Fig 1.
Trends in short-term outcomes for the Gait Deviation Index (GDI) show modest treatment effects (∼ 5 pts) that are stagnant and unpredictable.
The LOESS smoothing spline (span = .7) reflects the moving average.
Fig 2.
The notoriously difficult Step Two (Copyrighted artwork ScienceCartoonsPlus.com. All materials used with permission).
Fig 3.
The frequency of surgical procedure per limb reflects the underlying treatment philosophy at our center plus the nature of patients referred for gait analysis.
Table 1.
Limb characteristics.
Fig 4.
Concomitant treatment for limbs undergoing FDO or TDO.
Fig 5.
Overall performance of the IGA and CLI models.
Fig 6.
Comparison of CLI (top) and IGA (bottom) models. The predicted (x-axis) and measured (y-axis) results are shown for mean stance-phase foot progression angle. Columns reflect FDO and TDO status. Visually, it is apparent that the IGA model is more accurate and captures a wider range of treatment responses. These subjective impressions will be quantified below.
Fig 7.
RMSE lower (good) and r-squared higher (good) for IGA.
Fig 8.
IGA model does better at every level of complexity.
Fig 9.
No evidence of meaningful optimistic bias in the IGA model.
In fact, for some categories of limbs the repeated data exhibited slightly lower accuracy.
Fig 10.
Precision (CI) of CLI and IGA models.
A smaller value reflects a more precise prediction, and thus indicates superior performance.
Table 2.
Torsion profile.
Fig 11.
Note speed RMSE plotted at 100x to put it on similar scale as other outcomes.