The Blueprint for Safe(ish) Velocity
How physics helps us understand injuries, even if we can't predict them
Every other week I write an email discussing what I learn launching and growing Reboot Motion. If you would like to receive it directly in your inbox, subscribe below.
The following is the first post in a series on the Wake Forest Bridge Seminar, which was put on by Wake Forest Dec 1-3, 2023.
While nothing matches being there, I wanted to share what I learned for those who missed it, and to set the stage for an ever bigger turnout in 2024.
(As was the case last year, this first post may be a bit self serving, as Reboot’s own Jimmy Buffi is the presenter.)
—
We can’t predict injuries is an odd claim to start a presentation.
However, that is where Jimmy Buffi, CEO of Reboot Motion, started. In his opinion using motion capture to predict injuries is a fools errand.
While motion capture is a valuable tool, it is not all-encompassing. It doesn’t touch on workload, muscle volumes, or a variety of other factors that may help measure risk.
However, while motion capture data may not help teams predict injuries, it can enable analysts to learn more about them, which is exactly what Jimmy did by:
Building an actionable model that shows associations between 1) features coaches and front office personnel understand and 2) injuries, and
Combining this model with a similar one for fastball velocity- highlighting specific attributes that can simultaneously raise performance while deceasing injury risk.
Building the Model…
With Physics Based Metrics
While many biomechanists like using joint angles at key points, Reboot’s preference is to rely on physics-based metrics. These features are 1) more individualizable 2) easier to understand (especially as it relates to cause an effect) and 3) most importantly, more coachable.
Additionally, by building models with features that match the metrics used in Reboot reports, partners can easily interpret results and put them into action.
And Good Data
There is no such thing as perfect injury data. In order to get to the best place possible, here is what Jimmy did:
He collaborated with Sports Info Solutions, who did a lot of manual work reviewing video footage and annotating injury details, giving the group confidence that this is the most comprehensive injury dataset available.
He paired the data with MLB play information, enhancing the depth and context of the dataset. Two labels were assigned to a multitude of pitches in the database. The first label, referred to as "non-injured," was assigned to pitches thrown by pitchers who hadn't experienced an elbow injury throughout their career. The second label, "injured," was applied to any pitch thrown by a pitcher who had a history of elbow injuries.
Considering the dynamic nature of injuries, this is far from perfect. However, it serves as a good starting point.
The Results
In the image below, we see a list of features, ranked by their relative importance to impacting injury risk.
While there is a lot more that goes into the Shapley values displayed below, the TLDR is features that have blue on the left and red on the right are positively correlated with injury risk; while those that go from red to blue are inversely correlated.
For this model specifically, we see speed is the biggest factor- which is something Jimmy included as a sanity check. More interestingly, we saw a variety of features that can directly be brought to the field.
For example, lead arm momentum, torso/pitching hand alignment, and torso momentum were all inversely correlated with injury risk.
In other words: all else equal, pitchers that 1) use the lead arm to accelerate the torso, 2) get the torso moving fast and 3) line it up with the pitching arm, are less likely to get injured.
On a macro level, this makes sense. If the biggest concern is overuse of the pitching arm, it is the pitchers who are using the rest of their body to generate velocity that will be the best off.
A Blueprint For Safe Velocity
We all understand that increased velocity is associated with increased injury risk. However, that doesn’t necessarily mean all things that increase velocity will do the same for injury risk.
Therefore, Jimmy went one step further and combined the injury risk model with a similar one for fastball velocity, looking for features that were associated with both lower injury risk and higher fastball velocity, to create a blueprint for safe velocity.
Jimmy explains in detail below:
Thinking from first principles, these aren’t ground breaking results. And that’s exactly the point- it is their intuitiveness that makes sense them so useful.
If the goal is to increase velocity, and do so in a way that doesn’t increase injury risk- we have to generate momentum throughout the body and transfer to it the pitching arm efficiently, so the arm itself doesn’t have to do all the work.
We know what matters. We now have to measure (and coach) what matters.