Measuring a bone’s response to impact has traditionally been done using strain gauges that are attached directly to the bone. Accelerometers have also been used for this purpose because they are reusable, inexpensive and can be attached easily. However, little data are available relating measured accelerations to bone injury, or to judge if accelerometers are reasonable surrogates for strain gauges in terms of their capacity to predict bone injuries. Impacts were applied with a custom designed pneumatic impact system to eight fresh-frozen human cadaveric radius specimens. Impacts were repeatedly applied with increasing energy until ultimate failure occurred. Three multiaxial strain gauge rosettes were glued to the bone (two distally and one proximally). Two multiaxial accelerometers were attached to the distal dorsal and proximal volar aspects of the radius. Overall, peak minimum and maximum principal strains were calculated from the strain-time curves from each gauge. Peak accelerations and acceleration rates were measured parallel (axial) and perpendicular (off-axis) to the long axis of the radius. Logistic generalized estimating equations were used to create strain and acceleration-based injury prediction models. To develop strain prediction models based on the acceleration variables, Linear generalized estimating equations were employed. The logistic models were assessed according to the quasi-likelihood under independence model criterion (QIC), while the linear models were assessed by the QIC and the marginal R2. Peak axial and off-axis accelerations increased significantly (with increasing impact energy) across all impact trials. The best injury prediction model (QIC = 9.42) included distal resultant acceleration (p < 0.001) and donor body mass index (BMI) (p < 0.001). Compressive and tensile strains were best predicted by separate uni-variate models, including peak distal axial acceleration (R2 = 0.79) and peak off-axis acceleration (R2 = 0.79), respectively. Accelerometers appear to be a valid surrogate to strain gauges for measuring the general response of the bone to impact and predicting the probability of bone injury.
The use of prediction models for eliminating effects due to regression-to-the-mean in road accident data
