As a complement to macroscopic aging methods, or when necessary macroscopic elements are damaged or absent, human age can be estimated through histological examination of remodeling events in cortical bone. During the last half century the femoral midshaft has been the most commonly employed site for histological studies; however, a consensus is still lacking on where to best quantify remodeling, as different methods employ various regions of interest (ROIs) that differ in size, number, and location. This study employed geographical information systems (GIS) software to elucidate patterns in the spatial distribution of remodeling across the entirety of the femoral midshaft, to identify which region(s) produce the most accurate age estimates.

Thirty complete cross-sections from modern cadaveric donors were used, 15 of each sex, ranging from 21--97 years old. Using ArcGIS software, cross-sectional images of each sample were used to build maps depicting macroscopic and microscopic features of the femoral cortex. Polygon features were created to overlay cortical areas, while all remodeling events were digitally annotated with point features. A total of 10 different sampling strategies were employed, investigating the entire cortex, separating the cortex into quadrants and octants, and separating each of these into periosteal, middle, and endosteal thirds. Osteon population density (OPD), calculated as the number of remodeling events per square millimeter of each ROI, was used as the main variable to estimate age at death.

GIS density maps revealed that remodeling events are not evenly distributed throughout the femoral cortex. A striking pattern emerged where the highest density of remodeling events was nearly always in the anterolateral region of the cortex. Density maps also allowed examination of the femur for the OPD asymptote. Unlike smaller skeletal elements, an entire femoral cross-section will fail to hit an OPD asymptote due to its greater cortical area and spatial variation in remodeling density.

Stepwise linear regression was used to determine which ROIs from each segmentation method could predict age with the greatest accuracy. All regression models produced indicate a robust relationship between OPD and its ability to predict age at death, with the best model explaining more than 93% of the variability in age with a standard error of only 5.82 years. ROIs selected by stepwise linear regression for each segmentation method were predominantly in the anterior, lateral, or anterolateral cortex.

To further evaluate the performance of the predictive regression models, jackknife age estimates were constructed. Jackknife estimates from the most promising regression models yielded average differences between estimated and known age (bias) of less than one year, and average absolute differences between estimated and known age (inaccuracy) of less than six years. Remarkably, even individuals in the tenth decade of life had bias and inaccuracy values less than four years. Overall this research demonstrates the great utility of GIS software in skeletal biology research, and provides anthropologists with a powerful tool to accurately estimate age at death across the entirety of the adult aging spectrum. Such a tool may be useful in both paleodemography and forensic anthropology.