
TYPICALLY, WHEN WE TALK about analytics in sports, the discussion is focused
on advanced player statistics – and more recently, on performance
analytics through things like wearable sensor technologies and high-speed
cameras. Advanced analytics and AI in sports are continuing to grow and
evolve, even into the injury prevention and management arena. A recent
research study I was involved in explored how machine learning analytics
could be of use in modeling how high school sports
concussion symptom(s) resolve.
While we know that
concussion in sports is prevalent, clinicians, physicians and researchers also know that it's
not a one-size-fits-all injury. Some of the most frequently asked questions
when I'm evaluating and treating an athlete with concussion are: "How
long before I can return to play?" and "Do you think I'll
be ready for the next game/tournament?" Clinical management of concussions
has come a long way. We're much more aggressive with symptom management
and earlier physical activation, but timing of return-to-play protocols
ultimately depends on the individual and the unique course of resolution
his or her particular injury takes. The most accurate response to the
common questions is, "We'll have to see how things go."
That can be frustrating for patients, athletes, parents, trainers, coaches,
teammates – and their physicians. So, insight into a clearer path
for the anticipated resolution of concussion symptoms can be a huge benefit
to patients and the sports medicine community as a whole. The purpose
of this study was to implement a machine learning-based approach to model
the estimated time for resolution of symptoms in high school athletes
who suffered a concussion while playing sports.
We took three years of high school
concussion data from high school athletes (mostly football players) who were diagnosed
with concussion and examined the effectiveness of using an algorithm to
predict their concussion symptom (headache, dizziness and difficulty with
concentrating) resolution time. In effect, machine learning was being
tested in its ability to predict how long concussion symptoms would take
to resolve based on combinations of symptoms (and other variables) present
in injured high school athletes. The main finding and importance are that
machine learning and advanced analytics do have potential for clinical
utility in managing concussion. Even with the limited data set available,
computer modeling was able to accurately predict symptoms associated with
prolonged
recovery after a concussion.
So, what's the big deal here? (We already know that number and severity
of presenting symptoms, as well as presence of certain modifying factors,
are associated with prolonged recovery. A skilled and experienced physician
with knowledge of existing research on
concussion can evaluate a patient and make a reasonable estimate of severity as well
as predicted recovery.) Well, there are significant clinical and research
implications. This is an important first step. We believe the utilization
of machine learning and AI is a modern approach that will become more
and more accurate with the inclusion of more and more data. With time
(and appropriate input from subject matter experts), computer analysis
will be able to take into account the experiences of an unlimited number
of concussed individuals with interdependent variables, complex and comprehensive
data points at levels of complexity that would be impossible to achieve
by a singular individual tasked with evaluating and managing that patient.
Eventually, access to this kind of analysis will be available to the skilled
physician (via an app or hand-held device), thereby dramatically improving
the quality and accuracy of medical care relative to treatment approach,
and predicted recovery. Also, machine learning with this kind of modeling
and predictive analytics can accelerate traditional research by allowing
division into groups that differ in the expected duration of symptoms
and time to resolution – resulting in "cleaner" and more
reliably certain outcomes. Traditional approaches to research would take
years (even decades) to bear fruit.
The value of machine learning and big data analytics will be of increasing
value in sports neurology, sports medicine, and all of medicine in general.
From a practical application perspective, it can improve clinical management
of injury and illness. From a research angle, machine learning can rapidly
increase the efficiency of medical research. Once data is collected, scrubbed
and organized, super-computing enables enhanced analysis, and modeling
allows faster/cheaper research and testing. Importantly, the accuracy
and utility of machine learning output will improve over time as more
data is collected and added. Analytics, AI and big data are revolutionizing
almost every aspect of sports. I'm excited to consider the future
benefits to sports neurology.