The use of Bayes Factor for model selection for predicting age: an example on clavicular histomorphometry

Introduction: Histomorphometry has been extensively applied for estimating age-atdeath. Although ribs have been commonly used for histological age, other bones need to be considered in order to contemplate possible bone specific remodelling rates and various recovery scenarios. Materials and Methods: Thirty-two left clavicle midshaft fragments were obtained from routine autopsies conducted at the Institute of Forensic Medicine (Albania, 2014-2015). Thin-sections of 0.5-1mm were prepared using standard protocols. Thirteen histomorphometric variables were collected using a Reflected Light Microscope and data were analysed using JASP 9.0.1. Results: Bayes factor showed that the existence of positive correlation of OPD with age is 348,050 times more likely than the lack of correlation. In addition, Mean Perimeter and Mean area are 1.047 and 1.357 times more likely to have a positive correlation with age that the alternative variables. Bayes linear regression was used to compare different predicting models. The best model includes OPD and Osteon Perimeter (R²=0.678, SEE=7.5years) which is 2.388,000 times more likely to explain age than the null model. Discussion: The Bayesian approach allowed the evaluation of the fitness of the model compared to the null model and to better explain the causative relationship of the predictors with age. This study is only the fourth worldwide performed on clavicular histomorphometry and provides unique population specific standards for age estimation of future forensic cases in Albania. The use of RLM as a new technique on forensic bone histology can be further tested and applied on a variety of skeletal elements.