A formal study was used to test the validity of the DVS Scoring System.  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.  

data collection

Career data was collected and video analysis was performed on 449 randomly sampled former and current professional pitchers born between 1921 and 1996. Pitcher demographic data included birth year, reliever versus starter status, total number of innings pitched, major upper extremity injury, and total innings until injury.

video collection

To analyze the mechanical patterns for all subjects in the initial study, DVS obtained both a side angle and rear angle of every pitcher’s delivery.  To ensure consistency, all video obtained from each pitcher was actual game footage.  Game footage ensures all pitchers are performing at a max intensity level.  Therefore, the components analyzed within the pitching delivery remain constant. Multiple pitches were recorded from both angles at the same point in time to verify each player’s mechanical pattern. 

formal analysis

To have any validity, the DVS Scoring System must be statistically related to the risk of major pitching-arm injury.  This was done using a survival analysis model called a Cox Proportional Hazard Model.  The Cox Proportional Hazard model is a widely-accepted and widely-used statistical model for the discovery of relationships between time-to-event data and the variables associated with that data. 

The Cox Proportional Hazard Model has two main components: the baseline hazard function and the effect parameters.  The baseline hazard function describes the hazard function at baseline levels of the covariates and effect parameters describes how the hazard function reacts to changes in the values of the covariates.  The Proportional Hazards Assumption, on which the Cox model rests, means that changing values of the covariates are multiplicatively-related to the baseline hazard function.  This is convenient because it offers an easy interpretation of the coefficients: all other covariates held constant, changing the value of a single covariate multiplicatively changes the risk of the event occurring at the next increment of time (i.e. x% more or less likely to occur). 



The following is a summary of the model output from R:

    Concordance= 0.709(se = 0.017 )

    Likelihood ratio test= 208.5on 9 df,   p=0

Overall, the model takes into account three main factors: pitching mechanics, birth year, and major pitching-arm injury history.  This isolates the impact of mechanics on the risk for major pitching-arm injury because it accounts separately for the known effects of youth overthrowing (through the birth year) and injury history.