Sexual dimorphism and population variation of a selection of vertebrae in a South African sample

Abstract

In the discipline of forensic anthropology, numerous methods exist to establish a biological profile. However, many methods rely on established standards created from a limited number of skeletal elements. If these skeletal elements are not recovered or are fragmentary, the established methods cannot be applied, and the biological profile cannot be estimated accurately. As such, further exploration into the variation of numerous alternative skeletal elements and their potential use in forensic analyses is required. The purpose of this study was to explore the osteometric variation of a selection of vertebrae to examine the population affinity and sexual dimorphism in a South African sample. Additionally, this study aimed to establish if a universal discriminant analysis can be used to estimate population affinity and sex without the prior determination of the specific number of vertebrae within the vertebral column. The current study consisted of two components, centred firstly around the exploration of vertebral variation and secondly on its use in predictive models. The variation exploration component quantified the osteometric variation among vertebrae to determine if there are statistically significant differences among vertebrae to see if non-specific, universal formulae - regardless of the specific number of a vertebra (e.g., T1, T2 etc.) - can be created. Secondly, the classification component aimed to apply the vertebral differences to develop classification standards to estimate population affinity and sex in a sample of black and white South Africans. For the variation exploration component of the study, a series of 20 measurements were collected to assess the variation within and among the cervical, thoracic, and lumbar vertebrae of a sample of 30 black South African males. Out of the original 20 measurements, 14 were observed to be sufficiently repeatable and were retained for the creation of predictive models. The vertebrae within each type were compared using ANOVA and Tukey’s HSD, which revealed that the atypical vertebrae (i.e., C7, T1, T12) were significantly different from the typical vertebrae within each vertebral region. As such, the atypical vertebrae are too different from the remaining vertebrae in each of the relevant subgroups and are not useful for non-specific, universal formulae. For the classification component, a selection of cervical, thoracic and lumbar vertebrae of 180 black and white South African adult males and females were measured. Both univariate models, as well as multivariate models (combining all of the measurements), were created for each vertebra to assess how accurately the vertebrae can predict population affinity and sex when using linear discriminant analysis. Finally, combined models were also created using the measurement means for all vertebrae in each subtype (i.e., a universal cervical model, universal upper thoracic model, universal lower thoracic model, and universal lumbar model). The univariate models using each measurement collected for each sub-type of vertebra separately, resulted in classification accuracies of 50% to 90% for population affinity and 35% to 92% for sex when considering all vertebrae. Overall, the cervical vertebrae (C3-C6) presented with the highest accuracies for sex estimation but yielded poor results for population affinity estimation, where both the thoracic (T2-T11) and lumbar (L1-L4) vertebrae performed comparably well. Among the measurements, the vertebral body, spinous process and transverse process lengths and height measurements performed the best across all vertebrae. The universal univariate models assessing each measurement collected for the combined cervical, thoracic and lumbar vertebrae within each sub-type presented with classification accuracies ranging between 78% and 82% for population affinity and between 73% and 84% for sex estimation, where the upper thoracic vertebrae (T2-T6) and lower thoracic vertebrae (T7-T11) performed the best. Multivariate models using all of the measurements per sub-type vertebrae resulted in classification accuracies between 73% and 94% for population affinity and 70% and 89% for sex estimation. Finally, the universal multivariate model assessed all the vertebrae together (cervical, thoracic and lumbar vertebrae), and demonstrated classification accuracies between 81.7% and 87.2% for population affinity estimation, and 81.7% and 87.2% for sex estimation, where the lower thoracic (T7-T11) vertebrae performed the worst for both biological parameters.