May 30, 2014

Working the Count

The Five-Day Pitcher Injury Zone

by Noah Woodward

Last week, I wrote about Jose Fernandez, his attorney’s comments about the cause of his season-ending injury, and warning signs that preceded his exit from his final start of 2014. After I wrote that article, I began to think more about the righty’s second-to-last start. Could this injury have been prevented a week earlier? I also took to heart a popular criticism of the “injury zone” research supporting that article. Some argue that it doesn’t identify injury risk in time to actually prevent a ligament from tearing, and that it instead picks up on injuries that have already occurred.

I completely agree with this criticism. One thing that we’re repeatedly told by team doctors and medical experts is that while traumatic injuries for pitchers (like ligament tears) might seem like sudden tragedies, they actually may tend to occur through small amounts of incremental damage done over time.

When I revisited Josh Kalk’s Injury Zone piece, Dr. Aaron Gray (who works as a team physician for the University of Missouri baseball team and has experience treating youth pitching injuries) had this to say in the comments section:

When looking at ulnar collateral ligament (Tommy John) injuries, most of these are overuse injuries that gradually occur over time. Sometimes the ligament is relatively fine and then abruptly tears but usually it is a much more gradual occurrence.

I think the “Injury Zone” is more of a slow downward slide into the “Injury Pit” that occurs over many games.

We can’t stick a camera in the elbow of every major league pitcher to monitor the state of his ulnar collateral ligament after every pitch. Instead, we’re stuck monitoring changes in the things that we can measure.

Stan Conte, who serves as Vice President of Medical Services for the Los Angeles Dodgers, stated on a recent episode of Effectively Wild that he believes we aren’t too far away from the use of in-game biomechanical sensors. This tool would be placed directly on pitchers, and it would measure mechanical changes that might be indicative of pain or injury.

We don’t have access to biomechanical data yet, but we do have PITCHf/x data. The information captured by PITCHf/x cameras contains nothing about pitching mechanics, but these cameras can be used to infer certain things about a pitcher’s delivery. If these numbers stray too far from what we’d call “normal” for a particular pitcher, something might be wrong—as Conte put it on the “Medical Analysis and Injury Prevention” panel at the SABR Analytics Conference in March, “Change is bad.”

On the podcast, he stressed that while simple fatigue or pitch count thresholds aren’t injury indicators, sudden mechanical changes might be. It makes sense that these changes could come at the tail end of a start, but they could also come a start before the “injury start.” For all we know, they could come weeks before the “injury start.”

Regardless, I thought it might make sense to step back and consider a snapshot that is larger than the 15 pitches before the exit forced by injury, which is the sample my “Injury Zone” model considers. Instead of looking at the 10 or 15 or 30 pitches that preceded an injury exit, I’ll look at a starter’s entire second-to-last start.

In order to conduct this type of analysis, I had to throw out a lot of injury data. A pitcher’s velocity or release point readings mean nothing without a relevant baseline to compare to, so I decided to place the same limitations on this study that I imposed on the Injury Zone piece. I looked only at pitchers who made five starts within five weeks prior to the “injury start,” and I created a baseline pool (for each pitcher) of all starts made from 2011-2013 prior to these five starts. All injuries that occurred within a few weeks of spring training or in the offseason are also obviously omitted.

These data requirements resulted in a final pool of pitchers that includes 32 shoulder injuries and only 17 UCL injuries. Hesitant to create a full-blown model out of nothing, I decided instead to examine the PITCHf/x numbers behind each of these injuries. We can’t draw any real conclusions from a sample this small in size, but we can at least look for trends. This second-to-last start pool deserves a look it has yet to be given.

Changes in Release Point
As I did in the 15-pitch Injury Zone model, I’ll be measuring differences from pitcher-specific baselines in standard deviations here. That might sound like an unnecessarily complicated thing to do, but standard deviation differences create a level playing field through which we can compare pitchers. For example, Bronson Arroyo varies his fastball release point a lot more than Jose Fernandez does (or did?), so any minor release point shift for Arroyo would look like a major change for Fernandez.

One hypothesis I’ve thrown around is that pitchers attempt to shift their release slots when they begin to feel elbow discomfort. This claim is backed by nothing but personal experience, but it is something that we can evaluate with PITCHf/x data.

When I asked Dr. Gray about this theory, he agreed and pointed me to this paper. Its authors argue that when the elbow is straighter and as Gray says, “more out to the side” (as it would be in a ¾ type of delivery), the force on the UCL increases. It only makes sense, then, that a pitcher experiencing UCL pain would opt for a bent arm and a higher release point. Do we see that shift in the start before these 17 UCL injuries?

PITCHf/x release points are measured 55 feet away from home plate, so the system’s inferences aren’t perfect. This small sample appears to be fairly evenly distributed, with the exception of a small cluster of increases in release point height in the second-to-last start. This cluster is composed exclusively of seven of the 17 elbow injuries. This is a small subset of injuries, but we might be able to increase the size of the elbow injury pool quite a bit if we include relief pitchers or expand our time frame. If a larger pool of elbow injuries does indeed support the “raised release point” theory, we’d have an indicator that would be of value to major league clubs and also consistent with existing sports medicine literature.

Changes in Release Point Variance
Another way to examine release point shifts is through the measurement of release point variance. If you’re at all familiar with the release point graphs at Brooks Baseball, you can think of this number (the size of one release point standard deviation) as the spread between release points.

For example, here’s low-variance Cliff Lee:

And high-variance Roenis Elias:

High variance is sometimes the result of shifting on the rubber, so I’ll measure only variance of the vertical release height. The variance differences below capture the gap between a pitcher’s normal amount of variance and his variance in the second-to-last start. Does it look like pitchers shift around more often?

Not really. While a few injuries were preceded by sharp increases in variance, most weren’t far off from each pitcher’s baseline numbers. Release point variance tends to spike just before the injury exit, but it doesn’t appear to be a strong indicator in the second-to-last start.

Changes in Velocity
Some would argue that the clearest warning sign preceding an injury is a velocity drop-off. If we accept a lot of false positives, this claim is generally true in extreme cases. Jose Fernandez experienced a dramatic decline in velocity just before his exit, but he also lost a bit in his second-to-last start. Is that a common trend?

It looks like fastball velocity changes aren’t a universal sign of elbow or shoulder injury risk in the second-to-last start. One peak is centered around the + .5 standard deviation mark, while another sizable peak is centered on -1.5 standard deviations. Most injured pitchers don’t experience dramatic increases in velocity in that second-to-last start, but we can’t say that all (or even most) elbow and shoulder injuries are preceded by velocity decreases in the same start. That’s something worth considering when we start playing the post-surgery blame game.

Conclusions
PITCHf/x data just doesn’t provide a strong case for predicting injury exits a full start before they happen. The theory of pain causing mechanical shifts makes sense, but the things that we can measure right now (release point, variance, and velocity) don’t back it up. It’s possible that I’m still taking too narrow a view of these injuries—shifts might occur in the three starts preceding the exit, or maybe even five starts prior.

In the meantime, we increase our knowledge base by asking Tommy John victims about what they felt prior to exiting that last start (and at what point they began to feel pain). I’d also ask about mechanisms pitchers use to cope with pain. If we hear that pitchers do tend to make small mechanical changes to mitigate elbow pain, we should pursue any PITCHf/x data that these pain-induced adjustments might affect. Otherwise, we’re grasping at straws until the moments before the injury exit.

Thanks to Jon Roegele’s Tommy John surgery list and the injury database compiled by baseballheatmaps.com

Noah Woodward is an author of Baseball Prospectus. Follow @woodwardps
Click here to see Noah's other articles. You can contact Noah by clicking here