Bootstrap prediction and confidence bands: a superior statistical method for analysis of gait data

Area of Science:

• Biomechanics
• Orthopedics
• Statistical analysis in healthcare

Background:

• Gait analysis commonly uses continuous data curves over the gait cycle.
• Standard statistical methods for single data points are insufficient for curve analysis.
• Accurate statistical methods are crucial for reliable gait research.

Purpose of the Study:

• To evaluate the performance of two statistical methods for constructing prediction and confidence bands for continuous gait data curves.
• To compare point-by-point Gaussian theory intervals with simultaneous bootstrap intervals.
• To determine the most appropriate statistical method for gait data curve evaluation.

Main Methods:

• Cross-validation methodology was employed using a typical gait analysis dataset.
• Performance evaluation of point-by-point Gaussian theory intervals.
• Performance evaluation of simultaneous bootstrap intervals (Sutherland et al., Olshen et al.).

Main Results:

• Bootstrap bands demonstrated appropriate coverage for continuous gait data (86% observed vs. 90% targeted).
• Gaussian bands provided inadequate coverage (54% observed vs. 90% targeted).
• Gaussian method deficiencies can lead to inaccurate conclusions in gait studies.

Conclusions:

• Simultaneous bootstrap intervals are recommended for gait data curve analysis due to their non-parametric nature and reliable coverage.
• Bootstrap bands maintain nominal coverage levels for entire gait data curves, unlike Gaussian methods.
• Adoption of bootstrap bands as a standard method will enhance the accuracy and reliability of gait analysis research.