A method of sexing the human os coxae based on logistic regressions and Bruzek's nonmetric traits.

Abstract

This study aims at proposing a visual method for sexing the human os coxae based on a statistical approach, using a scoring system of traits described by Bruzek (2002). This method is evaluated on a meta-population sample, where the data were acquired by direct observation of dry bones as well as computed tomography (CT) scans. A comparison with the original Bruzek's (2002) method is performed. Five hundred and ninety two ossa coxae of modern humans are included in the reference dataset. Two other samples, composed respectively of 518 ossa coxae and 99 CT-scan images, are both used for validation purposes. The individuals come from five European or North American population samples. Eleven trichotomic traits (expressing female, male, or intermediate forms) were observed on each os coxae. The new approach employs statistical processing based on logistic regressions. An R package freely available online, PELVIS, implements both methods. Both methods provide highly reliable sex estimates. The new statistical method has a slightly better accuracy rate (99.2%) than the former method (98.2%) but has also a higher rate of indeterminate individuals (12.9% vs. 3% for complete bones). The efficiency of both methods is compared. Low error rates were preferred over high ability of reaching the classification threshold. The impact of lateralization and the asymmetry of observed traits are discussed. Finally, it is shown that this visual method of sex estimation is reliable and easy to use through the graphical user interface of the R package.