Evaluation of effectiveness of cranial morphological characteristics for sex estimation in a Brazilian sample

Forensic physical anthropometry allows the determination of animal species and estimates sex, ancestry, age and height. Aim: To analyze the effectiveness of a cranioscopic/ morphological evaluation for sex estimation with a sample of the Brazilian mixed-race population by conducting a qualitative visual assessment without prior knowledge of sex. Methods: This is a blind cross-sectional study that evaluated 30 cranial characteristics of 192 skulls with mandible, 108 male and 84 female individuals, aged 22 to 97 years, from the Osteological and Tomographic Biobank. The qualitative characteristics were classified and compared to the actual sex information of the Biobank database. The statistical analysis was used to calculate de Cohen’s kappa coefficient, total percentage of agreement, sensitivity and specificity of visual sex classification. Results: Of the 30 cranial variables analyzed, 15 presented moderate degree of agreement, achieving value of Kappa test between 0.41–0.60: Glabella (Gl), Angle and lines (At), Mental eminence (Em), Mandible size (Tm), Cranial base (Bc), Mouth depth (Pb), Nasal aperture (Anl), Supraorbital region (Rs), Orbits (Orb), Mastoid processes (Pm), Alveolar arches (Aa), Zygomatic arch (Az), Orbital edge (Bo), Supraorbital protuberances (Pts), and Supramastoid crests and rugosity (Crsm). The Facial physiognomy (Ff) presented substantial reliability (0.61-0.80) with 89.8% sensitivity for male sex and 70.2% specificity. Conclusions: Cranial morphological characteristics present sexual dimorphism; however, in this study only 15 variables showed moderate degree of agreement and can be used in sex estimation. Only one variable (Ff) 81.2% total agreement with substantial reliability. Quantitative methods can be associated for safe sex estimation.


Introduction
In an anthropological examination for forensic purposes, the determination of sex, species, ethnic group and estimated age and height are essential as such information help build an individual's biological profile 1,2 and subsequent identification.
Musilová et al. 2 (2016) e Durić et al. 3 (2005) portrayed that the pelvis is the structure of the human skeleton that presents the highest degree of sexual dimorphism, being the most reliable bone for sex estimation. According to Musilová et al. 2 (2016) the pelvis responds with evolutionary adaptation to bipedal locomotion and birth mechanisms, enabling parturition of children with relatively big brains.
But, in situations where the skeleton is not complete 4 or when the pelvis is not fully recovered, sex estimation can be achieved by performing a cranial analysis. As mentioned by Spradley and Jantz 4 (2011), the skull has a high correct classification of sex, of 90-91%.
Lewis and Garvin (2016) 7 , Biancalana et al. (2015) 8 , Godde (2015) 9 , Tallman and Go (2018) 10 , Walker (2008) 11 and Langley et al. (2018) 12 are some authors which estimated the sex based on the skull morphology. This qualitative analysis is based on visual examination of the presence or degree of expression of morphological characteristics 7 , but despite its subjectivity [8][9][10][11]13 , may be the only possible method in cases of bone fragmentation 7 . An examination of non-metric traits also ensures an easy and fast analysis, without requiring any devices 7,10,12 .
In general, bone aspects such as prominences, crests and apophyses are more notable in men, while women have more delicate and less pronounced characteristics 14 . Walker (2008) 11 stated that the accuracy of sex determinations based on visual inspection depends on the osteologist's familiarity with the population being studied. And Franklin et al. (2013) 15 mentioned that the forensic practitioner should access an osteological database for their specific geographic jurisdiction.
This changes in skull shape and size are population-specific 11,15 , and can be explained because each population is submitted to its own forces of evolution 1,9,10 . Environmental interventions 8,10,15 , nutritional status 8,9 , temporal changes [8][9][10][11][12] and biomechanical processes related to neck, face and head movement 9 can alter cranial morphological aspects, smoothing or enhancing the robustness of some characteristics.
In view of these situations, this study aimed to analyze the effectiveness of a cranioscopic/morphological evaluation (a qualitative visual assessment without prior knowledge of sex) with a sample of the Brazilian mixed-race population, for sex estimation.
The main sample consisted of 192 human skulls without alterations that impaired the analysis of morphological characteristics, 108 were male and 84 were female skulls, aged between 22 and 97 years, with median age of 57 years, from the Osteological and Tomographic Biobank. The year of death of this sample varied from 2006 to 2010.
The skulls of individuals who were 22 years or older at the time of their death were analyzed, excluding skulls of individuals who had not reached puberty as they show slightly pronounced qualitative characteristics, providing little information for sex estimation 12,14 .
All analyses were performed by a single rater. The researcher was previous calibrated analysing all the dichotomous variables (male or female) in 10 skulls, obtaining 100% of consensual agreement between itself and a gold rater. The calibration has not been made through a statistical test. And the sample used for calibration was not included in the main sample.
Then the main sample of 192 skulls was evaluated, of which 30 cranial anatomical structures that were analyzed by this rater through visual inspection, using a non-metric method without prior knowledge of sex. Table 1 shows the morphological characteristics analyzed in this study. The percentage of correct sex was calculated using frequency tables (Crosstab) and the Cohen's kappa coefficient, total percentage of agreement, sensitivity, specificity visual sex classification were calculated. It was used SPSS Statistics version 25 statistical package (IBM Corporation, Chicago, USA) for data processing.

Results
The frequency and percentage of skull's real sex are described in Table 2. The sample was relativily balanced with 56.3% of male and 43.8% female.
32.8% of the sample consisted of skulls which age range was between 26-50 years old, 33.3% between 51-70 years and 30.7% over 70 years old. And 96.5% of the skulls analysed presented the year of death 2010. Due to the sample consisted of skulls aged older than 22 years, the authors performed the skull concordance tests without separating them by age. Table 3 shows the frequency and percentage of sex classification based on a subjective analysis of different variables.
Through Kappa test, variables Pe (k=0.08) and Ct (k=-0.04), should not be used to determine sex, as they do not present statistical significance in the agreement test (p>0.05). Although the variable tooth size (Td) has been visually analyzed, its statistical analysis was not made due to several skulls had no tooth.
The other variables showed statistically significant agreement from slight to fair level.    Table 5 shows the percentage of correct sex through Crosstab of 16 (sixteen) variables that presented statistically significant moderate and substantial agreement (k>0.40; p<0.05). The visual sex classification of Ff, At and Gl variables presented more than 80% of total agreement. The visual sex classification of Ff variable showed 89.8% sensitivity and 70.2% specificity for male sex classification. The At showed 87% sensitivity and 72.6% for male sex specificity and Gl showed 89.8% sensitivity and 69.0% specificity. The visual classification was more sensitivity for male sex, whereas for the female classification the sensitivity was lower. The variable Orb showed 92.6% sensitivity for male sex, however, 51.2% of specificity. The Ff, Gl and At were the variables with the highest index of sensitivity and specificity.
The percentages of correct sex for the 10 morphological variables with the best agreement degrees are highlighted with a hash in the same Table 5 and are presented in the Figure 1; all of them had higher percentages in the determination of male in relation to female sex, that is, such characteristics are more evident and facilitate identification in male sex.

Discussion
In our study, using a balanced sample, although 30 cranial characteristics were analysed, only 15 variables achieved moderate agreement and one substantial agreement, but according to Cicchetti and Feinstein 17 (1990) a low kappa can occur at a high agreement.
It is remarkable that age is a variable that can influence the quantitative measurement of bone size. However, the present study classified each skull qualitatively and the sample consisted of skulls aged older than 22 years, not dealing with the stage before puberty, between 10 and 21 yearsold which can be a confounding bias in the identification of the gender. For this reason, the skulls concordance tests were performed without separating them by age.
As the sex of all skulls was indexed in the Biobank, it was possible to estimate the percentage of correct sex for female and male population, which was not obtained in the study by Biancalana et al. 8 (2015). Correct sex percentages for the 10 characteristics presenting the best agreement results were high, ranging from 47.6% to 92.6%. Being observed that this qualitative classification is more sensitivity for male sex.
As observed in other studies 2,7,13,18,19 , glabella presented a high sexual dimorphism index, reaching 89.8% sensitivity and 69.0% of specificity for male sex classification, and a total agreement of 80.7%.
In a qualitative analysis of the glabella region similar to ours, Abdel Fatah et al. 13 (2014) and Walker 11 (2008) found correct sex classification of 82% and 82.6%, respectively.
Langley et al. 12 (2018), when analyzing non-metric cranial traits, observed that mental eminence was the only variable that did not present a reasonable to moderate agreement for sex estimation, and should be avoided for such purpose. Low accuracy of 45.03% was also obtained in the study by Durić et al. 3 (2005) for the size of mental eminence. In contrast, in our study, mental eminence was among the 10 best variables, with total agreement of 79.1%, having 84.3% of sensitivity for male sex and 53.6% for female sex. Similarly, Lewis and Garvin 7 (2016) and Walker 11 (2008) found correct sex classification for this variable of 75.0% and 76.6%, respectively.
Nikita and Michopoulou 19 (2018) also analyzed the mastoid process profile and found for this variable, up to 75.2% and 74.5% correct classification for both sexes, similar to the percentages found by Walker 11 (2008), while our study found for this characteristic, 85.2% of sensitivity for male sex and 58.3% for female sex.
In a study conducted by Graw et al. 20 (1999), the analysis of the supraorbital margin shape allowed correct identification of sex, with about 70% accuracy, an index that is similar to what was observed by Walker 11 Keen 22 (1950) reports that the angle of the mandible did not present a high index in sex differentiation.
In 2018, Tallman and Go 10 evaluated Asian skulls, qualitatively analyzing nuchal crest, mastoid process, supraorbital margin, glabella and mental eminence, and obtaining 57.9% correct classification for female and 92.4% for male sex.
Similarly, was observed in our study that the visual classification of the cranial characteristics was more sensitivity for male sex. And to increase the probability of correct sex determination, we agree with Loth and Henneberg 23 (1996) when they advise that a complete examination should be made of all available bones known to belong to an individual, combining qualitative and quantitative methodologies, to ensure improved certainty and reliability of forensic anthropological reports.
The percentages showed here are helpful for forensic practioners according to which preserved cranial trait is available in the skull that they are working on, and according to the population the skull is originated.
However, this study presents some limitations, as the analyses were performed by only one examiner, and although calibrated, by being a qualitative analysis it can be influenced by subjectivity, so the authors suggest that future studies use some examiners. Another limitation was common lighting and tables to analyze the skulls, being evaluated with the naked eye.
In conclusion, the visual classification of Ff variable presented the best sensitivity and specificity to male sex with substantial reliability. Next in decreasing order for the best qualitative evaluation of sex were the variables Gl, At, Em, Tm, Bc, Pb, Anl, Rs, Orb Pm, Aa, Az, Bo, Pts and Crsm which presented moderate agreement (41% to 60%). The visual classification was more sensitivity for male sex. However, for improved certainty and reliability in sex estimation, quantitative methods are recommended.