Harvard Medical embraces NBA sports science
Artificial intelligence is coming to your hip
|Dec 27, 2019||5||2|
BY HENRY ABBOTT
Part 4 of TrueHoop’s series on preventing NBA injuries.
Dressing up doesn’t come naturally to Marcus Elliott, M.D. In his hometown of Santa Barbara, he’s no lock to wear shoes; sometimes he’s barefoot even at the high-tech gym where he works. But for the Augustus Thorndike Grand Rounds orthopedic lecture series at Boston’s Massachusetts General—arguably the most prestigious hospital in the nation—Elliott’s tweed jacket and tie were straight Ivy League. That was one sign that he had been preparing for this honor for months, or in many ways, decades.
Of course if you get the chance, you go to Harvard Medical School, as Elliott did. But there is at least one subtle downside: Institutions steeped in tradition are… steeped in tradition. And Elliott wanted to do something that the nation’s finest doctors didn’t value much a couple of decades ago. He took his Harvard M.D. and used it to work with elite athletes. By now, his P3 clinic may work with a who’s who of the NBA, NFL, MLB, global soccer, and the like. But in a recent phone call describing the decades since Harvard, Elliott used the phrase “waste of a world-class education” seven times. It’s a direct quote from one of his Harvard instructors, and a general sentiment that has evidently tugged at Elliott.
So, he confesses, it meant a lot when he became Harvard Medical School’s 12th annual Augustus Thorndike visiting lecturer a few months ago. (Thorndike pioneered things like having doctors attend football games. In 1938, Thorndike wrote America’s first book on athletic injuries. He reportedly introduced the idea of taping.) Nadine, Elliott’s wife, and their three children don’t typically accompany Marcus on business trips. They did on this one.
Elliott prepared for the three associated presentations like an athlete. He wanted to present with impact. His work is a little harder to explain than Thorndike’s taping, but it is inarguably uncharted and fundamental territory with big potential to alter the future of sports medicine. His talks included all kinds of novel NBA findings, including:
Elliott’s team has long used three-dimensional motion capture technology to assess factors they suspected might lead to NBA injuries. (For example, in some players, the long bones of the leg rotate as the knee bends. That’s an important predictor of knee injuries.) More recently, though, the P3 team has subjected their growing datasets to the scrutiny of machine learning algorithms, which can see correlations that humans might not expect, especially between multiple factors. It has been revelatory, mostly in ways Elliott has yet to discuss. One little finding: being overweight is not a big predictor of knee trouble if you move well. But in combination with other movement red flags, like that bone rotation thing, extra weight makes injuries more likely.
One year into what Elliott expects will eventually be a published study with two years of data, P3 now has an incredible ability to predict NBA knee injuries. “If we said you were at risk of knee injury,” he said by phone, “there was a 70 percent likelihood you had an acute injury within the next year. We are adding another year of data, then we will publish.”
As NBA players land, about two thirds of them “yield” to the floor, in P3’s lingo. That means their ankles, knees, and hips bend in concert to accept the force of the floor. The other third of players bend their ankles and knees, but leave the hips out of the party. Elliott says that in the data, it looks like they are trying to land with their ankles and knees, but that with their hips they are trying to jump. Those “non-yielders,” Elliott reported in video of his talk to Mass General, are 250 percent more likely to have back trouble. “If you have an athlete that has back issues,” he told his audience of doctors, “it might not be that his core is too weak. It might be because the way he lands is creating too much anterior shear in his lumbar spine and he needs to be retrained.” The next slide shows, essentially, that P3 has had real success taking just a short time to teach players a different way to land. This is evidence no one has seen before, on players no one has studied in these numbers before, with therapies most doctors would never know to try.
NBA ankle injuries can seem random—landing on someone’s foot is one common unlucky cause. “We tend to put things up as being random or freak when we don’t have the answers as to why they happen,” Elliott told the Grand Rounds audience. “We certainly don’t have all the answers, but I can’t imagine a world in which there wasn’t some Newtonian physics behind the pattern.” To that end, in a study of 306 NBA players, P3 found big differences between those who went on to have ankle injuries and those who didn’t. The measures have names like “initial impact spike,” “ankle flexion,” “ankle translation,” and “maximum ankle eversion.” Basically, these numbers show how a foot moves upon landing. The key takeaway: well before anyone is injured, the numbers are very different. Some players show signs of being much higher risk for NBA ankle injuries. In plain English: It’s bad to land with your toes down. Your heels then slap the floor and big forces shoot up your leg. (On the other hand, if you land in the mid-foot, you can absorb the impact into the big muscles of the leg and butt.) Landing toe-heel like that also tends to impair the next jump. Like most of what P3 finds, the thing that hurts your peak performance also increases the odds of injury. But here, too, Elliott has news for the doctors: “You can train these things.”
In 1999, as a medical student, Elliott worked on a hamstring injury study in which he tested NFL players’ leg strength three times a season on a Cybex machine. Then he’d scan through the data, looking for connections between this measurement and hamstring injury.
The field has evolved mightily in the decades since. P3 has assessed more than half of the current NBA, most more than once. The standard assessment includes hundreds of data points. Put it all together and it’s a massive database—the spreadsheets literally have so many columns that they exceed the capacity of Excel or Google Sheets. (P3 had to identify a vendor with even bigger spreadsheets.) Who knows what a machine might learn, loosed to swim in that ocean of data? Now every time a player gets hurt, the machine can dig back through all the past measurements, and see what correlations might exist. Does something in the measurement of a 17-year-old predict things in 25-year-olds?
Hell yes, and in powerful ways. The machine learning discovered that NBA players tend to fall into groupings. The P3 staff has given them names. Some of the categories are straight from the scouting report: “hyper-athletic guards,” “traditional bigs,” “athletic bigs.” One group is simply called “specimens.”
They tend to move like each other, and to have injury profiles like each other. Many NBA players are assessed long before they reach the NBA. In his Grand Rounds lecture, Elliott said some clusters “are full of athletes that were injured in their first two or three years.”
This is where AI can get a little scary. The computer has an idea, right now, who among next summer’s draftees are likely to get hurt. This is why people rightly worry about who controls the data. There will have to be meaningful rules and oversight. The norms of the business have yet to be established. (In the meantime, Elliott says P3 attempts to serve the athlete first, and in all other matters to be “neutral like Switzerland.”) In general, Elliott says almost nothing about specific players. In a phone interview about this, he was only vague.
But I found one of his recent talks to doctors had been filmed and posted online (Elliott’s response: “yikes!”), and it included particulars interesting to the NBA. For instance: the AI has identified a type of player P3 had labeled “kinematic movers.” They tend to do very well in the NBA, avoiding injury while producing nicely, despite being underrated coming into the league. This is real data with far more impact than most draft data. CJ McCollum, Bradley Beal, Trae Young, Grayson Allen, and Landry Shamet are kinematic movers. Elliott told the doctors that in tests, these players are effective moving laterally, are more smooth than explosive, and are generally below average in height and reach. They are average to above average jumpers, and average vertically. Something about that combination is, Elliott said, “really successful.”
A lot of teams are now hiring their own people to collect and analyze this kind of data. Many a front office has a data science person, a biomechanist, or a sports scientist. Elliott knows many of them personally, and says he likes and respects a good number. But he also says the league is in an “awkward phase.” The people who understand the data tend to be weak on the day-to-day “blocking and tackling” of training strong athletes. (A lot of the job really is about coaching a good jumping squat or rhythm box.) On the other hand, the team programs led by seasoned trainers tend not to have anything resembling P3’s team of Stanford-trained data scientists. Elliott singles out the Nets and Bucks as exceptions who are doing it well. He believes the Spurs were once well ahead, but have come back to the pack. I asked him about a few other teams, he scoffed noticeably at mention of the Lakers. By and large, he said that right now, at the team level, it’s “the blind leading the blind.”
One of the giant shortcomings for teams is that they test only one roster of players—too few to draw meaningful conclusions. P3’s database is approaching a thousand NBA players. That’s worth an almost infinite amount; the real-world implications to teams, gamblers, and doctors are off the charts. (Elliott says P3 has paid a lot of attention to elite data security.) “We haven’t studied movement that much,” pointed out Elliott. But already P3 has confidence about who might injure their knees this year, how to train someone out of ankle injury risk, and which players are “kinematic.”
This brave new world will only get braver. All these kinds of measurements come from a very limited number of high-tech labs—at P3, Sparta Science, and a few specialized gyms with force plates. But soon, computer vision will make these kinds of assessments possible straight from the court. There’s an infrared depth sensor in your iPhone. Whether or not you like the idea of databases full of motion data, there will be databases full of motion data.
It might make all of our lives better. In the recorded talk at Massachusetts General, Elliott addressed the doctors personally: “We should know that our left hip is really unstable and that in 20 years we’re going to have a problem with our left knee,” he said. Essentially every factor that leads to injury is what he calls “plastic,” in that it can be improved with carefully targeted prevention work. In some ways, he is the glorified trainer some at Harvard always feared he would be.
A medical honor from Yale or the Mayo Clinic would have been nice, Elliott said. But from Harvard, it was sweet vindication. Elliott hadn’t changed anything about his approach, he just kept after a vision that he had in high school. At long last, in his fifties, the world of medicine came to him.
He hadn’t only been doubted by stuffy medical academics. People who worked in gymnasiums also expected him to fail. “I don’t believe in science,” he reports an NFL head strength coach told him. “I believe in God and luck.”
“You don’t want to be too early to things,” he said by phone, talking about it later. “A lot of the time the audience just throws tomatoes. Everyone at Harvard saw me go off to work with athletes and thought it was a waste of a world class education. If you’re early to something a lot of times you don’t stick around long enough to see anything but the tomatoes. Now they want to learn about this space. It’s incredibly gratifying.”
And the doctors ate it up, which is a sign that the future is going to be fundamentally different for Elliott, NBA players, and the rest of us. Elliott’s time in Boston included any number of banquets, dinners, and ceremonies—lots of time to rub elbows with the elite of orthopedics. “It was really compelling,” he said, “hanging out with these orthos and feeling that interest they have. They want something more.”
Elliott sees a continuum of human performance. At one end are elite athletes at their peak—at the other are the profoundly injured. Almost all of us are somewhere in the middle. Orthopedics, though, has traditionally only happened at the injured end. “They’re just focused on this tiny band of I can’t do what I want to do anymore, can you help me doc? But there’s so much white space in front of that. There’s such a big continuum. It’s a problem to have no resources before I’m screwed up,” he said.
His Grand Rounds lecture ended with his saying, “I guarantee this is coming to your world. Your professional world, but also your personal world. I’d encourage you to engage.”
Orthopedics will change fundamentally before it can apply these findings. Doctors might believe in the concepts now, but who knows when insurance companies will pay for, say, training to land better, to prevent an injury that hasn’t happened yet. (Who pays for all this was one of the first questions after Elliott finished his lecture.) Elliott’s answer to how P3 is financed was disappointingly vague, but the main point was: players. That’s where the NBA has a magical advantage. Preventing an NBA injury is worth a fortune to the league, to the team, to the players. The league can afford to lead.
Thank you for subscribing to TrueHoop! Happy New Year!