Gait scores commonly summarize someone’s gait into a single, convenient metric. Oftentimes, researchers consider these measures when we want to find a simple way to communicate the findings of the kinematic signals. This method is particularly helpful when explaining results to someone who might not know the specific biomechanics. This single metric typically categorizes someone’s gait as “Normal” vs “Different”, or perhaps “Asymmetric” vs “Symmetric”.
If, for instance, we have a group of control subjects, we can summarize their performance by calculating average lower limb kinematics per stride. This is illustrated in the figure below. We can compare a single subject to this control group by calculating the root mean square (RMS) difference in lower limb kinematics. We call this measure the Gait Profile Score (GPS). The GPS uses a single value to indicate the difference between an individual’s gait and a control group’s average.
A similar measure, called the Gait Deviation Index (GDI), identifies independent lower limb joint rotation patterns, or feature components, from a large sample of strides. The GDI measures how an individual’s gait deviates from the control group’s average. It uses a scaled linear combination of differences in feature components to calculate the deviation.
Both the GPS and GDI demonstrate great reliability. So if I were to calculate these measures on my own data, what would the scores mean?
Classifying Data with Single Metric Scores
We report the GPS in the same units as the kinematic measures used to compute the RMS differences. So in this case, we would report the GPS in degrees. We can infer, then, that as long as your modeling decisions and kinematic measures are consistent across participants, higher GPS values indicate a “less normal” gait pattern.
The GDI is a unitless scaled value that centers around 100, rather than presenting in degrees. It is typical to use a GDI score of 100 as a threshold for categorizing individuals. A GDI above 100 indicates that the subject’s gait is as “normal” as any randomly selected individual from the control group. Every 10 points below a score of 100 is equivalent to being one standard deviation away from the control group. So, a lower GDI score indicates a “less normal” gait pattern.
Comparing GDI and GPS
So we find the GPS score by comparing lower limb kinematics of individuals to a normal population. And, we find the GDI score by comparing lower limb kinematics of individuals to a normal population. If it isn’t obvious yet, these measures are quite similar… and let me show you why.
The GPS and GDI use the same kinematic features to compare an individual against a control group. Not only do they tell us the same information, we can find one measure from the other.
Thanks to Richard Baker and Michael Shawartz, we learned that we can derive the GDI directly from the GPS measure. This greatly simplifies the calculation of the GDI. Schwartz et. al. published this in 2014 as a method of showing how the GPS and GDI are substantially equivalent by using a scaled GPS value.
Deriving the GDI from the GPS
So I went ahead and tested this myself! First I calculated the traditional GDI and GPS on my dataset using Richard Baker’s publicly available spreadsheet. I then found each subject’s average kinematic features in the control group compared to the overall average of the control group. I then found a mean and standard deviation of this variability across the control group to scale the GPS score of the subject. This gave me a derived GDI, which Baker and Schwartz called the GDI*.
My own comparison between GDI and GDI* shows a strong linear regression. The figure below shows a slope of 0.99 and an R-squared value of 0.99. These results align with the trend seen by Schwartz et. al., and supports the claim that we can derive the GDI measure from the GPS.
Rasmussen et. al. not only showed that these measures were reliable, but they also had very strong agreement in classifying an individual’s gait. And as we have shown throughout this blog, these measurements are telling us the same information. So do we need both measures? The answer is no, they contain the same information. Your choice should be based on your audience and how you can most effectively communicate with them.
This entire breakdown of the GDI from the GPS appears in Richard Baker’s 3 part blog post. I highly recommend reading his blog if you are wanting a deep dive into these measures.
Computing Gait Scores in Sift
The field of biomechanics uses the GPS and GDI to summarize lower limb variance between individuals and a normal dataset. Both measures achieve this with a single metric score. With the recent introduction of Normal Databases in Sift, it only made sense to enable users to leverage their own control datasets. Users can now measure these community-standard scores within Sift. Our documentation guides you through all gait measures in Sift’s toolbox and highlights the relevant papers to get you started.
If you’re interested in testing out the GPS, GDI, or both, be sure to check out our new tutorial. Liked this post? Let us know what you think of the GPS and GDI, or other gait scores you have found!