The first step to a good product is the ability to complete strong reliability studies with a vast database across many different specialized tests. Sparta has been able to do this time and time again and we check our data annually to update our reliability metrics and search for new metrics that might be useful to our partners. Reliability research is completed using an aggregation of all data across all our partner organizations in order to account for any variability in ability and skill. This data is then analyzed by a third-party statistician to ensure an unbiased analysis.Learn More
- Is the data reliable? And how are these studies performed?
- Is the data valid? And how are these studies performed?
In addition to the reliability of our data, we strive to ensure that what is presented to the coaches, athletes, and medical personnel is applicable and important. In our validity studies, we aim to characterize sport- and gender-specific patterns in the data to provide each team, level, and gender with personalized predictions. We also use these studies to predict the risk and type of injury based on each type of test and an athlete's outcomes. Similarly, the data is used to predict performance within a season and starter status. A few of these studies mentioned above can be seen in the attachments.Learn More
- What did the BYU research tell us about knee injury prediction?
The odds of having a knee injury (vs no injury) increase by 0.57 for every 1 t-score increase in LOAD or average eccentric rate of force. This predictive model was then confirmed using cross validation, a statistical method used to estimate the skill of machine learning, where half of the aforementioned dataset was withheld from building the model so it could be used to test the accuracy of the predictions, which came to around 97%.Learn More
- What research does Sparta Science have regarding concussions?
This document focuses on the relationship between the BESS Concussion Screening Test and our Balance test at Sparta. We found that the Sparta Balance test is significantly correlated to the BESS Test results and that our measure of resultant sway velocity can predict the number of errors obtained on the BESS test.Learn More
- What is the point of Sparta Science?
Sparta Science is a software and training company based around the use of a force plate to determine force profiles, movement patterns, balance measures, injury risks, readiness, talent identification, and classification for sports, military, and medical branches. Our software is used by over 80 organizations which range from hospitals to insurance companies, professional sports teams to high schools, and many branches of the military special forces. Each organization uses Sparta in a unique way to get the most out of what they want to know about their participants. Sports teams use the platform for injury risk monitoring and talent identification as well as athlete tracking throughout a session and during training in order to prescribe them workouts that help address their movement signature weaknesses. Hospitals use the platform to monitor subjects after surgery or during rehabilitation processes to fast track recovery and address any weaknesses as they arise. Military groups use Sparta to not only monitor soldier's progress through training and rehabilitation, but also as a classification tool to better organize their soldiers and allow the weaker ones to receive pointed training in order to improve to a group's needs.
Our system would act as an assessment tool to target weaknesses and strengths and help as an assistive tool for trainers, doctors, physical therapists, and coaches in streamlining training and rehabilitation. We can provide readiness metrics to help assess ability to return to activities (sport, walking, etc.). Our platform consists of 3 tests which all have their benefits and collectively lead to a complete picture of how the subject is moving, recovering, and excelling.Learn More
- What are the three tests? And how are they performed?
The three tests are as follows: Jump Scan (6 countermovement jumps used to characterize a subject's movement signature and therefore how they complete any force-requiring movement from jumping to running to hitting etc as well as to identify injury risks and locations, talent identification/classification, and readiness), Balance Scan (2 trials on each leg of eyes closed balance used to characterize the first stage of return-to-activity and to track progress during rehabilitation as well to target injury risks and locations and general stability discrepancies between limbs), and the Plank Scan (2 trials on each arm in plank position used to characterize the first stage of return-to-activity for upper body injuries and to track progress during rehabilitation as well to target injury risks and locations and general stability discrepancies between limbs).Learn More
- What do the T-scores mean?
T-scores represent the raw variables' relationship to the population average. A score of 50 would represent a score equal to the population average. 10 T-scores are equal to 1 standard deviation along the normal bell curve. The T-scores are calculated off of the following variables for each test: Percentage Body Weight Relative to Resultant Sway Velocity for the Plank Scan, Resultant Sway Velocity for the Balance Scan, Average Eccentric Rate of Force Development for Load from the Jump Scan, Average Relative Vertical Concentric Force for Explode from the Jump Scan, and Average Relative Vertical Concentric Impulse for Drive from the Jump Scan.
- How are decision trees used within the software?
Within our software, we use decision trees in a variety of ways, ranging from workout application to injury risk assessment. We use CART (Classification and Regression Tree) models to predicted injury risk and games played in a season. Models like these are used in our software to create gender, sport, and position specific predictive outputs so as to allow each person to have the most accurate predictive ability for their situation.
Similar decision trees are used throughout our software in order to allow the user to easily navigate their training or rehabilitation to the best possible outcome. Decision trees are used, based off of their scan results, to create workouts or rehab programs that best address their weaknesses while maintaining their strengths. Once again, these are specific to sport and position as well.Learn More
- What kind of studies are performed internally at Sparta Science?
Our data scientists perform routine reliability and validity studies on previously collected data in order to check ourselves as well as come up with new insights. We use a retrospective study approach to our work due to the nature of the data we are collecting. We use an approach that may seem counterintuitive, however allows us to both create new conclusions while simultaneously checking and re-checking our old ones. We gather seasonal data from our many sites to create a database that is both diverse and also has the ability to be very specific to sport, gender, and condition. By using a retrospective study, we are able to allow the natural progression of an athlete's season to happen and then see how their scores were affected by the various outcomes from that season.
We use this method across most of our statistics, however we do have prospective studies that outside researchers perform with our guidance. These studies are used to also find new relationships and back up what we already have discovered internally.
The benefit of retrospective research like this is not to just throw everything at the wall and see what sticks, but to test hypotheses we have seen or heard of first hand about athletes and their injuries and performance on a much bigger scale without the worries of a prospective study.Learn More
- What is our r-squared?
One of the challenges of answering this question is because r-squared has fallen out of favor with many statisticians & data scientists. While R-squared will give an estimate of the relationship between movements of a dependent variable based on an independent variables' movements, R-squared doesn't tell you whether your chosen model is good or bad, nor will it tell you whether the data and predictions are biased. A high or low R-square isn't necessarily good or bad, as it doesn't convey the reliability of the model, nor whether you've chosen the right regression. You can get a low R-squared for a good model, or a high R-square for a poorly fitted model, and vice versa.
Instead of r-squared, the data teams we work with at universities now use Beta coefficients**. Beta coefficients are essentially the odds of having an injury as you increase by one score. Interpreted as:
The odds of having a injury location injury vs no injury will increase/decrease by β for a one-unit increase in scan metric (or with the presence of one of the additive effects).
P-value significance can still be shown for these relationships. Beta would act as the slope of the regression in this case. So for this data set, you could say the average Beta coefficient (linear not log odds) is 1.013. For the test statistic, we are looking for the absolute value of number to be greater than or equal to 2. For a sample model, the average test statistic value is -1551.08, therefore highly significant.
**Definition of Beta Coefficients: The beta coefficients can be negative or positive, and have a t-value and significance of the t-value associated with each. The beta coefficient is the degree of change in the outcome variable for every 1-unit of change in the predictor variable. The t-test assesses whether the beta coefficient is significantly different from zero. If the beta coefficient is not statistically significant (i.e., the t-value is not significant), the variable does not significantly predict the outcome. If the beta coefficient is significant, examine the sign of the beta. If the beta coefficient is positive, the interpretation is that for every 1-unit increase in the predictor variable, the outcome variable will increase by the beta coefficient value. If the beta coefficient is negative, the interpretation is that for every 1-unit increase in the predictor variable, the outcome variable will decrease by the beta coefficient value.
- What data do we have around professional soccer players?
With the use of the Sparta Score (a variable describing athlete injury risk and performance readiness), we are able to predict injury types based on positions and sport. Using the Sparta Score for players in the Professional Soccer, we were able to determine a strong correlation between Minutes Played per Game and Sparta Score.
In this case, we can use the Sparta Score to predict season performance outcomes. For example, for Professional Soccer Goalies and Defenders, as Sparta Score increases, minutes played per game will increase at a 12% greater rate than other positions, providing a metric by which to potentially determine a starter vs a bench player as well as monitor a player's potential playing capacity through the Sparta Score.
Similarly, using the Sparta Score for the Professional Soccer population (academy level included), a significant correlation was found between thigh muscle (quad and hamstring) injuries and Sparta Score with the average number of thigh muscle injuries predicted to decrease by about 13% for each increase of 1 in Sparta Score. There is also a threshold of 79 for Sparta Score where average number of thigh muscle injuries per year decreased by 91.58% as the player crosses this threshold into a Sparta Score of 79 or above.Learn More
- What insights could Sparta give into professional basketball?
Here are some outputs from the G League Combine looking at NBA specific metrics for injury prediction and performance. We found there to be a 67% injury risk sensitivity based on our injury risk score, as well as an increased risk of lower back and hip injuries in the G league population as compared to the general basketball database.
- How can I reduce costs for my team using Sparta?
Here is an analysis done for a financial model of the NFL, looking at position specific profiles and the financial impact of high risk individuals. In general, keeping players out of the high risk profile groups would allow a team playing 16 games a year to save an average of $4 million a season.Learn More
- What sensitivity and specificity metrics does Sparta have?
From the NFL Combine data, we found that for ACL injuries occurring within the first year of play after being tested at the NFL Combine, our injury prediction sensitivity was 73%, specificity was 38%, PPV was 54%, NPV was 58%, and an AUC of 52.3%
- What are the population stats for the ACL injury prediction study from the University of Pennsylvania?
The population used was only from the University of Pennsylvania over the last 4 and a half years and included over 250 lower extremity injuries and, specifically, 16 ACL injuries. Degree of exposure was not taken into account in this analysis. Subjects were matched by age, sex, and sport against uninjured controls. ACL injuries were dispersed across 8 sports, with the highest prevalence occurring in football players (8). When this paper is published, more information on this will be available.Learn More
- What is the calibration process?
The force plate is tared/zeroed in three different cases: 1) Prior to each assessment 2) In the case that the software ever detects a value below -20N or above 50N at rest 3) each time the force plate is physically plugged in. When each force plate is made, the strain gauges in the feet are initially calibrated to a standardized weight. Once this initial calibration is completed, the data is stored in the firmware of the hardware itself, allowing the software to tare or zero the plate when drift is detected.
- Why use LOAD if it seems to be unreliable?
The LOAD variable is not unreliable; the intra-trial reliability of LOAD is acceptable. An example of a commonly used test despite low intra- and inter-rater reliability is the Functional Movement Screen (FMS), which shows how testing set ups with even lower reliability than our test are still used as a gold-standard. To improve upon the reliability of the raw data, Sparta averages best 3 out of 6 trials before transforming into a T-Score, which is the form that is presented to the customer. The above study shows that day-to-day there is greater variability in LOAD relative to the other variables, but this can easily be "controlled" for by understanding the application of the smallest worthwhile change. The smallest worthwhile change (SWC) for Explode and Drive is 2, while the median worthwhile change (MWC) for Load is 6 because it is less than the typical error. A recent study has shown that LOAD is related to post-training impairments and neuromuscular fatigue.Learn More
- Why does Sparta use such few metrics for their assessments?
1) Consistent, longitudinal data on a few choice metrics is infinitely more valuable for prediction than hundreds of metrics reported on irregularly. The value of data depth vs width is demonstrated across healthcare. For example, there is growing evidence that dental records can help detect diabetes and pre-diabetes (and other diagnoses) in patients. The simplification and accessibility of testing, like with Sparta, allows for more frequent meaningful screenings.
2) Reliability / Validity. A metric such as peak power looks at an instantaneous point in time, where movement occurs over the course of time, and it's unsurprising that single-time point metrics have little evidence of correlation to real outcomes (injury, in-game performance).
3) Actionability. If the metric one is measuring can't be changed, why waste time measuring it? The field has become paralyzed by data. Sparta wants to present only the metrics that can be influenced by practitioners through targeted training/recovery. We feel it is unethical to report on a metric that we haven't found to affect injury risk and waste precious athlete/practitioner time.
- I feel that Sparta is a "black box" of calculations; what is really happening under the surface?
The only "black-box" algorithm that is used is the Sparta Score, which was developed as a request from dozens of customers seeking a single number to answer the question "is this scan better than the last?". Load, Explode, and Drive as well as the Vertical Jump protocol are clearly defined in the research. Sparta utilizes T-scores (similar to Z-scores) which is a very standard practice in order to normalize data and understand meaningful changes.
- I don't trust your research because it was not done in a randomized clinical trial setting; how can you convince me that your research is legitimate?
There is a fundamental difference in approach as, unfortunately, the hope of only utilizing randomized controlled studies with athletic populations on a large enough scale for statistical and practical significance is a fantasy. Athletes should not be used as experimental subjects due to the possibility of negatively affecting an athlete's performance or injury risk. Even if one were to conduct a viable study, it would likely not be performed with high-level hockey athletes, rather a small population of non-athletes, of which the results would most likely not translate to the athlete population. When retrospective studies are performed and there is high integrity to the data, then the study results in more actionable insights as compared to a prospective study which simply monitors athletes and allows injuries to happen.
- I only trust peer-reviewed published research; why don't you have more supporting your product?
It is Sparta's goal to continue to show validity in their own and others' research, with the referenced Penn study and the elbow study as examples. Unfortunately because of a combination of privacy restrictions (HIPAA/IRB) and antiquated journal editors, the road to peer-reviewed publications has been tougher than anticipated. Sparta's goal is to not only continue to create more peer-reviewed validation but also continue to learn from their own and others' research to improve the product.
- What is an example of sites that Sparta is getting data from in order to run these validation studies?
Some example of our partner sites are as follows:
- NFL combine, NFL referees, international academies
- Naval Special Warfare (Navy Seals) and the FBI Hostage Rescue Team (Domestic version of Seal Team 6)
- All of the US Marines Recruiting centers
- Stanford Health System, USC Hospitals, Cleveland Clinic, NYU, HSS
- 30+ professional teams including but not limited to the Baltimore Ravens, Pittsburgh Steelers, Washington Redskins, Detroit Lions, Anaheim Ducks, Cleveland Cavaliers, Colorado Rockies, San Diego Padres, Miami Marlins, Portland Timbers
- International Professional Teams & Governing Bodies like English National Soccer, Australia & South Africa & Japan Rugby, India National Cricket
- 40+ universities inc Clemson, Ohio State, Univ of Texas, Auburn, Penn, Rutgers, Kansas, Cal, UCLA, Georgia, Michigan, Oklahoma, Rutgers, etc
- Exos Performance Gyms & Physical Therapy Centers (formerly Athletes Performance)
- Can Sparta be used for testing of health and wellness rather than just sports performance and injury prediction?
Sparta can be used in many different settings and can be used to measure many physical states outside of the sports performance realm.
One such state is movement variability, which is linked to longevity and health, with greater movement variability correlated to better stability and movement coordination.
With this in mind, one of the best ways to measure movement variability is through the use of a single-leg balance test. At Sparta, we use a single-leg balance test to predict lower body injuries as well as concussions and to monitor people as they progress through rehabilitation. Using Sparta to monitor movement variability and thus potential long-term health of an individual would allow for easy use and interpretation of results and the ability to track progress over time. Our balance test can be done in a progression from double-leg eyes open, double-leg eyes closed, single-leg eyes open, to single-leg eyes closed. This allows for metrics of health and variability while also accounting for ability.
- What pro soccer specific insights does Sparta have?
The average number of thigh muscle injuries were predicted to decrease by 13% for each increase of 1 in Sparta Score, after adjusting for player position. A Sparta Score cut-off of 79 is a good cut-off for identifying players at risk for thigh muscle injuries, with the average number of thigh muscle injuries per year decreasing by about 0.3 as you cross this threshold.
- What pro football specific insights does Sparta have?
DT, LB, OL who have a Sparta Score of at least 85 are 22% less likely to incur an injury than those with Sparta Scores less than 85. DE, DB, TE, WR, and RB who have a Sparta Score of at least 91 are 14% less likely to incur an injury than those with Sparta Scores less than 91. LB, RB, DB, and TE with Sparta Scores of at least 81 played an average of three more games than those with scores below 81. WR, OL, DE, and DT with Sparta Scores of at least 78 played an average of two more games than those with scores below 78. TE and DB with Sparta Scores below 87 were much more likely to get any injury (94% below vs. 78% above). Below 87, TE and DB have a 23% risk of getting a thigh muscle injury.
- What pro basketball specific insights does Sparta have?
Basketball players with a Sparta Score greater than or equal to 85 were predicted to play 36% more minutes a season than players with Sparta Scores below 85.
- What pro baseball specific insights does Sparta have?
For MLB pitchers, FIP decreases by an average of 0.06 for every increase of one in Sparta Score, after adjusting for age. Veteran players (age >=28) with a Sparta Score above 85 had the lowest FIP on average (3.4), while young players (age < 28) with a Sparta Score below 81 had the highest FIP on average (5.6). For batters, there was an average predicted increase of 1 Stolen Base for every increase of 3 Sparta Scores, after adjusting for age and plate appearances. Pitchers with Sparta Scores of 85 or above were almost twice as likely to stay healthy as pitchers with Sparta Scores below 85 and less than 4 times as likely to sustain elbow injuries. Pitchers with Sparta Scores of between 79 anf 85 were more than three times as likely to stay healthy as pitchers with Sparta Scores below 79 and more than two times less likely to sustain an elbow injury.
- What college volleyball specific insights does Sparta have?
For college volleyball players, Load was positively correlated with Digs and Service Aces. Explode was positively correlated with Points, Kills, Blocks, Digs, Service Aces, and Matches Played. Drive was positively correlated with Points, Kills, Blocks, and Matches Played and negatively correlated with Ball Handling Errors for Setters. An Explode cut-off of 45 is a good cutoff to use in identifying strong players at all positions; players with explode values in this range or higher tended to score more points (Right Side, Outside Hitter, Middle Blocker), get more digs (Defensive Specialist, Setter, Outside Hitter), and play more matches (all positions). A Drive cut-off of 55 is a good cut-off for identifying high point scorers in the Outside Hitter, Right Side, and Middle Blocker positions as well as good ball handlers in the Setter position although excessively high drive scores (above 58) may not be desirable in Setters. A Load cut-off of 62 is a good cut-off for identifying strong Outside Hitters, Defensive Specialists, and Setters as measured by digs or service abilities. Sparta Score is significantly correlated with blocks with suggested cut-offs of 82 for Middle Blockers and 90 for Outside Hitters, Right Sides, and Setters.