Sex estimation using adult femora and humeri trained on a Portuguese reference sample, and tested on Portuguese and South African samples

Forensic anthropology involves analysing human skeletal remains to assist legal investigations, with sex estimation being a key component of constructing a biological profile. This study evaluated the feasibility of developing metric methods for sex estimation using femora and humeri across different populations. Logistic regression models were created from Portuguese training samples (360 femora; 329 humeri) and tested on independent Portuguese (200 femora; 150 humeri) and South African (66 femora; 66 humeri) samples. The primary goal was to develop sex estimation models using pooled data from Portuguese samples and assess their reliability across Portuguese and South African populations. Femoral models for the Portuguese testing sample achieved an accuracy between 82.0% and 85.5%, with sex bias from 0.0% to 8.0%. Humeral models showed higher accuracy, between 88.0% and 94.0%, with sex bias from 2.4% to 13.1%. In the South African sample, both femoral and humeral models exceeded 75.0% accuracy but exhibited notable sex bias. The results suggest that using a heterogeneous training set, combining data from diverse populations, offers a broader understanding of sexual dimorphism compared with a homogeneous dataset from a single group. The study emphasizes that relying solely on skeletal samples from one population is insufficient for modern forensic anthropology standards.