A multivariable model combining skeletal maturation and bone length may improve forensic age estimation accuracy compared with traditional single-indicator methods, particularly for identifying patients at the legal age of 18 years or older, according to a recent study.
In the retrospective analysis involving 666 chest radiographs from participants aged 10 to 30 years at three large Italian hospitals, researchers evaluated medial clavicular epiphysis (MCE) ossification stage and clavicle length measured on conventional imaging. They developed three regression models—MCE stage plus sex, clavicle length plus sex, and MCE stage plus clavicle length and sex—to estimate the participants' chronologic age and classify them as minors or adults at the 18-year threshold.
The combined model including all three variables (MCE stage, clavicle length, and sex) showed the strongest performance, with an area under the curve of 0.94 and overall accuracy of 87%. The model also achieved 91.4% sensitivity and 80.1% specificity in identifying adult participants, compared with 62.5% sensitivity and 98.1% specificity for the commonly used threshold of MCE stage 4 or higher.
Single-variable models were unable to achieve the same performance. For instance, the model based on clavicle length alone showed lower accuracy and poor specificity, while the MCE-only model performed well but didn't match the combined approach.
Subgroup analyses showed that clavicle length was associated with age in patients aged 10 to 15 years but wasn't a significant predictor in patients aged 16 to 20 years, the range most relevant for determining legal adulthood.
The researchers noted that conventional radiography allowed evaluation in most cases but may be limited by overlapping anatomical structures. The study population was limited to Italian patients, and the use of a single anatomical region resulted in wider prediction ranges compared with multifactor approaches that combine hand-wrist maturity and dental markers.
The findings supported the use of statistical modeling over rigid thresholds when evaluating existing imaging data. “These findings suggest that transitioning from rigid cut-offs to statistical modeling provides a more balanced risk assessment, reducing false negatives,” wrote lead study author Giorgio De Donno, MD, of the Interdisciplinary Department of Medicine in the Section of Legal Medicine at the University of Bari Medical School “Aldo Moro,” and colleagues.
The study authors reported no competing interests.
Source: Journal of Forensic Sciences
