What does Risk represent?

Risk is a machine-learning derived value assigned to an individual based on the results of their most recent Jump Scan leveraging our 12+ years of aggregate data.

Risk is a representation of an individual’s relative risk of suffering a musculoskeletal injury compared to their peers. Risk is only available on our current cloud platform. It is a replacement for the MSK Health metric, which was an older version of the model, and utilizes our most up-to-date model and data.

Read more in about risk modeling in our White Paper.

 

 

How does the scale work?

Risk is a categorical metric, with 5 unique values of risk on a scale: Low, Below Average, Average, Above Average, and High. As Risk ranges from one value to another (eg Low to Below Avg), the relative change or increase in relative risk can be interpreted as approximately 20%. Average risk can be thought of as average or typical risk for our database, with the two values on either side being interpreted as risk greater than (Above Avg, High) or less than (Below Avg, Low) average.

 

Sparta ML Image

How do we know this?

Sparta has been consistently collecting ground reaction force data from numerous assessments for over a decade from 80k+ unique individuals across all ranges of ages, backgrounds, and abilities.  Along with that data, we have ingested relevant contextual outcome data such as injuries, performance, and exercise programs into our data lake. With this depth of data, Sparta continuously develops and deploys iterative machine learning models into our product that can be customized for customer organizations, populations, and outcomes.

The base out-of-the-box Risk model in our platform is iteratively developed leveraging our accumulating data lake of force plate assessment data and securely shared injury data provided by our partners primarily in high school, collegiate, and professional sports. Our Force Plate Machine Learning (FPML) engine analyzes each individual’s Movement Signature and compares this data to our historical data of force, injury, exposure, and assigned sex to assign an Injury Risk value for each individual. This model has shown significant predictive accuracy in these populations, while the results of the Risk metric for novel populations should be interpreted with caution.

Future iterations and population-specific models are in constant development and given sufficient data can be developed and deployed for specific use cases and/or customer organizations.

What goes into generating Risk

The current input features that contribute to the model are listed below. These features will evolve and grow over time to produce more accurate models for a variety of populations ranging from sport athletes to tactical athletes to industrial athletes.

Current Features:

  • Assigned Sex at Birth
  • Load (Average Eccentric Rate of Force)
  • Explode (Average Relative Concentric Force)
  • Drive (Relative Concentric Impulse)
  • Mass

What's next?

Future injury-specific (eg ACL) and population-specific (eg Basic Military Trainee) models are currently in development, though sufficient data needs to be shared/aggregated to enable to development of these models.  

The current Risk model is a general "athlete-based" model (aggregating multiple types of athletes/injuries) which has shown predictive ability and a significant level of predictive accuracy. 

Retrospective analyses of injury prediction accuracy are available for customers to encourage transparency and improvements in modeling outcomes.

In addition to the development of new injury prediction models, our general model will continue to evolve and improve over time.

 

 Near term and future additions

  • Additional biomechanical metrics from Jump Scan
  • Data-driven (eg machine learning) metrics from Jump Scan
  • Biomechanical/data-driven metrics from Balance Scan
  • Additional individual characteristics (eg Age, Height, Sport, MOS, etc)
  • Previous injury history
  • Additional (continuous stream) outcome data (injury, performance)