Let’s go back to 2012, when the Washington Nationals made one of the most controversial decisions in recent memory by shutting down pitcher Stephen Strasburg late in the season, even though it meant that Strasburg, though not injured at the time, would not pitch for the Nationals in their Division Series. The Nationals lost that series to the St. Louis Cardinals three games to two, and Lana del Rey wrote “Summertime Sadness” as a result (no, not really). The Nationals justified that decision by saying that they wanted to keep Strasburg below 160 innings pitched for the season to prevent him from further injury. In 2011, Strasburg only pitched in five games, spending most of the season recovering from Tommy John surgery. He was healthy through most of 2013 and has been so far through 2014.
Flash forward to 2014, where a laundry list of starters (Jarrod Parker, Patrick Corbin, Jameson Taillon, Kris Medlen, Brandon Beachy, Josh Johnson, and Matt Moore) have already reported to Dr. James Andrews’s office, and we’re not even out of April. Could something have been done to save them? Perhaps the Nationals had it right. Maybe taking it easy with Strasburg in 2012 was the right move.
Today, we’re going to look at the tricky subject of innings limits for pitchers and whether or not they actually “work.” That is, do they show any effect on a pitcher’s health later in his career? But before we jump in, let’s discuss something first. I’m going to be looking at this from a public health perspective. Injury prevention is likely best done with actual X-rays of a pitcher’s elbow and shoulder and a firm understanding of his mechanics and how those mechanics put stress on his joints. But, in the same way that knowing a man is overweight tells me something (but not everything) about his chances for developing heart disease, we can look at innings workloads as an indicator. In the aggregate, do they help or hinder a pitcher’s health? If the answer is yes, then we can at least say that teams should have a bias toward innings limits or not.
Warning! Gory Mathematical Details Ahead!
I gathered stats on the number of innings that pitchers threw, whether in MLB or the minors, from 2000-2013, along with the number of games in which they appeared, the number of batters that they faced, and the number of pitches that they threw. I cross-referenced that with BP’s injury database, which lists both day-to-day injuries for players (when they are reported) as well as trips to the disabled list for that same time period.
For these analyses, I looked only at shoulder and elbow injuries (separately) that required a trip to the disabled list. Because we know that a really good predictor of whether you will suffer an elbow injury this year is whether you suffered one last year, I coded whether the pitcher had experienced an elbow (or shoulder, as the analyses dictated) injury last year or two years ago.
I took all pitchers in their age-23 season who pitched in MLB that year and coded for whether they made a trip to the DL for a shoulder-related issue during that year. That’s a binary outcome (yes/no), so I used a binary logistic regression. I entered as a predictor their previous shoulder injury status, along with their innings pitched stats for their ages 19-22 seasons (again, with inning counts from both the minors and majors)—that is, all data that would have been available up to that point. I also included the total number of cumulative innings pitched up to that point. As expected, last year’s shoulder injury held up as a good predictor of this year’s shoulder injury risk, but innings pitched in the age-22 season also had a significant effect.
I repeated the exercise for all players in their age-24 season, again with previous shoulder injuries and innings data, this time for ages 19-23 as my predictors. Again, last year’s injury status and last year’s (that is, age-23’s) innings count had a significant effect. I took the exercise all the way up to the age-28 season and found roughly the same result each time. Last year’s innings were predictive of this year’s shoulder injury chances, over and above last year’s injury. I played around with the particulars, most notably re-running the same analyses, but using only innings data starting at age 22 as a predictor (some pitchers don’t enter the professional system until they are drafted out of college). The regression kept spitting out the same answer: look at his injury history from last year and how many innings he threw last year. It held when I stuck only to starters and when I looked only at shoulder injuries that required a trip to the 60-day disabled list. Score one for innings limits?
Well, not so fast. It could be said that not all innings are the same. One common refrain during the summer of Strasburg was that while innings limits were nice, why frame the discussion in terms of innings? Why not limit him by the number of batters he faced or the number of pitches he threw? Well now, we have information on that.
I re-ran all of the above analyses, but instead of using innings tallies as a predictor, I substituted the number of batters faced. Then I tried number of appearances (and number of appearances strictly as a starter). I also used pitches thrown, although I should probably point out that sometimes, these data are a little spotty for minor league seasons. What’s interesting is that while last year’s innings total was pretty consistently a significant predictor of shoulder injuries, other stats from last year show much less predictive power. Number of pitches might not be working because the data are not well-collected, but it’s easy to count the number of batters a pitcher faced. While innings was commonly a significant predictor in the regression equations, batters faced was not. Maybe the Nats were on to something.
On to the elbow. I replicated the above analyses but did a find and replace, swapping out shoulder for elbow. The results again followed a visible pattern. Again, inning count, rather than appearances, batters faced, and pitch count, was the most commonly significant predictor (I should probably point out, in the direction that you would expect) by a wide margin. Unlike shoulders, it wasn’t just the innings tally from last year that predicted elbow injuries. Of course, elbow injuries from the last year continued to have a significant predictive effect, but innings counts, not only from last year, but commonly from the past three or four years were all coming up as significant predictors.
A high innings count might affect a pitcher’s shoulder next year, but it might affect his elbow even three years later. As a pitcher reached age 28 (as high as I went with my analyses), the effects started to get less strong. There were still some effects of innings counts from earlier seasons, but older pitchers were less likely to feel the effects of their previous workload. That might be survivorship bias or it might be that by that point, the elbow is stronger and more developed. I’m not able to make that call with these data.
The question, of course, is how much risk a team is taking by increasing that innings total. After all, those innings might be critical, say in a playoff series. Sometimes a risk is worth taking, even if it backfires. Of course, I’ve run a million regressions, so it’s hard to pin down an exact number, but let’s look at the marginal effect of increasing a pitcher’s workload from 160 innings to 200 innings in his age-24 season, and how it affects his chances for an elbow injury requiring DL time at ages 25, 26, and 27. We know that a previously un-elbow-injured pitcher goes on the DL for an elbow injury about two percent of the time. The additional 40 innings at age 24 adds about an extra one percent chance of an elbow injury at age 25, a very slight bump up at age 26, and about seven-tenths of a percent at age 27. The effect size for shoulder injury at age 25 was also about an extra one percent.
Yeah, that’s it. Now, before we go any further, repeat after me: “This is for the ‘average’ pitcher.” Still, for the few percentage points of injury risk that the Nats saved by holding back Strasburg in 2012, it seems that they gave up a lot by not pitching him in the playoffs, comparatively.
I’m the Doctor, Run for Your Life
If there’s something to be gained from the publicly available injury research, dating all the way back to the original Pitcher Abuse Points research done by Rany Jazayerli and Keith Woolner, it’s the knowledge that pitcher injuries are not random events. The way that pitchers are handled, both within a season and over time, makes a huge contribution to whether or not they will remain healthy. However, not all “abusive” practices are created equal. For example, 40 extra innings in a season does put a pitcher at a slightly greater risk for injury in the next few years, but the effect size is actually smaller than letting a pitcher hit a pitch count in one game above 120 pitches.
Before I hit “run” on the program that gave me these numbers, I figured that with all the talk of innings limits, the effect sizes would be a lot bigger. Before I put too much stock into these numbers, we should first talk about “the average pitcher.” One of the problems with regressions as analytical tools is that they throw everyone into the same mixing bowl and treat them all as equals. I can control for some variables, but certainly not all of them. I would be comfortable saying that these results suggest that, in general, innings limits have a statistically significant effect in keeping pitchers healthy, but statistically significant is not the same as “big,” and this is certainly not big.
If the innings that a prospect or a young pitcher might pitch will be essentially meaningless to his development and his team, then shutting him down to preserve his health makes sense, but if either could be served by keeping him on the mound, then keep him out there! I’d argue that the burden of proof that this particular pitcher is a special case is on the person arguing that he is a special case. It’s not that I couldn’t be convinced, it’s just that I’d enter the conversation looking to be convinced.
To go back to the example of Strasburg, there is the fact that he had undergone Tommy John surgery in late 2010. I don’t have access to his elbow X-rays from that time (nor would I have any idea what I was looking at if I did), and I don’t have enough expertise on pitching mechanics and elbow stress and the niceties of how the UCL is attached to the… other thing. As I often remind my mother, I’m not that kind of doctor. Maybe understanding anatomy and biomechanics is the key to really understanding pitching injuries and an aggregate public health perspective just isn’t all that informative. It happens.
If there’s one thing buried in these results that surprised me, it’s that innings pitched, rather than batters faced (which would seem to be more salient and more fine-grained) was the variable that had the most predictive power. Perhaps we’ve been a little too focused on pitch count, thinking that the part of pitching that was so dangerous was the consistent throwing of the ball. Maybe the bigger issue is the number of times that he has to sit down, then stand back up and go out for another 10-15 minutes of throwing, and then sit down again. I don’t think we have enough evidence to say one way or the other here, but at least it does lend some credence to denominating work limits in terms of innings, rather than batters faced.
So should we blame all of the recent Tommy John madness on teams who are pushing their young pitchers too hard in the minors and having them rack up too many innings? I’d argue that no, we shouldn’t, or at least that if we do, we should point out that the increased risk that comes with extra innings can certainly be justified, especially for teams in contention or for young pitchers who need the extra reps to work on something.