0.51
1.04
0.014
9–9.9
31
9.50 (0.28)
9.29 (1.13)
-0.21
1.46
0.243
10–10.9
32
10.48 (0.30)
10.53 (1.60)
0.05
1.99
0.855
11–11.9
22
11.42 (0.31)
11.41 (1.57)
-0.01
1.63
0.978
12–12.9
26
12.51 (0.30)
13.31 (1.91)
0.80
2.72
0.040
13–13.9
30
13.57 (0.28)
13.63 (1.03)
0.06
1.53
0.698
14–14.9
25
14.44 (0.28)
14.53 (1.13)
0.09
1.43
0.762
15–16
15
15.52 (0.39)
15.87 (0.52)
0.35
0.52
0.024
Table 5 Number and percentage (%) of individuals whose estimated age fell within the 94 % Demirjian’s percentile range (3rd–97th) Group ages (years)
Females
Males
6–6.9
0
6 (30.0)
7–7.9
12 (26.1)
13 (68.4)
8–8.9
12 (31.6)
28 (68.3)
9–9.9 10–10.9
13 (29.5) 19 (25.0)
33 (80.5) 10 (31.3)
11–11.9
2 (7.7)
8 (47.1)
12–12.9
18 (62.1)
12 (57.1)
13–13.9
0
8 (27.6)
14–14.9
6 (37.5)
0
15–16
0
0
population. Clearly, Demirjian’s method is not appropriate for forensic age estimation and should be abandoned altogether. Regardless of the important limitations in Demirjian’s method, there have been recent attempts to improve and/or correct the method to increase its success rate in estimating age. These studies cannot be considered revised, corrected or improved versions of the method per se, but instead new
123
Mean EA (SD)
methods in their own merit that are based on Demirjian’s general approach, in particularly the dental scoring system. For example, Willems [16] studied 2523 orthopantomograms from a Belgian population, and reported an overestimation of chronological age using Demirjian’s original method. However, their revised version of the approach using an adapted scoring system resulted in a higher accuracy in the sample. Similarly Chaillet et al. [6] compared the accuracy of Demirjian’s original method with a revised version using differently weighted scores, polynomial functions and a Belgian sample. Their results showed that the polynomial functions were highly reliable, and that the percentile method using population-specific weighted scores was very accurate, supporting the notion that a revised Demirjian approach (not revised method) can provide better results. In 2004, Chaillet et al. [17] also tested Demirjian’s original method with a revised model using polynomial functions and a Finn sample, and reported a significant increase in accuracy. Demirjian’s original method was deemed most useful for dental health clinicians. As a comparison, Nolla and Haavikko’s method underestimated age [18], and seems to be unreliable in a forensic context as well. Although these studies do support how useful Demirjian’s general approach can be for devising new methods for age estimation, they actually
Forensic Sci Med Pathol
reinforce the conclusion that Demirjian’s original method is indeed not suitable for forensic age estimation. In addition, despite the merits of these new methods, they have yet to report their performance when age is estimated as a probable interval. A clear understanding dental maturity and how different methods establish a relationship with chronological age is of the utmost importance to forensic scientists since it can provide, not only the means for age estimation in a variety of scenarios, but also the information that allows them to decide which methods to apply in which circumstances. Although different methods have been proposed to estimate age from assessments of dental maturity [1, 10, 19– 29], the accuracy of these methods has yet to be determined in detail. Particularly because little attention has been given to the calculation of confidence intervals for the estimated age, which is the only valid approach in any forensic investigation. Consequently, methods that have been considered reliable based on point age estimate comparisons may actually be revealed as inaccurate and unreliable. In addition to testing the available methods further, several authors suggest that differences between estimated and chronological age in test samples are due to populationbased variation [4, 7–10, 18, 30, 31], arguing for the use of specific-population standards. However, Liversidge [12] has argued that the existence of population differences is still not clear, since many studies have different age ranges and uneven age distributions, which may explain some of the differences found. According to Liversidge [12], dental maturity curves are similar in different regions and population specific data do not reduce age variation in dental maturation, suggesting that population differences may actually not be important and population specific methods are unlikely to improve the accuracy of age estimation.
Conclusions Our conclusions from this study can be summarized are as follows: A systematic bias was found in estimating age from dental development using Demirjian’s original method, particularly in boys. In addition to the systematic bias, the true chronological age almost always fell outside of the 3rd–97th percentile range of dental age, particularly in girls. Globally, Demirjian’s original method was found unsuitable for age estimation in the study sample and results suggest it is a method that should never be used for age estimation in a forensic context. Other methods must be systematically tested or developed specifically for the purpose of age estimation.
Key Points 1. 2. 3.
4. 5. 6. 7.
Demirjian’s method was not developed for forensic age estimation purposes. The purpose of this study was to test Demirjian’s method when a probable age range is obtained. A systematic bias and consistent inaccuracy was found in estimating age from dental development using Demirjian’s method. Age overestimation occurred in boys in every age group. In girls, age was overestimated in the 10–15 year-old age groups. Estimating age as a 94 % confidence interval proved unreliable. Demirjian’s method was deemed unsuitable for age estimation.
References 1. Demirjian A, Goldstein H, Tanner JM. A new system of dental age assessment. Hum Biol. 1973;45(2):211–27. 2. Krailassiri S, Anuwongnukroh N, Dechkunakorn S. Relationships between dental calcification stages and skeletal maturity indicators in Thai individuals. Angle Orthod. 2002;72(2):155–66. 3. Sahin Saglam AM, Gazilerli U. The relationship between dental and skeletal maturity. J Orofac Orthop. 2002;63(6):454–62. 4. Bagherpour A, Imanimoghaddam M, Bagherpour MR, Einolghozati M. Dental age assessment among Iranian children aged 6–13 years using the Demirjian method. Forensic Sci Int. 2010;197(1–3):121 e1–4. 5. Celikoglu M, Cantekin K, Ceylan I. Dental age assessment: the applicability of Demirjian method in eastern Turkish children. J Forensic Sci. 2011;56(Suppl 1):S220–2. 6. Chaillet N, Willems G, Demirjian A. Dental maturity in Belgian children using Demirjian’s method and polynomial functions: new standard curves for forensic and clinical use. J Forensic Odontostomatol. 2004;22(2):18–27. 7. Koshy S, Tandon S. Dental age assessment: the applicability of Demirjian’s method in south Indian children. Forensic Sci Int. 1998;94(1–2):73–85. 8. Lee SS, Kim D, Lee S, Lee UY, Seo JS, Ahn YW, et al. Validity of Demirjian’s and modified Demirjian’s methods in age estimation for Korean juveniles and adolescents. Forensic Sci Int. 2011;211(1–3):41–6. 9. Tunc ES, Koyuturk AE. Dental age assessment using Demirjian’s method on northern Turkish children. Forensic Sci Int. 2008;175(1):23–6. 10. Willems G, Van Olmen A, Spiessens B, Carels C. Dental age estimation in Belgian children: Demirjian’s technique revisited. J Forensic Sci. 2001;46(4):893–5. 11. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33:159–74. 12. Liversidge HM. The assessment and interpretation of Demirjian, Goldstein and Tanner’s dental maturity. Ann Hum Biol. 2012;39(5):412–31.
123
Forensic Sci Med Pathol 13. Flood SJ, Mitchell WJ, Oxnard CE, Turlach BA, McGeachie J. A comparison of Demirjian’s four dental development methods for forensic age assessment. J Forensic Sci. 2011;56(6):1610–5. 14. Blenkin MR, Evans W. Age estimation from the teeth using a modified Demirjian system. J Forensic Sci. 2010;55(6):1504–8. 15. Goya HA, Satake H, Maeda T, Tanaka S, Akimoto Y. Dental age in Japanese children using a modified Demirjian method. Pediatr Dent J. 2009;19(1):82–8. 16. Willems G. A review of the most commonly used dental age estimation techniques. J Forensic Odontostomatol. 2001; 19(1):9–17. 17. Chaillet N, Nystrom M, Kataja M, Demirjian A. Dental maturity curves in Finnish children: Demirjian’s method revisited and polynomial functions for age estimation. J Forensic Sci. 2004; 49(6):1324–31. 18. Maber M, Liversidge HM, Hector MP. Accuracy of age estimation of radiographic methods using developing teeth. Forensic Sci Int. 2006;159(Suppl 1):S68–73. 19. Foti B, Lalys L, Adalian P, Giustiniani J, Maczel M, Signoli M, et al. New forensic approach to age determination in children based on tooth eruption. Forensic Sci Int. 2003;132(1):49–56. 20. Gleiser I, Hunt EE Jr. The permanent mandibular first molar: its calcification, eruption and decay. Am J Phys Anthropol. 1955;13(2):253–83. 21. Gustafson G, Koch G. Age estimation up to 16 years of age based on dental development. Odontol Revy. 1974;25(3):297–306. 22. Haavikko K. The formation and the alveolar and clinical eruption of the permanent teeth. An orthopantomographic study. Suom Hammaslaak Toim. 1970;66(3):103–70.
123
23. Hagg U, Matsson L. Dental maturity as an indicator of chronological age: the accuracy and precision of three methods. Eur J Orthod. 1985;7(1):25–34. 24. Harris MJP, Nortje CJ. The mesial root of the third mandibular molar. A possible indicator of age. J Forensic Odontostomatol. 1985;2:39–43. 25. Hunt EE Jr, Gleiser I. The estimation of age and sex of preadolescent children from bones and teeth. Am J Phys Anthropol. 1955;13(3):479–87. 26. Kullman L, Johanson G, Akesson L. Root development of the lower third molar and its relation to chronological age. Swed Dent J. 1992;16(4):161–7. 27. Liliequist B, Lundberg M. Skeletal and tooth development. A methodologic investigation. Acta Radiol Diagn. 1971;11(2): 97–112. 28. Moorrees CF, Fanning EA, Hunt EE Jr. Age variation of formation stages for ten permanent teeth. J Dent Res. 1963;42: 1490–502. 29. Nolla CM. The development of the permanent teeth. J Dent Child. 1960;27:254–66. 30. Frucht S, Schnegelsberg C, Schulte-Monting J, Rose E, Jonas I. Dental age in southwest Germany. A radiographic study. J Orofac Orthop. 2000;61(5):318–29. 31. Chen JW, Guo J, Zhou J, Liu RK, Chen TT, Zou SJ. Assessment of dental maturity of western Chinese children using Demirjian’s method. Forensic Sci Int. 2010;197(1–3):119 e1–4.
