External validation of femoral sex estimation equations: Evidence supporting population-specific standards in forensic anthropology

Background

Sex estimation from skeletal remains represents a fundamental step in forensic identification, with femoral measurements offering reliable alternatives when traditional pelvic and cranial elements are unavailable. Population-specific variation significantly affects accuracy, yet no external validation has examined cross-regional applicability of Thai femoral sex estimation equations. This study evaluated the performance of established Thai regional equations when applied to independent samples.

Materials and methods

External validation was conducted using 200 femoral specimens (100 male, 100 female) from northeastern Thailand, representing an independent collection not used in previous equation development. Five femoral parameters were measured following standardized protocols. Performance of sex estimation equations from northeastern, central, and northern Thai populations, plus one Chinese equation, was assessed through comprehensive validation metrics including accuracy, sensitivity, specificity, Cohen's kappa, and area under the curve (AUC) from ROC analysis.

Results

The northeastern Thai equation demonstrated optimal performance with 93.0 % accuracy, κ = 0.86, and AUC = 0.975 when applied to the same population. Cross-regional Thai equations showed moderate accuracy reductions: central Thai achieved 91.5 % and northern Thai achieved 90.5 %. The Chinese equation exhibited the greatest decline at 87.0 %, representing a 6.5 % reduction from optimal performance. Male classification consistently outperformed female classification across all equations with sensitivity ranging from 92.0 to 95.0 % versus specificity ranging from 82.0 to 91.0 %. Performance decline correlated with increasing population genetic distance.

Conclusion

External validation confirms the importance of population-specific standards in forensic anthropological practice. While Thai regional equations maintain acceptable cross-regional performance, optimal accuracy requires population-appropriate equation selection. The substantial performance decline with ethnically distant equations emphasizes careful equation selection for reliable forensic identification.