Evidence of Age Estimation Procedures in Forensic Dentistry: Results from an Umbrella Review

1. Introduction

Age estimation is a key forensic and archeological element. Often useful for human remains forensic identification, legal assistance involving minors or clinical diagnosis and planning, [1,2,3,4], is also helpful in mass migration and lack of valid identification [1,2,5]. Several methods have been developed to this end, among them, skeletal and dental development, sexual maturation or height/weight ratios [6,7].

Although the skull, the skeleton or the hand-wrist are used on age estimation as maturity indicators, they often present lack of good conditions for a correct identification or estimation [8,9]. Teeth are the hardest human organs and often found in adequate conditions [10,11,12]. Furthermore, dental measurements and indices are considered more useful and reliable, due to less variability during development as well as the greater resistance of teeth to systemic, environmental or destructive factors [6,13].

Estimating dental age may be achieved through several strategies depending on whether tooth development (around 20 years of age) or body development have been completed. On the one hand, methodologies based on teeth development are more accurate and with less margin of error [8,14]. On the other hand, the biological age of the individual is being estimated, always understood in a period of time, with some level of precision, and according to the method used. Chronological age will be included, at best, in this age range [12,14].

Several systematic reviews have been published with numerous dental methods based on radiographic (panoramic radiographs or otherwise) and non-radiographic approaches, most of them only evaluate one or two methodologies. Considering the variety and the discrepancy of the methods, it is helpful to compare and summarize the evidence previously published regarding age determination in Forensic Dentistry. This umbrella review intended to appraise the existing evidence on age determination procedures in forensic dentistry. Our focus was two-fold: to ascertain the quality of evidence and the overall clinical accuracy of each procedure.

2. Materials and Methods

We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline [15] (Appendix S1) and the guide for systematic reviews of systematic review [16]. The review protocol was approved a priori by all authors and registered on Open Science Framework (DOI 10.17605/OSF.IO/CPBZY).

The Review question was: “What is the current evidence on age determination approaches in Forensic Dentistry?”.

Eligibility criteria

To answer the proposed research question, the inclusion criteria were: (1) systematic review (with or without meta-analysis); (2) addressing age determination in Forensic Dentistry; (3) absence of data duplication within the included studies in the meta-analysis. No restrictions on year of publication or language were applied.

Information sources search

Electronic data search was performed in four electronic databases: PubMed, Cochrane Database of Systematic Reviews, LILACS (Latin-American scientific literature in health sciences) and Web of Science. We merged keywords and subject headings in accordance with the thesaurus of each database and applied exploded subject headings, with the following syntax “((age determination) OR (age determination forensic) OR (age estimation) OR (dental age estimation) OR (forensic age estimation) OR (age estimation methods) OR (age prediction) OR (dental age prediction)) AND ((tooth) OR (teeth) OR (dental) OR forensic OR (forensic dentistry) OR (forensic odontology)) AND ((Systematic Review) OR (Meta-analysis))”. Grey literature was searched through three appropriate databases (opensigle.inist.fr, https://www.ntis.gov/, https://www.apa.org/pubs/databases/psycextra).

Study selection

Two researchers (JAN and LBL) independently screened titles and abstracts. The agreement between the reviewers was assessed by Kappa statistics. Any paper classified as potentially eligible by either reviewer was ordered as a full-text and independently screened by the reviewers. All disagreements were resolved through discussion with a third reviewer (JB).

Data extraction process and data items

Two researchers (JAN and LBL) independently extracted: authors and year of publication, objective/focused question, databases searched, number of studies included, type of studies included, main results and main conclusions. All disagreements were resolved through discussion with a third reviewer (JB).

Risk of bias assessment

Two researchers (JAN and LBL) employed the A Measurement Tool to Assess Systematic Reviews (AMSTAR 2) to determine the methodological quality of the included systematic reviews [16]. AMSTAR 2 is a comprehensive 16-item tool that ranks the overall methodological quality of a systematic review. Accordingly, the quality is ranked as follows: High means ‘Zero or one non-critical weakness’; Moderate means ‘More than one non-critical weakness’; Low means ‘One critical flaw with or without non-critical weaknesses’; and Critically low means ‘More than one critical flaw with or without non-critical weaknesses. The estimation of the AMSTAR quality rate for each study was calculated through the AMSTAR 2 online tool (https://amstar.ca/Amstar_Checklist.php).

3. Results

Study selection

Electronic searches retrieved a total of 738 titles. After manual assessment of title/abstract and deletion of duplicates, 34 potentially eligible full-texts were screened (Figure 1). Full-text screening excluded 16 studies with reasons (Supplementary File 1), resulting in eighteen systematic reviews that fulfilled the inclusion criteria. Inter-examiner reliability at the full-text screening was recorded as high (kappa score = 1.00).

SR characteristics

Overall, fourteen SRs [6,17,18,19,20,21,22,23,24,25,26,27,28,29] with meta-analysis and four without [30,31,32,33] were included (Table 1). Multiple sub-topics were investigated, such as imagiology methods based on panoramic x-rays [6,19,20,21,22,24,26], CBCT and CT [30,32] and MRI [17,30] (Table 1).

One SR failed to report a defined timeframe [33]. Seven SRs described their studies eligibility criteria in the review [20,24,25,30,31,32,33], while four could not describe in adequate detail [18,19,22,30], and eight had no funding information [17,18,23,26,28,29,30,32].

Methodological Quality

We observed excellent inter-examiner reliability at the RoB assessment (kappa score = 0.93; 95% confidence interval: 0.91-0.95).

None of the included SRs fully satisfied the AMSTAR2 Criteria (Table 2). Four studies were rated as of ‘high quality’ [6,21,26,27], three as of ‘moderate quality’ [17,22,25], six as of ‘low quality’ [18,20,28,29,31,33] and five as of ‘critically low quality’ [19,23,24,30,32].

When analyzing the major inconsistencies detected by the AMSTAR2: all studies did not report on the sources of funding for the studies included in the review (100%, n=18)[6,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33]; 44.4% did not have data selection in duplicate (n=8)[18,22,23,24,25,28,30,32]; 38.9% of studies did not explain selection literature search strategy (n=7)[19,24,25,30,31,32,33]; and 33.3% did not have study selection in duplicate (n=6)[18,22,23,24,30,32].

Synthesis of Results

Overall, three main topics of research were found among the included SRs: panoramic radiographs-based methods; three-dimensional imaging methods; and artificial intelligence (AI)-based methods.

Panoramic radiographs-based methods

Overall, the level of evidence of the SRs focusing forensic methods based on panoramic radiographs was of low quality. Three main methods were the aim of research: Demirjian’s [6,20,21,22,23,24,26,33], Willems’ [21,22,25,26,28,29] and Cameriere’s [19,22,27,32].

As regards to Demirjian’s method, all studies are in agreement of an overestimation that varies between 4 to 9 months [6,20,21,22,24,26]. The majority affirm that this method is geographically sensitive [6,20,21,24] except the studies that are single-population oriented [22,26]. With respect to sex, two studies tend to overestimate the females [21,24], two showed more overestimation in males and lastly, one study reported that there were no differences between sexes [6] and one did not analyze sex subgroups [22]. Demirjian’s method was also applied solely to the 3rd molar in two papers. Haglund et al. [23] defined the accuracy of the method for the 18 years old threshold as 71% and Rolseth et al. [33] identified that the different development ranged from 4 to 7 years of the 3rd molar mineralization.

As for the Willems’ method, most studies showed a slight age overestimation, varying between 1 to 5 months [21,22,25,28,29], aside from one [26] that concluded underestimation by a month. They also conclude that males are more susceptible to this overestimation [25,26,28,29], except one [21], which concludes that females are more sensitive. One other study [22] does not analyze sex subgroups. Respecting age and geographic subgroups, some authors [21,25,28,29] reported differences between populations as the rest only studied a single population [22,26]; Esan et al., Yosuf et al. and Wang et al., also stated that a few age subgroups are more prone to overestimation [21,25,29].

Regarding Cameriere’s method, studies demonstrated overestimation that varied from 3 months to one year, making this variation slightly greater in males, but without statistically significant difference [19,22,32]. Marroquin et al. [32] also reported that in the Indian subpopulation, this variation can be as high as 10 years of overestimation. Cameriere also developed an index to assess the threshold of 18 years old with a percentage of correct classification ranging from 72 to 96%, with a better accuracy in males [27].

Kvall’s method [32] overestimates age between 1 to 2 and 12 to 13 years old and Chaillet’s method [18] within 6 to 8 months. Both methods overestimate more females and present different results when analyzing various geographical subpopulations.

Other methods included in this overview were only investigated by Franco et al. [22] for the Brazilian population. Nolla’s method demonstrated 2 to 3 months of overestimation, Lilequist and Lundberg’s method 1 to 2 months of underestimation, Mornstard’s 3 to 4 months of overestimation and lastly, Haavikko’s method underestimates between 10 to 12 months.

Three-dimensional imaging methods

Overall, the level of evidence of the SRs relating to forensic methods based on 3 dimensions imaging was of low quality.

Cone Beam Computed Tomography (CBCT) and Computed Tomography (CT) Scans reported different margin of errors, according to the method applied, ranging from 3,5 to 28 years [32]. When using pulp/tooth ratio, individuals were correctly identified between 30 to 90%, being the majority around 60% [30]. Both studies [30,32] agree that these values differ depending on the sex of the individual and the type of the tooth.

As for Magnetic Resonance Imaging (MRI), all studies concluded that results from images obtained through MRI are equated to panoramic radiographs, without the ionizing radiation [17,30].

AI-based methods

The level of evidence on AI-based methods for sex prediction using dental measures was collectively based on low quality SRs. AI displayed precision and accuracy similar to trained examiners, overcoming the observer subjectivity. Although the accuracy, real-life testing and validation are yet to be proved [31].

4. Discussion

This umbrella review was able to sum up the evidence provided by the available SRs on age estimation methods in Forensic Dentistry. The collective knowledge is currently based on low to moderate-confidence evidence-based studies, at best, ranging between critically low to high quality. Overall, these results show that some forensic tools used nowadays might be outdated or misused, and some results must be analyzed with care, due to the lack of quality of many of the studies here referred.

Age estimation is a major step in forensic and archeological investigation [4]. Despite several methods have been developed to this end, approaches using dental tissues are among the most useful and reliable, due to their low changeability and greater resistance to degradation [7,13]. Most studies rely on methods developed upon bidimensional radiographic images, such as the Demirjian’s and Willems’ methods that have seen validation worldwide [6,18,19,20,21,22,24,25,26,27,28,29,32,33]. Both methods tend to overestimate age [6,20,21,22,23,24,25,26,28,29,33]. Cameriere’s method was first studied for several age groups and later as an index (I3M) that differentiates the legal threshold of 18 years old [19,22,27,32]. Other methods included are Chaillet’s, Lundbergs, Nolla, Mornstad, Haavik and Kval’s methods [18,22,32]. More recently, a few studies investigated the applicability of 3-dimensional imaging in forensic dentistry [17,30,32]. AI technology has already been tested for age estimation and seems a potential forensic tool [31].

Demirjian’s method obtained global acceptance and became the most widely used for dental age estimation [21]. Nonetheless, most studies concluded that this method is geographically sensitive and varies according to the subpopulation studied [6,20,21,24]. A possible reason for such a thing was the origin of this dataset (Caucasian subpopulation) with low heterogeneity, different from the other subpopulations studied [6,20,21,24]. Demirjian’s method tends to overestimate the age of the individuals, regardless of the sex of the subject. For such reasons, Demirjian’s method renders as a poor forensic tool when misapplied [20,24]

Willems’ method is also geographically sensitive [21,25,28,29]. Sehrawat et al. [28], for instance, determined that this method overestimates in the majority of the countries, except China and India. Sex seems to be a factor to consider because this method tends to overestimate more males than females [21,25,26,28,29].

Three studies comparing both Demirjian and Willems’ methods are all in agreement that the latter is more accurate and less prone to overestimation [21,22,26].

Demirjian’s method was introduced in the 1970′s and Willem’s method in the 2000′s. Since these methods (and most of the remaining methods) are based on the tooth maturation and this characteristic is growth dependent, it is likely to need to be updated on a regular basis. Growth patterns have evolved with the improvement of healthcare, nutrition and genetics so new methods must be developed to accompany the evolution of times [21,34].

Not only in Forensic Dentistry, dental methods and indexes have been misused. In Orthodontics, indexes such as Bolton, developed from a specific subpopulation, have been proven incorrect when generalized to other populations [35] .

Cameriere’s method is one of the geographically stable methods and tends to overestimate by 4 months, both boys and girls, without statistically significant difference [19,22,32]. According to Hostiuc et al. this method seems to outperform several others, including Dejirmijian and Willems methods [19].

Santiago et al. concluded that I3M was validated worldwide in several subpopulations since it has high accuracy in discriminating if an individual has reached 18 years of age, regardless of population studied. Regarding sex, a tendency of better results was observed for males, but high values of accuracy, sensibility, and specificity were also achieved for females [27].

Forensic age estimation based on dental measures has been, until recently, based on bidimensional imaging. 3-Dimensional imaging is gaining more relevance in all dentistry areas, including forensic dentistry with the first study that reported the use of this technology was published in 2004 [36]. CBCT and CT scans reconstructions allow the investigators to analyze the pulp/tooth volume ratio. The volume ratio might be an interesting tool for predicting age after root maturation of the 3rd molar, around the second decade of life [30]. Despite Micro-CT scans requiring the use of extracted teeth, the images are of greater quality and produce accurate measures because of more spatial resolution that of a CBCT; however, its application in live subjects is not viable and a model based in this type of imaging may not be replicable for forensic proposes in live individuals [30,32]. One major limitation of this method are the artefacts produced by adjacent metal structures and restorations, such as implants and amalgam fillings [30]. Also, the difficulty in reproduction of the site for measurements might lead to inaccurate analysis; the lack of a simple method of investigation should be the focus for the next researchers [30].

Due to the ethical implications of the usage of ionized radiation for other than diagnostic indications, MRI arose as a valid alternative since it uses strong magnetic fields and radio waves to generate imagens. It’s a relatively new tool in age estimation, published for the first time in 2015 [37]. Regarding the methodology itself, MRI can be used associated with other methods previously stated, such as Demirjian’s. MRI tends to be more accurate in the yearly stages of development but to be more challenging and inaccurate in the latter stages because of the lack of contract between dental and bone tissue. It also takes more time and is more expensive than an ordinary panoramic radiograph [30]. Due to the scarcity of research on this subject, the inter-ethnic variability is not yet proven [17]. Discrepancies between the MRI approaches make it inappropriate to pool data together and perform a proper systematic review with meta-analysis. Also, future age estimation methods based on MRI will probably be based on multifactorial sites and measures [17].

AI based automated systems have been developed to surpass the examiners subjectivity. The AI model that best performed was the Deep Learning Convolutional Neural Network approach, with similar accuracy when compared with trained researchers. AI models that combine a dual Convolutional Neural Network, first to predict sex and afterwards age, outperformed a single Convolutional Neural Network approach. However, AI based models have not proved themselves in the field to be routinely applied [31].

With the purpose of improving the quality of research, the checklist from TRIPOD [38] should be followed. The research question should be clearer with special emphasis on selection of study designs and reasons for exclusion. It would also enrich future SRs if they considered the RoB of individual studies as well as the number of authors that performed data extraction, to prevent bias.

Strengths and limitations

The present umbrella review has several strengths. Overall, these results provide a comprehensive overview of the available SRs on REP using a transparent and evidence-based methodology. We commend a cautious interpretation, as the individual studies included in each of the present SR were not explored. Thus, the conclusions lean on the interpretation of the systematic review’s authors.

5. Conclusions

Current evidence on forensic dentistry methods for age estimation is supported by low to moderate confidence systematic reviews. Willems’ method is more accurate and less prone to overestimation. Most methods seem to be geographically sensitive despite some authors attributing this heterogeneity to methodological errors. Cameriere’s index has high accuracy regardless of the population studied. 3D imaging an AI technology, although in the rise still lack field validation. This umbrella review reports the most common mistakes performed in SRs and will pave the way for more robust evidence-based research in the future.