The physiologist and coach Jack Daniels once filmed a bunch of runners in stride, then showed the footage to coaches and biomechanists to see if they could eyeball who was the most efficient. “They couldn’t tell,” Daniels later recalled. “No way at all.” Famously awkward-looking runners like Paula Radcliffe and Alberto Salazar sometimes turn out to be extraordinarily efficient. Smooth-striding beauties sometimes finish at the back of the pack.
The act of running, it turns out, is surprisingly complicated. The bob of your head, the rotation of your hips, the angle of your foot—all these factors and many others can vary in endless ways. So it’s a more or less hopeless task to simply watch someone run past and diagnose problems with their stride, whether it’s inefficiencies or vulnerabilities to certain types of injury. Amid the endless variables, we can’t possibly zero in on the ones that matter in real time.
One solution to this problem is to slow it all down. Film a runner and watch the footage in slow motion. Or better yet, attach a bunch of markers to key joints, feed the data into a computer, and create a three-dimensional model of the runner’s stride, so that you can analyze every joint angle and acceleration at your leisure. That’s what biomechanics researchers have been doing for years now, trying to link certain gait characteristics—a knee that rotates inward more than usual, say—with particular injuries like patellofemoral pain or IT band syndrome. They’ve had hints of success, but overall the results have been somewhat muddled and hard to interpret.
So another solution is more radical: call in our robot overlords, let them sort through the mountains of data, and see what they come up with. That, in essence, is the approach in a new study from researchers at the University of Jyväskylä in Finland and the University of Calgary in Canada. They ran the data from 3D gait analysis of a bunch of runners, some injured and some healthy, through a form of artificial intelligence called unsupervised machine learning, to see if it could group the runners into categories based on their strides, and whether those categories would reflect the types of injuries the runners were subject to. The answers—yes to the first question, no to the second—are both worth thinking about.
The study, published in the Scandinavian Journal of Medicine and Science in Sports, involved 291 runners whose gait had previously been analyzed by Reed Ferber of the University of Calgary’s Running Injury Clinic. They had an average age of 39.5 with a roughly even split between men and women, and based on their most recent race times were a mix of recreational and competitive runners. Of these subjects, 266 had some form of injury, including patellofemoral pain (44), iliotibial band syndrome (29), Achilles tendinopathy (15), plantar fasciitis (14), medial tibial stress syndrome (12), and others. Their gaits were analyzed by affixing markers on seven lower-body segments, then filming them with an eight-camera set-up to digitize their motions.
In all, each gait analysis produced 62 variables, including things like knee and foot angles, vertical oscillation, and stride rate. After some further manipulation, this data was fed into the computer for a “hierarchical cluster analysis,” which is a way of dividing the subjects into groups with shared characteristics. Crucially, this process doesn’t involve telling the computer in advance what to look for or what variables are most important. This sort of machine learning offers a way of detecting hidden patterns in complex data. (I wrote a few years ago about some related research that used a similar approach to distinguish between recreational and competitive runners.)
The cluster analysis divided the runners into five groups, each of which was distinct from all the other groups. For example, one group had runners whose knees collapsed and flexed the most during running. Another was characterized by stiff limbs, as indicated by knees and hips that bent less than usual. A third had a pronounced heel strike and big changes in foot angle through the stride. Picking out these subgroups with the naked eye, or even by manually combing through the gait analysis data, would have been all but impossible—but in retrospect, the patterns are quite distinct.
The researchers started with the hypothesis that these groupings would map onto the runners’ injury diagnoses. You might think that, say, the group with collapsing and flexing knees would rack up the most knee injuries. But there was no pattern of that sort. The various types of injuries, as well as the ratio of injured and healthy runners, were distributed pretty much randomly across all the groups. How you run, according to the biomechanics equivalent of Deep Blue, doesn’t determine whether, where, or how you get injured.
This has some interesting implications. If the results are confirmed in further trials, it suggests that, as the researchers put it, “there is not a single ‘protective gait pattern’ reducing the likelihood of developing [running-related injuries].” On Twitter, Rod Whiteley, a prominent physiotherapist at the Aspetar Sports Medicine Hospital, floated the suggestion that each of us adapts to the idiosyncrasies of our own running style. Injury risk, in this view, comes from changes in your training load, rather than, say, the angle of your knee. That echoes retired University of Calgary biomechanist Benno Nigg’s take: 80 percent of running injuries, he used to say, result from training errors like increasing your mileage too quickly or not taking enough recovery.
In the end, even sophisticated artificial intelligence algorithms don’t guarantee that these results are correct. Maybe the runners in the sample were too similar to each other; maybe they were too diverse. Or maybe there just weren’t enough of them. But if it turns out that supercomputers really are just as powerless as humans at predicting future injuries based on running form, then maybe we can dream of an idyllic future where we stop arguing about footstrike and cadence and shank angle and so on—and instead just agree to take a rest day every once in a while.
My recent book, Endure: Mind, Body, and the Curiously Elastic Limits of Human Performance, with a foreword by Malcolm Gladwell, is now available. For more, join me on Twitter and Facebook, and sign up for the Sweat Science email newsletter.
Support Outside Online
Our mission to inspire readers to get outside has never been more critical. In recent years, Outside Online has reported on groundbreaking research linking time in nature to improved mental and physical health, and we’ve kept you informed about the unprecedented threats to America’s public lands. Our rigorous coverage helps spark important debates about wellness and travel and adventure, and it provides readers an accessible gateway to new outdoor passions. Time outside is essential—and we can help you make the most of it. Making a financial contribution to Outside Online only takes a few minutes and will ensure we can continue supplying the trailblazing, informative journalism that readers like you depend on. We hope you’ll support us. Thank you.