Data prevents injury

Good stuff.  I had a narrow gap between two classes at noon yesterday so was able to hear a visiting speaker pitching at the Sports Science side of our Department. It's barely a week since we had the last external speaker talking with bounce and clarity about turning kids from sacks into six-packs. If this level of inspiration keeps up, I'm going to start running . . . between the fridge and the sofa. Yesterday, it was Stephen Smith, a graduate from our Sports Rehabilitation degree about 8 years ago. He admitted to have been a less than stellar student here, especially when it came to number-crunching. That is just bonzer because he's done well for himself and that will buck up some of our struggling students. He left college looking for work as a coach and fitness expert and found that he and his opposite numbers were making observations and delivering recommendations in a way that seemed to be wholly unreproducible seat of the pants stuff based on 'experience' and tradition.

He then started off on a mammoth three year study to measure what could be measured in an effort to find predictors of injury. Preventing injury is big money in sports because if your stars are man-down with a sproinged hamstring you don't win the match and your corporate sponsors look elsewhere. In US NFL pro-football the bill for injury is $500+ Million a year; more for baseball. Even with 64 independent variables clocked in numerous athletes, however, Smith couldn't find any variable or combination of variables which explained the distribution of subsequent injury.  It's a bit like predicting obesity from a mammoth population genome-wide association study GWAS: each fat person is over-weight because a particular set of circumstances that will be different from the guy on the next door treadmill at weight-watchers. It's also possible that using GWAS to look for mutations across the human genome to predict obesity is looking in the wrong haystack - try looking at the intestinal flora instead! Having shed-loads of data which is unable to offer any insight is galling but not uncommon in science. Smith & Co. took their data to the Head of Stats at TCD who set a couple of PhD students on it and they advised 'machine-learning' approaches for future work.

In the old days [2005?!], you'd have to ask each athlete how much they could stretch or bend in the directions demanded and that sort of data was more-or-less useless - biased, fuzzy and unreliable. As well as getting a better statistical foundation, Smith and his partners looked to develop an App which would grab motion-capture data [L] in an efficient, objective manner. That needed ca$h and they were able to pull down $4million in Silicon Valley venture capital to float Kitman Labs in October 2012. From three employees three years ago, they now have 30 people of the pay-roll! A key piece of kit is a single high-def camera coupled with a depth sensor [L] that can capture the 3-D location of various joints (knee, shoulder, ankle, hip etc.) to a precision of 0.5cm. That's extraordinarily elegant and parsimonious: a typical rival set would require 6-8 cameras and a slab of software to integrate the information. They also slid their test-subjects into an MRI scanner so that they can extrapolate reliably from the video-captured external surfaces to the actual joints within. By taking measurements of the distance between shoulder-elbow and elbow-wrist you can infer the position of the elbow joint even when, say, the fist occludes direct line of sight.

And it is lickedy-spit: you can put an entire rugby team including substitutes through their paces before a training session taking 60-90 seconds each. While the lads are running intervals or practicing line-out strategy on the field, the boffins in the club house can identify the members of the team who most need work from the fitness trainers. They can tell, for example, if the full-back is favoring his left leg when doing a standing jump for the camera. That should be worked on: asymmetry is a flag for stress.  But the ease of data capture also allows longitudinal studies. If the full back is favoring his left leg today but not last week you might want to pull him off the field right now to rest it.  The longitudinal information incorporates reports from each athlete about sleep-deficit, diet, 'well-being' and other issues that might impact on stress and so be likely to result in injury.  That exposes some interesting variation among people: some are steady-as-you-go tug-boats through life: sleep eight hours every-night, happy in their own skins, regular in their habits and clockwork in their bowel-movements.  A small-small glitch in one parameter for those people is an orange-light warning. Others are racing yachts - caught in the doldrums one day and romping home with the wind abaft the next. Their mood-swings have also to be accommodated by team-mates and management.

We live in a data deluge. Kitman Labs seem to be doing something interesting with it all.  One aspect of their business package is that they undertake to integrate all the relevant information which has been traditionally kept in separate file-cabinets: the team doctor releases nothing because of medical confidentiality; the coach is dyslexic and has no written records; the team are allowed to self-report their alcohol consumption; omerta rules on grassing up your team-mates. Break all those impediments down and put everything into a gurt big computer and you might get some replicable data. And throw those stool samples into the hopper too!