Are Baseball Statistics an Appropriate Tool for Assessing Return to Play in Injured Players: An Analysis of Statistically Variability in Healthy Players Using a Machine Learning Approach

摘要

Objectives: Traditional pitching statistics (ERA, WHIP, etc) have been used as surrogates for pitcher performance without being validated. Even amongst healthy pitchers, the normal variability of these parameters has not yet been established. The purpose of this study was to determine the normal variability of basic and advanced pitching statistics in non-injured Major League Baseball (MLB) pitchers. It is our hope that this work will serve as the foundation for the identification and implementation of validated, pitcher dependent statistical measures that can be used to assess return to play performance following injury. Methods: Publicly available data from MLB Statcast and Pitch/Fx databases was used to analyze all non-injured MLB pitchers during 2015 and 2016 seasons who pitcher greater than 100 innings each season without injury. Traditional and advanced baseball pitching statistics were analyzed. The variability of each parameter was assessed by computing coefficient of variation (CV) between individual pitchers and across all pitchers. A CV below 10 is typically indicative of a relatively constant parameter, and parameters with a CV & 10 are generally considered inconsistent and unreliable. Results: A total of 118 pitchers met all inclusion criteria. For each of these healthy pitchers, 38 basic/traditional parameters and 17 advanced parameters were analyzed. Of the traditional pitcher statistics, only 1 (3%) demonstrated a CV value & 10 (average fastball velocity [FBv]; CV 1.5) (Figure 1). In advanced statistics, 9 of 17 (53%) variables demonstrated acceptable consistency as evidenced by a CV value & 10 (Figure 2). Release position from plate (release_pos_y) along with velocity from the plate (vy0) where the two most constant advanced parameters. When separated by pitch type, these two parameters were the most constant (lowest CV) in every pitch type. Conclusion: The validity and variability of baseball statistics as surrogate markers for performance after injury/surgery have not yet been evaluated. It is critical that baseball statistics undergo proper vetting prior to being used to assess recovery. This study reveals average fastball velocity and release position from the plate to be the least variable basic and advanced baseball statistics in MLB pitchers. In total, only 10 of the 55 statistics analyzed demonstrated acceptable consistency and reliability. This study can be further used to determine the minimum time that each of these variables needs to be followed to ensure an appropriate sample size is obtained to detect significant differences in pre- and post- injury performance. Figure 1. Variability for each traditional MLB pitcher statistic sorted in ascending order. Lower values indicate less variation. Figure 2. Variability for each advanced MLB pitcher statistic. Lower values indicate less variation.