In January 2018, our research into predicting throwing-related injuries among professional baseball pitchers, and the efficacy of the DVS Model, was published in the Orthopedics Medical Journal. A formal study was conducted to test the validity of the DVS Scoring System, a proprietary component of the DVS Model that is used to quantify the efficiency within any pitcher’s delivery. The subjects were collected for the study by taking an approximately random sample of major and minor league pitchers from different eras. This random sample was conducted as blindly as possible; injury history was not known prior to assigning a DVS Score. Overall, the DVS model takes into account four main factors: pitching mechanics (DVS Score), birth year, major pitching-arm injury history, and role (starter vs. reliever).
Orthopedics. Predicting Injury in Professional Baseball Pitchers From Delivery Mechanics: A Statistical Model Using Quantitative Video Analysis. 2018;41(1):43-53 Link: https://doi.org/10.3928/01477447-20171127-05
Ultimately, the usefulness and power of any predictive model is measured by its ability to make accurate predictions on data points not in the model-building sample. The DVS model is able to generate predictions and probabilities regarding time-to-injury data from over 1,200 pitchers that reside within our data set. The database consists of current and active pitchers across multiple decades, including most recently all pitchers who made an MLB roster at the start of the 2018 season. This vast range of data, in addition to the number of overall data points, allows the DVS Model to have an unparalleled level of predictive power.
The power of our predictive results are currently measured through two primary statistical measures: the Brier Score and Concordance. Both measures are adjusted for model-overfitting to give a measure of out-of-sample performance. No set of predictions is perfect, especially in a field as uncertain as injury prediction, but the DVS Model gives one the ability to stack the odds in their favor.
The Brier score is a measure of the accuracy of a probabilistic function (in this case, the probability of sustaining a major shoulder or elbow injury). The closer the Brier score is to 0, the more accurate the model predictions. When assessing the ability of the model to predict major throwing-related injury, the overall Brier score is 0.134, which is 46% better than flipping a coin.
The Concordance yields a measure of predictive performance for making decisions between pairs of pitchers. The Concordance for the DVS Model is 63%, which means that when comparing two pitchers the DVS Model is able to accurately predict the pitcher who will undergo a major throwing-related injury first 63% of the time. This is approximately 2-to-1 odds of making the right decision between two pitchers.
The results below are just a sample of recent trends that the DVS Model has produced.
Pitchers born after 1990 have an average time to first major injury of only 363 innings, compared to 657 innings for pitchers born between 1980 and 1990 and 1109 innings for pitchers born between 1970 and 1980.
Pitchers with a DVS Score of 10 or less have an average career length of 1269 innings, whereas pitchers with a DVS Score of 17 or more have an average career length of 2419 innings.
Pitchers with a DVS Score of 10 or less have an average time to first major injury of 696 innings, whereas pitchers with a DVS Score of 17 or more have an average time to first major injury of 1013 innings.
Pitchers are 33.7% more at risk for major re-injury after a first major injury and 84.0% more at risk for major re-injury after a second major injury.