Data processing and statistical analysis

The distribution of kinetic measurements (i.e. bending, torsion and axial) for the different motions were analyzed in the same manner as Drew and Izykowski et al. [17] by finding the peak forces and moments then averaging them over three trials. This was completed for each prosthetic model and then compared to the intact model at the different amputation levels (i.e. 25%, 50% and 75%, Fig 1). Log transformation of the observed measures created a more approximately normal distribution of the outcomes. A multivariate linear mixed effect regression approach was taken for all comparisons, which accounts for the multiple comparisons issue automatically, to simultaneously model the three force measurements. Comparisons were made to assess the difference between 1) each prosthetic model and the intact model, 2) prosthetic models, and 3) amputation lengths within each model. For each comparison, estimated percent differences in the measurements were calculated between 1) the four prosthetic devices (Table 1) and intact model for each combination of motion and amputation level, 2) each pair of prosthetic devices for each combination of motion and amputation level, and 3) each pair of amputation levels for each model and given motion. A total of 810 statistical comparisons were made to capture 4 prosthetic models at 3 amputation lengths for 6 activities.

Finally, a one-sided one-sample Wilcoxon Signed-Rank Test was performed to compare the median of observed loading values to the physical implant loading values from Drew and Taylor et al [22]. That study examined yield and ultimate failure of the interface between the bone and a porous coated transhumeral OI endoprosthesis at time-zero before bone ingrowth (not the implant failure itself), both with and without the use of stabilizing screws in axial, torsion, and bending loading modalities for proximal (30%) and distal (65%) amputation levels. Yield values indicate a threshold above which micromotion could initiate at the bone-implant interface and disrupt osseointegration, leading to early post-operative aseptic loosening [19, 20]. Ultimate failure values indicate a threshold where fracture between the bone and endoprosthesis is a dangerous possibility. The modeled 25 and 75% amputation levels were compared to the closest anatomical counterpart: proximal (30% amputation) and distal (65% amputation), respectively. This analysis was performed for all activities and prosthetic models.

Significance level was set at p≤0.05 and all analyses were performed using the statistical programming language R 3.6.3 (2019-07-05) [23]. The main functions applied were ggplot (from ggplot2 package), lme (from nlme package), and other Base R functions.